Thèse Igname
Thèse Igname
THÈSE
pour obtenir le grade de docteur délivré par
Denis CORNET
14 janvier 2015
Jury
Mme. Monique BODSON, Professeur, Université de Liège, Belgique Rapporteur
M. Emmanuel FROSSARD, Professeur, ETH Zurich, Suisse (Président de Jury) Rapporteur
Mme Aurélie METAY, Maître de Conférences, Montpellier SupAgro Examinateur
M. Philippe LETOURMY, Chercheur, CIRAD Examinateur
M. Benoît GABRIELLE, Professeur, AgroParisTech (Directeur de thèse) Examinateur
AgroParisTech
M. Jorge SIERRA, Directeur de Recherche, INRA (Encadrant) Examinateur
Cirad, UMR AGAP,
F-34398 Montpellier, France
Page | 1
à Bertrand Ney (1956-2013)
J’ai rencontré Bertrand lors de mon premier comité de thèse. Je me disais qu’avec le niveau de
responsabilité qui était le sien, son implication serait minime. Surtout sur une plante aussi
marginale que l’igname. A la fin de ce premier comité, j’avais non seulement un directeur de
thèse intéressé, curieux et intéressant, mais surtout une pêche d’enfer, une motivation à toute
épreuve et l’envie. Bertrand était comme ça, capable de s’intéresser sincèrement aux sujets
scientifiques mais surtout aux personnes qui les abordent ; et par là même, de donner l’envie. Il
était capable de faire confiance sans cesser de s’impliquer. Son humanité lui permettait
d’exprimer une exigence scientifique à l’abri des rivalités, d’être écouté et d’écouter.
Par la suite, nous nous sommes revus à plusieurs reprises, à Paris, ou lors de ses séjours en
Guadeloupe. Avec le recul, je mesure quelle chance j’avais de soustraire Bertrand à ses
obligations durant ces séjours. Scientifiquement, les échanges étaient riches. Bertrand arrivait à
être pédagogique sans être condescendant. Mais une fois la journée de travail terminée, il avait
plaisir à laisser le travail de côté, à discuter avec mes filles et mon épouse, à s’intéresser, encore, à
d’autres sujets et d’autres personnes.
Quand le moral est au plus bas, je l’entends me dire « dis, mon p’tit chou, tu oublies la biomasse
racinaire dans le calcul de la RUE… ». Et je souris, car il n’y avait que lui pour appeler un grand
belge de deux mètres, « mon p’tit chou ». Merci Bertrand pour ces moments partagés, pour tes
conseils et pour ta confiance qui, malgré la tristesse, continuent de me donner envie.
Page | 2
Page | 3
Remerciements
Je tiens en premier lieu à remercier mes encadrants, Raymond Bonhomme en coulisse et Jorge
Sierra en couleur. Merci Raymond pour ta disponibilité, ton franc-parler et ton humanité ces 10
dernières années. Merci à Jorge qui, au-delà des incompréhensions levéyes par le dialogue belgo-
argentin argentino-belge, a réussi brillamment le passage quotidien de la science du sol
caribéenne, à l’ésotérique étude bayésienne des ignames africaines (et ce, même pendant la coupe
du monde !). Vous m’avez tous les deux apporté énormément, scientifiquement mais aussi
humainement.
Merci à Benoît Gabrielle qui a accepté au pied levé l’encadrement d’une thèse qui, comme son
sujet, était orpheline. Merci également à Emmanuel Frossard, Monique Bodson, Aurélie Metay et
Philippe Letourmy pour avoir accepté de faire partie de mon jury. Mes remerciements vont aussi
à Corinne Fiers et au personnel de l’école doctoral ABIES pour leur disponibilité et leur
compréhension.
Enfin, j’adresse mes tendres remerciements à ma famille. Mes parents qui sont en quelque sorte
les coupables indirects de ce qui suit. Mes filles Nell, Fany et Léa pour leur patience et leur
admiration qui donne confiance (si si, elles arrivent les crêpes). Et bien sûr Kim, mon épouse,
pour sa vigilance, son amour et sa compréhension qui m’ont permis ces derniers mois de me
plonger dans la rédaction de la thèse tout en conservant une place douillette où me ressourcer les
jours de cafards ; et qui, bientôt, connaîtra le même sort (es-tu vraiment bien sûre de le
vouloir ? )…
Page | 4
Page | 5
Influence des premiers stades de croissance sur la variabilité du
rendement parcellaire de deux espèces d'igname (Dioscorea spp.)
cultivées en Afrique de l'Ouest
Résumé :
En Afrique de l’Ouest, l’igname est traditionnellement une culture itinérante, installée après
défriche-brûlis d’une jachère longue. L’état actuel des connaissances sur le fonctionnement de la
culture est insuffisant pour permettre d’élaborer des alternatives durables à ce mode de conduite
pénalisant pour l’environnement. Des études récentes montrent que la variabilité du rendement
entre plantes d’une même parcelle constitue un verrou majeur à l’amélioration de la culture
d’igname. Actuellement, la cause de cette variabilité reste inexpliquée et freine les tentatives
d’intensification des systèmes de culture. Dans ce contexte, l'objectif de la thèse était d’étudier la
variabilité interplante des deux espèces majeures d’igname en Afrique de l’Ouest (Dioscorea alata
et D. rotundata), et plus particulièrement : de la quantifier, d’en identifier les causes et d’en estimer
les conséquences physiologiques, agronomiques, et économiques.
Une série de 11 expérimentations menées de 2006 à 2009 sur deux sites au Bénin nous a permis
de démontrer que la variabilité de rendement entre plantes d’igname était nettement supérieure à
celle des autres cultures à tubercules, telles que la pomme de terre (CV > 40% et CV < 20%,
respectivement) et s’exprimait dès les premiers stades de croissance. Les causes de cette
variabilité ont été identifiées et leurs effets directs et indirects ont été quantifiés en utilisant un
modèle graphique de type réseau bayésien. Ces résultats montrent qu’aux densités de plantations
traditionnelles en Afrique de l’Ouest (0,4 à 1 plante m-²), il n’y a pas de compétition entre plantes.
En revanche, le matériel de plantation utilisé introduit une forte variabilité initiale entre individus
via la taille des semenceaux et la date d’émergence.
Nous avons mis au point un modèle à l’échelle de la plante permettant de simuler la croissance
de l’igname en fonction des facteurs responsables de la variabilité du rendement, et son
développement en fonction de la photopériode et de la température. Le modèle prédit bien les
rendements individuels observés. Les observations ainsi que les prédictions montrent clairement
que les plantes émergées plus tardivement, ou issues de semenceaux de faible qualité, auront une
croissance ralentie, une tubérisation initiée trop rapidement et finalement, un rendement plus
faible. En comparaison avec l’utilisation d’un matériel de plantation théorique, plus homogène,
calqué sur celui de la pomme de terre, ces pertes sont estimées, en moyenne, à 22 et 27% du
rendement pour D. alata et D. rotundata respectivement. En tenant compte du prix de vente
saisonnier et du coût des semenceaux, cette perte de rendement se traduit par une réduction du
profit de l’agriculteur de 30 et 40%, respectivement.
Pour la culture d’igname en Afrique de l’Ouest, les conséquences agronomiques et économiques
de la variabilité de la croissance et du développement entre plantes d’une même parcelle, plaident
en faveur du développement d’une filière de production de semenceaux de qualité.
Abstract:
In West Africa, shifting cultivation using the slash and burn technique is the traditional yam
cropping system. Current knowledge is insufficient to elaborate and recommend sustainable
alternatives to this cropping system, which threatens the environment. Recent studies have
shown unexplained interplant yield variability to be an important brake on the improvement of
yam production system that hinders intensification attempts. In this context, the objective of the
thesis was to study yield interplant variability of the two main yam species (Dioscorea alata and D.
rotundata) and more specifically to quantify this variability, to identify its causes and to estimate its
physiological, agronomic and economic consequences.
A set of 11 experiments carried out between 2006 and 2009 at two locations in Benin, enabled us
to demonstrate that interplant variability of yam yield is much higher than that of similar tuber
crops, such as potato (CV > 40% and CV < 20%, respectively) and that it appears at an early
growth stage. Causes of this variability were identified and their direct and indirect effects were
quantified using a Bayesian network. At common planting densities (0.4-1 plants m-²), these
results showed that there was no competition between plants. In contrast, the use of traditional
planting material introduced important initial variability between individual plants, especially in
the observed emergence date and seed-tuber weight.
We developed a plant model to simulate yam growth as a function of factors affecting yield
variability and yam development as a function of daily photoperiod and temperature. The model
gives good estimates of observed plant yield. Observations and simulations both demonstrate
that delayed emergence and lower seed-tuber quality (lower weight) lead to a slower vegetative
growth, unduly early tuber initiation and finally a lower yield. In comparison with the use of a
theoretical planting material based on the potato, the use of traditional planting material resulted
in a 22 and 27% yield loss for D. alata and D. rotundata respectively. Taking into account the
seasonal selling price of the crop and the cost of the seed tubers, the yield loss represents a
decrease of 30 and 40% in profit for the producer.
Agronomic and economic consequences of the interplant yield variability would suggest the
development of a quality seed-tuber production system to the benefit of yam cropping in West
Africa.
Keywords: yams (Dioscorea spp.), yield variability, emergence, plant modeling, West Africa,
planting material
Page | 8
Page | 9
Table des matières
INTRODUCTION GENERALE ..................................................................................... 14
Page | 10
2.2. HYPOTHESES DE TRAVAIL ET OBJECTIFS DE L’ETUDE ................................................ 51
2.3. ORGANISATION DU MEMOIRE ................................................................................... 52
Page | 13
Introduction générale
Les ignames appartiennent au genre Dioscorea. Elles sont cultivées depuis plus de 12000 ans et la
consommation de leurs tubercules remonte au paléolithique moyen. Leur protoculture,
consistant à planter des ignames sauvages près des habitations et à les exploiter, est souvent citée
comme une des premières formes de transition vers l’agriculture (Coursey 1976). Il est
intéressant de noter que si l’on connaît surtout les ignames comme culture vivrière en Afrique et
en Océanie, leur tubercule était aussi consommé au Japon, bien avant l’introduction du riz, et
qu’en France, il existe une production d’igname depuis plus d’un siècle. Le genre Dioscorea est en
effet ubiquiste et cultivé à travers le monde à des fins principalement alimentaires. Les ignames
alimentaires regroupent une dizaine d’espèces principalement réparties en zones intertropicales ;
ces espèces participent à la sécurité alimentaire de nombreuses populations en Océanie et dans la
Caraïbe. Cependant, c’est en Afrique de l’Ouest que plus de 90% de la production mondiale est
actuellement réalisée. Deux espèces contribuent majoritairement à cette production, il s’agit de
D. rotundata, espèce originaire d’Afrique de l’Ouest, et D. alata, espèce originaire d’Asie du Sud-
Est et introduite en Afrique dans le courant du 16ème siècle.
En 2013, l’Afrique de l’Ouest (Nigéria, Bénin, Togo, Côte d’Ivoire et Ghana principalement)
produisait 55 millions de tonnes, faisant de l’igname la deuxième production alimentaire de la
région. Sa culture reste extensive et consomme peu, voire aucun intrant. Pourtant, elle est jugée
exigeante en matière de fertilité des sols et se place prioritairement en tête de rotation après une
jachère longue, obligeant l’agriculteur à chercher toujours plus loin les zones propices à sa
culture. A terme, ce mode de production itinérant, fortement dommageable pour
l’environnement, est condamné par l’épuisement des ressources naturelles. La demande
croissante en ignames accélère la déforestation des dernières zones boisées, et fait de la
sédentarisation de l’igname une problématique centrale pour la sécurité alimentaire en Afrique de
l’Ouest. Ces 40 dernières années, la production d’igname en Afrique de l’Ouest a plus que
quadruplé, démontrant ainsi que l’intérêt pour ce tubercule dépasse largement les populations
traditionnellement consommatrices et inclut maintenant les grands centres urbains. Cette
augmentation de la production s’est principalement faite par extension des surfaces cultivées :
l’augmentation de rendement restant modeste (de 8,5 à 12 t ha-1) et fortement éloignée de son
potentiel, estimé à plus de 75 t ha-1 (Zinsou 1998). Cet écart de rendement est en partie
imputable au manque de connaissance sur la plante, sa culture et ses contraintes de production.
Page | 14
En effet, si le nombre de publications sur une culture donnée reflète l’état des connaissances,
notre connaissance actuelle des deux espèces majeures d’ignames (2023 publications) équivaut à
celle sur la pomme de terre avant 1973 !
Il est donc urgent de développer, avec les producteurs, des alternatives techniques permettant
une production durable d’igname dans le cadre de systèmes conduits sans recours à la pratique
de défriche-brûlis. Les recherches sur les systèmes de cultures durables à base d'igname sont peu
nombreuses et les tentatives d'intensification, avec forte utilisation d'intrants et mécanisation,
vulgarisées dans les années 60-70, se sont révélées infructueuses et inadaptées. Plusieurs
techniques, mieux adaptées à l'agriculture africaine, ont été proposées depuis quelques années par
le Cirad dans le domaine de la production durable d'ignames : l’agroforesterie, les jachères
améliorées et l’association culturale à base de légumineuses. Celles-ci ont montré, notamment au
Bénin, des premiers résultats encourageants en milieu paysan (Cornet 2005). Les systèmes de
culture utilisant les plantes de couverture et les techniques de semis direct sur couverture végétale
(SCV) constituent une autre voie prometteuse pour la mise au point d’une agriculture durable en
zone tropicale. Ces techniques sont bien connues et maîtrisées pour les plantes à graines
(céréales, coton...).
Cependant, ces techniques restent à mettre au point pour des systèmes de culture à base de
plantes à racines et tubercules. De plus, l’option finalisée revendiquée par les projets portant ces
alternatives ne permet pas actuellement une compréhension des mécanismes d’élaboration du
rendement de l’igname. Les connaissances sur la physiologie de la plante sont si fragmentaires et
incomplètes qu’un pilotage de ces systèmes sur la base de connaissances agronomiques reste
limité ; les chercheurs en sont donc réduits à un apprentissage par l’expérience, similaire à celui
des producteurs. Par exemple, les études consacrées à l’intensification de la culture se heurtent à
une grande variabilité de rendement entre plantes d’une même parcelle, sans que les causes de
cette variabilité et leurs conséquences sur la culture de l’igname soient connues. Ce travail
d’acquisition de connaissances est un préalable indispensable à l’amélioration de cette plante
orpheline, dont la culture traditionnelle est condamnée à moyen terme.
Page | 15
Chapitre 1 Etat de l’art
Page | 16
1.1. Les ignames
Le genre Dioscorea comprend plus de 600 espèces avec une répartition majoritairement
pantropicale et subtropicale. Les Dioscorea sont cultivées dans le monde à des fins
pharmaceutiques (D. dumetorum, D. villosa, D. zingiberensis…), ornementales (D. discolor, D.
elephantipes…) mais surtout alimentaires (D. rotudata, D. cayenensis, D. alata, D. esculenta…), et il
existe plus d’une vingtaine d’espèces cultivées pour la consommation de leur tubercule. La
consommation d’igname remonte à plus de 100 000 ans (Coursey 1976) et sa culture remonte à
9 000 ans en Océanie, 12 000 ans en Asie et 10 000 ans en Afrique de l’Ouest (Haudricourt 1964,
Coursey 1976, McConvell 1990, Matsui et Kanehara 2006).
Les 600 espèces appartenant au genre Dioscorea présentent une grande diversité
d’adaptations écologiques (Wilson 1977). La majorité de ces espèces se rencontre dans les zones
intertropicales (entre 20°N et 20°S), mais on en cultive également en zone tempérée, comme en
Argentine, en France ou au Japon (Figure 1).
Figure 1. Production annuelle d’ignames dans le monde par pays (t) et par habitant
(kg habitant -1 ).
Page | 17
Cependant, si l’igname se cultive sur plusieurs continents, c’est l'Afrique de l'Ouest, avec
les cinq grands pays côtiers producteurs (Côte d’Ivoire, Ghana, Togo, Bénin et surtout Nigeria)
qui forme de loin la zone de production la plus importante du monde. Cette région couvre 95%
des superficies cultivées, d'où son nom de "civilisation de l'igname" (Coursey 1976). Avant
l’introduction des céréales en Afrique de l’Ouest, l’igname était la principale source d’hydrates de
carbone ; aujourd’hui, avec une production de près de 55 millions de tonnes par an, il s’agit de la
deuxième plus importante production, après le manioc (Figure 2). Dans cette région, l’igname
séduit de plus en plus les populations des grands centres urbains non traditionnellement
consommatrices, et devient donc une source de revenus agricoles (Bricas et al. 2003). Bien que
l’igname soit considérée comme un aliment de luxe, cette demande croissante en zone urbaine
laisse à penser qu’elle pourrait être, dans un proche avenir, un recours fiable pour répondre aux
besoins alimentaires grandissants.
90 12
80 Production
10
Production (Mt)
70 Valeur
Valeur (M$)
60 8
50
6
40
30 4
20
2
10
0 0
Figure 2. Tonnage (en millions de tonnes) et valeur (en millions de dollars) des principales
productions alimentaires végétales en Afrique de l’Ouest en 2012 (FAOSTAT 2014).
Page | 18
1.1.2. Etat des connaissances
Malgré son importance et sa large répartition, l’igname reste une culture « orpheline », c’est-à-dire
une culture qui participe peu aux échanges commerciaux internationaux, alors qu’elle joue un
rôle important à l’échelle de la sécurité alimentaire régionale en Afrique de l’Ouest (Naylor et al.
2004). Le corollaire de ce constat, est qu’une culture orpheline ne reçoit pas ou peu d’attention
de la part de la recherche : en plaisantant, Donald Kennedy, éditeur en chef de la revue Science,
recommandait aux chercheurs travaillant sur l’igname de ne pas prendre le même avion pour aller
à leur prochaine conférence (Kennedy 2003). Ce trait d’humour illustre malheureusement assez
bien la situation de la recherche sur l’igname. Pour souligner ces propos, la Figure 3 présente le
nombre de publications recensées par le Web of Science © entre 1970 et 2014 pour les ignames
(D. alata et D. rotundata) et pour la pomme de terre (Solanum tuberosum). En 2014, il y a quarante
fois plus de publications sur la pomme de terre que sur l’igname (85214 et 2023 respectivement).
En moyenne, la recherche produit chaque année autant de publication sur la pomme de terre
qu’en 40 ans sur l’igname. Si la production de connaissance sur l’igname continue à ce rythme, il
faudra attendre plus de 800 ans pour atteindre le niveau de connaissance actuel sur la pomme de
terre.
90000 85214
80000
Igname
Nombre de publications
70000
Pomme de terre
60000
recensées
50000
40000
30000
20000
10000
2023
0
1970 1975 1980 1985 1990 1995 2000 2005 2010
Figure 3. Nombre de publications recensés dans le Web of Science © entre 1970 et 2014 pour
l’igname (Dioscorea alata et D. rotundata en bleu) et la pomme de terre (Solanum tuberosum en
rouge).
Page | 19
1.1.3. Ecologie des ignames
Les ignames (Dioscorea spp.) sont des monocotylédones, géophytes à port lianescent (Figure 4).
Leurs stratégies de croissance sont conditionnées par les caractéristiques associées à
l’organisation anatomique des monocotylédones, et en particulier, par l’absence d’un cambium
vasculaire dans les axes aériens et donc l’impossibilité d’effectuer une croissance secondaire
(McKey et al. 1998). Les tiges des lianes doivent donc posséder un système vasculaire très
performant. L’absence de croissance secondaire chez les ignames empêche la réparation des
éléments conducteurs endommagés et ne permet pas l’addition de nouveaux éléments de
phloème et de xylème pour alimenter un appareil aérien. Ces facteurs limitent la durée de vie de
l’appareil aérien des ignames (Di Giusto et al. 2001) : toutes les variétés cultivées et de
nombreuses espèces sauvages sont annuelles, l’appareil aérien n’est maintenu qu’une seule
saison ; d’autres espèces sauvages sont bisannuelles ou pluriannuelles, l’appareil aérien étant
renouvelé tous les deux ans ou plus. Les espèces d’ignames des forêts tropicales humides sont
des plantes héliophiles (Johnston et Onwueme 1998, Onwueme et Johnston 2000). En effet,
l’origine (peri-)forestière de plusieurs espèces d’ignames ne doit pas faire oublier que son port
grimpant lui permet d’atteindre la canopée (McKey et al. 1998). Une des fonctions principales du
tubercule souterrain est donc d’alimenter la croissance du nouvel axe aérien jusqu’à la canopée.
Figure 4. Les lianes de D. rotundata utilisent tous les supports accessibles pour leur croissance .
Page | 20
Pour les espèces d’importance alimentaire, le développement d’un tubercule correspond
également à une adaptation aux saisons sèches marquées. En cours de cycle, l’igname peut tolérer
de courtes périodes de stress hydrique, mais sa tolérance vient en grande partie des réserves
disponibles dans le tubercule semence. En effet, la forte teneur en eau dans le tubercule rend sa
germination relativement indépendante du statut hydrique du sol (Onwueme 1976, Figure 5). La
culture de l’igname se pratique avec succès dans des zones où la pluviométrie varie entre 1000 et
plus de 3000 mm (Onwuerme 1975, Onwueme 1978a), cependant, les besoins hydriques réels
des ignames sont encore mal connus.
L’igname est avant tout un taxon d’espèces tropicales qui a besoin de températures
élevées. La germination du tubercule est optimale entre 25 et 30 °C, alors que des températures
inférieures à 15°C ou supérieures à 35°C retardent la germination (Onwueme 1978b). Selon
Copeland (1916), le taux de croissance est fortement ralenti en dessous de 20°C et continue
d’augmenter entre 25 et 30°C ; ces valeurs n’ont cependant été vérifiées que pour la croissance
de la partie aérienne il y a maintenant près d’un siècle. Plus récemment, Marcos et al. (2009) ont
estimé les températures cardinales de l’espèce D. alata durant les principales phases de
développement ; ils montrent que la température optimale dépend de la phase de développement
et varie entre 19 et 28°C. La photopériode influence également son développement (Njoku 1963,
Vaillant et al. 2005) : les jours courts, aux alentours de 12 heures, caractéristiques des zones
tropicales productrices d’igname, favorisent la tubérisation (Orkwor et al. 1998). A l’opposé, les
jours longs (> 12 h) semblent favoriser le développement de la partie aérienne (Coursey 1967).
La sensibilité à la température et à la photopériode de D. alata et D. rotundata sera discutée plus en
détails dans le chapitre 6.
Page | 21
La croissance du tubercule en formation est réalisée par l’allongement continu de l’apex
(Onwueme 1979) ; il ne s’agit pas de l’épaississement d’un stolon ou d’une racine préexistante,
comme c’est le cas pour la pomme de terre ou le manioc (Flach 1979). Il est donc important que
le sol soit meuble et présente de bonnes capacités de drainage afin de limiter sa compaction et la
stagnation de l’eau, deux phénomènes perturbant la croissance du tubercule (Ferguson et Gumbs
1976, Figure 6). Ensuite, au-delà des propriétés physiques, les sols convenant le mieux à la
culture de l’igname se distinguent par leur capacité d’échange cationique plus élevée et un
potentiel nutritionnel important (Obigbesan 1977, Nwinyi 1981, Ohiri et Nwokoye 1983, Diby et
al. 2009). D. alata semble plus tolérante aux sols pauvres que les autres espèces d’ignames
alimentaires, ce qui explique en partie sa plus large distribution géographique (Irvine 1969).
Pourtant, même si l’agriculteur considère l’igname comme une culture exigeante en termes de
fertilité, les sols qui les accueillent traditionnellement ne présentent pas de propriétés chimiques
particulièrement favorables à la culture (Asadu et al. 1990, Cornet et al. 2005b).
Figure 6. Tubercules de D. rotundata cv. Laboko développés dans un sol meuble (à gauche) et
tassé (à droite).
Page | 22
1.1.4. Morphologie et cycle de développement
Le cycle végétatif des ignames alimentaires est annuel et en adéquation avec l’alternance des
saisons dans les régions intertropicales (Zinsou 1998). La phase de croissance a lieu pendant la
saison humide, et est suivie par la sénescence de l’appareil aérien et l’entrée en dormance du
tubercule qui coïncide en général avec le début de la période sèche (Wickham et al. 1997). La
durée effective de la phase de croissance (de l’émergence à la sénescence) varie entre 6 et 12 mois
selon le génotype, le matériel de plantation, et le milieu. En Afrique de l’Ouest, le semenceau
« dormant » est planté en fin de saison des pluies de l’année précédente ou en début de saison.
Cette période de pré-émergence (de la plantation à l’émergence) n’est habituellement pas prise en
compte dans la phase de croissance puisque sa durée varie fortement (de 2 à plus de 12
semaines) selon le degré de maturité et la dormance du tubercule semence (Onwueme 1975).
Ainsi, le cycle de l’igname peut donc être divisé en cinq grandes phases, sans tenir
compte de la date de plantation. Ces phases sont représentées schématiquement sur la Figure 7.
Page | 23
a. De la maturité du tubercule à la levée de la dormance (phase 1)
La maturité biologique du tubercule est atteinte lorsque la translocation des produits de la
photosynthèse s’arrête. Dans la pratique, la maturité du tubercule est estimée à partir de la
sénescence complète du feuillage (Campbell et al. 1962, Waitt 1965), même si ce phénomène ne
correspond pas toujours avec l’époque de rendement maximal en tubercule. James (1953) montre
que le tubercule peut continuer à croître jusqu'à complète dessiccation de la végétation. Lorsque
la maturité est atteinte, le tubercule d’igname connaît une période de « dormance » pendant
laquelle il est incapable de germer (Craufurd et al. 2001). Une baisse de respiration et une
augmentation de composés cellulosiques caractérisent ce repos végétatif qui prend fin avec
l’apparition de micro-protubérances à la surface du tubercule (Figure 5). La durée de cette phase
est d’environ deux à quatre mois selon l’espèce et la variété.
e c
Figure 8. Détails de la levée des premiers organes (à gauche) : (a) la tige, (b) le complexe
nodal primaire, (c) le pré-tubercule en stagnation, (d) semenceau, et (e) racine coronaire ; et
émergence de D. rotundata cv. Morokorou (au centre) et D. alata cv. Florido (à droite).
Page | 24
c. La croissance, le développement végétatif et l’initiation du tubercule
(Phase 3)
La croissance commence par une phase dite hétérotrophe, où la tige et les racines sont
principalement alimentées par les réserves du tubercule semence. Selon les espèces et les variétés,
pendant cette phase, la tige ne porte aucune feuille ni ramification, mais porte des cataphylles
(feuilles réduites en écailles, Figure 9). A l’aisselle de chaque cataphylle se trouve un méristème
latéral, mais celui-ci ne se développe qu’ultérieurement. L’apparition tardive des feuilles chez D.
rotundata est un avantage à l’état sauvage : elle lui permet d’atteindre la canopée, et donc de sortir
du sous-bois ombragé avant d’émettre les premières feuilles. En revanche, l’apparition de
cataphylles est rare chez D. alata.
Cataphylles
1 cm Méristème latéral 3 cm
Figure 9. Morphologie caulinaire de D. rotundata cv. Morokorou durant les premiers stades de
croissance.
Par la suite, les appareils racinaires et aériens deviennent fonctionnels : c’est la phase
d’autotrophie (Degras 1986). Ferguson (1973) montre que, plus le tubercule mère est petit, plus
la plante passe rapidement à l’autotrophie (avec des taux d’assimilation nette plus élevés que les
gros semenceaux). Ce passage à l’autotrophie détermine la date à laquelle la plante devient
sensible aux facteurs exogènes. Simultanément, les tiges et leurs ramifications s’allongent et des
feuilles se mettent en place. Durant cette phase, l’axe principal croît seul pendant quelques jours,
ensuite, des bourgeons axillaires se développent. Les deux types d’axe présentent une courbe de
croissance en longueur sigmoïdale et entrent simultanément en phase linéaire (Trouslot 1983).
Page | 25
Le système racinaire de l’igname est mal connu. Les racines de plants provenant de
tubercules sont toutes, par définition, adventives. La première phase de croissance est
principalement marquée par le développement d’un système racinaire extensif. Celui-ci est
fibreux et se développe à partir du complexe nodal primaire (Onwueme 1978b). Le système
racinaire reste majoritairement superficiel (< 0,3 m de profondeur) tout au long du cycle (Njoku
et al. 1973, Hgaza et al. 2011, Figure 10). Clairon et Zinsou (1980) et Njoku et al. (1973) ont
observé que le maximum de développement racinaire apparaît environ un mois avant celui des
parties aériennes.
Figure 10. Excavation du système racinaire de D. rotundata cv. Gnidou, sept semaines après
plantation, à plat (à gauche) et en butte (à droite).
Page | 26
d. La phase de tubérisation rapide (phase 4)
Cette phase démarre avec l’entrée en compétition de la partie aérienne et du tubercule, pour
l’utilisation des assimilats (Onwueme et Charles 1994). Ceci entraîne rapidement
l’infléchissement des courbes de croissance des tiges et des feuilles. La partie aérienne alimente
alors exclusivement le tubercule, qui commence à se remplir activement suivant une loi de
croissance exponentielle. Le taux de remplissage durant cette période est maximal : de 21 à plus
de 51g de matière sèche par plante et par semaine, respectivement pour D. alata (Ferguson 1973)
et D. rotundata (Enyi 1973, Kpeglo et al. 1982, Njoku et al. 1984). Au cours de cette dernière
phase, on peut voir apparaître la floraison sur les plantes florifères (Figure 11). L’entrée en phase
rapide de tubérisation coïncide avec le passage par le développement foliaire maximal (Njoku et
al. 1973, Trouslot 1983); la durée de cette phase varie entre 60 et 90 jours.
Figure 11. Floraison chez l'igname : inflorescence mâle (à gauche), inflorescence femelle (au
centre) et graines (à droite).
Page | 27
1.2. L’igname en Afrique de l’Ouest
Figure 12. Installation d’une culture d’igname par défriche-brûlis après une jachère longue (>
15 ans).
Une fois la parcelle ainsi nettoyée, les tubercules ont besoin pour se développer d’un sol meuble
offrant peu de résistance à leur pénétration : la culture de l’igname nécessite donc un travail du
sol important. Le plus fréquemment, l’agriculteur confectionne des buttes qui favorisent le
drainage et facilitent la récolte (Figure 13). La hauteur des buttes varie généralement entre 0,3 et
1 m (Orkwor et al. 1998), mais on trouve parfois en zone inondable des buttes de plus d’un mètre
de hauteur et de 3 m de diamètre (Onwueme 1978b). La densité de plantation varie fortement en
fonction de la taille des buttes et des objectifs de l’agriculteur (culture pure ou en mélange) mais
reste beaucoup plus faible qu’ailleurs dans le monde : de 0,3 à 1 plant m-² en Afrique de l’Ouest
contre 2 à 3 et 2 à 6 plants m-2 aux Antilles et au Japon respectivement. La confection des buttes
est réalisée en fin de saison des pluies de l’année précédente si le calendrier cultural le permet, ou
en début de saison des pluies.
Page | 28
Figure 13. Buttes d'igname avant plantation en zone Bariba à Fo-Boure, Nord Bénin (à
gauche) et en zone Mahi à Ouédémé, Centre Bénin (à droite) .
Il existe deux types de variétés : les variétés à récolte unique, comme la majorité des D.
alata et certaines D. rotundata, et les variétés à deux récoltes. Le matériel de plantation varie en
fonction du type de variété et de la production précédente. Pour les D. alata et les D. rotundata à
une récolte, l’agriculteur utilise prioritairement des fragments (de 0,2 à 1,5 kg) de gros tubercules
et plus rarement des tubercules entiers de petite taille. En revanche, certaines D. rotundata ont une
tubérisation suffisamment rapide pour permettre une double récolte : (i) la première, appelée le
sevrage, est réalisée à un stade précoce et permet d’obtenir des ignames de table qui peuvent être
vendues à des prix élevés car elles sont proposées à une période où la demande est supérieure à
l'offre, (ii) la deuxième est réalisée en fin de saison et permet la récolte de petits tubercules qui
serviront de matériel de plantation. Le sevrage se pratique environ 4 à 5 mois après l'apparition
des tiges : le sol autour des tubercules est dégagé délicatement afin de ne pas blesser les racines
de la plante, puis les tubercules jugés suffisamment gros sont détachés à hauteur du
collet (Figure 14). La butte est ensuite remodelée. Si le nombre de semenceaux produit en
deuxième récolte n’est pas suffisant pour couvrir les besoins de plantation, l’agriculteur utilisera,
comme pour les ignames à une récolte, des fragments de tubercule. Cependant, la perte de
dominance apicale et le gradient physicochimique longitudinal présent dans les tubercules
fragmentés conduisent à un matériel de plantation particulièrement hétérogène (Ferguson 1973).
Page | 29
Figure 14. Sevrage de D. rotundata cv. Morokorou.
Enfin, la pratique « d’ennoblissement » (Dumont et al. 2005) permet dans certains cas de
fournir des semenceaux : les agriculteurs collectent dans la savane ou la forêt des ignames
sauvages aux tubercules filiformes qu’ils replantent dans leurs champs. Après plusieurs années de
culture, les plantes qui développent un tubercule de morphologie proche de variétés cultivées
sont choisies par le cultivateur pour être multipliées. Les processus biologiques impliqués dans la
transformation morphologique du tubercule et son maintien à travers les générations sont encore
inconnus. L’ennoblissement a pour résultat l’intégration d’ignames d’origine sauvage et de
variétés hybrides dans le portefeuille variétal de l’exploitation (Chaïr et al. 2010) ; cette pratique
est observée dans différentes régions écologiques et ethnolinguistiques d’Afrique de l’Ouest.
Pour toutes ces raisons, l’utilisation de champs semenciers reste à l’heure actuelle
anecdotique : la semence reste majoritairement un investissement de longue durée et, à ce titre,
fait partie intégrante de l’héritage familial. Elle ne représente donc pas une dépense répétée à
l’occasion de chaque cycle cultural (Dumont 1998).
La date de plantation varie fortement en fonction du début de la saison des pluies, mais
aussi de la variété. Afin de conserver le semenceau jusqu’à la plantation, l’agriculteur recourt
parfois à un égermage, technique qui consiste à casser la tige germée en cours de stockage. A la
plantation, l’agriculteur privilégiera les variétés de D. rotundata à double récolte, car cette espèce
présente une durée de dormance plus courte, une moins bonne conservation et une plus faible
tolérance à l’égermage en cours de stockage. Les variétés de cette espèce sont plantées au milieu
de la saison sèche et produisent de longues tiges initiales aphylles qui ne s’affaissent pas sur les
sols surchauffés. Les autres variétés seront plantées par ordre de priorité de l’agriculteur, en
Page | 30
plantant souvent les D. alata en dernier. Après la plantation, la butte est protégée par un
« chapeau » fabriqué avec des chaumes de poacées ou des débris végétaux trouvés sur place
(Figure 15). Le chapeau permet de retenir l’humidité et de limiter l’élévation de la température à
l’intérieur de la butte durant la saison sèche (Onwueme 1977).
Figure 15. Réalisation des chapeaux sur les buttes d’igname, à partir des déchets végétaux
trouvés sur place.
Malgré un effet positif reconnu (Hgaza et al. 2010, Hgaza et al. 2012), l’irrigation et la
fertilisation restent anecdotiques principalement à cause de leur indisponibilité ou de leur coût. Il
en va de même pour le tuteurage qui, reconnu bénéfique, se marginalise et ne concerne plus que
les variétés de D. rotundata à double récolte. En effet, l’utilisation du tuteurage est limitée par le
coût de leur installation et la sélection de variétés adaptées à la culture sans tuteur. Dans ces
conditions, le matériel de plantation représente souvent le seul « intrant » en culture traditionnelle
d’igname.
En Afrique de l’Ouest, l’igname contribue à l’apport journalier de plus de 200 calories par
personne, pour plusieurs millions de consommateurs (Amagbeto et al. 2000). Les populations
africaines des pays du Golfe de Guinée lui restent culturellement très attachées et elle constitue
l'un des produits de base de leur alimentation (Figure 16).
Figure 16. Marché saisonnier de bord de route à Dassa, Centre Bénin (à gauche) et marché de
Glazoué, Centre Bénin (à droite).
Les principales préparations culinaires peuvent être classées en deux grandes catégories :
les produits à base d’ignames fraîches comme l’igname pilée (fufu), bouillie ou frite et les
produits à base de farine de cossettes comme l’amala (pâte d’igname) ou le wassa-wassa (semoule
d’igname). L’igname pilée est le mode de consommation le plus apprécié : c’est la forme
traditionnelle de consommation au Bénin et au Nigeria (Figure 17). Sa préparation est
fastidieuse, et les variétés utilisées ne sont disponibles qu’une courte partie de l'année. C'est
pourquoi, on observe en ville le développement de modes de consommation alternatifs, à la fois
plus faciles à réaliser et pour lesquels un plus grand nombre de variétés sont utilisables : igname
bouillie à l'eau ou en ragoût, et frites à partir de tubercules frais (Bricas et al. 2003). Mais face à la
concurrence des produits importés sur les marchés urbains (riz, maïs, pomme de terre), l’igname
fraîche reste désavantagée par son coût élevé et des prix fluctuants en cours de saison. La
transformation artisanale de l'igname en cossettes permet de stabiliser ce produit et de le
consommer plus régulièrement au cours de l’année (Vernier et al. 1999, Figure 18).
Page | 32
Figure 17. Igname pilée (fufu).
Page | 33
Figure 18. Transformation de l’igname en amala : (a) séchage, (b) vente de cossettes sur le
marché, (c) pilage des cossettes, (d) farines de cossette, (e) mélange de la farine de cossette
dans l’eau bouillante, et (a) pâte de farine de cossette (amala).
Page | 34
1.2.3. Amélioration des systèmes de production
En Afrique de l’Ouest, la production a connu une croissance soutenue durant les trois dernières
décennies, estimée à plus de 3% par an. Cependant, cette croissance s’est maintenue
essentiellement par défriche-brûlis de nouvelles zones boisées : entre 1961 et 2004, les surfaces
plantées ont quadruplé alors que le rendement n’est passé que de 8 à 11 t ha-1 (Figure 19). Ces
rendements sont jugés très inférieurs au potentiel de production, estimé à 75 t ha-1 pour D. alata
(Zinsou 1998) et 140 t ha-1 pour D. rotundata (Doku 1983). L’écart entre les rendements observés
et potentiels témoigne des contraintes biotiques et abiotiques non maîtrisées dans les systèmes de
cultures traditionnels à base d’igname. L’absence d’irrigation et de fertilisation, la réduction des
temps de jachère, la faible qualité physique et chimique des sols, l’absence de matériel de
plantation de qualité, l’abandon du tuteurage, la difficulté d’entretien des champs et la difficulté
de contrôler les bioagresseurs sont autant de contraintes limitant la production. Mais au-delà des
contraintes de productions agronomiques, les objectifs définis par l’agriculteur imposent aussi
des contraintes socio-économiques. Ainsi, le producteur privilégie clairement la production de
gros tubercules via la réalisation de buttes à faible densité et la plantation de gros semenceaux.
Ces choix de production se font souvent au détriment du rendement surfacique (Oriuwa et
Onwueme 1980, Rodriguez-Montero et al. 2001).
800
700 Surface
Rendement
600
Augmentation relative (%)
Production
500
400
300
200
100
-100
Page | 35
et al. 2005a, Cornet et al. 2006, Figure 20). Cependant, l’orientation très empirique de ces travaux
ne permet pas actuellement une compréhension des mécanismes d’élaboration du rendement de
l’igname. Or, si la problématique de sédentarisation de la culture est commune à la sous-région
de l’Afrique de l’Ouest, les solutions techniques à apporter sont susceptibles de varier selon les
conditions locales. Les connaissances sur la physiologie de la plante sont trop fragmentaires pour
élaborer et piloter ces systèmes (Onwueme et Haverkort 1991).
Figure 20. Culture associée d’igname et de Pueraria phaseoloides (a), installation d’une jachère
améliorée de P. phaseoloides (b), tuteurage vivant de l’igname avec du Gliricidia sepium (c), et
jachère améliorée de Mucuna pruriens (d).
Afin de diminuer l’impact des contraintes de production, de nombreuses études sont consacrées
à l’intensification de la culture (Vernier 1998, Cornet et al. 2005b), mais ces études se heurtent à
une grande variabilité de rendement entre plantes d’une même parcelle (Akoroda 1984, Ferguson
et al. 1969, Okoli et al. 1999). Les causes de cette variabilité et leurs conséquences sur la culture de
l’igname sont encore inconnues.
Page | 36
1.3. La variabilité de rendement entre plantes
La variabilité interplante est une mesure de l’inégalité entre plantes voisines en terme de taux de
croissance, de stade de développement ou de rendement (Martin et al. 2005). Elle peut avoir
plusieurs causes, exogènes ou endogènes. Les causes exogènes résultent le plus souvent d’une
mauvaise maîtrise de l’environnement et des bioagresseurs : labour irrégulier, résidus de culture
hétérogènes, ombrage localisé, application d’engrais irrégulière ou attaque de parasites (Weiner et
Solbrig 1984). Ces différentes situations conduisent à des inégalités d’accès aux ressources, dans
l’espace ou dans le temps (eau, nutriments, rayonnement), et donc à des croissances variables.
Chez l’igname en Afrique de l’Ouest, la confection de buttes diminue l’hétérogénéité du milieu,
en déstructurant le sol et en mélangeant ses différents horizons. En revanche, l’utilisation de
parcelles en jachère depuis plusieurs années, et l’habitude de laisser certains arbres en place pour
servir de tuteurs, augmente la variabilité spatiale des ressources (matière organique, nutriments,
eau, rayonnement…).
En plus de ces causes exogènes, il existe des causes endogènes, propres à la culture : la
variabilité génotypique, la plasticité phénotypique et le matériel de plantation hétérogène. La
variabilité génotypique implique une croissance et un développement différent selon les
génotypes. Elle peut résulter d’un mélange variétal, de reproductions sexuées non-maîtrisées ou
de l’accumulation de mutations somaclonales (Elias et al. 2001, McKey et al. 2010). Dans le cas de
l’igname, Chair et al. (2010) ont montré que les variétés traditionnelles sont fortement
polyclonales, y compris au sein d’une exploitation. Ainsi, sur une parcelle donnée, les différents
génotypes d’une même variété représentent une source de variabilité supplémentaire. Cette
variabilité génotypique peut être modulée (renforcée ou atténuée) par la plasticité phénotypique
résultant de l’interaction entre le génotype et son microenvironnement. Les plantes grimpantes,
comme l’igname, sont connues pour présenter une plasticité morphologique et architecturale
plus importante que les plantes à port érigé (Poorter et al. 2012). Mais surtout, le processus
toujours actuel d’ennoblissement renforce la sélection de génotypes présentant une
extraordinaire plasticité phénotypique (McKey et al. 2012). Cette plasticité continue à s’exprimer
même dans le milieu homogénéisé qu’est la parcelle agricole.
Enfin, la compétition entre plantes est connue pour générer et accentuer la variabilité
(Damgaard et al. 2002). Lorsque la compétition provoque une diminution de l’accès individuel à
des ressources limitées (rayonnement, nutriments, eau), on parle alors de compétition ressource-
dépendante. En revanche, les modifications d’architecture ou de morphologie induites par la
présence de plantes voisines avant que les ressources ne deviennent limitantes (stratégie
d’évitement de l’ombrage…) sont appelées compétitions ressources-indépendantes (Harper
1977).
Page | 38
1.3.2. L’utilisation des réseaux bayésiens comme méthode d’identification
des causes de variabilité et de leur interaction chez l’igname
L’identification des causes de variabilité interplante est complexe, du fait du grand nombre de
variables pouvant intervenir et interagir à des stades différents. Ces variables peuvent être de
natures différentes (discrètes, continues, binaires…). Les réseaux bayésiens permettent d’aborder
ce type de problème : s’ils sont encore peu connus en agronomie, les réseaux bayésiens trouvent
de multiples applications en biologie, médecine, écologie ou épidémiologie (Porth et al. 2013,
Ward 2013). Leur utilisation présente un double avantage. Dans un premier temps, ils permettent
d’identifier la structure des relations de dépendance entre variables. Ainsi, lorsqu’une analyse
classique par régression multivariée permet d’identifier les covariables associées avec la variable
dépendante (le rendement par exemple), le réseau bayésien permet de séparer empiriquement ces
relations en dépendances directes et indirectes (Lewis et McCormick 2012). Dans un deuxième
temps, les réseaux bayésiens permettent de calculer les probabilités conditionnelles de chaque
variable en tenant compte des dépendances identifiées. Au final, ce type de modèle graphique
permet de représenter les variables sous la forme d’un graphe orienté acyclique dans lequel les
nœuds correspondent aux variables aléatoires, et les arcs (le graphe est donc orienté) reliant ces
dernières sont associés à des probabilités conditionnelles. Dans ce graphe, les relations de
dépendance entre les variables ne sont pas déterministes, mais probabilisées : l'observation d'une
variable a priori n'entraîne pas systématiquement l'effet ou les effets qui en dépendent, mais
modifie seulement la probabilité de les observer. Une description plus détaillée de la construction
de ce type de modèle sera présentée dans le chapitre 4.
Page | 39
de la variabilité : il indique si la distribution est symétrique ou non par rapport à la moyenne.
Autrement dit, il indique s’il y a plus de petits individus que de gros (et vice versa). En revanche, il
n’apporte aucune information sur la contribution relative des individus extrêmes à la valeur totale
de la population et son interprétation peut prêter à confusion (Hipple 2005).
Figure 21. Courbes de Lorenz pour deux populations contrastées : une population où la
variabilité est principalement due à un petit nombre de très grand s individus (rouge) et une
population où la variabilité est principalement due à un grand nombre de petits individus
(bleu). La ligne en pointillé représente l’absolue égalité entre individus. G’ est le coefficient de
Gini et S le coefficient d’asymétrie de Lorenz.
Page | 40
Comme toute statistique synthétique, le coefficient de Gini ne contient pas toute
l’information apportée par la courbe de Lorenz. En effet, les courbes présentées dans la Figure
21 possèdent le même taux d’inégalité pour deux populations très différentes : une population où
l’inégalité est principalement due à un petit nombre de très grands individus contribuant
fortement à la biomasse totale (ligne rouge) et une population où l’inégalité est principalement
due à un grand nombre de petits individus contribuant faiblement à la biomasse totale (ligne
bleu). La différence entre ces deux populations peut être quantifiée en mesurant l’asymétrie de la
courbe de Lorenz perpendiculairement à la diagonale et plus spécifiquement, en mesurant le
point pour lequel la pente de la courbe est égale à 1 (Damgaard et Weiner 2000) : cette mesure
s’appelle le coefficient d’asymétrie de Lorenz (S). Elle est supérieure à 1 lorsque l’inégalité est
principalement due à un petit nombre de très grands individus et inférieure à 1 lorsque l’inégalité
est principalement due à un grand nombre de petits individus.
Page | 41
Chapitre 2 Contexte expérimental et objectifs de
l’étude
Un expert,
c'est quelqu'un qui a fait toutes ses erreurs dans un champ réduit d'applications.
Niels Bohr
Page | 42
2.1. Expérimentations
Le Bénin possède un relief peu accidenté mais constitué du sud au nord de trois types de
modelés : tabulaire, vallonné et chaotique. Notre étude a été conduite sur deux sites au Bénin : le
premier site d’expérimentation est situé à Glazoué dans la région des Collines au Centre Bénin
(Figure 22). Le département des Collines est une région traditionnellement productrice d’igname
qui profite d’un climat de type soudano-guinéen : la pluviométrie annuelle varie entre 800 mm et
de 1200 mm avec une répartition inégale. Les sols sont sableux, principalement de type
ferrugineux tropical à concrétions sur socle cristallin, relativement riches en éléments minéraux.
La végétation se compose de savane claire très ouverte au Sud allant vers une forêt semi-décidue
au Nord-Ouest. Le maïs, l’arachide, le manioc et l’igname constituent les principales cultures
vivrières, tandis que le coton, le soja et l’anacarde (noix de cajou) constituent les principales
cultures de rente. Sur ce site, les expérimentations ont été menées en milieu paysan, sans
possibilité d’irrigation.
Figure 22. Carte d’Afrique et localisation des sites d’étude sur la carte administrative du Bénin
(latitudes en ordonnées et longitudes en abscisses).
Page | 43
Le deuxième site est situé sur la station expérimentale de l’IITA/AfricaRice à Cotonou,
sur des sols communément appelés terres de barre. Ces sols sont ferralitiques, faiblement
désaturés et se sont développés sur des matériaux sédimentaires argilo-sableux. Le climat est
soudano-guinéen, caractérisé par une pluviométrie annuelle comprise entre 1000 et 1200 mm. A
Cotonou, les expérimentations se sont déroulées en station expérimentale et ont pu être irriguées
en fonction des besoins.
Page | 44
Cotonou Glazoué
Figure 23. Climatogramme de Cotonou (en 2007 (a), 2008 (b), 2009 (c)) et de Glazoué (en
2007 (d), 2008 (e) et 2009 (f)). Les températures moyennes mensuelles et les cumuls de
précipitations mensuels sont mis en graphique avec un rapport d’échelle de 1:2 afin de
permettre une appréciation rapide des mois secs (aires en pointillés rouges), humides (aires
hachurée en bleu) et très humides (aires colorée en bleu). L’échelle des précipitations change
au-delà de 100 mm par mois. Les températures annuelles minim ales et maximales sont
présentées à côté des ordonnées. La température moyenne annuelle et la pluviométrie totale
annuelle sont présentées en titre pour chaque climatogramme (Walter et al. 1975).
Page | 45
Le Tableau 1 présente les caractéristiques pédologiques des deux sites d’expérimentation.
Les sols sur la station de Cotonou sont fortement acides. La teneur en matière organique est
faible sur les deux sites. Les sols de Glazoué présentent une plus grande disponibilité en
nutriments (par exemple la quantité de phosphore assimilable et la capacité d’échange cationique).
Au Bénin, comme en Afrique de l’Ouest, les ignames D. rotundata sont, de loin, les plus
cultivées. Elles couvrent environ 90 % des surfaces consacrées à l’igname dans chacun des
différents pays producteurs d’Afrique occidentale, exception faite toutefois pour la Côte d’Ivoire.
Les ignames D. alata, moins fréquentes, existent partout dans la zone de production ouest-
africaine de l’igname. Elles ont une importance majeure en Côte d’Ivoire où elles représentent
environ 60 % des surfaces cultivées en igname.
Les expérimentations menées dans le cadre de cette thèse ont été réalisées par l’unité mixte de
recherche AGAP du Cirad, en collaboration avec l’IITA. La direction de thèse était assurée par
AgroParisTech et l’encadrement au quotidien par l’unité de recherche Astro de l’INRA. Les
expérimentations de thèse font partie d’un ensemble d’expérimentations plus vaste mené au
Bénin entre 2006 et 2009 sur l’écophysiologie de l’igname. Le Tableau 2 présente un résumé des
expérimentations réalisées et précise comment ont été utilisés les jeux de données au sein des
Chapitres 3, 4 et 5.
Page | 46
Tableau 2. Expérimentations de thèse. N éch et N rec présentent respectivement le nombre d’échantillons utilisés pour les observations
additionnelles et le nombre de plants récoltés en fin de saison pour la mesure des composantes du rendement. Ch.3, Ch.4 et Ch.5 indiquen t les jeux
de données utilisés dans les chapitres 3, 4 et 5 respectivement. Fl=Florido, Mo=Morokorou, Ko=Kokoro, Gn=Gnidou.
Essai Année Site Cultivar Plantation Densité Traitement Nb de Observations additionnelles Néch Nrec Ch.3 Ch.4 Ch.5
plant récolte
m -2
1 2006 Cotonou Fl, Mo, Ko 11-avr. 0,63 Azote 2 Interception 10 418 Oui
2 Cotonou Fl, Gn 4-avr. 0,11 Densité 6 Topologie aérienne, Architecture 68 50 Oui Oui
racinaire
3 Glazoué Fl, Mo, Ko 14-janv. 0,71 - 6 Recouvrement 144 308 Oui
4 2007 Cotonou Fl, Gn 19-févr. 0,08 Azote 5 Architecture racinaire 68 11
5 Cotonou Fl, Mo, Ko 23-févr. 0,83 - 4 Interception, [N, P, K], Recouvrement, 29 109 Oui Oui
Litière
6 Cotonou Fl, Mo, Ko 21-févr. 0,67 Azote 5 [N, P, K], Recouvrement, Litière 115 122 Oui Oui
7 Glazoué Fl, Mo, Ko 24-avr. 0,71 Azote 10 [N, P, K] 220 40 Oui
8 Glazoué Fl, Mo, Ko 24-avr. 0,67 Azote 5 [N], Recouvrement 120 120 Oui Oui
9 2008 Cotonou Fl, Mo 20-avr. 0,67 - - Dynamique racinaire, DVR 13 76
10 Cotonou Fl, Mo 21-avr. 0,67 Azote 4 SPAD, DVF, [N, P, K], Nitrates, 96 139 Oui Oui Oui
Litière
11 Cotonou Fl, Mo 4-avr. 1,25 Azote, 1 - 0 64
Potassium
12 Glazoué Fl, Mo 04-mai 0,67 Azote 4 - 96 337 Oui Oui
13 2009 Cotonou Fl, Mo 24-mars 0,67 Azote 5 Interception, DVF, [N, P, K], Nitrates, 96 162 Oui Oui Oui
Litière, Recouvrement
14 Cotonou Fl, Mo 06-mai 0,67 - 5 Architecture racinaire, Masse linéique 37 0
15 Cotonou Fl, Mo 20-avr. 0,67 - Dynamique racinaire 12 0
16 Glazoué Fl, Mo 22-mars 0,67 Azote 3 DVF 72 149 Oui
DVF : Durée de vie foliaire (jours).
DVR : Durée de vie racinaire (jours).
SPAD : mesure de teneur en chlorophylle avec un chlorophylle-mètre (SPAD 501, Konica Minolta, Osaka, Japan).
[N, P, K] : concentration tissulaire en azote, phosphore ou potassium.
Page | 47
Les 16 expérimentations présentées ont toutes été réalisées en blocs aléatoires complets
avec de deux à quatre répétitions par unité expérimentale et par date d’échantillonnage. L’unité
expérimentale est composée de une à trois plantes, entourées en tout sens d’une plante de
bordure. Certaines observations ont été réalisées systématiquement sur tous les essais et pour
chaque plante du dispositif : il s’agit du poids du semenceau, de la date d’émergence individuelle,
du nombre de tiges principales, du nombre de nœuds portant des cataphylles, des biomasses
fraîches et sèches (70°C durant 72h) de tiges et de feuilles, et des composantes du rendement
(nombre et poids de tubercules). Pour les plantes échantillonnées, la surface foliaire a été soit (i)
mesurée à l’aide d’un planimètre (Li-Cor 3100, LI-COR Inc., Lincoln, NE, USA), comme c’est le
cas à Cotonou en 2008 et 2009, (ii) soit estimée à l’aide d’une relation allométrique entre la
biomasse sèche et la surface des feuilles, dans les autres situations ; le choix du modèle
allométrique fait l’objet du Chapitre 5.
D’autres observations ont été réalisées ponctuellement. Elles sont précisées dans la
colonne « observations additionnelles » du Tableau 2 et illustrées à la Figure 24 :
- Interception : des mesures d’interception du rayonnement ont été réalisées à l’aide d’un
ceptomètre (AccuPAR, LP-80, Decagon Devices, Pullman, WA, USA).
- Topologie aérienne : la topologie aérienne (longueur d’axes, ramification, nombre de
nœuds, nombre de feuilles) a été caractérisée dans le cadre d’une expérimentation
tuteurée, de type « corde à linge », permettant de séparer les axes et de les observer de
manière non destructive.
- Recouvrement : le recouvrement au sol par la plante a été estimé indirectement, soit
visuellement à Glazoué (Marnotte 1984), soit par analyse d’image à l’aide du logiciel
ImageJ et du logiciel R (Schneider et al. 2012, R Development Core Team 2013).
- DVF : la Durée de Vie Foliaire a été suivie dans certaines expérimentations par marquage
de feuilles à différentes dates d’émissions et suivi hebdomadaire.
- Litière : les feuilles mortes tombées au sol ont été ramassées tous les trois jours et leur
biomasse sèche a été mesurée.
- Architecture racinaire : l’architecture racinaire (topologie, ramification, profondeur, masse
linéique…) a été caractérisée via des excavations successives (Figure 10).
- Dynamique racinaire : les dynamiques racinaires (vitesse de croissance, durée de vie
racinaire…) ont été suivies à l’aide de rhizotrons de plein champ (Jourdan et Chopart
1998).
- [N, P, K] : les concentrations en azote, phosphore et potassium des différents organes
(semenceaux, feuilles, tiges, feuilles mortes, tubercules) ont été mesurées.
- Nitrates : les concentrations en nitrates dans les feuilles, tiges et pétioles ont été mesurées.
Page | 48
Figure 24. Dispositifs de terrains installés à Cotonou : mesure de l'interception du
rayonnement solaire par la culture d'igname à l'aide d'un ceptomètre (a), étude de l'architecture
aérienne de l’igname à l'aide d'un dispositif de type "corde à linge" (b), estimation du
recouvrement du sol par une culture d'igname à l'aide de l'analyse d'image (c), observation de
l'architecture racinaire (d), et étude de la dynamique racinaire de l'igname à l'aide des
rhizotrons de terrain (e).
Page | 49
Ces observations n’ont pas toutes été utilisées dans le cadre de cette thèse, mais elles nous ont
permis d’obtenir des informations plus complètes sur les résultats mis en avant.
Afin de représenter les deux espèces majeures d’igname en Afrique de l’Ouest, deux variétés
ont été retenues dans le cadre de la thèse. Il s’agit de Florido pour l’espèce D. alata et Morokorou
pour l’espèce D. rotundata. Florido est une variété introduite en Afrique de l’Ouest au début des
années soixante-dix depuis Puerto Rico. Comme la majorité des D. alata, c’est une variété à une
seule récolte ; elle produit de deux à cinq tubercules de formes arrondies (Doumbia et al. 2004).
Morokorou est une variété traditionnelle à double récolte, cultivée au Nord et Centre Bénin. La
première récolte produit le plus souvent un seul tubercule allongé de forme cylindrique (Dansi et
al. 1999). D’autres variétés de D. rotundata (Kokoro, Kpouna, Gnidou) ont été étudiées
ponctuellement au cours des quatre années d’expérimentations, mais n’ont pas été utilisées dans
le cadre de la thèse.
Page | 50
2.2. Hypothèses de travail et objectifs de l’étude
Le Chapitre 1 a montré que l’état actuel des connaissances de la culture d’igname est insuffisant
pour permettre d’élaborer et piloter des systèmes de culture durables en Afrique de l’Ouest. Les
études plus mécanistes menées récemment soulignent l’importance de la variabilité de rendement
entre plantes d’une même parcelle. A l’heure actuelle cette variabilité reste inexpliquée et freine
les tentatives de compréhension du fonctionnement de la plante. Il nous semble donc important
de la quantifier, d’en comprendre les causes et d’en mesurer les conséquences.
Les faibles densités de plantation pratiquées en Afrique de l’Ouest et la plantation en
buttes minimisent les interactions entre plantes d’igname. Il nous paraît donc peu probable que la
compétition au sein de la parcelle et la variation des ressources expliquent cette variabilité. En
revanche, l’utilisation par l’agriculteur de semenceaux d’origine et de tailles variables comme
matériel de plantation, nous apparaît comme une cause possible de cette variabilité. Le
fonctionnement de la plante et de la culture durant les premiers stades de croissance (de la
plantation à l’apparition des premières feuilles vraies) tient donc potentiellement un rôle clé dans
le déterminisme de la variabilité interplante. La sensibilité de l’igname à la photopériode aura
tendance à hâter l’initiation de la tubérisation des plantes « retardataires » ; et l’imminence de la
saison sèche ne permet pas de débordement de la saison culturale. Le retard pris par certaines
plantes durant ce premier stade est donc susceptible de pénaliser directement le rendement. C’est
pourquoi nous faisons les hypothèses que (i) la variabilité de taille entre individus s’exprime dès le
premier stade, (ii) il n’existe pas (ou peu) de compétition entre plantes d’igname dans les systèmes
de culture traditionnels en Afrique de l’Ouest, (iii) la variabilité de taille générée à la plantation
n’est pas compensée par la suite et entraîne la variabilité de rendement observée à la récolte, et
(iv) cette variabilité entraîne une perte de rendement à l’échelle de la parcelle. L'objectif de la
thèse est donc d’étudier la variabilité interplante des deux espèces majeures d’igname en Afrique
de l’Ouest, et plus particulièrement :
- de caractériser cette variabilité et d’en identifier les causes principales,
- d’identifier la structure de dépendance entre les différentes causes et quantifier leur
contribution respective,
- de quantifier les conséquences agronomiques et économiques de cette variabilité à l’aide
d’un modèle de croissance et de développement à l’échelle de la plante.
Page | 51
2.3. Organisation du mémoire
La suite du manuscrit est organisée en quatre chapitres et une conclusion. Les Chapitres 3, 4 et 6
correspondent respectivement aux trois objectifs fixés. Le Chapitre 5 est une mise au point
préalable et nécessaire à la calibration du modèle-plante développé dans le chapitre 6. La
présentation de ces quatre chapitres sous forme de publications implique inévitablement quelques
débordements thématiques, c’est pourquoi, en tête de chapitre, un encart résume les points
importants abordés et leur lien avec les objectifs fixés. La conclusion permet d’agréger ces
différentes contributions et de les mettre en perspective.
Page | 52
Page | 53
Chapitre 3 Caractérisation de la variabilité de taille
et de rendement en culture
d’igname, en relation avec la date
d’émergence individuelle
When a yam doesn't grow well,
we don't blame it.
Proverbe Tshi (Ghana)
Page | 54
Yams plant size hierarchy and yield
variability: emergence time is critical
Abstract
Yam crops (Dioscorea spp.) present a very high and unexplained interplant variability which
hinders attempts at intensification. This paper aims to characterize the plant-to-plant variability in
yield and to identify its underlying causes for the two major yam species (Dioscorea alata and
Dioscorea rotundata). Four field experiments were carried out between 2006 and 2009 in Benin.
Yams were grown using a traditional cropping method (i.e. in mounds at 0.7 plants m-2) without
biotic or abiotic stresses. In order to test interplant competition, a low density treatment (0.08
plants m-2) was included for D. alata in the 2006 experiment. Throughout four years of
experimentation, yields varied from 12 Mg ha-1 to 21 Mg ha-1. Both yam species presented a high
interplant coefficient of variation (CV) for tuber yield (42 to 71%). The unbiased Gini coefficient
(G’) was used to measure how steep a hierarchy is in an absolute sense. CV and G’ of individual
plant biomass both confirm clear plant size hierarchies from early growth. However, no
difference in the CV of plant size and plant tuber yield was observed between high and low plant
density. This implies that, despite early interaction between neighbouring plants, competition was
not the driving factor controlling plant variability. In fact uneven emergence proved to be the
primary cause. Yam emergence takes place over a long period (e.g. it took 51 and 47 days for the
90% central range to emerge for D. alata and D. rotundata respectively), creating an early inter-
plant size hierarchy which later affected tuber production. For both species, plants which
emerged early initiated their tuberization earlier in the growing season and reached higher
maximum yield regardless of weather conditions (e.g. 1200 and 764 g plant-1 for early-emerging
D. alata and D. rotundata plants respectively, and 539 and 281 g plant-1 for late-emerging plants).
Plant size hierarchization together with its observed left-skewed distribution, led to reduce total
and marketable yield by increasing the proportion of small tubers. These results highlight the
need to better understand the underlying mechanisms controlling the yams’ uneven emergence
before attempting to improve traditional cropping systems.
Keywords: Dioscorea, yam, emergence, plant size hierarchies, tuber yield, cohort
Page | 55
3.1. Introduction
Dioscorea is a C3 monocotyledonous genus grown throughout the tropical world for food,
pharmaceutical products, and ornamental purposes (Ayensu 1972, Cornet et al. 2007). It is an
important food crop in at least ten tropical countries from Nigeria to Jamaica and the Solomon
islands, accounting for 155 million inhabitants. More than 85% of world yam production takes
place in West Africa (FAOSTAT 2014). Despite its economic value, only limited literature is
available on the physiology of the crop and on the effect of agricultural practices on yields
(Onwueme and Haverkort 1991). This is particularly important because of the tremendous yield
gap (Nin-Pratt et al., 2011) and the wide range of crop management practices applied to yams; e.g.
staked or non-staked, multiplication by sets of various size and origins, and varying fertilizer rates
and planting dates (Orkwor et al. 1998).
Attempts to better understand yam crop physiology and reduce yield gap come up against
the very high and still unexplained intra-field yield variability (Akoroda 1984, Ferguson et al. 1969;
Okoli et al. 1999). Coefficient of variation (CV) of individual plant yield is often used to estimate
this variability (Weiner and Solbrig 1984). While plant yield CV reported in the literature for
potato (Solanum tuberosum L.) or sweet potato (Ipomea batatas L.) hardly reached 30%, yams have a
CV as high as 60% (Akoroda 1984, Alcoy et al. 1993, Okoli et al. 1999, Wurr et al. 1993).
Consequences of this plant-to-plant yield variability are already known for other root and tuber
crops: (i) lower mean yield, (ii) lower marketable production due to tiny and oversized tubers, and
(iii) difficulties for intensifying cropping systems because of the high variability of the individual’s
response to agronomic practices (Alcoy et al. 1993, Mackerron et al. 1988).
Beneath an apparently uniform cover, yam fields contain individual plants of variable size.
In fact, most plant populations exist as size hierarchies of individuals (Weiner and Thomas 1986).
The size hierarchy refer to the frequency distribution of individual plant sizes in which relatively
few individuals contribute most of the population's biomass and most individuals are relatively
small. These unequal distributions are appropriately thought of as size (usually biomass)
inequalities (Weiner and Thomas 1986). In controlled environment and monospecific stand,
Weiner and Thomas (1986) identified two main factors driving strong individual size hierarchies.
First, they showed that these size distributions could be a direct consequence of interplant
competition. Here competition refers to the negative effects caused by the presence of neighbors
by reducing the availability of resources (Harper 1977). Furthermore, Weiner and Thomas (1986)
demonstrate the existence of a positive relationship between size inequality and population
density as a consequence of increasing interplant competition with increasing plant density.
Page | 56
However interplant competition might be limited in the traditional cropping systems of West
African farmers because of the low yams planting density.
The second factor driving strong individual size hierarchies highlighted by Weiner and
Thomas (1984) is the uneven emergence, leading to plant age differences (Black and Wilkinson
1963, Damgaard and Weiner 2000). Unlike potatoes, yams do not benefit from a well-structured
certified tuber seed production system allowing fast and uniform sprouting. Because yams are
grown by planting pieces of tuber, or small whole tubers (called seed yams) saved from the
previous season, many factors may affect emergence. Accordingly uncontrolled seed yam
physiological age, size and tuber portion origin (proximal, middle or distal) may influence
emergence time (Orkwor et al. 1998) while seed quality (e.g. dry matter and nutrient content) may
affect sprouting vigour (Ferguson 1973, Orkwor et al. 1998). These uncontrolled conditions
may lead to a population with disparate plant ages. Hence we hypothesize that for West African
conditions yam yield variability in the field is rather a direct consequence of individual emergence
variability but not an effect of interplant competition.
The objective of this study was to test that hypothesis. The aims were to assess plant-to-
plant yield variability for the two major yam species (Dioscorea alata L. and Dioscorea rotundata
Benth.), and to identify the underlying mechanisms controlling it in the field (emergence and
competition). The consequences of these findings are discussed in terms of further improvement
of yam cropping systems.
Page | 57
3.2. Materials and methods
The data used in this study come from four trials carried out at AfricaRice/IITA-Cotonou
Station (Benin, 6°25 N, 2°19 E, 23 m asl) between 2006 and 2009, using the two main yam
species D. alata and D. rotundata. For D. rotundata the cultivar Morokorou was used. It is a
traditional early-maturing variety originating from north Benin, producing a few (1-3) cylindrical
tubers. For D. alata the cultivar Florido was used. It was introduced into West Africa from Puerto
Rico in the early 1970s and produces two to five round tubers (Doumbia et al. 2004).
The soils used in this study are arenosols (FAO-ISRIC-ISSS 1998) with good physical
characteristics but low nutrient levels (Table 3). The field experiments were located in the forest-
savannah transition zone. The climate is sub-equatorial with a bimodal rainfall pattern and a
mean annual rainfall of 1200 mm, falling mainly during the rainy seasons from March to July and
from September to October. The mean annual temperature was about 27°C with a low diurnal
variation of 7–10°C. All trials followed traditional planting systems used in West Africa. Small
entire seed tubers with a constant weight of 400 g were planted without staking in mounds at a
density of 0.7 plant m-2. In order to test interplant competition a low density treatment of 0.08
plant m-2 was added for D. alata in the 2006 experiment. And to completely avoid competition in
this treatment, plants were staked and a rope network allowed each stem to grow individually
without any plant interaction.
Compound fertilizer was applied at a rate of 60 kg N ha-1, 30 kg P ha-1 and 140kg K ha-1
one month after emergence. Additional N was applied two months after emergence as urea at a
rate of 60 kg N ha-1. The crop was irrigated to field capacity at planting. Afterward it was irrigated
according to a water balance, and further supplementary irrigation (totaling between 80 to
150 mm depending on the cropping season) was applied to replace estimated evapotranspiration
Page | 58
using overhead sprinklers (Marcos et al. 2009). The plants did not show any visual evidence of
water stress or nutritional imbalance. Weed control was done by hand approximately every two
weeks. All experiments were completely randomized design (the yam species being the only
factor with two level: D. rotundata and D. alata) using four replications of 25 plants per treatment.
Individual emergence was recorded when a tuber sprout emerged from the soil. Individual plant
emergence dates were recorded in all trials until final plant stand. Final emergence rate as defined
by the percentage of emerged plants was always higher than 95%. Mean emergence date and its
standard error between the four years were May 7th ± 10 days. For each experiment the date of
field emergence was determined by the date where 50% of individual plants had emerged. This
date was used to define further plant sampling dates. In this study time to plant sampling is
always expressed in number of days after 50% emergence (DAE). Five or six plant samplings
were made during the crop’s lifespan (typically 30, 60, 80, 125, 180 and 230 DAE). At each
sampling date, an area of 1.2 m² was totally harvested. The sampled plants were always bordered
by at least one plant in each direction. At each sampling date the plant above-ground fresh and
dry biomass, the number of tubers per plant and the fresh and dry biomass of each individual
tuber were measured. The trials were ended at 230 DAE with the last sampling, long after tuber
maturity.
Because of the prostrate habit of yams, only sampling dates up to 90 DAE allowed
individual measurements of above-ground and tuber biomass. Later it was not possible to
separate individual plants and above-ground measurements were based on soil area. So after 90
DAE, yield components (i.e. tuber number per plant and tuber weight) were the only individual
measurements to rely on.
In order to ensure the absence of biotic stress a visual estimation was made for the main
foliar diseases (anthracnose for D. alata and viruses for D. rotundata) and for the main tuber
diseases (nematodes and mealy bugs) at harvest. Only in those few instance, plants exhibiting
disease symptoms were discarded from our samplings.
Page | 59
3.2.3. Statistical analysis
Differences in yield between years and species were tested using ANOVA in conjunction with
Tukey’s test to determine the significance of differences between means.
The means and coefficients of variation (CV) were used to characterize plant size
distribution. Beyond traditional dispersion measures of distribution, this study focused on
inequality rather than on asymmetry in describing the size hierarchy (Bendel et al. 1989).
Therefore the Gini coefficient proposed by Bendel et al. (1989) was used to measure how steep a
hierarchy is in an absolute sense. The Gini coefficient is population- and scale-independent. It
means that its value does not change if each individual's size is raised in the same proportion.
Thus, it works equally well on a small group as it does on a large group (Bendel et al. 1989) and it
allow the comparison of inequalities between populations with different means (Weiner and
Solbrig 1984). It has a minimum value of 0 when all individuals are equal in weight and a
maximum of 1 when all individuals but one have zero weight. Steepness of hierarchy is measured
with a Gini coefficient ordered as proposed by Damgaard and Weiner (2000):
∑𝑛
𝑖=1(2𝑖−𝑛−1)𝑥′𝑖
𝐺= (1)
𝑛²𝜇
where n is the number of plants, 𝑥′𝑖 is the size of the plant i and µ is the average plant size.
Glasser (1962) shown that sample Gini coefficient calculated by Eq. (1) should be multiplied by
n (n-1)-1 to reach unbiased estimates (G’).
Populations with different size distribution can have the same G’, depending on whether
most of the inequality is due to large or small individuals (Weiner and Solbrig 1984). So the
Lorenz asymmetry coefficient (S) was used to estimate the degree of asymmetry of the
distribution of plant variables (Damgaard and Weiner 2000). The Lorenz asymmetry coefficient is
defined as the point at which the slope of the Lorenz curve is parallel to the line of equality
(Weiner and Solbrig 1984). So, when S>1 the inequality is due mostly to a small number of very
large individuals, while when S<1 the inequality is due mostly to a large number of very small
individuals. G’ and S were calculated using the ineq extension package for the software
environment R (R Development Core Team 2013) and the Lorenz.R function developed by
Buckley and Daamgard (2012).
Page | 60
In order to test the influence of emergence on yield, the emergence dates were ordered
and their distribution was divided into five classes (pentiles), each containing a fifth (20%) of the
population: first 20 percentile (P0-20), 20th to 40th percentile (P20-40), 40th to 60th percentile (P40-
60), 60th to 80th percentile (P60-80) and last 20 percentile (P80-100). Differences of mean tuber
biomass between emergence classes were tested using ANOVA.
The time course of plant tuber weight for each emergence class was modelled using non-
linear regression curve fitting and plotting using nls function, drc and ggplot2 extension packages
for the software environment R (R Development Core Team 2013). In a preliminary study
logistic, Gompertz, Richards and Weibull models and the beta growth function described by Yin
et al. (2008) were compared. The quality of fit for each model was estimated using the Akaike
Information Criterion and the extra sum of squares (Akaike 1973). The beta growth model gave
the best fit over all the trials (data not shown). The beta growth function is a three-parameter
growth equation with an initial weight of zero:
𝑡𝑒
𝑡 −𝑡 𝑡 𝑡 −𝑡𝑚
𝑤 = 𝑤𝑚𝑎𝑥 (1 + 𝑡 𝑒−𝑡 ) (𝑡 ) 𝑒 with 0 ≤ 𝑡𝑚 < 𝑡𝑒 and 𝑤 = 𝑤𝑚𝑎𝑥 if 𝑡 > 𝑡𝑒 (2)
𝑒 𝑚 𝑒
where tm (DAE) is the time at maximum growth rate and Wmax (g tuber plant-1) is the maximum
value of tuber weight w (g tuber plant-1), which is reached at time te (DAE).
To verify the suitability for fitting individual emergence classes (full model) compared
with a single fit for all the classes (reduced model), an ANOVA was used to test whether
reduction in the residual sum of squares are statistically significant or not (Lomax and Hahs-
Vaughn 2012). Differences in model parameters (te, tm, Wmax in Eq. (2)) between each emergence
class were tested using the compParm function of the drc extension package for the software
environment R (R Development Core Team 2013).
Page | 61
3.3. Results
The statistics of yam yield variability throughout the four years demonstrated the high variability
between and within experiments (Table 4). For traditional plant stands (i.e. 0.7 plant m-²), yams
yielded between 6.9 and 23.8 Mg ha-1 of fresh tubers over the 4 years of experiments. In 2008 and
2009 D. rotundata had higher yield than D. alata. The CV of individual plant yield was always
above 40% and was highly correlated (R²=0.98, P<0.001) with G’. In all experiments, S was
always less than 1, indicating that plant yield inequality is due mostly to a large number of small
individuals. There was no difference in the CV and G’ of the yield of D. alata between traditional
planting density (0.70 plant m-²) and isolated plants (0.08 plant m-²).
Table 4. Individual yield inequality statistics: number of samples (N), co efficient of variation
(CV), unbiased Gini coefficient (G’) and Lorenz asymmetry coefficient (S). Yield values having
different letters are significantly different at the 5% level
Density Yield CV
Year Trial Species N G’ S
(plants m-²) (Mg ha-1) (%)
2006 T1 0.08 D. alata 50 5.5a 45 0.25 0.90
0.70 D. alata 50 19.1cd 42 0.24 0.88
0.70 D. rotundata 50 16.1bd 55 0.31 0.67
2007 T2 0.70 D. alata 95 22.1d 57 0.32 0.91
0.70 D. rotundata 95 23.8d 59 0.33 0.78
2008 T3 0.70 D. alata 72 6.9a 65 0.36 0.85
0.70 D. rotundata 72 14.0bc 60 0.33 0.89
2009 T4 0.70 D. alata 72 12.4ab 67 0.36 0.96
0.70 D. rotundata 72 20.7d 52 0.28 0.89
The individual yams’ emergence distribution in the four trials was typical of an uneven emergence
(Figure 25). Emergence distribution was left skewed and the 90% central range was spread
over 51 and 47 days for D. alata and D. rotundata, without any significant distinction between
species. The 90% central range is defined by the numerical difference between the values of the
5th and 95th percentiles.
Page | 62
Figure 25. Distribution of the weekly proportion of emergence for D. alata and D. rotundata
averaged over trials carried out from 2006 to 2009. Error bars indicate standard deviations
between years.
As an example, Table 5 presents a summary of individual plant weight and yield statistics
over time for trial 2 in 2007. For both species individual plant size inequality started from the
very beginning of crop life (Table 5). The CV of individual total dry matter was very high as
soon as the first month after emergence for both species. The final plant tuber dry matter had an
even higher CV. Other trials showed the same trend.
Table 5. Summary of individual plant weight statistics over time. DAE is the number of days
after 50% emergence, N is the number of observation s and CV is the coefficient of variation
Page | 63
3.3.3. Influence of emergence date on subsequent plant yield
The model statistics and parameters summary by species highlight some significant differences
between emergence classes (Table 6). For both species, the main differences between emergence
classes concerned the maximum tuber weight (Wmax) and time to reach it (te): the earlier the plant
emerged, the higher the plant yielded and the faster this yield was achieved. Difference in final
plant yield amongst emergence class was always significant for D. rotundata while it was only
significant for extreme emergence class for D. alata. For both species, differences in the time to
achieve the final yield were only significant for extreme emergence class. Only for D. rotundata
were there some significant differences in the time to maximum growth rate (tm) between the first
emergence class (P0-20) and the others.
Table 6. Beta growth function parameters fitted to emergence classes and yam species. P0 -20
is the first 20 percentile class, P20-40 is the 20th to 40th percentile class, P40-60 is the 40th to
60th percentile class, P60-80 is the 60th to 80th percentile class, P80-100 is last 20 percentile
class. Wmax is the maximum value of tuber weight, te is the time at maximum tuber weight and
tm is the time at maximum growth rate. Model parameters with different letters are
significantly different at the 5% level for a given species.
Model parameters
Species Emergence class Wmax te tm
(g plant-1) (day of the year)
P0-20 764a 239a 231a
P20-40 734ab 247 a 229a
D. alata P40-60 569bc 253a 236ab
P60-80 439c 313a 288ab
P80-100 281c 293a 271b
P0-20 1200a 239a 218a
P20-40 1056b 269 b 240b
D. rotundata P40-60 929b 279ab 246ab
P60-80 733c 254ab 237ab
P80-100 539d 269ab 247b
Plant tuber dry weight over time shown significant differences between species (Figure
26). Whatever the emergence class, D. rotundata presented an earlier tuberization than D. alata.
For the sake of simplicity, Figure 26 only shows three emergence classes e.g. P0-20, P40-60 and
P80-100. For both species, the earlier the emergence class, the earlier the tuberization and the
higher the yield.
Page | 64
Figure 26. Plant tuber weight as a function of time of D. alata (in red) and D. rotundata (in
blue) for three different emergence classes: P0-20 is the first 20 percentile emergence class
(solid line), P40-60 is the 40th to 60th percentile class (dotted line) and P80-100 is last 20
percentile class (dashed line). Lines correspond to the fitted beta growth model.
The influence of emergence class on yam yield components was significant (Figure 27). For
easier readability Figure 27 only presents data for extreme emergence classes: the top 20
percentile (P0-20) and the last 20 percentile (P80-100). For both species, the plant tuber dry
weight of the late-emerged plants showed significantly narrower distributions skewed towards
smaller values compared with the early-emerged plants (Figure 27a and Figure 27b). Despite
this similar behaviour of D. alata and D. rotundata for total plant yield, significant differences
between species were observed for tuber number and tuber dry weight. For D. alata, differences
between emergence classes concerned the tuber numbers per plant (Figure 27c and Figure
27e). In contrast, for D. rotundata the main difference between emergence classes concerned the
mean tuber weight (Figure 27d and Figure 27f).
Page | 65
Figure 27. Frequency distribution of: plant tuber dry weight (a) and (b), number of tubers per
plant (c) and (d), and tuber dry weight (e) and (f) of early (P0 -20 is the top 20 percentile
emergence class) and late (P80-100 is last 20 percentile emergence class) emergence class for
two yam species (D. alata and D. rotundata). Lines correspond to the adjusted empirical
probability density function for early (solid line) and late (dotted line) emergence classes.
Page | 66
3.4. Discussion
This study confirmed the tremendous variability in individual yam plant yield on the field scale.
The three major findings of this study are: (i) plant yield variability is mainly due to uneven
emergence rather than interplant competition, (ii) the earlier the plant emerges, the higher the
tuber yield, and (iii) there is a species effect in the response of yield components to different
emergence classes. Throughout four years of experiments, both yam species (D. alata and D.
rotundata) exhibited very clear plant size hierarchies. Weiner and Solbrig (1984) have proposed the
use of CV and G’ together to define and quantify plant size hierarchies. In the present study, yam
plant yield had a CV of over 40% and G’ up to 0.35 while commonly observed interplant CV for
tuber crops such as potato or sweet potato scarcely reached 20% (Smith 1938, Storck 2011). To
our knowledge, there is no available information on G’ for root and tuber crops. Sadras and
Bongiovanni (2004) reported G’ values from 0.04 to 0.19 for maize (Zea maïs L.), and Ellison and
Rabinowitz (1989) found G’ value of 0.20 for pea (Pisum sativum L.). The observed high values of
CV and G’ highlight the variation of yam plant yield. Moreover CVs of total plant biomass
confirm this observation.
For West African conditions, yam plant variability is not a direct consequence of
interplant competition. Weiner (1985) showed that G’ of plant size increases sharply with
increasing density due to increase competition. For D. alata in 2006, no difference was found in
this coefficient between virtually isolated plants and the traditional African plant stand density.
Because there was no interference between plants in the isolated plants treatment, there was also
no competition. So for the isolated plant stand, none of the interplant variability (as measured by
CV or G’) could be assigned to competition. Therefore because both coefficients (CV and G’)
are similar between the two treatments it can be concluded that there is no competition in this
study. Our results differ from those obtained by Vega and Sadras (2003) and Weiner and Thomas
(1986), who observed increasing competition with increasing plant density for many crops. The
lack of competition in this study could be due to the low density traditionally practiced in the yam
fields of West African farmers. Indeed, in order to get the biggest tubers at harvest farmers plants
yams at 0.7 plant m-² in individual big mound, limiting interactions between plants. So yams
grown in mounds at traditional West African planting density can be assumed to behave like
isolated plants. Nevertheless, yam interplant competition should not be ignored in other
situations. In fact, as a prostrate crop with very shallow rooting system, interplant competition
may be considerable at higher plant density. Elsewhere in the world, yams are grown at a very
Page | 67
wide range of plant density: medium in Caribbean (2-3 plants m-²) to very high in Japan (6-9
plants m-²) (Orkwor et al. 1998). With so much variation in cropping density, competition effects
are likely to be specific to each cropping system.
High CVs of total plant biomass can be observed from the first weeks of growth. Such a
high and early plant size inequality results from a heterogeneous emergence and/or initial growth
rates. Van der Zaag and Van Loon (1987) showed that for potato, 90% of seed tubers emerge
within 10 to 20 days depending on storage conditions. The results of this study indicated that
yams took twice as long to emerge.
For both species, early-emerged plants initiated their tuberization earlier in the growing
season and reached higher maximum yield. Photoperiod sensitivity may explain why late-
emerging plants yielded less. Indeed yams are photoperiodic crops (Marcos et al. 2009, Marcos et
al. 2011, Shiwachi et al. 2002) and a daylength greater than 12 hours favours growth of vines while
satisfactory tuber growth takes place during shorter daylengths. Late emergence means a shorter
period of vegetative growth and less leaf area index for tuber growth, resulting in lower yields.
The results of this study are in accordance with these observations.
Results showed that while plant tuber yield reduction linked to late emergence for D. alata
was mainly due to smaller tuber number per plant, plant yield reduction for D. rotundata resulted
from smaller mean tuber weight. This indicates a different strategy for yield formation, which
could be associated with a different sensitivity to photoperiod. Differences between species for
the response of the yield components to delayed emergence have already been observed for
potato (Rykbost and Locke 1999).
3.4.2. Consequences
Managing a crop consisting of individual plants of various ages and sizes has important
constraints: (i) for the farmers, plant size variation leads to reduced total and marketable yields,
(ii) for researchers, it affects both statistical analysis and interpretation of experimental results.
The detrimental effect of delayed emergence may be exacerbated in traditional cropping systems
where no irrigation is available. Indeed in West Africa irrigation is scarce and the end of the
cropping season is determined by the rainfall pattern. Hence a delay in emergence might not only
affect the length of the growing period, but could also reduce the growth rate at the end of
vegetative phase due to water deficits. Shortening of the growing season or untimely tuberization
can both lead to decreased plant and crop yield. Reduced marketable yield could be another
Page | 68
consequence of plant size hierarchy because yam prices on West African markets are based on
tuber size (i.e. bigger tubers attract a higher price) (Orkwor et al. 1998). D. rotundata shown a
higher proportion of smaller tubers in late-emerging plants compared to early ones. Uneven
emergence thus leads directly to reduced profitability for this yam species. Because tuber size of
D. alata is less affected by emergence delay, this species may represent a more profitable option.
In order to reduce emergence variability further research are needed to determine the cost benefit
ratio of producing quality seed tubers from alternatives multiplication techniques as vine cuttings,
vitroplant or microtuber (Olivier et al. 2012). Improved storage techniques could also reduce the
emergence variability (Van Loon 1987).
From an agronomic perspective, the high plant variability affects crop response to
agronomic practices (e.g. fertilization) because of different responses of each plant’s physiological
age or size. From a statistical point of view, skewed distributions make it difficult to apply
classical statistical analysis. Moreover, highly variable plant size induces large standard error and
obvious difficulties to obtain statistically significant results. For instance, Cornet et al. (2005)
found that only 22% of 200 field experiments dealing with yam fertilization showed a significant
increase in crop yield. This emphasizes that high plant size inequality may hinder attempts at crop
intensification because the experimental error is too great to detect any positive effect of the
agronomic practices tested.
These findings stress the need to better understand and control the causes of uneven
emergence of yams before attempting to improve traditional cropping systems. A better control
Page | 69
of tuber seed characteristics (e.g. physiological age, nutrient content, size, origin) could lead to
more uniform and higher crop yields.
Acknowledgements
Page | 70
Page | 71
Chapitre 4 Relations entre pratiques culturales,
croissance initiale de l’igname et
composantes du rendement
Pour le couteau,
il n'y a point de secret à l'intérieur de l'igname.
Proverbe Fon (Bénin).
Ce chapitre est présenté sous la forme d’un article soumis à la revue Field Crop
Research; les co-auteurs sont J. Sierraa, R. Tournebizea B. Gabrielleb et F. I. Lewisc. La
première partie de l’article aborde la structure des relations de dépendance entre les
variables explicatives et les composantes du rendement à l’aide d’un réseau bayésien. La
deuxième partie quantifie la contribution de chaque variable en calculant l’ensemble des
probabilités jointes. La présentation des deux parties sous forme d’un graphe acyclique
orienté permet une représentation simple et complète d’une réalité complexe.
___________
a INRA, UR1321 ASTRO, Agrosystèmes Tropicaux
b AgroParisTech, UMR1091 INRA AgroParisTech
c Section of Epidemiology, VetSuisse Faculty, University of Zürich
Page | 72
Bayesian network modeling of early
growth stages explains yam interplant
yield variability in West Africa
Abstract
Yams (Dioscorea spp.) are neglected, yet important species mainly for resource-poor farmers of
West Africa. The crop experiences early plant size hierarchy leading to reduced marketable yield.
Yams, as other vegetatively propagated crops, have heavier planting material, liable to induce
such interplant variability. This study identifies and compares for the first time the direct and
indirect dependency between planting practices, early growth variables and yield components of
the two main yam species, D. rotundata and D. alata.
Data set came from six trials carried out in Benin at two locations between 2007 and
2009. Additive Bayesian Network modeling was used for structure discovery; its directed acyclic
graph offers an ideal background for discussing complex systems with lack of theoretical
knowledge such as for yams.
Here we show that emergence date is the only direct cause of plant yield variability
common to both species. For D. rotundata we observed a direct contribution of the cataphyll
number to the plant tuber weight. These two results suggest the existence of some uncontrolled
latent variables (i.e. physiological age and nutrient content of the seed-tuber). For D. alata the
model did not show any effect of seed-tuber size, despite strong effect for D. rotundata. We
suggest that traditional practices transposed from native D. rotundata may have led to oversize
seed-tuber of D. alata. This study demonstrates a serious weakness in the traditional cropping
systems concerning the uncontrolled and wide range of physiological age and reserves content in
seed-tuber lots.
Keywords: Additive Bayesian network modeling, cataphyll, early growth, vegetatively propagated
crops, yam (Dioscorea spp.), yield variability.
Page | 73
4.1. Introduction
Yams (Dioscorea spp.) belong to a C3 monocotyledonous genus grown for food, pharmaceutical
products, and ornamental purposes (Ayensu 1972, Cornet et al. 2007). Food yams are grown for
their underground tubers, which are an important source of carbohydrates in many regions of the
world (Kennedy 2003, Asiedu and Sartie 2010). Projections showed that consumption of and
demand for yams will increase rapidly, particularly in West Africa, leading to an even higher
production growth rate for yam than for cassava (Scott et al. 2000). Of the ten economically
important species, Dioscorea rotundata and Dioscorea alata are the two main ones. Although D.
rotundata is the most cultivated species in West Africa where 90% of the world production is
located, D. alata is the most ubiquitous yam species, grown from Japan to West Africa and
throughout Central America (Orkwor et al. 1998). Despite its importance in food security and in
household income, particularly for the resource-poor famers of the planet, little is known about
the plant and crop physiology (Kennedy 2003). While over the past 45 years the Web of
Science® reference has nearly 500 000 publications for the single species of maize, it has 35 times
fewer publications for the whole yam genus Dioscorea spp.
In a recent study, Cornet et al. (2014) highlighted the strong plant size hierarchy in the
yam fields of West Africa and its adverse effects on total and marketable yield. In West African
yam cropping systems, competition between neighboring plants is negligible. This absence of
competition allowed the authors to study the crop yield variability at the individual plant scale.
Cornet et al. (2014) pointed out the role of individual emergence date to explain part of this
interplant variability and concluded that other complex interacting processes involved in early
growth stages might play an important role. Unlike potatoes (Solanum tuberosum), yams do not
benefit from a certified tuber seed production system allowing fast and uniform sprouting. Many
factors may affect the timing of yam emergence and vigor. In the absence of quality seed tuber
production, some of these factors are stochastic (i.e. physiological age and nutrient content of
seed-tubers) while others can be managed or controlled: seed-tuber size, seed-tuber state
(presprouted or not) and planting date (Ferguson 1973, Orkwor et al. 1998).
The objective of this study was to assess how early growth variables (until the appearance
of the first true leaf) affect the plant yield formation of the two major yam species. Because of the
number of variables involved in early growth stages and the complexity of the interactions,
Bayesian network modeling was used for structure discovery and parameter learning (Heckerman
et al. 1995, Korb and Nicholson 2004). Bayesian network analysis is a form of graphical modeling
focused on structure discovery: determining an optimal statistical model, i.e. graphical structure,
Page | 74
directly from observed data. Whilst relatively uncommon in plant development studies, Bayesian
network analysis is increasingly finding application in areas of biology, medicine, ecology or
epidemiology (Porth et al. 2013, Ward 2013). Lewis and McCormick (2012) show that while
multivariable regression seeks to identify covariates associated with some output variables (e.g.
plant yield), Bayesian network analysis goes much further in also empirically separating these into
those directly and indirectly dependent upon the output variable. Bayesian network has the
potential to reveal far more about key features of biological complex systems than existing
commonly used approaches (Lewis and McCormick 2012). Moreover, in Bayesian network
modeling no attempt is made to reduce dimensionality (as in exploratory data analysis using
principal components) which makes biological interpretation of results easier. Its probabilistic
formalism provides a natural treatment for stochastic nature of biological systems and
measurements (Needham et al. 2007).
The aims of this study were (i) to discover and quantify the structure of dependences
amongst practices at planting (i.e. planting date, seed-tuber state and weight), early growth
variables (i.e. emergence date, stem and cataphyll number) and plant yield components (i.e. tuber
number and weight), (ii) to compare these Bayesian network models for the two major food yam
species, and (iii) to discuss implications of these findings both for farmers and scientists.
Page | 75
4.2. Materials and methods
The dataset used in this study came from six trials carried out between 2007 and 2009 at two
locations: AfricaRice-Cotonou Station (Benin, 6°25 N, 2°19 E, 23 m asl) and Glazoué (Benin,
7°56 N, 2°15 E, 200 m asl). We used the two cultivars belonging to the main yam species D. alata
‘Florido’ and D. rotundata ‘Morokorou’. Morokorou is a traditional early-maturing variety
originating from north Benin, producing 1-3 cylindrical tubers. Florido was introduced into West
Africa from Puerto Rico in the early 1970s and produces two to five round tubers (Doumbia et al.
2004). The field experiments were both located in the forest-savannah transition zone. The
climate is sub-equatorial with a bimodal rainfall pattern, falling mainly from March to July and
from September to October. Administrative maps of Benin with the two locations, their soil
characteristics and weather data are presented in Figure 22, Figure 23 and Tableau 1.
All trials followed traditional planting systems used in West Africa. Small entire seed
tubers were planted in mounds, without staking, at a density of 0.7 plants m-2. The planting dates
were February 20th, March 25th, and April 25th in 2007, 2009 and 2008 respectively. Each year
seed-tubers were bought from one farmer, coming from the same field under the same climatic
and cropping conditions. Fertilizer was applied at a rate of 60 kg N ha -1, 30 kg P ha-1 and
140 kg K ha-1 one month after emergence. Additional N was applied two months after emergence
as urea at a rate of 60 kg N ha-1. Although the experiments in Glazoué were not irrigated, in
Cotonou the crop was irrigated to field capacity at planting. Afterwards it was irrigated according
to a water balance, and further supplementary irrigation (totaling between 80 to 110 mm
depending on the cropping season) was applied to replace estimated evapotranspiration using
overhead sprinklers (Marcos et al. 2009). The plants did not show any visual sign of water or
nutritional stress. Weed control was done by hand approximately every two weeks. All
experiments had a completely randomized design with yam species being the only treatment, with
two levels (D. rotundata and D. alata), and using four replications of 25 plants per treatment.
Page | 76
4.2.2. Explanatory variables
Table 7. Explanatory variables used to build the Bayesian model explaining plant yield.
Page | 77
Figure 28. Graphical presentation of the dispersion of the main variables for D. alata (red) and
D. rotundata (blue): (A) the count of sprouted and unsprouted seed-tubers, (B) the seed-tuber
weight per mound, (C) the emergence date, (D) the number of main stems per mound, (E) the
number of nodes carrying cataphyll per mound, (F) the number of tubers per mound, (G) the
mean tuber fresh weight per mound, and (H) the plant fr esh yield per mound. The colored box
corresponds to the inter-quartile range (IQR). The upper and lower whisker extends from the
box to the highest and lowest value that is within 1.5 * IQR. Data beyond the end of the
whiskers are outliers and plotted as points.
Page | 78
1st true leaf
Cataphyll
1 cm Axillary bud 3 cm
Figure 29. Morphology of yam (D. rotundata) early stage (from the planting date until
appearance of the first true leaf).
The influence of emergence characteristics over plant yield is typical to multivariable regression
modeling (Klmeke and Moll 1990, Rebetzke et al. 2007, Fayaud et al. 2014). By extending this
approach to an analogous multivariate regression model, in which all variables are simultaneously
considered as potentially mutually statistically dependent, substantially enhanced insight into the
plant yield formation system under study may be gained (Lewis and Ward 2013). We sought to
determine the globally optimal model for each yam species (i.e. a model with best goodness of fit
to the observed data). These models describe formally the interrelationships between the
explanatory variables and plant yield. The Bayesian network modeling approaches presented here
are similar to those used by Lewis and McCormick (2012) using an additive Bayesian network.
When the analytical task is to identify statistical dependencies with one or more response
variables, the additive Bayesian network structure discovery approach is well suited. Bayesian
network are graphical models comprising a set of conditionally independent generalized linear
models combined in such a way as to be probabilistically coherent (i.e. no cycles in the graph)
while maximizing fit to observed data. In these models, the goodness of fit (network score) and
model parameters must be estimated numerically rather than analytically, using Laplace
approximations at each node (Tierney and Kadane 1986).
Page | 79
All modeling was carried out in R, using the abn library (R Development Core Team
2013). Some dependence relationships between variables were banned from the structure
discovery analysis to maintain a logical timeframe (e.g. planting date cannot be dependent on
emergence date while the converse could be). Prior to model fitting, each variable was
standardized to a mean of zero and standard deviation of one. This transformation had no effect
on the identification of dependencies between variables.
A Bayesian network can be represented visually by a directed acyclic graph, a set of nodes
connected by directed links (arcs). The usual notation is to use squares for discrete nodes and
circles for continuous nodes. Each node denotes a random variable and arcs define a given
factorization of the joint probability distribution of all the random variables. For example, if X, Y
and Z are three random variables then a directed acyclic graph with serial connection between
nodes X, Y, Z (i.e. arc from X to Y, and from Y to Z) implies that P(X,Y,Z)=P(X)P(Y|X)P(Z|Y).
In structure discovery the objective is to identify that factorization which best represents the
study data, i.e. to determine an optimal directed acyclic graph structure. In data analyses, once the
joint probability structure is known then we have complete information, i.e. given this we can
then directly estimate any desired parameter or variable effect. In a graphical model all variables
in the same component (collection of connected arcs -ignoring direction) are jointly statistically
dependent. This means that knowing the value of one variable in this component can potentially
provide information about likely values of any other variable in this component. If a variable has
no arcs, either emanating from it or terminating at it, then it is statistically independent. Arcs in a
Bayesian network model only denote statistical dependency, unless otherwise stated. Causal
dependency can only be asserted using obvious real interpretation (like logical timeframe binding
variables) or expert knowledge.
In order to identify a globally optimal directed acyclic graph, an exact structure discovery
approach was used (Koivisto and Sood 2004). The main purpose of Bayesian network structure
discovery is to estimate the joint dependency structure of the random variables in the available
data. It is assumed that all structures are equally supported in the absence of any data (an
uninformative prior on structures) and so comparing Bayes factors collapses to comparing the
Page | 80
marginal likelihoods, which is done on a log scale. The log marginal likelihood (MacKay 1992),
typically referred to as the network score, was used to compare all models. Uninformative
parameter priors were used throughout, specifically Gaussian distributions with means of zero
and variance of 1000 for the marginal mean effects in each individual multivariable regression. By
using uninformative priors, the structure discovery process effectively commenced from a point
equivalent to no prior knowledge. Specifying and justifying informative parameter priors is
impractical for every combination of variables across every model under comparison (Firestone et
al. 2013).
Once the globally optimal model has been identified then the next task is to check this model for
over-fitting. In common with any model selection metric in multi-model selection, the marginal
likelihood may identify structural features, which, if the study was repeated many times, would
likely only be recovered in a tiny fraction of instances (Babyak 2007). To correct for such over-
fitting, a parametric bootstrapping approach can be used in Bayesian network modeling
(Friedman et al. 1999). This is conceptually straightforward, although computationally intensive,
as for each simulated (bootstrap) dataset we need to repeat the exact same model search as used
with the original data. We take our chosen model - that identified by applying the exact structure
discovery search to the study data - and then simulate datasets from this, the same size as the
original observed data, and see how often the different structural features are recovered. These
simulations were generated using the open source JAGS software (Plummer 2003) and the rjags
library in R. We further remove all dependencies (arcs in the directed acyclic graph) which have
insufficient statistical support to be considered robust, i.e. which were not recovered in at least a
majority (50%) of the bootstrap results (Poon et al. 2007).
The result of the analyses is a statistically robust additive Bayesian network (henceforth
called the final best directed acyclic graph) with the parameters in this model being exactly those
in a standard multivariable logistic regression, except we now have many more of these i.e. a set
for each variable (Firestone et al. 2013). The parameters have the usual interpretation as posterior
marginal mean effects for each covariate. The effects of the various variables in our study were
estimated, along with their posterior 95% confidence intervals.
Page | 81
Figure 30. Directed acyclic graph of globally optimal multivariate regression model of D. alata
early growth variables on plant yield and yield components using an exact search additive
Bayesian model. Red arcs indicate a negative relationship between two variables while green
arcs denote a positive relationship. Dashed arcs represent arcs which were not recovered in at
least 50% of the directed acyclic graphs based on the 10000 bootstrapping replicates.
Page | 82
4.3. Results
The complete dataset comprised 439 observations for D. alata and 388 for D. rotundata. Because
cataphyll occurrence in D. alata species was scarce (less than 2%), estimation of marginal
posterior densities for Cataphyll number and Cataphyll number per stem were not possible. Indeed
these estimations presuppose that the data contains sufficient information to accurately estimate
the joint probability distribution of the variables in the data. Consequently, these last two
variables were dropped from the directed acyclic graph structure discovery of D. alata. So the
globally optimal model (i.e. the model with the maximum goodness of fit over all possible
directed acyclic graphs structure) had a total of 10 nodes and 17 arcs for D. alata and 12 nodes
and 18 arcs for D. rotundata (Figure 30). Ten thousand bootstrap datasets were generated and
fitted using an identical exact model search. Pruning off all arcs from our two globally optimal
directed acyclic graphs which were not recovered in at least 50% of the directed acyclic graphs
based on the bootstrapping resulted in the removal of 3 arcs for D. alata (Figure 30, Figure 31).
Figure 31. Number of arcs recovered in 10000 bootstrapping replicates for D. alata (red) and
D. rotundata (blue).
Page | 83
The final best directed acyclic graphs for D. alata and D. rotundata comprised 14 and 18
arcs respectively, with log marginal likelihoods of −4259 and –3 631 (Figure 32, Figure 33 and
Figure 34). In these directed acyclic graphs, white nodes belong to the category of variables
linked to practices at planting, black nodes belong to variables of plant yield components and
grey nodes belong to variables linked to the plant’s early growth. Dashed arcs indicate a negative
relationship between two variables while solid arcs denote a positive relationship. Each arc is
labelled with the standardized marginal posterior median and the frequency at which each arc was
recovered in bootstrapping in brackets. The standardized median is the effect size indicator and
gives indications on the direction (positive or negative) and the strength of the dependency
between the two variables linked by the arc. The frequency at which each arc was recovered in
the bootstrapping gives us the level of directional support between two variables (the maximum
value possible being 10000: 100% support). All the selected variables used in both models were
justified using log marginal likelihood and bootstrapping. They showed at least one dependency
(Figure 33, Figure 34).
Figure 32. Comparison of goodness of fit for different parent limits for D. alata (red) and D.
rotundata (blue).
Page | 84
Figure 33. Directed acyclic graph of final best multivariate regression model of D. rotundata
early growth variables on plant yield and yield components using an exact search additive
Bayesian model. White nodes belong to the category of variables linked to practices at planting,
black nodes belong to variables of plant yield components and grey nodes belong to variables
linked to the plant’s early growth. Dashed arcs indicate a negative relationship between two
variables while solid arcs denote a positive relationship. Arcs are labelled with the standardized
median marginal posterior density and the frequency at which each arc was recovered in
bootstrapping in brackets .
Page | 85
Figure 34. Directed acyclic graph of final best multivariate regression model of D. alata early
growth variables on plant yield and yield components using an exact search additive Bayesian
model. White nodes belong to the category of variables linked to practices at planting, black
nodes belong to variables of plant yield components and grey nodes belong to variables linked
to the plant’s early growth. Dashed arcs indicate a negative relationship between two variables
while solid arcs denote a positive relationship. Arcs are labelled with the standardized median
marginal posterior density and the frequency at which each arc was recovered in bootstrapping
in brackets.
Page | 86
The statistical confidence attributable to each dependency is given by its 95% confident
interval (CI); that is, the intervals containing the true value of the effect of one variable upon
another with a probability of 95% (Figure 35). All of the effect parameters have 95% CI which
do not pass through the origin (Figure 35; Table 8). These would therefore typically be
considered as having a strong degree of statistical support.
Figure 35. Forest plot of the median effect (dot) of the distribution of the marginal posterior
for D. rotundata (blue) and D. alata (red). Quantile-based 95% probability intervals are given
using horizontal lines.
Page | 87
Table 8. Marginal posterior quantiles for each effect parameter (precision and intercept terms
not shown) in the final best model associating early growth variables with yield components of
D. alata and D. rotundata.
Results of the analysis of the dry matter and nutrient content confirmed that dry matter and
nutrient content were highly variable within the seed-tuber lot (Figure 36). Nutrient content of
D. rotundata seed-tuber was always higher but also more variable than that of D. alata.
Page | 88
Figure 36. Reserves content in seed-tubers of D. alata (red) and D. rotundata (blue). The
proportion (%) corresponds to the percentage of dry matter on a fresh weight basis and to the
percentage of nitrogen, phosphorus and potassium on a dry weight basis respectively. The
filled box corresponds to the inter-quartile range (IQR). The upper and lower whisker extends
from the box to the highest and lowest value that is within 1.5 * IQR. Data beyond the end of
the whiskers are outliers and plotted as points.
For both species, the chain of events leading to emergence showed the following logical
pattern: presprouted tubers had shorter latency which, with an early planting, allowed early
emergence. For D. rotundata the cataphyll number depended on the length of the latency phase
(Figure 33, Figure 34). For D. alata, the tuber state was dependent on the planting date: despite
large 95% probability intervals (Figure 35), late planting showed more presprouted seed-tubers.
Thereafter, the stem number depended upon emergence date for D. rotundata and upon planting
date for D. alata. The stem number and seed-tuber weight determined the seed-tuber weight per
stem of both species. While the seed-tuber weight per stem of D. alata did not influence any
other variable, for D. rotundata it determined the cataphyll number per stem, together with seed-
tuber weight, latency and cataphyll number.
Page | 89
4.3.3. Plant yield components
The multivariable models provide statistical evidence that for both species, the emergence date
and the tuber number were directly dependent on the tuber weight (Figure 33, Figure 34). The
more the tuber number and the later the emergence date, the lower the plant tuber weight. In
Figure 33, the multivariate model provides evidence that additionally the seed-tuber weight and
the cataphyll number per stem were also directly dependent on the tuber weight, and therefore it
was also obviously in the Markov blanket (i.e. the set of nodes composed of the observed node’s
ascendants, its descendants, and its descendants’ other ascendants) of the plant tuber weight of
D. rotundata. This then suggests that cataphyll number per stem and seed-tuber weight should be
included along with these other variables for further investigation into their potential biological
significance for controlling plant tuber weight. Similarly, D. alata also had an additional variable to
the common set of ascendants, i.e. seed-tuber state (Figure 34).
For D. rotundata, the tuber number did not show any ascendant, while for D. alata, it was
dependent on seed-tuber state directly and indirectly through latency. Final plant yield of both
species depended on tuber number and tuber weight. For D. rotundata the cataphyll number also
had a direct effect on plant yield.
Page | 90
4.4. Discussion
Given the intrinsic property of directionality in Bayesian networks (Friedman et al. 2003), the
directed acyclic graphs presented a general framework allowing for (i) some new insight into the
biological framework of emergence patterns leading to yield formation of yams, (ii) the
identification of opportunities for improvement and action priorities that will lead to improved
yam cropping practices in West Africa, and (iii) consideration of the agronomic importance of
yam seed-tuber quality.
Directed acyclic graphs offer a way of studying the influence of cultural practices and early plant
growth on plant yield formation. This kind of systemic analysis is scarce in plant development yet
it enables us to tackle complex systems with lack of expert knowledge, which is typically the case
with yams. Both directed acyclic graphs confirmed the importance of planting practices and early
growth on yield formation. Such a strong influence is in contrast with what is known about other
tuberous crops such as potatoes. In fact the size range of the seed-tuber is also much wider for
yams than for potato: between 200 and 1000 g depending on available planting material, while it
is more homogenous and much smaller (around 50g) for potato seed-tubers (Orkwor et al. 1998,
Van Ittersum 1992).
The emergence date is directly dependent on the planting date and latency. Given our
effect size indicator (i.e. standardized median posterior), these dependencies were strong and
positive. The later the planting date and the longer the latency, the later the emergence date. But
for D. alata the directed acyclic graph allowed us to understand and quantify a more complex
situation. Indeed, the planting date also had an indirect effect on emergence date through the
seed-tuber state. Later planting increased the number of presprouted tubers and thus decreased
the latency. Thereafter, this shorter latency resulted in an earlier emergence date. Thus, until
breaking of seed-tuber dormancy, the planting date had a single direct and positive effect on
emergence date. Afterwards, planting date may have opposing direct and indirect effects that may
lead to counter-intuitive results. For the descendent variables, the directed acyclic graph showed
us that emergence date is the only direct cause of plant tuber weight variability common to both
species. This is in accordance with previous studies (Cornet et al. 2014, Marcos et al. 2011). But
the directed acyclic graph highlights the direct effect of emergence rather than planting date. This
means that, for a given planting date, the unevenly emerging stand still influenced the plant tuber
Page | 91
weight. Given its influence on yield and yield variability in the field, this demonstrates a serious
weakness in the traditional cropping systems in West Africa, namely the uncontrolled and wide
range of physiological age in seed-tuber lots.
For D. rotundata, emergence date also exhibited some complex indirect effects on plant
tuber weight: the later the emergence, the larger the number of main stems, the less the seed-
tuber reserves per stem, the less the cataphyll number per stem and finally the less the plant tuber
weight. This last direct contribution of the cataphyll number to the plant tuber weight and yield is
a novel result that questions the emergence functional morphology of D. rotundata. As for cassava
(Manihot esculenta), emergence functional morphology is the result of selection and can be better
understood by looking at wild relatives (Pujol et al. 2005). The D. rotundata's wild relatives (D.
prahensilis) originated from forested zones of West and Central Africa where, once initiated, the
shoot must grow through a poorly illuminated understory (Di Giusto et al. 2001). Thus, cataphylls
provide a source of buds without placing a high respiration load on the seedling which would
result if a lamina developed (Wright et al. 2000). The number of nodes carrying cataphylls is thus
dependent on the seed-tuber reserve, i.e. the seed-tuber size but also seed-tuber content
(carbohydrates and mineral nutrients). This seed-tuber content varied within a seed-tuber lot,
even between equally sized seed-tubers (Figure 36). In the model, the cataphyll number could
thus represent the nutritional status of the plantlet related to the seed-tuber reserve unexplained
by seed-tuber weight. As the plants develop the cataphylls could be an indicator of the quantity
and the time of seed-tuber reserve utilization by the plant. Applications of this relationship are
not only of interest in functional morphology but could also find useful applications as an
indicator for phenotyping early growth vigor or as an indicator of the end point of early plant
growth (Hanley et al. 2004).
Seed-tuber weight influenced plant tuber weight for D. rotundata but not for D. alata, which does
not agree with the results reported by Njoku et al. (1984). Ferguson (1973) showed an asymptotic
response of D. alata to seed-tuber size from 80 up to 250 g while Kayode (1984) showed that
seed-tubers larger than 400 g could be used for maximum D. rotundata tuber yield. In fact D. alata
is native from south-east Asia and was introduced later into West Africa (Doumbia 2004). It
seems that traditional practices transferred from D. rotundata may have led to oversized D. alata
seed-tubers. For instance, in the Caribbean, D. alata is planted using much smaller seed-tubers
Page | 92
(i.e. 100g; Marcos et al. 2011). In order to avoid wastage, it could be necessary to look for the
optimum seed-tuber size for D. alata in West Africa.
As to the influence of the seed-tuber state on plant yield component, it seems that
delayed planting (i.e. after seed-tuber sprouting) might drastically decrease marketable yield.
Indeed, presprouted seed-tubers of D. alata caused an increase in tuber number and a decrease in
tuber weight. Yam prices on West African markets are based on tuber size, bigger tubers
attracting a higher price (Orkwor et al. 1998). On the other hand, the farmers’ seed-tuber stock
could be profitably managed on the basis of seed-tuber state. Farmers could use unsprouted
seed-tubers for ware yam production. Keeping the presprouted ones for seed-tuber production
could enable farmers to produce more and smaller tubers for next year’s planting.
One novelty associated with this study lies in the use of a Bayesian network to represent complex
systems to non-experts in a way that facilitates automated analysis. The importance of a better
control of the seed tuber quality comes out of this discussion. This is obvious both for the
explanatory variables like seed-tuber weight and the uncontrolled latent variables such as
physiological age and seed-tuber content. The need to develop seed-yam systems in West Africa
is currently well understood and tackled by donors and international development agencies
(WECARD 2011, DFID 2014, IITA 2014). Although most projects focus on clean (healthy) seed
yam production, directed acyclic graph's representation of the consequences of using
uncontrolled planting material clearly supports clean yam seed-tubers but also good quality ones
(i.e. calibrated, uniform in terms of physiological age and nutrient content).
Cornet et al. (2014) claimed that until we have a more effective production system of
quality seed-tubers, cohorts could be a suitable experimental unit for analyzing processes in yam
field populations. A cohort can be defined as a group of individuals at the same phenological
stage (Deaton and Winebrake 2000). They advised building cohorts based on emergence date.
Based on our results we advise refining the yam cohort concept by grouping plants based on their
belonging to similar categories of emergence date and cataphyll number.
Page | 93
4.5. Conclusions
One novelty associated with this study lies in the use of a Bayesian network to represent complex
systems to non-experts in a way that facilitates automated analysis. Directed acyclic graphs offer a
way of studying the dependency between cultural practices, early plant growth variables and plant
yield components. This kind of systemic analysis is scarce in plant science. All the selected
variables used in both models were statistically justified and could be considered as having a
strong degree of statistical support. The directed acyclic graphs present a general framework
allowing for both, new insight into the yams biological framework, and the identification of
opportunities for cropping system improvements.
For D. rotundata, the direct contribution of the cataphyll number to the plant tuber weight
and yield is a novel result that questions the emergence functional morphology. Cataphyll
number could be an indicator of the quantity of seed-tuber reserve and the timing of its
utilization by the plant. Applications of this relationship are not only of interest in functional
morphology but could also find useful applications in other discipline (e.g. phenotyping).
For D. alata the model did not show any effect of seed-tuber size. It is suggested that
traditional practices transposed from native D. rotundata may have led to oversize seed-tuber of
D. alata and that reducing seed-tuber size may lead to less wastage. We demonstrate the influence
of seed-tuber state of D. alata on yield components, directly and indirectly through latency. This
relationship explains why delayed planting (i.e. after seed-tuber sprouting) might drastically
decrease marketable yield and also provides some reflective backing to improve the farmers’
seed-tuber stock management.
Both directed acyclic graphs confirmed the influence of planting practices and early
growth on yam (Dioscorea spp.) yield formation in West Africa. Given the influence of early
growth on plant yield variability, this study demonstrates a serious weakness in the traditional
cropping systems (i.e. the uncontrolled and wide range of physiological age and reserves content
in seed-tuber lots), and highlights the importance of a better control of the seed tuber quality.
Acknowledgments
Page | 94
Page | 95
Chapitre 5 Mise au point de modèles
allométriques permettant l’estimation
de la croissance végétative des ignames
La science est un jeu,
dont la règle du jeu consiste à trouver quelle est la règle du jeu.
François Cavanna
Nous proposons dans ce chapitre à la fois (i) un modèle linéaire spécifique à chaque
espèce, permettant l’estimation de la surface foliaire à partir de la biomasse sèche en
feuilles, et (ii) un modèle allométrique commun aux deux espèces d’igname permettant
l’estimation non destructive de la surface foliaire, à l’échelle de la feuille, à partir de
mesures linéaires de son limbe.
_________________
a
INRA, UR1321 ASTRO, Agrosystèmes Tropicaux.
Page | 96
Assessing allometric models to predict
vegetative growth of yams in different
environments
Abstract
Yams are neglected crops, mostly grown in West Africa by resource-poor farmers. Little is
known about the physiology of the crop and researchers lack practical and efficient tools to
conduct growth analysis. The objective of this study was to develop allometric models able to
predict yam leaf area and leaf and stem dry mass with acceptable accuracy. The models were
calibrated using a dataset comprising 10 cultivars belonging to the two main species (Dioscorea
alata and D. rotundata) grown at two locations in Guadeloupe (Lesser Antilles) and two locations
in Benin (West Africa). The best models were selected based on Akaike Information Criteria and
validated against independent data sets. A power regression was best for predicting leaf area from
leaf measurements while linear relationships were sufficient to predict the relationship between
crop leaf area and leaf and stem mass. The use of species-specific models for the estimation of
leaf and stem mass significantly improved the models’ performance. Models predicting yam leaf
area and leaf mass proved to be reliable and accurate (no significant deviation and adjusted
R² > 0.95). For stem mass, overestimation always occurred during validation (+9%). To
overcome this discrepancy a methodology was proposed that allows the user to calibrate the
model by tailoring the sampling size in order to obtain the required precision. The use of the
selected models provides a nondestructive and reliable alternative to estimate leaf area and leaf
and stem biomass for different cultivars and sites.
Keywords: allometry; yam; Dioscorea spp.; non-destructive leaf area estimation; dry mass; model
selection
Page | 97
5.1. Introduction
Yams (Dioscorea spp.) are grown throughout the tropical parts of the world and are important
staple crops in at least ten countries (ranging from Nigeria to Jamaica and the Solomon Islands)
accounting for 155 million consumers (FAOSTAT 2014). Despite their economic value, research
on yams is still scarce (Varshney et al. 2012) and there is little literature available on the crop’s
physiology (Onwueme and Haverkort 1991, Marcos et al. 2011). In order to address this lack of
knowledge, there is a clear need for practical methods allowing basic growth analysis. An essential
aspect of crop growth analysis lies in the accurate and nondestructive estimation of key variables
such as leaf area or crop biomass (Normand and Lauri 2012). To perform such estimations,
modern equipment (e.g. hand scanners or laser optic apparatus) may be expensive and too
complex for basic and simple studies. Moreover, 85% of world yam production is in rural areas
of West Africa, where modern facilities and instruments may be unavailable, yet simple practical
tools and methods may suffice for researchers. The use of predictive allometric models allows
this difficulty to be circumvented by quantifying statistical relationships between key variables
and some easily measured predictors (Niklas 2004).
Estimation of leaf area is a key part of classical growth analysis and is necessary in many
plant physiological studies (Sestak et al. 1971). Models for the nondestructive estimation of leaf
area are useful tools for researchers in agronomic experiments. The linear measurement method
allows the leaf area to be estimated without disturbing the growth of the plant (Sestak et al. 1971).
This simple, non-destructive and accurate method is based on linear regressions of leaf area on
various leaf measurements. For heart-shaped leaves, the most frequently used measurements are
leaf maximum length, leaf maximum width, midrib length, or their combination (Sinoquet and
Andrieu 1993). Although simple methods exist for estimating leaf area of some common crops
(Kandiannan et al. 2002, Lu et al. 2004, Pompelli et al. 2012), only fragmentary studies are reported
for yam (Gumbs and Fergusson 1976, Ravi and Chowdhury 1989, George 1992). Unfortunately,
these published reports on yams fail to show that any one specific approach for deriving leaf area
is adequate across cultivars and environmental conditions. Moreover, sometimes the adequacy of
the model’s assumptions for estimating leaf area has not been carefully examined (e.g. there may
be no validation step).
Predictive models for estimating key variables at the crop level have been developed for
agronomic or ecological purposes (Muller et al. 2000, Shipley and Meziane 2002, McCarthy and
Enquist 2007). Some relate the leaf area to the leaf biomass through the specific leaf area while
others investigate allometric biomass partitioning. These classical allometric models differ from
Page | 98
the models at the level of the leaf because they integrate all the growth processes of the plant
(Normand and Lauri 2012) and are thus often less accurate.
The objective of this study was to develop predictive models to estimate leaf area and leaf
and stem mass of 10 cultivars belonging to the two main yam species (D. alata and D. rotundata).
This study focuses on two levels: the individual leaf and the whole crop. At the level of the
individual leaf this study evaluates the currently-used models and proposes a reliable and accurate
model using non-destructive linear measurements for the estimation of the leaf area of yams. At
the crop level this study links the stem and leaf mass to the leaf area of the crop.
Page | 99
5.2. Materials and Methods
Ten data sets from field experiments with a wide range of cultivars and cultural practices (Table
9) were selected. The experiments were carried out in Guadeloupe (Lesser Antilles) at the INRA
Experimental Stations of Duclos (16°12’ N, 61°39’ W, 250m asl) and Godet (16°24’N, 61°29’W,
30 m asl) and in Benin (West Africa) at the IITA Experimental Stations of Cotonou (6°25’ N,
2°19’ E, 23 m asl) and Glazoué (7°56’ N, 2°15’ E, 200 m asl). Each field experiment consisted of
300–550 plants.
At the crop level, plants were sampled from an area of 1.5–3 m2 every two to three weeks
from vine emergence to the beginning of tuber growth (approximately 110 days) in order to
measure leaf and stem biomass and crop leaf area index. There were four replicates of 1 to 4
plants at each sampling date, depending on plant density. Crop leaf area was measured with a
planimeter (Li-Cor model 3100, LI-COR Inc., Lincoln, NE) and expressed as the area of leaves
(m2) per m2 of ground. Leaf biomass (LM in g m-²) and stem biomass (SM in g m-²) were
determined after drying the material at 70 °C. In order to study the allometric relationships
between crop leaf area and leaf or stem dry mass, field experiments were divided into two groups:
experiments 1 to 6 (520 samples) were used to select models and estimate their parameters, while
experiments 7 to 10 were used to test the chosen models (259 samples). All cultivars belonged to
the two main species (i.e. Florido, Belep, Oriental, Kinabayo, Boutou and Plimbite for D. alata
and Kpouna, Morokorou, Gnidou and Kokoro for D. rotundata).
At the individual leaf level, samples were collected in 2001, 2004 and 2007 (Table 9) to
study the relationship between linear measurements of leaf blade and leaf area. All samples
consisted of at least 40 leaves collected randomly from different positions on the vine. All
cultivars had heart-shaped leaves on which maximum length (L in cm), maximum width (W in
cm) and midrib length (M in cm) were measured. Afterwards, individual leaf area (A in cm²) was
measured with a planimeter. Samples from 2001 and 2007 were used for model calibration and
investigation on the effects of site and cultivar (456 samples) while samples from 2004 were used
for model validation (354 samples).
Page | 100
Table 9. Experiment characteristics of each data set used in this study.
Page | 101
5.2.2. Statistical analysis
Statistical analysis was performed in four steps. First, all the models were evaluated and the best
equations for each of the three allometric relationships selected. Second, the effects of the
covariates (cultivar, species, site and year) on allometric models were tested and the selected
models were calibrated. Third, the calibrated models were validated against an independent data
set and compared to other published models. Finally, if a significant deviation in the model
estimates was found during the validation step, the minimum sample size for the
parameterization of such a model was then evaluated. All statistical analyses were performed
using R software (R Development Core Team 2013).
For each allometric relationship, several linear and nonlinear regression models using
combinations of predictors and response variables were run. These were either chosen from the
literature or derived for the present study based on their logical mathematical relation to the
published models. Preliminary regression analysis showed that most of the relationships between
crop leaf area and leaf and stem dry biomass followed linear patterns, while the relationship
between linear measurements and leaf area was better modeled with power functions typical of
allometric models. Hence, in this study, only linear models are presented for the estimation of
leaf and stem mass while only power models are presented for leaf area. The log transformation
was used to make the assumption of homogeneous variances reasonable in the power models
(e.g. 𝑦 = 𝑎. 𝑥 𝑏 ):
ln y b ln x ln a ,
where y is the response variable, x the predictor and a and b the model parameters.
Models were selected using the Akaike Information Criterion (AIC). The AIC is a
common model selection criteria that deals with both the goodness of fit and the complexity of
the model (i.e. number of parameters) and enables several models to be compared simultaneously
(Johnson and Omland 2004). The lower the value of the AIC, the better the model.
Page | 102
Because the AIC tells nothing about the quality of the model in an absolute sense, the
̅̅̅̅) were also
adjusted coefficient of determination (R²) and the mean percent standard error (S%
calculated in order to illustrate the mean size of the error as a percentage of the estimated values:
100 n yˆ i yi
S%
n
i1 yˆ ,
i
where yi is the actual value, ŷi is the estimated value and n is the number of observations
(Schaeffer 1980). As the model estimates were computed on the logarithmic scale, they must then
be back-transformed to their original, arithmetic scale. Because this transformation is non-linear,
and there is variability in the observed data around the fitted relationship, a simple ‘naive’
exponential-based transformation will generate bias (Clifford et al. 2013). Consequently,
correction factors are typically calculated to remove this bias when back-transforming. As
proposed by Clifford et al. (2013) we calculated the uniform minimum variance unbiased
estimations described by Bradu and Mundlak (1970). The R codes required to fit a log–log
allometric model and to make predictions using the uniform minimum variance unbiased
estimations were adapted from Clifford et al. (2013).
Experimental factors (i.e. cultivars, sites, and year) can affect the slope and/or the intercept of
the relationships and, thus, were studied as covariates. In order to test the influence of these
covariates on the allometric relationships, a maximal model was fitted (i.e. a model including
explanatory variables, covariates and first order interactions). Furthermore the maximal model
was simplified according to Crawley’s (2007) procedure. The model summaries were inspected
and the least significant term was removed. The full and the simplified models were then
compared using the F-test. If no significant difference was observed, the simplified version was
kept and the least significant highest-order term was again deleted. If the F-test revealed a
significant difference (P<0.05) the previous model was kept. Finally a 1000 bootstrap resampling
in conjunction with automated variable selection methods was applied in order to check the
efficiency of the selected models (Austin and Tu 2004). Variables that were identified as
independent predictors of the response variable in at least 60% of the bootstrap samples were
kept and this last model was called Minimal Adequate Model. The relative importance of each
term of the Minimal Adequate Model was estimated using the Lindeman, Merenda and Gold
Page | 103
metric (Lindeman et al., 1980) allowing an estimation of the percentage of variance explained by
each term. The confidence intervals for relative importance were estimated using a bootstrap
procedure on 1000 samples (Groemping 2006). Differences in model performance (i.e. ΔR²,
ΔAIC and Δ S % ) were calculated between the Minimal Adequate Model and the classical pooled
regression.
To validate the calibrated models, leaf area, leaf mass and stem mass were estimated using the
selected allometric models on independent data. The estimated values were compared to
observed ones using S % , R² and the bias ( b% ). The bias is defined as the scaled mean error (i.e.
the mean difference between estimated and observed values expressed as a percentage of the
mean observed value). The bias indicates whether the model, on average, overestimates ( b% > 0)
yˆ i yi
b% 100i 1
n
.
yi
Paired t-tests were run to determine whether the differences between the estimated and
the observed values were statistically significant at P<0.05. When this difference was significant,
we estimated the minimum sample size allowing user-recalibration with minimum bias. In this
case, parameter values and bias were calculated on randomly-selected subsamples of increasing
size. All sample sizes were submitted to a bootstrapping procedure with 100 re-samples with
replacement. The mean bias was then expressed as a function of the sample size.
Page | 104
5.3. Results and Discussion
The domain of validity of the allometric models is defined by the experimental range for crop
leaf area, leaf and stem mass and individual leaf area used for model calibration. A summary of
the data sets is given in Table 10.
Table 11 summarizes the performance of candidate models during the calibration phase.
Models were fitted to the calibration data set and ranked according to their performance.
Individual leaf area was best explained by a power function of the square of maximum leaf width
multiplied by the maximum leaf length ( S % of 5.4% of mean leaf area and R² of 0.987). Leaf and
stem mass were better explained with a linear function of crop leaf area. The best allometric
models for leaf and stem mass were not as good (i.e. S % of 10.5 and 15.3, and R² of 0.977 and
0.935 respectively) as for leaf area but similar to those reported in previous studies (Normand
and Lauri 2012, Pompelli et al. 2012, Kandiannan et al. 2002). Higher S % and lower R² for leaf
mass estimation may be due to the large variation in specific leaf area within a single stand, plant
(Cooper and Qualls, 1967; Reddy et al., 1989) or even leaf (Sestak et al. 1971). Moreover the
specific leaf area could have been influenced by the growth stage of the plant (Koller 1972) and
by the time of the day (Chatterton et al. 1972) when samples were taken. Cornet et al. (2014)
showed that yam emergence may occur over more than one month creating a very heterogeneous
population of individuals of different ages with different specific leaf area. This variability in
specific leaf area may explain some of the scatter of leaf mass estimates with the evaluated model.
Page | 105
Table 10. Summary (minimal, maximal and mean ± SD values) of the calibration and validation datasets used to build predictive allometric models
to assess vegetative growth of the two main species of yam.
Dataset Species Site N Min Max Mean ± SD Min Max Mean ± SD Min Max Mean ± SD
Leaf and stem mass estimation -
Crop Leaf area (m² m ²) -2
Leaf dry weight (g m ) Stem dry weight (g m-2)
Calibration D. alata Cotonou 160 0.03 3.70 0.68 ± 0.75 1.44 234.00 40.50 ± 43.29 1.78 198.00 30.54 ± 34.51
Glazoué 70 0.01 3.33 0.76 ± 0.59 0.40 188.00 45.02 ± 33.46 3.00 141.00 31.24 ± 25.14
D. rotundata Cotonou 200 0.02 3.00 0.92 ± 0.76 1.00 160.00 45.00 ± 36.66 2.03 122.00 40.31 ± 27.28
Glazoué 90 0.03 1.69 0.51 ± 0.45 1.70 88.00 25.91 ± 22.78 1.70 74.00 22.10 ± 17.56
Validation D. alata Duclos 100 0.10 3.20 0.68 ± 0.71 5.21 117.19 28.76 ± 25.95 4.17 77.29 19.92 ± 16.76
Godet 60 0.10 2.00 0.79 ± 0.52 6.67 91.67 41.44 ± 24.31 5.00 66.83 31.98 ± 16.18
Cotonou 25 0.04 1.48 0.40 ± 0.34 2.33 108.21 26.51 ± 24.69 3.67 62.44 19.56 ± 14.07
Glazoué 25 0.06 1.18 0.36 ± 0.27 4.00 75.00 22.98 ± 17.03 1.00 49.00 14.27 ± 10.35
D. rotundata Cotonou 25 0.17 2.46 0.80 ± 0.59 8.66 121.59 39.33 ± 28.74 13.33 92.42 34.08 ± 18.06
Glazoué 24 0.11 0.91 0.36 ± 0.20 6.00 48.00 18.76 ± 10.59 4.00 32.00 13.35 ± 6.05
Leaf area estimation Leaf maximum length (cm) Leaf maximum width (cm) Leaf area (cm2)
Calibration D. alata Duclos 226 6.6 24.4 14.1 ± 3.6 3.6 13.2 7.53 ± 1.97 15.0 178.00 66.74 ± 32.79
Godet 230 5.2 18.6 11.9 ± 3.2 3.6 11.0 6.60 ± 1.61 7.0 164.00 45.82 ± 24.29
Validation D. alata Duclos 244 5.9 27.2 12.6 ± 4.1 2.7 16.2 6.64 ± 2.77 8.4 265.78 56.48 ± 46.48
Glazoué 55 7.6 17.5 11.4 ± 2.9 3.7 11.5 6.50 ± 2.20 13.9 116.79 47.43 ± 30.90
D. rotundata Glazoué 55 5.4 13.3 9.3 ± 1.8 3.7 11.3 6.50 ± 1.78 13.8 82.10 40.32 ± 17.49
Page | 106
Table 11. Candidate allometric models ranked by performance metrics: Akaike Information
Criteria (AIC), mean percent standard error ( S % ), adjusted coefficient of determination (R²)
and, the model bias ( b% ); L is the maximum length, M is the midrib length, W is the maximum
width, a and b are model parameters.
Model AIC S% R² b%
Individual leaf area (A) estimation
A = a (W² L)b -663 5.4 0.987 -0.08
A = a (W² M)b -663 5.6 0.987 -0.09
A = a (W L)b -615 5.4 0.985 0.15
A = a (W M)b -612 6.0 0.984 0.15
A = a (W L M)b -407 8.2 0.968 0.29
A = a Wb -311 11.2 0.955 -0.63
A = a Lb -39 15.1 0.884 0.04
A = a (L M)b -14 16.0 0.873 0.03
A = a Mb 44 18.0 0.845 -0.09
Leaf mass (LM) estimation from lea area (LA)
LM = b LA 1972 10.5 0.977 -1.26
LM = a LAb 1975 10.6 0.963 1.45
Stem mass (SM) estimation from lea area (LA)
SM = a + b LA 1956 15.3 0.935 0.00
SM = a LAb 1978 16.4 0.875 -1.73
The leaf area to stem mass allometric model showed the highest S % and lowest R². This
was also observed by other authors and may be due to changes in both specific leaf area and
leaf:stem ratio (Normand and Lauri 2012). In fact herbaceous self-supporting plants show an
ontogenetic trend towards decreasing leaf:stem allocation ratio of assimilates, with increasingly
larger investment in the support tissues needed by larger plants (Poorter et al. 2012). Because all
yam species grown for food have a climbing growth habit they are likely to behave differently.
Unfortunately literature on the functional anatomy of climbing species is scarce. Some studies
indicate that vines and lianas have a higher leaf:stem ratio, but also higher plasticity (i.e. greater
variation) in their anatomical characters in response to changing environmental conditions than
their self-supporting relatives (Gianoli 2002, den Dubbelden and Verburg 1996, Werth et al.
2013). Thus, the increasing competition between plants with age can modify leaf:stem ratio and
therefore increase the variability of related allometric relationships.
Page | 107
5.3.2. Year, site and cultivar effects on allometric models
Once the best model for each allometric relationship was selected, the relative importance of
cultivar and environmental factors (year and site) was investigated. A summary of all terms of the
minimally adequate model and their relative importance is presented in Table 12. Explanatory
variables (i.e. linear measurements, and crop leaf area for the estimation of leaf area or leaf and
stem mass respectively) always accounted for more than 83% of the variance explained by the
model. Among covariates, cultivar was the most influential and accounted for 17%, 4% and 3%
of explained variance for leaf area, leaf mass and stem mass respectively while year accounted for
0%, 4% and 6%, respectively (Table 12). The relative importance of sites and interaction terms
was always less than 1.5%. These results suggest that the effects of year, site and their interaction
had negligible impact on the performance of the models.
Table 12. Summary of the relative influence of the experimental factors on the allometric
models. Bootstrap confidence intervals for relative importance were based on 1000 samples.
Differences in model performance were calculated between the minimal adequate model and
the equivalent pooled regression (i.e. ΔR²: difference in adjusted coefficient of determination,
ΔAIC: difference in Akaike Information Criteria, Δ S % : difference in mean percent standard
error and, Δ b% : difference in the model bias). L is the maximum length, M is the midrib length
and W is the maximum width.
Page | 108
Despite the quite large fraction of the variance explained by cultivars for the estimation of
leaf area (17%), their inclusion only slightly improved model performance compared to the
pooled regression ( S % only decreased by 0.3). In view of the lack of bias when using a common
model for leaf area estimation, it seemed a reasonable approximation to apply a common
equation derived from the pooled data. Conversely, the use of cultivar-specific models decreased
S % by 3.1 and 4.1 for leaf and stem mass estimates respectively. However this study included 10
cultivars, whereas Dansi et al. (1999) reported that there were 560 cultivars just within Benin. So
the use of cultivar-specific models limits the scope of their application. In order to include
genotypic differences while keeping the method sufficiently generic we defined species-specific
models. Indeed, the studied species cover 90% of yam production worldwide (FAOSTAT 2014).
Hence, leaf and stem mass species-specific models were fitted and compared to the pooled
regression. Because they performed significantly better they were kept for the validation step.
Final model parameters and their confidence interval are summarized in Table 13. Leaf
area estimation was best described by a power model calibrated on pooled data. Conversely, leaf
and stem mass estimations were best fitted with species-specific linear models. Only in the
estimation of stem mass of D. rotundata was the intercept significantly different from zero.
Differences between species may be due to the heteroblastic growth of D. rotundata. Indeed
leaves at the first few nodes of this species may be reduced to form modified shield-shaped
cataphylls (Onwueme 1978). These cataphylls are thick, lack a distinct leaf lamina and are limited
in aerial spread. Cornet (unpublished results) observed that a D. rotundata plant can bear up to 16
nodes carrying cataphylls before appearance of the first true leaf, while cataphyll occurrence in D.
alata was scarce (less than 2%). Thus, plants bearing cataphylls produce significant stem mass
before developing any leaf area and thus could modify the intercept of the leaf area to stem mass
relationship.
Final models were validated against a completely independent data set (Table 10). The S % of
the model was distinctly higher for validation than for calibration (Table 13). However, the
adjusted coefficient of determination was always above 0.93 and except for stem mass ( b%
=9.3%) the bias was negligible (-0.8 and 1.2 respectively for leaf mass and leaf area estimation).
Comparing the leaf area model obtained in this study with available models from the literature
Page | 109
confirmed its greater accuracy (Figure 37). Indeed the model developed in the present study
always outperformed models from the literature. A paired Student’s t-test did not reveal any
differences between the estimated and observed leaf areas using our model but significant
differences were observed when using the published models.
Table 13. Summary of final model parameter values (mean and confiden ce interval, CI) and
performance at calibration and validation stage. Performance statistics: the adjusted coefficient
of determination (R²), mean percent standard error ( S % ) and, the model bias ( b% ).
Figure 37. Statistical analysis (Student’s t-test for paired data) of the deviation between the
estimated and observed leaf area, leaf mass and stem mass. Estimates were based on the final
best models (pooled or species-specific) of this study and models from literature where
available. Vertical bars denote the bias and spreads denote 95% confidence intervals. ns = non -
significant, * significant at P < 0.05, ** significant at P < 0.01 and *** significant at P < 0.001.
Page | 110
Models to estimate yam leaf and stem mass could not be found in the available literature.
No differences between estimated and observed leaf mass were observed using our model.
Moreover, the use of species-specific models decreased model bias for both leaf and stem mass
estimation (Figure 37). However, estimation of stem mass was very biased. The diagnostic plot
shows clear differences between the stem mass model and the leaf area model (Figure 38).
Estimated and observed values were closely correlated for both leaf area and stem mass models
(0.99 and 0.95 respectively) but the slope of the regression line was significantly different from 1
for stem mass but not for leaf area. This confirmed the formerly observed bias in stem mass
estimation and showed that this bias was greater for D. rotundata than D. alata (slopes of 1.09
against 1.05 respectively). Residuals of the stem mass model, particularly those for D. rotundata,
showed a biased and heteroscedastic pattern with a clear overestimation for stem mass above
30g. The potential number of cataphylls per stem depends on seed-tuber physiological age and
weight (Cornet, unpublished results). Because of differences in agricultural practices, the mean
physiological age and weight of the seed-tuber varied between trials. This could have led to the
presence of varying numbers of cataphylls per stem, which may explain the large differences in
leaf:stem ratio between trials. In order to overcome this lack of reliability we analyzed the
relationship between number of samples and the performance of the stem mass model. The
variation in the estimate of the model parameter b tended to decline as the sampling size
increased (Figure 39A). The inferred model bias decreased sharply with increased number of
samples and fell below 2% with 86 samples (Figure 39B). Depending on the user’s requirements
from the model, they may choose a suitable sample size to obtain the necessary precision.
Page | 111
Figure 38. Diagnostic plots of allometric models to predict leaf area (A) and stem mass (B) for
the two main yam species (D. alata in red and D. rotundata in blue). (A) Main plot presents the
regression of estimated vs. actual values (black line and equation) of the pooled allometric
model for leaf area estimation from linear measurement, while the inset presents the residuals
of the model compared to the data estimated by the model. (B) Main plot presents the
regression of estimated vs. actual values (lines and equations) of the species -specific allometric
model for stem mass estimation from crop leaf area, while the inset presents the residuals of
the model compared to the data estimated by the model.
Page | 112
The proposed models were tested and calibrated for a given range of validity (Table 12).
The leaf area estimation from linear measurements was carried out on heart-shaped leaves and
should be applied to species presenting different leaf shapes (e.g. D. dumetorum or D. trifida).
Moreover, while the leaf area estimation from linear measurements is independent of the growth
stage and could be applied throughout the growing season, the estimation of leaf and stem mass
from crop leaf area may depend on the growth stage and thus cannot be estimated with the same
expected accuracy after tuber initiation. Moreover senescence may change these relationships
drastically. Finally building models can take time and it is important to continue to test and
improve them as new data become available.
Figure 39. The main plot (A) presents the bootstrapped (n=100) estimate of the parameter b
of the stem mass model (SM = a LAb) as a function of the sample size. The dashed grey lines
of the main plot represent the 99% confidence intervals of the mean of the conversion factor
for each sample size. The inset (B), presents the model bias as a function of the sample size.
Page | 113
5.4. Conclusion
In this study, very close relationships were found between estimated and observed values of leaf
area at the leaf level and for leaf and stem mass at the crop level. Simple model were selected and
calibrated. A power model applicable to both yam species was found to best predict a leaf’s area
in cm² (A) from its linear dimensions: A=0.74(W² L)1.94, where W is the maximum leaf width and
L is the maximum leaf length both in cm. Conversely linear relationships were found between
crop leaf area (LA in m² of leaves m-² of ground) and leaf (LM in g m-2) and stem mass (SM in g
m-2). The use of species-specific models for leaf and stem mass estimation significantly improved
model performance: LMD. rotundata = 48.81 LA, LMD. alata = 57.52 LA, SMD. rotundata = 7.83+34.40 LA
and SMD. alata = 42.08 LA, where LA is the crop leaf area.
Models were all validated against a completely independent data set. Models predicting
the leaf area and leaf mass of yams (D. alata and D. rotundata) proved to be reliable and accurate.
Unfortunately such a relationship for stem mass was found to be less reliable and a systematic
overestimation of predicted values was observed during validation. To overcome the significant
differences in the leaf:stem ratio between years, experiments and cultivars a method that allows
the user to calibrate the model, tailoring their sample size to the desired precision is proposed. All
proposed models represent practical, cheap and reliable alternatives for an accurate and
nondestructive estimation of key variables of yam vegetative growth.
Acknowledgments
Page | 114
Page | 115
Chapitre 6 Modélisation des conséquences
économiques et agronomiques de la
variabilité interindividuelle à l’échelle
de la parcelle.
La théorie c'est quand on sait tout et que rien ne fonctionne.
La pratique c'est quand tout fonctionne mais personne ne sait pourquoi.
Ici nous avons réuni les deux, théorie et pratique,
rien ne fonctionne... et personne ne sait pourquoi.
Albert Einstein
Page | 116
Modelling physiological, agronomic
and economic consequences of yam
heterogeneous planting material in
West Africa
Abstract
Based on six experiments carried out in Benin at two locations in 2007, 2008 and 2009, a model
for individual plant growth and development is developed. It allows an accurate estimation of
yam (D. alata and D. rotundata) growth stage durations and tuber yield. This model simulates the
combined effect of emergence (through photoperiod and temperature) and seed-tuber weight on
yam plant growth and development. Its estimates were well correlated with observed leaf area
index and plant tuber yield (R²>0.83).
Keywords: yams (Dioscorea spp.), plant yield variability, photoperiod, model, West Africa
Page | 117
6.1. Introduction
Yam (Dioscorea spp.) is the third most important tropical root crop after cassava and sweet potato.
Most of its world production is in West Africa and is based on traditional cropping which
involves low density planting and extensive practices. This cropping system is faced with many
constraints, including a lack of quality planting material. Traditionally, farmers either plant whole
small tubers which arise during the process of ware-yam production, or cut large tubers into
smaller pieces of the desired size. In this way, up to 33% of yams otherwise available for food are
reserved for planting new crops (Orkwor et al. 1998). This process also results in highly variable
planting material, whether in size, physiological age or nutrient content (Cornet et al. submitted).
Accordingly, uncontrolled seed-tuber physiological age and size may influences emergence time
(Orkwor et al. 1998). Moreover, several studies showed that seed-tuber size strongly influence
aerial growth during the vegetative phase and consequently tuber yield (Ferguson 1973, Okoli et
al. 1999, Cornet et al. submitted). Cornet et al. (2014) showed that even equally-sized traditional
planting material may emerge unevenly, leading to individual plant size variation and high
interplant yield variability. Despite strong evidence, the cost of heterogeneous seed-tubers, or to
express it in a more constructive way, the expected benefit from improved planting material is
still unknown.
Page | 118
The present study aims at investigating the consequences of uneven seed-tuber weight
and emergence date on crop yield, using a simple and descriptive model of individual yam growth
and development. First the influence of environmental variables (temperature, absolute daylength
and rate of change of daylength) on model performance was estimated for the two main yam
species (D. alata and D. rotundata). The plant tuber yield was simulated for different emergence
dates and seed-tuber sizes to quantify their effect on the yield of the two main yam species.
Finally, the profitability of traditional and improved planting material was estimated.
Page | 119
6.2. Materials and Methods
The dataset used here came from six trials carried out between 2007 and 2009 at two locations in
Benin (West Africa): AfricaRice-Cotonou Station (6°25’ N, 2°19’ E, 23 m asl) and Glazoué (7°56’
N, 2°15’ E, 200 m asl). Both sites are located in the forest-savannah transition zone. The climate
is sub-equatorial with a bimodal pattern of rainfall, which occurs mainly from March to July and
from September to October. The soils used in this study are arenosols (FAO-ISRIC-ISSS 1998)
with good physical characteristics but low nutrient levels. Field experiments consisted of 300–550
plants. All trials followed traditional planting systems used in West Africa. Small entire seed
tubers coming from traditional farmer’s planting material and weighing between 220 and 600 g
were planted in mounds, without staking at a density of 0.7 plant m-².
We used two cultivars, “Florido” for D. alata and “Morokorou” for D. rotundata.
Morokorou is a traditional early-maturing variety originating from north Benin, producing 1-3
cylindrical tubers. Florido was introduced into West Africa from Puerto Rico in the early 1970s
and produces two to five round tubers (Doumbia et al. 2004). The planting dates were 20
February 2007, 25 March 2009 and 25 April 2008. The experiments in Glazoué were not
irrigated, but in Cotonou the crop was irrigated to field capacity at planting, and further
supplementary irrigation (totalling between 80 to 110 mm depending on the cropping season)
was applied to replace estimated evapotranspiration using overhead sprinklers (Marcos et al.
2009). The plants did not show any visual sign of water or nutritional stress. Weed control was
done by hand every two weeks. For more information on sites or agricultural practices, see
Cornet et al. (in press).
Individual plant emergence dates were recorded every three days in all trials until the final
plant stand was established. We defined individual emergence when tuber sprouts emerged from
the soil. Plants were sampled from an area of 1.5–3 m² every 2-3 weeks for leaf and stem biomass
and crop leaf area from vine emergence to the tuber maturity. There were four replicates of two
plants at each sampling date. Leaf areas at Glazoué and in the Cotonou in 2007 trials were
estimated using the allometric model proposed by Cornet et al. (in press). In the 2008 and 2009
trials in Cotonou, leaf areas were measured with a planimeter (Li-Cor model 3100, LI-COR Inc.,
Lincoln, NE, USA). Leaf, stem and tuber biomass was determined after drying the material for
72 h at 70 °C. Daily minimum and maximum air temperature were measured by a standard
meteorological station located at less than 500 m from the plots. Mean daily temperature (T) was
Page | 120
calculated as the average of daily minimum and maximum temperatures. Astronomical daylength
including twilight was calculated according to Keisling (1982).
Past studies showed that maximal leaf area index was reached soon after tuber initiation
(Campbell et al. 1962, Sobulo 1972, Marcos et al. 2011). We assumed that the tuber weight (TW,
g plant-1) increases proportionally to leaf area at tuber initiation (LATI, m² plant-1):
where HA and TI are the harvest and tuber initiation dates and β is a constant.
The leaf area at tuber initiation was assumed to be proportional to the length of the
vegetative phase (EM-TI), defined as the time between emergence and tuber initiation:
where EM is the emergence date and γ is a constant determining the leaf area growth rate
(m² day-1).
Past studies have shown that plants from smaller seed-tubers accumulate dry matter and
develop leaf area at much slower rates than others (Onwueme 1972, Nwoke et al. 1973, Cornet et
al. 2014). Because the seed-tuber weight in farmer’s planting material is not uniform, its influence
on vegetative growth was also included in the model. An asymptotic relationship between γ and
seed-tuber weight (SW, g plant-1) was used because past experiments have shown that the
increase in plant growth rate with increasing seed-tuber weight tends to diminish asymptotically
(Ferguson 1973, Rodriguez-Montero et al. 2001). In the current study this asymptotic relationship
was defined by a modified Gompertz curve with biologically meaningful parameters (Hunt 1982).
(δ∗(250−SW))
𝛾 = 𝜀 𝑒 −𝑒 (3)
where δ is the relative growth rate at inflection point and ε is a constant of proportionality. The
abscissa of the inflection point was set at 250g, which corresponds to the seed-tuber weight
necessary to obtain one third of the maximum plant yield found by Ferguson (1973).
The crop development was assessed using a daily cumulative photothermal age. The development
rates were then computed as summations of daily photothermal ratios. The method of Marcos et
al. (2011) was used to simulate the response to temperature and photoperiod by calculating a
dimensionless cultivar-specific factor that ranges from 0 to 1. To simulate the progression of the
crop through the growth stage, a rate of development (r, day-1) is computed daily by multiplying
the temperature (f(T)), absolute photoperiod (f(P)) and rate of change in photoperiod (f(C))
factors by the phase-specific daily maximum rate (rx, day-1).
The effect of daily mean temperature (T, °C) on development rate can be described by
using the beta function (Wang et Engel 1998) with a cardinal minimum, optimum and maximum
temperature (Tn, To and Tx, °C) in each phase:
0, 𝑖𝑓 𝑇 ≤ 𝑇𝑛 𝑜𝑟 𝑇 ≥ 𝑇𝑥
log(2)
𝑓(𝑇) = {2(𝑇−𝑇𝑛)𝛼((𝑇𝑜 −𝑇𝑛)𝛼−(𝑇−𝑇𝑛)2𝛼 , with 𝛼 = 𝑇 −𝑇 (5)
, 𝑖𝑓 𝑇𝑛 < 𝑇 < 𝑇𝑥 log( 𝑥 𝑛 )
(𝑇𝑜 −𝑇𝑛 )2𝛼 𝑇𝑜 −𝑇𝑛
The effect of daily absolute daylength (P, h day-1) on development rate can be described
by using a negative exponential law (Angus et al. 1981). There is no development if P is below the
critical photoperiod (Pc, h day-1). Above Pc, the response to P decreases according to the
sensitivity coefficient (Ps, h-1):
1, 𝑖𝑓 𝑃 < 𝑃𝑐
𝑓(𝑃) = { −𝑃𝑠 (𝑃−𝑃𝑐) (6)
𝑒 , 𝑖𝑓 𝑃 ≥ 𝑃𝑐
The effect of daily rate of change in daylength (C, s day-1) on development rate can be
described using the same negative exponential law as for the absolute daylength. There is no
development if C is below the critical rate of change of photoperiod (Cc, s day-1). Above Cc, the
response to C decreases according to the sensitivity coefficient (Cs, s-1):
1, 𝑖𝑓 𝐶 < 𝐶𝑐
𝑓(𝐶) = { (7)
𝑒 −𝐶𝑠 (𝐶−𝐶𝑐) , 𝑖𝑓 𝐶 ≥ 𝐶𝑐
The rate r (day-1) is initialized at emergence (EM) with a value of 0; tuber initiation (TI)
and tuber maturity (HA) are reached when the accumulated r exceeds 1, and 2, respectively:
r𝑥𝐸𝑀−𝑇𝐼 𝑓(T)𝑓(P)𝑓(C), for the vegetative phase
𝑟={ (8)
1 + r𝑥𝑇𝐼−𝐻𝐴 𝑓(T)𝑓(P)𝑓(C), for the tuberisation phase
Further details of the model can be found in Marcos et al. 2011. This cumulative method
implies that the plant progresses every day towards tuber initiation or maturity with a variable
rate, which depends upon temperature and photoperiod. Because there is traditionally no
Page | 122
irrigation in the yam fields of West Africa, a supplementary limitation, stopping ongoing growth
at the onset of the dry season (i.e. mid-November), was added to the model.
Combining Eqns 4 and 8 it is possible to estimate the plant tuber yield as a function of seed tuber
size and emergence date at a given date of harvest (HA). By activating f(T), f(P), f(C) or a
combination of these it is possible to model the individual and combined effects of these
environmental variables on the plant tuber yield. Four models were compared. The first, as for
potato development models, considered a combined effect of temperature and photoperiod on
tuber initiation (e.g. Griffin et al. 1993, Streck et al. 2007) and a single effect of temperature on
tuber maturity. The second model considered a combined effect of temperature and photoperiod
on both the vegetative and the tuberization phase (Marcos et al. 2009). The third model was
similar to the second but the photoperiod was replaced by the rate of change of photoperiod (Ile
et al. 2007). The last model corresponded to a “full” model with a combined effect of
temperature, photoperiod and rate of change of photoperiod on both growth stages. In further
discussion they are referred to respectively as the potato model (MPotato), the yam photoperiod
model (MYamP), the yam rate of change model (MYamR) and the full model (MFull).
The best model was selected by comparing models abilities to simulate plant tuber yield
observations on the basis of three summary statistics: (i) the root mean square error (RMSE), (ii)
the Nash-Sutcliffe model efficiency coefficient (ME) ranging between -∞ and +1, and (iii) the
Akaike Information Criterion corrected (AICc). The first two of these assess the absolute and
relative ability of the model to represent the modelled phenomenon. The AICc allows model
ranking on the basis of the performance / parsimony (number of variables). The root mean
square error (RMSE) is defined as
0.5
∑n
i=1(Oi −Pi )
2
RMSE = ( ) (9)
n
where Oi and Pi represents the observed and the predicted plant tuber yield respectively for a
dataset of n observations.
Page | 123
The model efficiency ranges between –inf and 1, with 1 indicating a perfect fit, 0 for a quality of
fit equivalent to the null (overall mean) model and a negative value for worse prediction than the
null model (Nash and Sutcliffe 1970).
where k is the number of parameters and SSres is the residual sum of squares (Burham and
Anderson 2004). For each species, the best model was chosen for prospective analysis, firstly
according to the RMSE value, therefore considering model robustness, and secondly according to
AICc, considering the ability / parsimony trade-off.
In order to reduce the number of parameters to be estimated during the main optimization, a
preliminary optimization was done on a subset of the data corresponding to plant samples
collected during the vegetative phase (i.e. samples without new tubers). For this, the plant
cardinal temperatures and photoperiod sensitivity parameters of the vegetative phase (i.e. Tn, To,
Tx, Pc, Ps and rx) and the parameter of the Gompertz model (δ) were first estimated by
minimizing the RMSE between the observed leaf area and a submodel vegetative index (SVI):
(δ∗(250−SW))
𝑆𝑉𝐼 = 𝑒 −𝑒 ∑𝐻𝐴
𝑖=𝐸𝑀 𝑓(𝑇)𝑖 (12)
In a second step, it was possible to estimate the remaining model parameters by minimizing the
RMSE between observed and predicted plant tuber yield using Eq. 4.
Page | 124
the R statistical software and the DEoptim library (Mullen et al. 2011, R Development Core Team
2013).
6.2.6. Simulations
Because there is no interplant competition at traditional plant densities in the yam’s fields of West
Africa (Cornet et al. 2014), it was possible to simulate the crop yield as the sum of individual plant
yields using the best plant model. In order to determine the consequences of using heterogeneous
planting material, the crop yield obtainable from traditional and improved planting material were
simulated. The simulated distribution of the traditional planting material was characterized by a
spread of emergence dates and variable seed-tuber weight, while the distribution of the improved
planting material was characterized by a narrower spread of emergence and more uniform seed-
tuber weight. The seed-tuber weight and emergence distributions for the traditional planting
material were defined from six trials carried out in Benin (Cornet et al. 2014) while it was based
on the potato for the improved treatment (O’Brien et al. 1983). In traditional cropping systems,
Cornet et al. (2014) observed a normal distribution of seed-tuber weight and a gamma distribution
of emergence, with shape and scale parameters equal to two (left skewed). The characteristics
(mean and standard deviation) of the traditional and improved planting material are presented in
Table 14. Climatic variables (i.e. temperature, photoperiod and rate of change of photoperiod)
are those measured in 2009 in Cotonou, Benin.
Table 14. Characteristics of the plant populations generated for the simulation study
Factor Distribution Level Mean SD IDR90 Min Max
Emergence (days Gamma Traditional 27 19.8 67 2 100
after planting) Improved 8 6.2 20 2 30
Seed-tuber size Normal Traditional 400 50.6 140 255 533
-1
(g plant ) Improved 400 5.3 20 389 416
SD : Standard Deviation.
IDR90 : 90% Interdecile range.
A rough estimate of both traditional and improved profitability was calculated for successive
planting dates and varying seed-tuber size. In West Africa yam ware tubers are sold on a “cash
and carry” basis and prices are mainly based on the variety, the season and, for D. rotundata, the
perceived size of tubers (Orkwor 1998). The profit was calculated as the proceeds from the sale
of the tubers minus the cost of the seed-tubers. The selling prices of both species were recorded
Page | 125
weekly at the yam market at Glazoué (Benin). For D. rotundata, different prices were recorded for
large (> 2.5kg) or small tubers (0.8-2.5 kg). Because yams in Benin are sold on a volume basis (in
basins or heaps), tubers were weighed to find the price per kilo (€ kg-1). When yams were not
being sold in the market, missing values were estimated by linear interpolation between adjacent
data. Figure 40 shows the mean selling prices over the three growing seasons (2007, 2008 and
2009). The tuber size category was inferred from the estimated individual plant tuber yield of the
plant model. The date of sale was estimated from the model using the end of the tuberization
phase as the date of maturity. The cost of the seed-tuber was calculated on the basis of its weight
in relation to its grade, as commonly practised in the seed-tuber market of Nigeria (Ogbonna et al.
2011, Kleih et al. 2012).
Figure 40. D. alata (red) and D. rotundata (blue) selling price at the market of Glazoué (Benin)
depending on the size of the tuber (circle: tuber size between 0.8 and 2.5 kg; triangle: tuber
size over 2.5 kg). Error bars are standard deviation of mean price over 2007, 2008 and 2009.
Page | 126
6.3. Results
Table 15 provides a comparison of overall and species-specific performance between the potato
model (MPotato), the yam photoperiod model (MYamP), the yam rate of change model (MYamR) and
the full model (MFull). All models showed much better performance than the null model (ME
always > 0.787). Generally the full model was the more robust (RMSE=134.6 g plant-1). But
while looking at the balance between model robustness and model parsimony, the yam potato
model performed better (AICc=1820.8). Despite small differences, the same pattern could be
observed for species-specific models: the potato model gave the best performance for D. alata
and D. rotundata (AICc of 1130.5 and 698.8 respectively). In the following sections we used the
potato model.
Page | 127
6.3.2. Parameter optimization for the potato model
The parameter estimates for the best species-specific models are shown in Table 16. D. rotundata
had a shorter critical photoperiod than D. alata (i.e. Pc of 12.18 and 12.68 respectively) but
particularly a much lower sensitivity to photoperiod (Ps of 0.04 and 0.94 respectively). Except for
the optimum temperature (To) during the tuberization phase, cardinal temperatures of D.
rotundata were always higher than those of D. alata. The range of cardinal temperature (i.e. Tx-Tn)
was always wider for the tuberization phase than for the vegetative phase. The minimum duration
phase (1/rx) was also much longer for the tuberization phase than for the vegetative phase and D.
rotundata showed longer minimum duration phases than D. alata.
Table 16. Parameter estimates for the selected species-specific models during vegetative and
tuberization phases
D. alata EM-TI 11.1 19.3 29.1 12.68 0.94 23.8 0.154 0.006
TI-HA 10.4 26.8 42.6 80.7
D. rotundata EM-TI 15.2 24.8 28.1 12.18 0.04 35.5 0.231 0.007
TI-HA 13.9 21.5 46.9 99.7
P is the absolute daylength (h day-1).
EM-TI and TI-HA are the vegetative phase and the tuberization phase respectively (day).
To, Tn and Tx are respectively the optimum, the minimum and the maximum temperature (°C) for a given
growth phase.
Pc is the critical absolute daylength (h day-1) and Ps is the sensitivity coefficients to the absolute daylength
(h-1).
1/rx is the minimum duration of the respective growth phase (day).
ε and δ are two constants.
The estimated and observed values of the model outputs were analysed. Figure 41
present (i) the relationship between observed plant leaf area and the submodel vegetative index
(SVI, Eqn 12) predictions of the model and (ii) the relationship between the predicted and
observed plant yield. This double-checking of the model was necessary because of the
considerable risk of over-fitting given the large number of parameters used during the
optimization process and the many variables used in the model. In order to check the
appropriateness of the model’s components, the observed leaf area index was used to test the
vegetative part of the model (Figure 41a and Figure 41b). Leaf area of sampled plants ranged
between 0 and 1.6 m² plant-1. A coefficient of correlation always higher than 0.85 was observed.
The relationship between observed and predicted plant tuber yield of the two main yam species is
given in Figure 41c and Figure 41d. The yams yielded up to 1500 and 2000 g of tuber dry matter
Page | 128
per plant for D. alata and D. rotundata respectively. The species-specific models gave good
estimates of observed plant tuber yield (both R² > 0.83). No systematic bias was observed.
Figure 41. Estimated against observed outputs of best model components: plant leaf area (a
and b) and plant yield (c and d) of the two main yam species: D. alata (in red) and D. rotundata
(in blue) as influenced by temperature and daylength from plant emergence.
The range of validity of the estimates is given by the density function of observed emergence
dates during the experiments and corresponds to traditional planting dates (from February to
May). Mean temperature during the growing season oscillated between 20.4 and 30.5°C (Figure
42a). The photoperiod ranged between 12.5 and 13.3h (Figure 42b). Figure 42c and Figure 42d
show the estimates of growth stage duration for successive planting dates as a function of
temperature and daylength. Most of emergence occurred at a time of increasing daylength.
Within this range of emergence, the vegetative phase took place during a warmer period than the
tuberization phase. D. alata always had a longer vegetative phase and a shorter tuberization phase
than D. rotundata. Emergence later than 265 days resulted in a sudden decrease in the duration of
the vegetative phase, whereas emergence later than 170 days resulted in a sudden decrease in the
duration of the tuberization phase.
Page | 129
Figure 42. Temperature (a), photoperiod (b) and estimated growth stage durations (c and d) as
a function of emergence date of the two main yam species (D. alata in red and D. rotundata in
blue). EM-TI is the vegetative phase, and TI-HA the tuberization phase. The density plot at
the top of the figure presents the density function of the observed emergence dates of both
yam species within the six experiments of the current stud y.
Page | 130
6.3.4. Simulated crop yield
Figure 43a compares the simulated yields of both yam species using traditional and improved
planting material. Both simulated and observed crop yields decreased with successive planting
dates. The simulated yield of D. rotundata was always higher than that of D. alata, but this
difference tended to diminish with later planting. Within the range of traditional planting dates,
the yield improvement due to narrower emergence spread and uniform seed-tuber weight was
always more than 18% (Figure 43b). For planting dates prior to 150 days, the yield improvement
is relatively stable and higher for D. rotundata than D. alata (i.e. 24±2.1 and 20±1.5% yield
increase respectively). Afterwards the effect of improved planting material increased drastically
but the absolute yield gain remained low (+ 1.1 and 1.3 t ha-1 for D. alata and D. rotundata
respectively at planting date 185).
Figure 43. Influence of the planting date on the estimated yield differences between improved
(solid line) and traditional (dashed line) cropping systems of the two main yam species ( D. alata
in red and D. rotundata in blue). The black arrow indicates the traditional planting date in West
Africa.
Page | 131
The comparison of the tuber dry matter yield (t ha-1) of both species in the traditional
cropping system as a function of planting date and seed-tuber weight is presented as a contour
plot in Figure 44a and Figure 44b. For a given planting date and seed-tuber weight, D. rotundata
always outperformed D. alata; e.g. at 110 days with a seed-tuber weight of 550g the yields of
traditional cropping system were 6.9 and 5.9 t of tuber dry matter per ha respectively. In general,
the later the planting date, the lower the yield; while the heavier the seed-tuber, the higher the
yield. But the yield gain attributable to seed-tuber weight decreased with later planting dates: e.g.
for D. alata in the traditional cropping system (Figure 44a) with a planting date of 90 days,
doubling the seed-tuber weight from 250g to 500g increased the yield by 3.0 t ha-1, while with a
later planting date of 180 days, doubling the seed-tuber weight only increased the yield by
1.2 t ha-1. The yield gain in tuber dry matter (t ha-1) attributable to the use of improved planting
material is shown in Figure 44c and Figure 44d. It varied between 0.3 and 1.4 t ha-1 for D. alata
and between 0.4 and 2.8 for D. rotundata.
Figure 44. Yield (t of tuber dry matter per ha) of the traditional cropping system (a and b) and
yield improvement linked to the use of improved planting material (c and d) for the two main
yam species, D. alata (in red) and D. rotundata (in blue) as a function of planting date and seed-
tuber weight. Contour lines represent level of equal yield (a and b) or yield improvement (c and
d).
Page | 132
6.3.5. Profitability of traditional and improved planting material
The profit (i.e. the receipts from the harvest sales minus the seed-tuber cost, k€ ha-1) of the
traditional cropping system was estimated for D. rotundata and D. alata (Figure 45a and Figure
45b). Within the range of the models’ validity, the profitability of early planting (i.e. March)
increased with increasing seed-tuber weight, and reached 1.7 and 8.5 k€ ha-1 for D. alata and D.
rotundata respectively. Profit fell with delayed planting (reaching zero after 180 days) but especially
rapidly when using large seed tubers. The use of improved planting material was always followed
by a gain in profitability (Figure 45c and Figure 45d). This gain was much higher with higher
seed-tuber weight and early planting (up to 0.5 and 3 k€ ha-1 for D. alata and D. rotundata
respectively). In both absolute and relative terms, profitability was always higher with D. rotundata.
Indeed, the mean gains in profitability over the traditional planting dates were 30 and 40% for D.
alata and D. rotundata respectively (data not shown).
Figure 45. Profitability (k€ ha -1 ) of the traditional cropping system for D. alata (a) and D.
rotundata (b) and profitability improvement linked to the use of improved planting material for
D. alata (c) and D. rotundata (d) as a function of planting date and seed-tuber weight. Contour
lines represent level of equal profitability (a and b) and profitability improvement (c and d).
Page | 133
6.4. Discussion
The best model (i.e. the potato model) assumed a combined effect of temperature and
photoperiod on growth during the vegetative phase but an effect of temperature alone during the
tuberization phase. Within the range of the experimental data, this individual plant model
provided a good estimate (low RMSE and high correlation with observed data) of the growth
stage duration and of the plant yield for both species. Cardinal temperatures of D. alata were
similar to those found by other authors: i.e. the minimum, optimum and maximum temperature
in the current study were 11.1, 19.3 and 29.1°C against 12.3, 19.7 and 32.5°C in the study of
Marcos et al. (2009). The corresponding values during the vegetative phase were 10.4, 26.8 and
42.6°C, as against 17.2, 28.3 and 42.3°C. For D. rotundata, there is only one study available
estimating the cardinal temperatures over the entire growth period (Ile et al. 2007), in which 12°C
was found to be the minimum temperature and 25.8 and 27.5°C the optimum temperature. The
same authors set the maximum temperature as being equal to that of cassava (i.e. 42°C). Results
of the current study were in the same range.
The observed and simulated influence of seed-tuber weight on the plant yield were
considerable important for both species (i.e. δ of 0.0060 and 0.0069, indicating a 56 and 58%
yield increase, for D. alata and D. rotundata respectively for seed-tuber weights ranging from 250
to 550 g). This result for D. rotundata is in agreement with previous studies (Onwueme 1972,
Nwoke et al. 1973, Oriuwa and Onwueme 1980). For D. alata this important effect of seed-tuber
weight up to 550g was not observed either by Ferguson (1973) or by Cornet et al. (submitted)
who found no substantial benefit of increasing seed-tuber weight beyond 400g. This discrepancy
could be explained by some confounded effects. Because we used entire seed tubers from the
previous cropping season, the smaller seed tubers may have come from weaker plants. So the
observed detrimental effect of smaller seed-tuber weight may have been confounded with lower
quality tubers from previous cropping season (i.e. lower nutrient or dry matter proportion).
Page | 134
6.4.2. Simulation of yam plant growth and development
At traditional planting dates, emergence occurs at a time of increasing daylength. Within this
period of emergence, the vegetative phase took place during a warmer period than the
tuberization phase. Since the cardinal temperatures of both species are lower for the vegetative
phase than for the tuberization phase, a shift in the planting date towards a cooler period (May-
June) could lead to improved yields for low-photoperiod-sensitive cultivars in irrigated cropping
systems. A trial of off-season yam production was carried out in a farmer’s field at inland valley in
Niger state, Nigeria (Shiwachi et al. 2008). In their study the tested cultivars belonged to D. alata
and gave a low yield, for which no explanation was offered. In the future, the use of low-
photoperiod-sensitive cultivars of D. rotundata could be a solution.
The minimum lengths of both vegetative and tuberization growth phases (1/rx) were
shorter for D. alata than for D. rotundata, in accordance with current knowledge of yam
physiology (Orkwor et al. 1998, Sartie et al. 2012). Values of rx for D. alata were in agreement with
a previous study (Marcos et al. 2009). Despite shorter minimum vegetative phase (1/rx), D. alata
always had a longer simulated vegetative phase than D. rotundata. This could be explained by its
higher sensitivity to photoperiod, which delayed tuber initiation when daylengths were greater
than the critical photoperiod threshold. While the tuberization growth stage of both species is
constant for planting dates until the end of June, the duration of the vegetative stage decrease
asymptotically until reaching a minimum for planting dates in July. The decreasing duration of
the vegetative phase with delayed planting is explained by the influence of both the photoperiod
and the temperature. Later on, a sudden decrease in the duration of the vegetative and
tuberization phases respectively after emergence dates of 265 and 175 days respectively was
observed because the model imposes a limitation prohibiting growth beyond the onset of the dry
season.
Within the range of the experimental data and for a mean seed-tuber weight of 400 g, the
yield improvement due to the use of improved planting material was always more than 18% and
could reach up to 40% depending on the species and the planting date. While looking at the
influence of both planting date and seed-tuber weight, the yield improvement associated with
improved practices varied between 0.3 and 1.4 t of tuber dry matter per hectare for D. alata and
0.4 and 2.8 for D. rotundata.
The yield advantage of D. rotundata over D. alata was complemented by its much higher
selling price (up to three times). This was obvious while looking at the profitability of both
species (up to 1.7 and 8.5 k€ for D. alata and D. rotundata respectively in traditional cropping
systems). Despite the higher cost of increasing seed-tuber weight, the use of larger seed tubers
was also profitable. This benefit decreased with delayed planting. The substantial advantage of
Page | 135
increasing seed-tuber weight, up to more than 600 g, may explain the lack of adoption of some
innovative technique (e.g. minisets). For the traditional planting dates, the use of improved
planting material enhanced the simulated profitability, on average, by 30 and 40 % for D. alata
and D. rotundata respectively. For D. rotundata, the large seasonal price fluctuation made the
contour lines of the profitability gain more complex. Linking the model with a market
information system could help the farmer to choose the best options before planting and at
harvest regarding the possibility of storing the crop. Moreover the model could also be used as a
decision-support tool for farmers, allowing them to choose the best seed-tuber weight,
depending on the species and the planting date. Finally, the model helps to shed new light on a
farming practice which, at first glance, seems surprising to agronomists. The very low planting
densities traditionally practiced in West Africa allow for an optimal plant development, free from
competition, leading to maximization of individual plant yield. The deliberate choice to favour
individual plants rather than yield per unit land is due to consumer preferences for large tubers
which fetch higher prices.
At present there are many projects in West Africa for improved yam seed-tuber
production, although they focus on clean (healthy) seed yam production (WECARD 2011, DFID
2014, IITA 2014). The current study shows that ongoing development projects on improved yam
planting material could have greater impact if seed-tuber healthiness, size and physiological age
are all tackled together. Moreover, seed-tuber production systems could also control varietal
purity, another well-known source of variability in vegetatively propagated root and tuber crops
(Elias et al. 2001, Chaïr et al. 2010).
The use of this model is restricted to West African cropping systems where no
competition occurs between neighbouring plants. Higher density cropping systems (e.g. as in the
Caribbean) are likely to be subject to some interplant competition that will reduce the individual
plant yield. In order to extend the range of validity of the model, further work is necessary,
especially to consider the potentially more profitable out-of-season cropping.
Acknowledgments
We thank M. Sodanhoun, C. Adiba, D. Damissi and J. Lawson E. for field assistance and
data acquisition, Dr R. Bonhomme for helpful comments on the manuscript, and A. Scaife for
English revision.
Page | 136
Page | 137
Conclusions et perspectives
La problématique centrale de ce travail de thèse était de quantifier et de comprendre la variabilité
individuelle du rendement par plante au sein d’une culture d’igname. En effet, les niveaux de
variabilité couramment observés chez l’igname compromettent la compréhension du
fonctionnement de la culture et les tentatives d’intensification. Ce travail d’acquisition de
connaissances est donc un préalable indispensable à l’amélioration de la conduite de cette plante
orpheline, dont la culture traditionnelle en Afrique de l’Ouest par défriche-brûlis est condamnée à
moyen terme.
Afin d’aborder cette problématique, une série de 11 expérimentations ont été conduites durant
quatre années sur deux sites au Bénin avec deux variétés représentatives des deux espèces les plus
cultivées en Afrique de l’Ouest, D. alata et D. rotundata. Ces expérimentations ont permis
d’acquérir des données aussi bien sur la croissance et le développement de la plante que sur son
architecture et sa physiologie. Les résultats de ces expérimentations, ainsi que les variables
environnementales qui leur sont associées ont été rassemblés dans un jeu de données cohérent,
qui constitue un outil unique de réflexion et d’analyse.
Autre fait marquant, un même niveau de variabilité est observé pour des plantes isolées ou en
culture ; ceci atteste qu’aux densités couramment pratiquées en Afrique de l’Ouest la variabilité
n’est pas imputable à la compétition.
Enfin, ces premiers résultats ont montré que la variabilité apparaît dès les premiers stades (moins
d’un mois après émergence), ce qui confirme notre hypothèse initiale.
Page | 138
Les causes de la variabilité
L’observation des dates individuelles d’émergence a permis de montrer qu’au cours des quatre
années d’expérimentation, il fallait en moyenne 67 jours pour que 90% des plantes émergent,
contre 10 à 20 jours pour la pomme de terre. La variabilité des dates d’émergence est donc une
cause importante de la structuration précoce de hiérarchies de taille et de rendement entre
plantes : plus l’émergence d’une plante est précoce, plus la tubérisation est initiée rapidement, et
plus le rendement final est élevé. Or, la date d’émergence dépend avant tout de la « qualité du
semenceau » : son âge physiologique et son contenu en nutriments. Malheureusement, cette
qualité de semenceau n’est pas connue dans le cadre des systèmes de culture traditionnels. En
effet, compte tenu des conditions de récolte et de stockage, il n’est pas possible lors de l’achat de
semenceaux en milieu paysan d’en connaître ne serait-ce que l’âge physiologique. En revanche,
nous avons démontré que la qualité du semenceau peut être affectée par des pratiques culturales
mesurables (date de plantation, prégermination, taille du semenceau), et que son expression peut
être mesurée indirectement durant les premiers stades de croissance (émergence, nombre de tiges,
présence de cataphylles). Sur la base de nos observations, les variables impliquées dans la
variabilité des composantes du rendement ont été rassemblées au sein d’un modèle graphique de
type réseau bayésien. Leurs effets directs et indirects sur les composantes du rendement
individuel ont ainsi pu être identifiés et quantifiés. L’utilisation des réseaux bayésiens nous a
permis, dans un premier temps, de rechercher la meilleure structure expliquant les données
observées (réseau de relations directes et indirectes) et, dans un deuxième temps, de quantifier ces
relations.
L’analyse des résultats confirme, pour les deux espèces d’igname, l’influence des pratiques à la
plantation (date de plantation et prégermination) et de la date d’émergence sur les composantes
du rendement de l’igname. Cependant, les résultats ont abouti à des réseaux de relations de
dépendance différents selon l’espèce. Ainsi, nous avons pu mettre en évidence un effet quantitatif
positif du nombre de nœuds portant des cataphylles sur le poids moyen en tubercule par plante
de D. rotundata. Puisque les cataphylles ne participent pas directement à la croissance de la plante,
nous émettons l’hypothèse que leur émission reflète la durée de la phase d’hétérotrophie pendant
laquelle la jeune plante se nourrit principalement du semenceau. Ces cataphylles pourraient donc
être un indicateur indirect de la qualité relative du semenceau via la durée de la phase
hétérotrophe. Plus le semenceau contient de ressources facilement mobilisables, plus longtemps
la jeune plantule s’en nourrira, et plus sa croissance initiale sera avantagée comparativement à
Page | 139
celle issue d’un semenceau de moins bonne qualité. L’absence de cataphylles chez D. alata ne
permet pas d’établir de comparaison entre les espèces à ce niveau.
Les deux espèces montrent un effet positif de la taille du semenceau sur le rendement de la
plante, mais dans le cas de D. alata, cet effet doit être nuancé. En effet, dans l’expérimentation sur
les réseaux bayésiens, l’effet de la taille du semenceau, présent initialement, est éliminé après
l’étape d’ajustement du modèle contre le surapprentissage (Chapitre 4). A l’opposé, on observe
un effet important de la taille du semenceau chez D. alata dans le cadre des expérimentations sur
la modélisation (Chapitre 6). Cette contradiction apparente s’explique par le type de matériel
utilisé et les distributions de poids de tubercules différents entre expérimentations. Dans le cas
des réseaux bayésiens, les poids moyens de semenceaux étaient élevés et peu variables ; aucun
effet n’est donc observé dans le réseau bayésien final. En revanche, les expérimentations menées
en 2006 et en 2007 à Glazoué présentent une plus grande variabilité et de plus petits semenceaux.
Ces petits semenceaux sont susceptibles de provenir des plantes les moins vigoureuses de la
culture précédente, introduisant ainsi un mélange d’effet (taille et qualité). En regroupant les deux
types de population, et compte tenu du manque de flexibilité de la relation utilisée, l’effet du
poids du semenceau apparaît artificiellement élevé dans le modèle développé au Chapitre 6. Nous
retiendrons ici que l’effet de la taille du semenceau est important pour D. rotundata pour toute la
gamme de poids étudiée (200-400 g), alors que pour D. alata, son influence est susceptible d’être
négligeable au-delà de 400 g.
Page | 140
Les conséquences agronomiques et économiques pour l’agriculteur
Pour l’agriculteur, la conséquence principale de la variabilité, est une perte directe de rendement.
Les observations ainsi que la modélisation montrent clairement que les plantes émergées plus
tardivement, ou issues de semenceaux de moins bonne qualité, auront une croissance réduite, une
tubérisation initiée trop rapidement et au final, un rendement plus faible. En l’absence de
compétition, cette perte de rendement chez l’igname n’est pas corrigée par un éventuel effet
compensatoire comme cela s’observe pour la majorité des grandes cultures ; il s’agit donc bien
d’une perte de rendement directe comme nous l’avions envisagé avec notre dernière hypothèse.
En tenant compte uniquement de la variabilité de la date d’émergence et de l’influence du poids
du semenceau, les simulations réalisées avec le modèle-plante permettent d’estimer cette perte de
rendement entre 0,3 et 1,4 pour D. alata et entre 0,4 et 2,8 tonnes de matière sèche de tubercules
par hectare pour D. rotundata, en fonction de la date de plantation et de la taille du semenceau.
Cela correspond donc à une perte moyenne de 22 et 27% de rendement pour D. alata et D.
rotundata respectivement.
A cette perte de rendement s’ajoute, pour D. rotundata, une perte de profit. En effet, le prix de
vente en Afrique de l’Ouest dépend du calibre, et les faibles rendements individuels se traduisent
en conséquence par un prix de vente inférieur. En tenant compte de l’ajustement des prix en
fonction du calibre et de la saison, les simulations réalisées montrent que cette perte de profit est
en moyenne de 0,5 k€ ha-1 pour D. alata et de 3 k€ ha-1 pour D. rotundata, soit 30 et 40% de perte
de profit, respectivement.
Enfin, même si elles n’ont pas été étudiées durant ce travail, il est probable que la variabilité
provoque également des pertes indirectes. A titre d’exemple, les plantes émergées tardivement
conduisent à un couvert tardif et moins vigoureux, permettant aux plantes adventices de se
développer plus facilement : le nombre et l’intensité des sarclages sont donc susceptibles
d’augmenter avec la variabilité.
Page | 141
Les conséquences pour la recherche agronomique
L’importante variabilité observée chez l’igname questionne le travail de recherche, que ce soit au
moment de concevoir une innovation technique ou d’en analyser et d’en interpréter les résultats.
Sachant que l’émergence s’étale sur plus de deux mois, comment, par exemple, concevoir un
même traitement à appliquer uniformément à des plants d’âge et parfois de stade très différents ?
On imagine assez bien qu’une fertilisation azotée apportée en cours de phase végétative
favorisera le développement d’un couvert important et permettra ensuite un meilleur rendement.
En revanche, si la fertilisation est appliquée en début de tubérisation, l’effet est moins évident et
peut même se révéler négatif, en retardant la tubérisation. Or, compte tenu de la variabilité
mesurée à l’échelle de la parcelle, il est fortement probable que les deux stades de développement
se côtoient selon les plantes. Dans ces conditions, il devient difficile d’imaginer une innovation
technique efficace en utilisant du matériel de plantation traditionnel.
En partant du même exemple de fertilisation, il est possible d’illustrer les difficultés d’analyse et
d’interprétation des résultats. En effet, comment évaluer l’efficacité d’un traitement sur la base de
sa moyenne, lorsque les effets peuvent être différents voire contradictoires d’une plante à l’autre.
De plus, la forte variabilité génère des écart-types importants qui ont tendance à masquer les
effets étudiés ; aussi, les analyses classiques de variance permettent difficilement de dégager des
effets significatifs. Enfin, l’utilisation du coefficient d’asymétrie de Lorenz pour étudier la
structure de la variabilité entre individus a clairement montré que les distributions de taille de
plante ou de rendement s’éloignaient de la loi normale. En termes d’analyse statistique, cela
implique l’utilisation de méthodes alternatives aux ANOVA, souvent plus complexes et moins
accessibles (transformation de variable, modèle mixte linéaire généralisé…).
Durant ce travail de thèse, plusieurs méthodes et outils ont été mis au point. Ces outils sont
génériques et leur utilisation dépasse le cadre de la thèse. Ainsi, nous avons montré que les
réseaux bayésiens, peu utilisés en agronomie, permettent de représenter un système complexe de
telle façon que la compréhension et l’analyse du système en est facilitée, même pour un néophyte.
Leur application aux deux espèces majeures d’igname offre un cadre de réflexion idéal sur les
interactions entre pratiques agronomiques, croissance initiale et composantes du rendement. Les
graphes orientés permettent ainsi de mieux comprendre la physiologie des ignames et d’en
dégager des opportunités pour l’amélioration des systèmes de culture à base d’igname : par
Page | 142
exemple, (i) l’utilisation d’une mesure de l’étalement de l’émergence pour estimer l’efficacité de
différentes techniques de production de semenceaux d’âges physiologiques homogènes, ou (ii)
l’utilisation du nombre de cataphylles comme indicateur de la qualité du semenceau et de la date
de passage à l’autotrophie.
De même, un modèle allométrique commun aux deux espèces d’igname permet l’estimation non
destructive de la surface foliaire à l’échelle de la feuille (A, cm²) à partir de mesures de la largeur
maximale du limbe (W, cm) et L sa longueur maximale (L, cm) : A=0,74(W² L)1,94.
Cette étude a permis également la mise au point d’un modèle de croissance et de développement
à l’échelle de la plante ; le modèle intègre l’effet de la taille du semenceau et de la date de
plantation. Ce modèle permet de mieux comprendre la physiologie de la plante et notamment sa
sensibilité aux températures moyennes et à la photopériode. Il s’agit donc d’un outil fiable pour
étudier la croissance et le développement de la plante, et sa variabilité. Cependant, le modèle
n’intègre pas les évènements et les facteurs pouvant affecter la croissance et le développement de
la plante au-delà des premiers stades (nutrition azotée, plasticité phénotypique, rapport carbone-
azote…). L’étude de l’influence de ces facteurs sur la variabilité résiduelle, non expliquée par le
modèle, devrait permettre d’améliorer le modèle et d’augmenter sa généricité.
Enfin, ce modèle peut servir d’outil d’aide à la décision pour les acteurs du développement
agricole en Afrique de l’Ouest. Le lien avec l’évolution des prix sur le marché permet de réfléchir
aux choix des tailles de semenceaux en fonction de la variété et de la date de plantation. La mise
Page | 143
en évidence de l’influence de la température, de la photopériode et de la saison sèche, permet de
réfléchir aux possibilités de désaisonnaliser la culture.
Page | 144
Perspectives pour une meilleure compréhension de l’écophysiologie des ignames
L’ensemble des résultats acquis ici sont généralisables aux systèmes de culture traditionnels de la
sous-région. En revanche, l’utilisation de plus fortes densités de plantation ailleurs dans le monde
(Asie, Caraïbe) est susceptible d’introduire de la compétition entre plantes. Pour répondre aux
enjeux de sécurité alimentaire et d’augmentation de la pression démographique, il est probable
que les systèmes traditionnels ouest-africains, en cours de sédentarisation, voient leur densité de
plantation augmenter afin d’optimiser le rendement surfacique. L’étude de la compétition entre
plantes d’ignames à plus forte densités reste alors à réaliser.
Page | 145
Bibliographie
Akaike H. 1973. A new look at the statistical model identification. IEEE Transactions on Automatic
Control 19: 716-723.
Akoroda M.O. 1984. Variability, repeatability, character correlation and path coefficient analyses
in yellow yam. Theoretical and Applied Genetics 69: 217-221.
Alcoy A.B., Garcia A.G., Baldos D.P., Robles R. P., Cuyno R.V. 1993. Influence of planting
material and time of harvest on plant to plant yield variability of Sweet Potato (lpomoea
batatas (L.) Lam). Philippine Journal of Crop Science 18: 187-193.
Amagbeto K., Asiedu R., Bokanga M., Cardwell K., Carsky R.J., Chikoye D., Hughes J., Manyong
V.M., Meerma J.C.N., Mignouna H.D., Ng N.Q., Ng S.Y., Okoe S.N., Notle C., Soyinka
S.A.C., Tian G., Vernier P., Wanyera N.M.W. 2000. Improving yam-based systems.
Annual Report, IITA, Ibadan, Nigeria.
Angus J. F., Mackenzie D. H., Morton R., Schafer C. A. 1981. Phasic development in field crops
II. Thermal and photoperiodic responses of spring wheat. Field Crops Research 4: 269-283.
Asadu C. L. A., Akamigbo F. O. R., Ezumah H. C., Nweke F. I. 1990. The characterization of
selected yam-growing soils in southeastern Nigeria. II. Chemical and mineralogical
properties. Nigerian Agricultural Journal 24: 71-86.
Asiedu R., Sartie A., 2010. Crops that feed the World [Link]. Yams for income and food
security. Food Security 2: 305-315.
Austin P., Tu, J. 2004. Bootstrap methods for developing predictive models. The American
Statistician 58: 131-137.
Babyak M.A. 2006. What you see may not be what you get: a brief, nontechnical introduction to
overfitting in regression-type models. Psychosomatic Medicine 66: 411-421.
Bendel R.B., Higgins S.S., Teberg J.E., Pyke D.A. 1989. Comparison of skewness coefficient,
coefficient of variation, and Gini coefficient as inequality measures within populations.
Oecologia 78: 394-400.
Page | 146
Black J.N., Wilkinson G.N. 1963. The role of time of emergence in determining the growth of
individual plants in swards of subterranean clover (Trifolium subterraneum L.). Australian
Journal of Agricultural Research 14: 628-638.
Bradu D., Mundlak Y. 1970. Estimation in lognormal linear models. Journal of the American
Statistical Association 65: 198-211.
Bricas N., El Moussahoui N., Kayode P., Nindjin C., Orkwor G. Et Hounhouigna J. 2003. La
consommation et les critères de qualité des ignames dans les villes du Bénin, de Côte
d'Ivoire et du Nigeria. In Post-récolte et consommation des ignames : réduction des pertes et
amélioration de la qualité des produits pour les marchés africains, Cerna-UNB, 16 au 20 juin,
Cotonou, Bénin.
Buckley H.L., Damgaard C. 2012. Lorenz.R: R code for drawing sample Lorenz curves and to
calculate Gini Coefficients and Lorenz Asymmetry Coefficients [Computer program].
Burnham K. P., Anderson D. R. 2004. Multimodel inference: understanding AIC and BIC in
Model Selection. Sociological Methods and Research 33: 261-304
Campbell J.S., Chukwueke V.O., Teriba F.A., Ho-A-Shu, H.V.S. 1962. Some physiological
investigations into the White Lisbon yam (Dioscorea alata L.) II. Growth period and out-
of-season production. Empire Journal of Experimental Agriculture 30: 232-238.
Castañera M.B., Aparicio J.P., Gürtler R.E. 2003. A stage-structured stochastic model of the
population dynamics of Triatoma infestans, the main vector of Chagas disease. Ecological
Modelling 162: 33-53.
Caviglia O.P., Melchioria R.J.M., 2011. Contribution of contrasting plant hierarchies to the
response to N fertilizer in maize. Field Crops Research 122: 131-139.
Chair H., Cornet D., Deu M., Baco M.N., Agbangla A., Duval M.F., Noyer J.L. 2010. Impact of
farmer selection on yam genetic diversity. Conservation Genetics 11: 2255-2265.
Chatterton N. J., Lee D.R., Hungerford W.E. 1972. Diurnal change in specific leaf area weight of
Medicago sativa L. and Zea mays L. Crop Science 12: 576-578.
Clairon M., Zinsou C. 1980. Etude de plantations échelonnées d'igname Dioscorea alata cv. Lupias:
effet du vieillissement du tubercule sur la croissance et le développement de la plante. In
Séminaire international, INRA, Pointe-à-Pitre, Guadeloupe.
Page | 147
Clifford D., Cressie N., England, J. R., Roxburgh S. H., Paul K. I. 2012. Correction factors for
unbiased, efficient estimation and prediction of biomass from log–log allometric models.
Forest Ecology and Management 310: 375-381.
Cooper C. S., Qualls M. 1967. Morphology and chlorophyll content of shade and sun leaves of
two legumes. Crop Science 7: 672-673.
Copeland E. B. 1916. Growth phenomena of Dioscorea. Philippine Journal of Crop Science 11: 227-41.
Cornet D., Amadji F., Dossou R.A., Maliki R., Vernier P. 2005a. Intérêts des légumineuses
herbacées pour une production durable d'igname en Afrique de l'Ouest. In Proceedings of
the forty first annual meeting of the Caribbean Food Crops Society (Farant M., Anais G., Ozier-
Lafontaine H., Zebus M.F., Diman J.L., Hammouya D., eds.), CFCS, Guadeloupe,
French West Indies.
Cornet D., Hammouya D., Bonhomme R. 2005b. Importance of yam crop physiology studies for
a better use of chemical manure. In Proceedings of the forty first annual meeting of the Caribbean
Food Crops Society (Farant M., Anais G., Ozier-Lafontaine H., Zebus M.F., Diman J.L.,
Hammouya D., eds.), CFCS, Guadeloupe, French West Indies.
Cornet D., Vernier P., Amadji F., Asiedu R. 2006. Integration of yam in cover crop-based
cropping system: constraints and potential. In Roots and tubers for sustainable development:
issues and strategies. Proceeding of the 14th triennial symposium of ISTRC (CTCRT, ed.),
Thiruvananthapuram, India.
Cornet D., Sierra J., Bonhomme R. 2007. Characterization of the photosynthetic pathway of
some tropical food yams (Dioscorea spp.) using leaf natural 13C abundance. Photosynthetica
45: 303-305.
Cornet D., Sierra J., Tournebize R., Ney B. 2014. Yams (Dioscorea spp.) plant size hierarchy and
yield variability: Emergence time is critical. European Journal of Agronomy 55: 100-107.
Cornet D., Sierra J., Tournebize R. 2015. Assessing allometric models to predict vegetative
growth of yams in different environments. Agronomy Journal 107(1) : 241-248.
Cornet D., Sierra J., Tournebize R., Gabrielle B., Lewis F.I. 2014. Bayesian network modeling of
early growth stages explains yam interplant yield variability in West Africa. Submitted to
Field Crops Research.
Page | 148
Coursey D.G. 1967. Yams. An account of the nature, origins, cultivation and utilisation of the
useful members of the Dioscoreaceae. Longmans, Green and Co Ltd, London.
Coursey DG. 1976. The origins and domestication of yams in Africa. In Origins of African
domestication (Harlan JR., ed.), Mouton Publisher, The Hague, Nederland.
Craufurd P.Q., Summerfield R.J., Asiedu R., Vara Prasad P. V. 2001. Dormancy in Yams.
Experimental Agriculture 37: 147-181.
Damgaard C., Weiner J. 2000. Describing inequality in plant size or fecundity. Ecology, 81(4):
1139-1142.
Damgaard C., Weiner J., Nagashima H. 2002. Modelling individual growth and competition in
plant populations: growth curves of Chenopodium album at two densities. Journal of Ecology
90: 666-671.
Dansi A., Mignouna H. D., Zoundjekpon J., Sangare A., Asiedu R., Quin F. M. 1999.
Morphological diversity, cultivar groups and possible descent in the cultivated yams
(Dioscorea cayenensis/D. rotundata) complex in Benin Republic. Genetic Resources and Crop
Evolution 46: 371-388.
Deaton M.L., Winebrake J.J. 2000. Modeling mobile source air pollution inventories. In Dynamic
Modeling of Environmental Systems (Deaton M.L., Winbrake J.J., Eds.), Springer-Verlag, New
York.
Degras L. 1986. L’igname, plante à tubercule tropicale. Maisonneuve et Larose, Paris, France
Den Dubbelden K. C., Verburg R. W. 1996. Inherent allocation patterns and potential growth
rates of herbaceous climbing plants. Plant and Soil 184: 341-347.
DFID 2014. Giving seed-yams the credit they deserve. Project CPP25.
[Link] (last accessed 21
March 2014).
Diby L.N., Hgaza V.K., Tie T.B., Assa A., Carsky R., Girardin O., Frossard E. 2009. Productivity
of yams (Dioscorea spp.) as affected by soil fertility. Journal of Animal and Plant Sciences 5(2):
494-506.
Page | 149
Di Giusto B., Anstett M.C., Dounias E., Doyle B.M. 2001. Variation in the effectiveness of biotic
defence: the case of an opportunistic ant-plant protection mutualism. Oecologia 129: 367-
375.
Dognimeton S., Koutoua A., Sidiky B., Daouda D., Kouadio Y.J., Dago G. 2014. Can Kponan
yam (Dioscorea cayenensis) full season tuber sprout if planted? Journal of Agricultural Science and
Technology, B, 4: 312-320.
Doku E.V. 1985. Production potentielle des principales plantes tropicales à racines et à
tubercules. In Plantes-racines tropicales : Culture et emplois en Afrique (Terry E.R., Doku E.V.,
Arene O.B., Mahungu N.M., eds). IDRC, Ottawa, Canada.
Dumont R., Dansi A.A., Vernier P., Zoundjihèkpon J. 2005. Biodiversité et domestication des
ignames en Afrique de l'Ouest. Pratiques traditionnelles conduisant à Dioscorea rotundata
Poir. Coll. Repères, CIRAD-IPGRI, Montpellier, France.
Doumbia S., Tshiunza M., Tollens E., Stessens J. 2004. Rapid spread of the Florido yam variety
(Dioscorea alata) in Ivory Coast. Introduced for the wrong reasons and still a success.
Outlook Agriculture 33(1): 49-54.
Elias M., Penet L., Vindry P., Mckey D., Panaud O., Robert T. 2001. Unmanaged sexual
reproduction and the dynamics of genetic diversity of a vegetatively propagated crop
plant, cassava (Manihot esculenta Crantz), in a traditional farming system. Molecular Ecology
10: 1895-1907.
Ellison A.M., Rabinowitz D. 1989. Effect of plant morphology and emergence time on size
hierarchy formation in experimental populations of two varieties of cultivated peas (Pisum
sativum). American Journal of Botany 76(3): 427-436.
Enyi B.A.C. 1973. Growth, development and yield of some tropical root crops. In Proceedings of the
3rd International Symposium of the ISTRC (Leakey C.L.A., ed.), IITA, Ibadan, Nigeria.
FAO-ISRIC-ISSS 1998. World reference base for soil resources. World Soil Resources. Rep. No.
84. Rome, Italy.
Page | 150
Fayaud B., Coste F., Corre-Hellou G., Gardarin A., Dürr C. 2014. Modelling early growth under
different sowing conditions: A tool to predict variations in intercrop early stages. European
Journal of Agronomy 52: 180-190.
Ferguson T. U., Haynes P. H., Springer B. G. F. 1969. A study of variability in yams (Dioscorea
spp.). In Caribbean Food Crops Society, Vol. 6, 50-58, University of the West Indies, St
Augustine, Trinidad.
Ferguson T. U. 1973. The effect of sett characteristics and spacing on growth, development and
yield of yams (Dioscorea spp). PhD Thesis, the University of the west Indies St Augustine,
West Indies.
Ferguson T. U., Gumbs F.A. 1976. Effect of soil compaction on leaf number and area, and tuber
yield of White Lisbon yam. In Proceedings of the 4th International Symposium of the ISTRC
(Cock J., Macintyre R., Graham M., eds.), CIAT, Cali, Colombie.
Firestone S.M., Lewis F.I., Schemann K., Ward M.P., Toribio J.A.L.M.L., Dhand N.K. 2013.
Understanding the associations between on-farm biosecurity practice and equine
influenza infection during the 2007 outbreak in Australia. Preventive Veterinary Medicine 110:
28-36.
Flach M. 1979. Ecological competition among the main moisture rich starchy staples in the
tropics and subtropics. In Proceedings of the 5th International Symposium of the ISTRC" (Belen
E.H., Villanueva M., eds.), ViSCA, Los Banos, Laguna, Philippines.
Friedman N., Goldszmidt M., Wyner A., 1999. Data analysis with Bayesian networks: a bootstrap
approach. In Proceedings of the Fifteenth Conference on Uncertainty in Artificial Intelligence
(Kaufmann M., ed.), San Francisco, USA.
Friedman N., Koller D. 2003. Being Bayesian about network structure. A Bayesian approach to
structure discovery in Bayesian networks. Machine Learning 50: 95-125.
George J. 1992. Leaf area estimation in dwarf white yam (Dioscorea rotundata) by linear
measurements. Journal of Root Crops 18: 134-135.
Gianoli E. 2002. Phenotypic responses of the twining vine Ipomoea purpurea (Convolvulaceae) to
physical support availability in sun and shade. Plant Ecology 165: 21-26.
Glasser,G. J. 1962. Variance formulas for the mean difference and coefficient of concentration.
Journal of the American Statistical Association 57: 648-654.
Page | 151
Goenaga R., Irizarry H. 1994. Accumulation and partitioning of dry matter in water yam.
Agronomy Journal 86: 1083-1087.
Groemping U. 2006. Relative Importance for Linear Regression in R: The Package relaimpo.
Journal of Statistical Software 17: 1-27.
Gumbs F. A., Ferguson T. U. 1976. Effect on tuber yield of the exposure of yam to light.
Agronomy Journal 68: 831-832.
Heckerman D., Geiger D., Chickering D.M. 1995. Learning Bayesian networks - The
combination of knowledge and statistical data. Machine Learning 20: 197-243.
Hgaza V.K., Diby L.N., Assa A., Ake S. 2010. How fertilization affects yam (Dioscorea alata L.)
growth and tuber yield across the years. African Journal of Plant Science 4: 53-60.
Hgaza V.K., Diby L.N., Tié T.B., Tschannen A., Aké S., Assa A., Frossard E. 2011. Growth and
Distribution of roots of Dioscorea alata L. do not respond to mineral fertiliser application.
The Open Plant Science Journal 5: 14-22.
Hgaza V.K., Diby L.N., Oberson A., Tschannen A., Tié T.B., Sangakkara U.R., Aké S., Frossard
E. 2012. Nitrogen use by yam as affected by mineral fertilizer application. Agronomy
Journal 104: 1558-1568.
Hunt R. 1982. Plant Growth Curves. London Great Britain: Edward Arnold Publisher.
IITA, 2014. Project YIISWA: Yam improvement for income and food security in West Africa.
[Link] (last accessed 21 March 2014).
Ile E. I., Craufurd P. Q., Asiedu R., Battey N. H. 2007. Duration from vine emergence to
flowering suggests a long-day or rate of change of photoperiod response in white yam
(Dioscorea rotundata Poir.). Environmental and Experimental Botany 60: 86-94.
Irvine F.R. 1969. West African Crops. Oxford University Press, U.K.
James L.J. 1953. Study of the growth and development of yams - Dioscorea alata Linn. - in relation
to planting methods. PhD Thesis, University of the West Indies, St Augustine, Trinidad.
Page | 152
Johnson J. B., Omland K. S. 2004. Model selection in ecology and evolution. Trends in Ecology and
Evolution 19: 101-108.
Johnston M., Onwueme I.C. 1998. Effect of shade on photosynthetic pigments in the tropical
root crops: yam, taro, tannia, cassava and sweet potato. Experimental Agriculture 34: 301-
312.
Jourdan C., Chopart J.L. 1998. Methods for studying root systems in the field at CIRAD. In
Congrès Mondial des Sciences du Sol, CIRAD, Montpellier, France.
Justel, A., Peña, D., Zamar, R. 1997. A multivariate Kolmogorov-Smirnov test of goodness of fit.
Statistics and Probability Letters, 35: 251-259.
Kandiannan K., Kailasam C., Chandaragiri K. K., Sankaran N. 2002. Allometric model for leaf
area estimation in black pepper (Piper nigrum L.). Journal of Agronomy and Crop Science 188:
138-140.
Kayode G.O. 1984. Effects of sett size and spacing on tuber yield of white Guinea yam (Dioscorea
rotundata) in the rainforest and savanna zones of Nigeria. Experimental Agriculture 20: 53-
57.
Kennedy D. 2003. Agriculture and the Developing World. Science 302: 357.
Keisling T. C. 1982. Calculation of the length of day. Agronomy Journal 74: 758-759.
Kleih U., Phillips D., Mignouna D., Ogbonna M., Siwoku, B. 2012. Nigeria – Scoping Yam Value
Chain Analysis. In: Yam Improvement for Income and Food Security in West Africa, Final Report.
IITA-NRI-NRCRI.
Klemke T., Moll A. 1990. Model for simulation of potato growth from planting to emergence.
Agricultural Systems 32: 295-304.
Koivisto M., Sood K. 2004. Exact Bayesian structure discovery in Bayesian networks. Journal of
Machine Learning Research 5: 549-573.
Koller H. R. 1972. Leaf area-leaf weight relationships in the soybean canopy. Crop Science, 12: 180-
181.
Korb K.B., Nicholson A.E. 2004. Bayesian Artificial Intelligence. Chapman and Hall/CRC, Boca
Raton, USA.
Page | 153
Kpeglo K.D., Obigbesan G.O., Wilson J.E. 1982. Physiological studies on the white yam,
Dioscorea rotundata Poir.: 2. Influence of plant age and leaf position on the translocation of
assimilates. Beitrage zur Tropischen Landwirtschaft und Veterinarmediziz 20: 23-29.
Lacointe A. 1984. Quelques aspects du comportement d'une plante tubérifère tropicale (igname)
issue de la culture in vitro, en plantations échelonnées. Doctorat de 3ème cycle spécialité
Physiologie Végétale, Université de Clermont II U.E.R. Sciences exactes et naturelles.
Leith H. 1974. Purpose of a phenology book. In Phenology and seasonality modelling (Leith H., ed.),
Chapman and Hall, London, UK.
Lewis F.I., Mc Cormick B.J.J. 2012. Revealing the Complexity of Health Determinants in
Resource-poor Settings. American Journal of Epidemiology 176: 1051-1059.
Lewis F.I., Ward M.P., 2013. Improving epidemiologic data analyses through multivariate
regression modelling. Emerging Themes in Epidemiology 10: 4-14.
Lindeman R. H., Merenda P. F., Gold R. Z. 1980. Introduction to Bivariate and Multivariate
Analysis. Glenview IL: Scott, Foresman.
Lomax R.G, Hahs-Vaughn D.L. 2012. Statistical concepts: A second course for education and
the behavioral sciences, Lawrence Erlbaum Associates Publishers, Mahwah, NJ, USA.
Lu H. Y., Lu C. T., Wei M. L., Chan L. F. 2004. Comparison of Different Models for
Nondestructive Leaf Area Estimation in Taro. Agronomy Journal 96: 448-453.
Mackerron D.K.L., Marshall B., Jefferies R.A. 1988. The distributions of tuber sizes in droughted
and irrigated crops of potato. II. Relation between size and weight of tubers and the
variability of tuber-size distributions. Potato Research 31: 279-288.
Marcos J., Lacointe A., Tournebize R., Bonhomme R., Sierra J. 2009. Water yam (Dioscorea alata
L.) development as affected by photoperiod and temperature: Experiment and modeling.
Field Crops Research 111: 262-268.
Marcos J., Cornet D., Bussière F., Sierra J. 2011. Water yam (Dioscorea alata L.) growth and yield
as affected by the planting date: Experiment and modeling. European Journal of Agronomy
34: 247-256.
Page | 154
Marnotte P. 1984. Influence des facteurs agro-écologique sur le développement des mauvaises
herbes en climat tropical humide. In 7ème colloque international sur la biologie des mauvaises herbes,
du 9 au 11 octobre 1984, Paris, France.
Martin K.L., Hodgen P.J., Freeman K.W., Melchiori R., Arnall D.B., Teal R.K., Mullen R.W.,
Desta K., Phillips S.B., Solie J.B., Stone M.L., Caviglia O., Solari F., Bianchini A., Francis
D.D., Schepers J.S., Hatfield J.L., Raun W.R. 2005. Plant-to-plant variability in corn
production. Agronomy Journal 97: 1603-1611.
Matsui A., Kanehara M. 2006. The Question of Prehistoric Plant Husbandry during the Jomon
Period in Japan. World Archaeology 38: 259-273.
McCarthy M. C., Enquist B. J. 2007. Consistency between an allometric approach and optimal
partitioning theory in global patterns of plant biomass allocation. Functional Ecology 21:
713-720.
McConvell P. 1990. The Linguistic Prehistory of Australia: Opportunities for Dialogue with
Archaeology. Australian Archaeology 31: 3-27.
McKey D., Di Giusto B., Pascal L., Elias M., Dounias E. 1998. Stratégies de croissance et de
défense anti-herbivore des ignames sauvages : leçons pour l’agronomie. In L’igname, plante
séculaire et culture d’avenir. Actes du séminaire international Cirad-Inra-Orstom-Coraf (Berthaud J.,
Bricas N., Marchand J.L., eds.), CIRAD, 3-6 juin 1997, Montpellier, France.
McKey D., Elias M., Pujol B., Duputié A. 2010. The evolutionary ecology of clonally propagated
domesticated plants. New Phytologist 186: 318-332.
McKey D., Elias M., Pujol B., Duputié A. 2012. Ecological Approaches to Crop Domestication.
In Biodiversity in Agriculture: Domestication, Evolution, and Sustainability (Gepts P., Famula
T.R., Bettinger R.L., Brush S.B., Damania A.B., McGuire P.E., Qualset C., eds),
Cambridge University Press, New York, USA.
Moody K., Ezumah H.C. 1974. Weed control in major tropical root and tuber crops, a review.
International Journal of Pest Management 20: 292-299.
Moorby J. 1970. The production, storage and translocation of carbohydrates in developing potato
plant. Annals of Botany 34: 297-308.
Mullen K., Ardia D., Gil D., Windover D., Cline, J. 2011. 'DEoptim': An R Package for Global
Optimization by Differential Evolution. Journal of Statistical Software 40: 1-26.
Page | 155
Müller I., Schmid B., Weiner J. 2000. The effect of nutrient availability on biomass allocation
patterns in 27 species of herbaceous plants. Perspectives in Plant Ecology Evolution and
Systematics 3: 115-127.
Nafziger E.D., Carter P.R., Graham E.E. 1991. Response of corn to uneven emergence. Crop Science
31: 811-815.
Nash, J. E., Sutcliffe J. V. 1970. River flow forecasting through conceptual models part I - A
discussion of principles. Journal of Hydrology 10: 282-290
Naylor R.L., Falcon W.P., Goodman R.M., Jahn M.M., Sengooba T., Tefera H., NelsonR.J. 2004.
Biotechnology in the developing world: a case for increased investments in orphan crops.
Food Policy 29: 15-44.
Needham C.J., Bradford J.R., Bulpitt A.J., Westhead D.R. 2007. A primer on learning in Bayesian
networks for computational biology. PLoS Computational Biology 3: 1409-1416.
Nelder J. A., Mead R. 1965. A simplex method for function minimization. Computer Journal 7: 308-
313.
Niklas K. J. 2004. Plant allometry: is there a grand unifying theory? Biological Reviews 79: 871-889.
Nin-Pratt A., Johnson M., Magalhaes E., You L., Diao X., Chamberlin J. 2011. Yield gaps and
potential agricultural growth in West and Central Africa. IFPRI Research Monograph,
Washington DC.
Njoku J.E. 1963. The propagation of yams (Dioscorea spp.) by vine cuttings. Journal of the West
African Sciences Association 8: 29-32.
Njoku J.E., Oyolu C., Okonkwo S.N.C., Nwoke F.I.O. 1973. The pattern of growth and
development in Dioscorea rotundata Poir. In Proceedings of the 3rd International Symposium of the
ISTRC (Leakey C.L.A., ed.), IITA, Ibadan, Nigeria.
Njoku J.E., Nwoke F.I.O., Okonkwo S.N.C. 1984. Pattern of growth and development in
Dioscorea rotundata Poir. Tropical Agriculture 61: 17-19.
Normand F., Lauri P. E. 2012. Assessing allometric models to predict vegetative growth of
mango (Mangifera indica; Anacardiaceae) at the current-year branch scale. American Journal
of Botany 99: 425-437.
Page | 156
Nwinyi S.C.O. 1981. Preliminary evaluation of soils of the major yam growing areas in Nigeria.
Report. NRCRI, Umudike, Imo State, Nigeria.
Nwoke F.I.O., Njoku E., Okonkwo S.N.C. 1973. The effect of size of seed yams on yield of
individual plants of Dioscorea rotundata. In Proceedings of the 3rd International Symposium of the
ISTRC (Leakey C.L.A., ed.), IITA, Ibadan, Nigeria.
Obigbesan G.O. 1977. Nutritional problems in Root Crops production. In Proceedings of the first
national seminar on root and tubers (NRCRI, ed.), Umudike, Nigeria.
O’Brien P.J., Allen E.J., Bean J.N., Griffith R.J., Jones S.A, Jones J.L. 1983. Accumulated day
degrees as a measure of physiological age and the relationships with growth and yield in
early potato varieties. Journal of Agricultural Science, Cambridge 101: 613-631.
Ogbonna M.C., Korieocha D.S., Onyenobi V.O., Njoku S.C. 2011. Profitability of minituber seed
yam production technique in south east agro-ecological zone: evidence from Abia state,
Nigeria. Journal of Agriculture and Social Research 11: 113-119.
Ohiri A.C., Nwokoye J.U. 1983. Soil physical and chemical properties suitable for yam (Dioscorea
rotundata) production in southeastern Nigeria. In Proceedings of the 6th International Symposium
of the ISTRC (Shideler F.S., Rincon H., eds.), CIP, Lima, Peru.
Okezie C.E.A., Nwoke F.I.O., Okonkwo S.N.C. 1986. Field studies on the growth pattern of
Dioscorea rotundata Poir. propagated by seed. Tropical Agriculture 63: 22-24.
Okoli O. O., Opara M. U., Anyaoha C. O. 1999. Effect of seed weight on yield determinant, yield
components and intra-plot variability in yield of yams (Dioscorea spp). African journal of
root and tuber crops 3: 44-48.
Olivier K. A., Konan K. N., Anike F. N., Agbo G. N., Dodo H. W. 2012. In vitro induction of
minitubers in yam (Dioscorea cayenensis-D. rotundata complex). Plant Cell, Tissue and Organ
Culture 109: 179-189.
Onwueme I.C. 1972. Influence of weight of the planted tuber on the vegetative performance of
white yam (Dioscorea rotundata, Poir.) plants. Nigerian Agricultural Journal 9: 170-173.
Onwueme I. C. 1973. The sprouting process in yam (Dioscorea spp.) tuber pieces. Journal of
Agricultural Science 81: 375-379.
Page | 157
Onwueme I.C. 1975. Influence of storage time on earliness of sprouting and tubering in Dioscorea
rotundata yams. Journal of Agricultural Science 84: 503-5.
Onwueme I.C. 1976. Performance of yam (Dioscorea spp.) setts planted without water. Journal of
Agricultural Science 87: 413-415.
Onwueme I.C. 1977. Field comparison of West African planting and harvesting practices in yam
(Dioscorea rotundata): pre-sprouting, dry season planting, and double harvesting. Journal of
Agricultural Science 88: 311-318.
Onwueme I.C. 1978a. Sett weight effects on time of tuber formation and on tuber yield
characteristics in water yam (Dioscorea alata L.). Journal of Agricultural Science 91: 317-9
Onwueme I.C. 1978b. The tropical tuber crop: yams, cassava, sweet potato, cocoyams. John
Wiley and sons, New York, USA.
Onwueme I.C. 1979. Tuber physiology in yams (Dioscorea spp) and its agricultural implications. In
Proceedings of the 5th International Symposium of the ISTRC (Belen E.H., Villanueva M., eds.),
ViSCA, Los Banos, Laguna, Philippines.
Onwueme I.C., Haverkort J.A. 1991. Modelling growth and productivity of yams (Dioscorea spp):
Prospects and problems. Agricultural Systems 36: 351-367.
Onwueme I.C., Charles W.B. 1994. Tropical root and tuber crops. Production, perspectives and
future prospects. FAO, Rome, Italia.
Onwueme I.C., Johnston M. 2000. Influence of shade on stomata density, leaf size and other leaf
characteristics in the major tropical root crops, tania, sweet potato, yam, cassava and taro.
Experimental Agriculture 36: 509-516.
Oriuwa L.O., Onwueme I.C. 1980. Determining the optimum spacing and set weight for growing
yam (Dioscorea spp.) without stakes. Quarterly journal of International Agriculture 19: 5-24.
Orkwor G.C., Asiedu R., Ekanayake I.J. 1998. Food Yams: Advances in Research, IITA, Ibadan
and NRCRI, Umudike, Nigeria.
Patetsos E. 2001. Etude du comportement de quatre variétés d’igname Dioscorea alata en fonction
de contraintes agropédoclimatiques différentes en Guadeloupe. [Link]. Thesis. Ecole
d’Agro-Economie Internationale, Cergy-Pontoise, France.
Page | 158
Plummer M. 2003. JAGS: A Program for Analysis of Bayesian Graphical Models Using Gibbs
Sampling, In Proceedings of the 3rd International Workshop on Distributed Statistical Computing,
March 20-22, Vienna, Austria.
Poon A.F.Y., Lewis F.I., Pond S.L.K., Frost S.D.W. 2007. Evolutionary interactions between N-
linked glycosylation sites in the HIV-1 envelope. PLoS Computational Biology 3(1): e11.
Poorter H., Niklas K. J., Reich P. B., Oleksyn J., Poot P., Mommer L. 2012. Biomass allocation to
leaves, stems and roots: meta-analyses of interspecific variation and environmental
control. New Phytologist 193: 30-50.
Porth I., Klapste J., Skyba O., Friedmann M.C., Hannemann J., Ehlting J., El-Kassaby Y.A.,
Mansfield S.D., Douglas C.J. 2013. Network analysis reveals the relationship among
wood properties, gene expression levels and genotypes of natural Populus trichocarpa
accessions. New Phytologist 200: 727-742.
Price K. V., Storn R. M., Lampinen J. A. 2006. Differential Evolution - A Practical Approach to Global
Optimization. Berlin Heidelberg: Springer-Verlag, Berlin, Germany.
Pujol B., Mühlen G., Garwood N., Horoszowski Y., Douzery E.J.P., McKey D. 2005. Evolution
under domestication: contrasting functional morphology of seedlings in domesticated
cassava and its closest wild relatives. New Phytologist 166: 305-318.
Ravi V., Chowdhury R. S. 1989. A method for leaf area determination in Dioscorea. Journal of Root
Crops 15: 45-48.
R Development Core Team 2013. R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. [Link] (last
accessed 21 March 2014).
Rebetzke G.J., Richards R.A., Fettell N.A., Long, M., Condon, A.G., Botwright, T.L. 2007.
Genotypic increases in coleoptile length improves stand establishment, vigour and grain
yield of deep-sown wheat. Field Crops Research 11: 10-23.
Reddy V. R., B. Acock D. N. Baker, Acock M. 1989. Seasonal leaf area-leaf weight relationship in
the cotton canopy. Agronomy Journal 81: 1-4.
Page | 159
Rodriguez-Montero W., Hilger T.H., Leihner D.E. 2001. Effects of seed rates and plant
populations on canopy dynamics and yield in the greater yam (Dioscorea alata L.). Field
Crops Research 70: 15-26.
Rousselle P., Robert Y., Crosnier J.C. 1996. La pomme de terre, production, amélioration,
ennemis et maladies, utilisations. INRA-ITCF, Paris, France.
Rykbost K.A., Locke K.A. 1999. Effect of Seed Piece Size on Performance of Three Potato
Varieties in the Klamath Basin of Oregon. American Journal of Potato Research 75: 75-82.
Sadras V., Bongiovanni R. 2004. Use of Lorenz curves and Gini coefficients to assess yield
inequality within paddocks. Field Crops Research 90: 303-310.
Sands P. J., Hackett C, Nix H. A. 1979. A model of the development and bulking of potatoes
(Solanum tuberosum L). 1. Derivation from well-managed field crops. Field Crops Research 2:
309-331.
Sartie A., Franco J., Asiedu R. 2012. Phenotypic analysis of tuber yield- and maturity-related traits
in white yam (Dioscorea rotundata). African Journal of Biotechnology 11: 3964-3975.
Schneider C.A., Rasband W.S., Eliceiri K.W. 2012. NIH Image to ImageJ: 25 years of image
analysis. Nature Methods 9: 671-675.
Scott G.J., Rosegrant M.W., Ringler C. 2000. Global projections for root and tuber crops to the
year 2020. Food Policy 25: 561-597.
Sestak Z., J. Catsky, Jarvis P. G. 1971. Plant photosynthetic production. Manual of methods.
Junk, The Hague.
Shipley B., Meziane D. 2002. The balanced-growth hypothesis and the allometry of leaf and root
biomass allocation. Functional Ecology 16: 326-331
Shiwashi H., Ayankanmi T., Asiedu R. 2002. Effect of daylength on the development of tubers in
yams (Dioscorea spp.). Tropical Science 42: 162-170.
Shiwachi H., Kikuno H., Fashola R., Asiedu R. 2008. Enhancing off-season yam production in
Nigeria. African Journal of Agricultural Research 3: 407-411.
Page | 160
Sinoquet H., Andrieu B. 1993. The geometrical structure of plant canopies: characterization and
direct measurement methods. In Crop structure and light microclimate, characterization and
applications (Varlet-Grancher C., Bonhomme R., Sinoquet H., eds.), INRA, Paris.
Smith H.F. 1938. An empirical law describing heterogeneity in the yields of agricultural crops.
Journal of Agricultural Science 28: 1-23.
Sobulo R.A. 1972. Studies on white Yam (Dioscorea rotundata) I- Growth analysis. Experimental
Agriculture 8: 99-106.
Sprugel D. G. 1983. Correcting for bias in log-transformed allometric equations. Ecology 64: 209-
210.
Storck L. 2011. Partial collection of data on potato yield for experimental planning. Field Crop
Research 121: 286-290.
Storn R., Price, K. 1997. Differential Evolution – A simple and efficient heuristic for global
optimization over continuous spaces. Journal of Global Optimization 11: 341-359.
Tierney L., Kadane J.B. 1986. Accurate approximations for posterior moments and marginal
densities. Journal of the American Statistical Association 81: 82-86.
Tixier P., Malezieux E., Dorel M. 2004. SIMBA-POP: a cohort population model for long-term
simulation of banana crop harvest. Ecological Modelling 180: 407-417.
Treche S. 1989. Potentialités nutritionnelles des ignames (Dioscorea spp.) cultivées au Cameroun.
PhD Thesis, Université de Montpellier II, Paris.
Treche S., Guion P. 1979. Etude des potentialités nutritionnelles de quelques tubercules
tropicaux au Cameroun : I. Influence de la maturité à la récolte. Revue Science et Technique 1:
55-71.
Trouslot M.F. 1983. Analyse de la croissance de l'igname Dioscorea complexe cayenensis rotundata.
Thèse de Docteur es Science. Serie E N° 326, Université de Clermont Ferrand II,
Clermont Ferrand.
Uduebo A.E. 1970. Physiological studies on the proliferation and development of the axillary
tissue of Dioscorea bulbifera. L. Journal of West African Science Association 15: 57-62.
Page | 161
Vaillant V., Bade P., Constant, C. 2005. Photoperiod affects the growth and development of yam
plantlets obtained by in vitro propagation. Biologia Plantarum 49: 355-359.
Van der Zaag D. E., Van Loon C. D. 1987. Effect of physiological age on growth vigour of seed
potatoes of two cultivars. 5. Review of literature and integration of some experimental
results. Potato Research 30: 451-472.
Van Loon C. D. 1987. Effect of physiological age on growth vigour of seed potatoes of two
cultivars. 4. Influence of storage period and storage temperature on growth and yield in
the field. Potato Research 30: 441-450.
Varshney R. K., Ribaut J. M., Buckler E. S., Tuberosa R., Rafalski J. A., Langridge P. 2012. Can
Genomics Boost Productivity of Orphan Crops? Nature Biotechnology 30: 1172-1176.
Vega C.R.C, Sadras V.O. 2003. Size‐dependent growth and the development of inequality in
maize, sunflower and soybean. Annals of Botany 91: 795-805.
Vernier P., Dossou R.A., Letourmy P. 1999. La fabrication de cossettes à partir d'ignames
Dioscorea alata : influence de la variété et du type de cossette sur le séchage, la conservation
et les qualités organoleptiques. African journal of root and tuber crops 3: 62-67.
von Hippel P. 2005. Mean, median, and skew: Correcting a textbook rule. Journal of Statistics
Education 13(2).
Waitt A.W. 1965. A key to some Nigerian varieties of yam (Dioscorea spp.). Fed. Dept. Agric. Res.,
Nigeria.
Walter H., Harnickel E., Mueller-Dombois D. 1975. Climate Diagram Maps of the Individual
Continents and the Climatic Regions of the Earth. Springer-Verlag, New York.
Walther B.A., Moore J.L. 2005. The concepts of bias, precision and accuracy, and their use in
testing the performance of species richness estimators, with a literature review of
estimator performance. Ecography 28: 815-829.
Wang, E., Engel T. 1998. Simulation of phenological development of wheat crops. Agricultural
Systems 58: 1-24.
Page | 162
Ward M.P. 2013. Bayesian graphical modelling: applications in veterinary epidemiology. Preventive
Veterinary Medicine 110: 1-3.
WECARD 2011. Promotion of improved yam minisett technology to improve productivity and
reduce excessive use of food yam for planting in West Africa.
[Link]
D/ 2009-11/ (last accessed 21 March 2014).
Weiner J., Solbrig O.T. 1984. The meaning and measurement of size hierarchies in plant
populations. Oecologia 61, 334-336.
Weiner J. 1985. Size hierarchies in experimental populations of annual plants. Ecology 66: 743-752.
Weiner J., Thomas S.C. 1986. Size variability and competition in plant monocultures. Oikos 47:
211-222.
Werth M., Spiegel A. K., Kazda M. 2013. Leaf allocation patterns and 13C and 15N natural
abundances of tropical lianas (Passiflora sp.) as dependent on external climbing support.
Isotopes in Environmental and Health Studies 49: 98-108.
Wickham L.D., Wilson L.A., Passam H.C. 1997. The anatomical origin of the primary nodal
complex during vegetative propagation in yam (Dioscorea spp.) tubers. In Séminaire
international sur l’igname. Les colloques de l’Inra (INRA, ed.), Pointe-à-Pitre, Guadeloupe.
Wilson L.A. 1977. Root Crops. In Ecophysiology of tropical crops (Alvin P.T., Kozlowski T.T., eds.),
Academic Press, New York, USA.
Wright I.J., Clifford H.T., Kidson R., Reed M.L., Rice B.L., Westoby M. 2000. A survey of seed
and seedling characters in 1744 Australian dicotyledon species: cross-species trait
correlations and correlated trait-shifts within evolutionary lineages. The Biological Journal of
the Linnean Society 69: 521-547.
Wurr D.C.E., Fellows J.R., Lynn J.R., Allen E.J. 1993. The impact of some agronomic factors on
the variability of potato tuber size distribution. Potato Research 36: 237-245.
Yin X., Goudriaan J., Lantinga E.A., Vos J., Spiertz H.J. 2003. A flexible sigmoid function of
determinate growth. Annals of Botany 91: 361-371.
Page | 163
Zinsou C. 1998. Physiologie et morphogenèse de l'igname (Dioscorea spp.). In L'igname, plante
séculaire et culture d'avenir. Actes du séminaire international (Berthaud J., Bricas N., Marchand
J.L., eds.), Cirad Inra Orstom Coraf, Montpellier, France.
Page | 164
Page | 165
Table des figures
Figure 1. Production annuelle d’ignames dans le monde par pays (t) et par habitant
(kg habitant-1). ................................................................................................................................... 17
Figure 2. Tonnage (en millions de tonnes) et valeur (en millions de dollars) des principales
productions alimentaires végétales en Afrique de l’Ouest en 2012 (FAOSTAT 2014). ........ 18
Figure 3. Nombre de publications recensés dans le Web of Science © entre 1970 et 2014 pour
l’igname (Dioscorea alata et D. rotundata en bleu) et la pomme de terre (Solanum tuberosum en
rouge). ................................................................................................................................................ 19
Figure 4. Les lianes de D. rotundata utilisent tous les supports accessibles pour leur croissance. 20
Figure 6. Tubercules de D. rotundata cv. Laboko développés dans un sol meuble (à gauche) et
tassé (à droite). .................................................................................................................................. 22
Figure 8. Détails de la levée des premiers organes (à gauche) : (a) la tige, (b) le complexe nodal
primaire, (c) le pré-tubercule en stagnation, (d) semenceau, et (e) racine coronaire ; et
émergence de D. rotundata cv. Morokorou (au centre) et D. alata cv. Florido (à droite). ...... 24
Figure 9. Morphologie caulinaire de D. rotundata cv. Morokorou durant les premiers stades de
croissance........................................................................................................................................... 25
Figure 10. Excavation du système racinaire de D. rotundata cv. Gnidou, sept semaines après
plantation, à plat (à gauche) et en butte (à droite). ...................................................................... 26
Figure 11. Floraison chez l'igname : inflorescence mâle (à gauche), inflorescence femelle (au
centre) et graines (à droite). ............................................................................................................ 27
Figure 12. Installation d’une culture d’igname par défriche-brûlis après une jachère longue (> 15
ans). .................................................................................................................................................... 28
Figure 13. Buttes d'igname avant plantation en zone Bariba à Fo-Boure, Nord Bénin (à gauche)
et en zone Mahi à Ouédémé, Centre Bénin (à droite). ............................................................... 29
Page | 166
Figure 15. Réalisation des chapeaux sur les buttes d’igname, à partir des déchets végétaux trouvés
sur place. ............................................................................................................................................ 31
Figure 16. Marché saisonnier de bord de route à Dassa, Centre Bénin (à gauche) et marché de
Glazoué, Centre Bénin (à droite). .................................................................................................. 32
Figure 18. Transformation de l’igname en amala : (a) séchage, (b) vente de cossettes sur le
marché, (c) pilage des cossettes, (d) farines de cossette, (e) mélange de la farine de cossette
dans l’eau bouillante, et (a) pâte de farine de cossette (amala)................................................... 34
Figure 20. Culture associée d’igname et de Pueraria phaseoloides (a), installation d’une jachère
améliorée de P. phaseoloides (b), tuteurage vivant de l’igname avec du Gliricidia sepium (c), et
jachère améliorée de Mucuna pruriens (d)........................................................................................ 36
Figure 21. Courbes de Lorenz pour deux populations contrastées : une population où la
variabilité est principalement due à un petit nombre de très grands individus (rouge) et une
population où la variabilité est principalement due à un grand nombre de petits individus
(bleu). La ligne en pointillé représente l’absolue égalité entre individus. G’ est le coefficient
de Gini et S le coefficient d’asymétrie de Lorenz. ....................................................................... 40
Figure 22. Carte d’Afrique et localisation des sites d’étude sur la carte administrative du Bénin
(latitudes en ordonnées et longitudes en abscisses)..................................................................... 43
Figure 23. Climatogramme de Cotonou (en 2007 (a), 2008 (b), 2009 (c)) et de Glazoué (en 2007
(d), 2008 (e) et 2009 (f)). Les températures moyennes mensuelles et les cumuls de
précipitations mensuels sont mis en graphique avec un rapport d’échelle de 1:2 afin de
permettre une appréciation rapide des mois secs (aires en pointillés rouges), humides (aires
hachurée en bleu) et très humides (aires colorée en bleu). L’échelle des précipitations change
au-delà de 100 mm par mois. Les températures annuelles minimales et maximales sont
présentées à côté des ordonnées. La température moyenne annuelle et la pluviométrie totale
annuelle sont présentées en titre pour chaque climatogramme (Walter et al. 1975). .............. 45
Page | 167
racinaire (d), et étude de la dynamique racinaire de l'igname à l'aide des rhizotrons de terrain
(e). ....................................................................................................................................................... 49
Figure 25. Distribution of the weekly proportion of emergence for D. alata and D. rotundata
averaged over trials carried out from 2006 to 2009. Error bars indicate standard deviations
between years. ................................................................................................................................... 63
Figure 26. Plant tuber weight as a function of time of D. alata (in red) and D. rotundata (in blue)
for three different emergence classes: P0-20 is the first 20 percentile emergence class (solid
line), P40-60 is the 40th to 60th percentile class (dotted line) and P80-100 is last 20
percentile class (dashed line). Lines correspond to the fitted beta growth model. ................. 65
Figure 27. Frequency distribution of: plant tuber dry weight (a) and (b), number of tubers per
plant (c) and (d), and tuber dry weight (e) and (f) of early (P0-20 is the top 20 percentile
emergence class) and late (P80-100 is last 20 percentile emergence class) emergence class for
two yam species (D. alata and D. rotundata). Lines correspond to the adjusted empirical
probability density function for early (solid line) and late (dotted line) emergence classes. . 66
Figure 28. Graphical presentation of the dispersion of the main variables for D. alata (red) and
D. rotundata (blue): (A) the count of sprouted and unsprouted seed-tubers, (B) the seed-tuber
weight per mound, (C) the emergence date, (D) the number of main stems per mound, (E)
the number of nodes carrying cataphyll per mound, (F) the number of tubers per mound,
(G) the mean tuber fresh weight per mound, and (H) the plant fresh yield per mound. The
colored box corresponds to the inter-quartile range (IQR). The upper and lower whisker
extends from the box to the highest and lowest value that is within 1.5 * IQR. Data beyond
the end of the whiskers are outliers and plotted as points. ........................................................ 78
Figure 29. Morphology of yam (D. rotundata) early stage (from the planting date until appearance
of the first true leaf). ........................................................................................................................ 79
Figure 30. Directed acyclic graph of globally optimal multivariate regression model of D. alata
early growth variables on plant yield and yield components using an exact search additive
Bayesian model. Red arcs indicate a negative relationship between two variables while green
arcs denote a positive relationship. Dashed arcs represent arcs which were not recovered in
at least 50% of the directed acyclic graphs based on the 10000 bootstrapping replicates. ... 82
Figure 31. Number of arcs recovered in 10000 bootstrapping replicates for D. alata (red) and D.
rotundata (blue)................................................................................................................................... 83
Figure 32. Comparison of goodness of fit for different parent limits for D. alata (red) and D.
rotundata (blue)................................................................................................................................... 84
Page | 168
Figure 33. Directed acyclic graph of final best multivariate regression model of D. rotundata early
growth variables on plant yield and yield components using an exact search additive
Bayesian model. White nodes belong to the category of variables linked to practices at
planting, black nodes belong to variables of plant yield components and grey nodes belong
to variables linked to the plant’s early growth. Dashed arcs indicate a negative relationship
between two variables while solid arcs denote a positive relationship. Arcs are labelled with
the standardized median marginal posterior density and the frequency at which each arc was
recovered in bootstrapping in brackets. ........................................................................................ 85
Figure 34. Directed acyclic graph of final best multivariate regression model of D. alata early
growth variables on plant yield and yield components using an exact search additive
Bayesian model. White nodes belong to the category of variables linked to practices at
planting, black nodes belong to variables of plant yield components and grey nodes belong
to variables linked to the plant’s early growth. Dashed arcs indicate a negative relationship
between two variables while solid arcs denote a positive relationship. Arcs are labelled with
the standardized median marginal posterior density and the frequency at which each arc was
recovered in bootstrapping in brackets. ........................................................................................ 86
Figure 35. Forest plot of the median effect (dot) of the distribution of the marginal posterior for
D. rotundata (blue) and D. alata (red). Quantile-based 95% probability intervals are given
using horizontal lines. ...................................................................................................................... 87
Figure 36. Reserves content in seed-tubers of D. alata (red) and D. rotundata (blue). The
proportion (%) corresponds to the percentage of dry matter on a fresh weight basis and to
the percentage of nitrogen, phosphorus and potassium on a dry weight basis respectively.
The filled box corresponds to the inter-quartile range (IQR). The upper and lower whisker
extends from the box to the highest and lowest value that is within 1.5 * IQR. Data beyond
the end of the whiskers are outliers and plotted as points. ........................................................ 89
Figure 37. Statistical analysis (Student’s t-test for paired data) of the deviation between the
estimated and observed leaf area, leaf mass and stem mass. Estimates were based on the
final best models (pooled or species-specific) of this study and models from literature where
available. Vertical bars denote the bias and spreads denote 95% confidence intervals. ns =
non-significant, * significant at P < 0.05, ** significant at P < 0.01 and *** significant at P <
0.001. ................................................................................................................................................ 110
Figure 38. Diagnostic plots of allometric models to predict leaf area (A) and stem mass (B) for
the two main yam species (D. alata in red and D. rotundata in blue). (A) Main plot presents
the regression of estimated vs. actual values (black line and equation) of the pooled
allometric model for leaf area estimation from linear measurement, while the inset presents
the residuals of the model compared to the data estimated by the model. (B) Main plot
Page | 169
presents the regression of estimated vs. actual values (lines and equations) of the species-
specific allometric model for stem mass estimation from crop leaf area, while the inset
presents the residuals of the model compared to the data estimated by the model............. 112
Figure 39. The main plot (A) presents the bootstrapped (n=100) estimate of the parameter b of
the stem mass model (SM = a LAb) as a function of the sample size. The dashed grey lines
of the main plot represent the 99% confidence intervals of the mean of the conversion
factor for each sample size. The inset (B), presents the model bias as a function of the
sample size....................................................................................................................................... 113
Figure 40. D. alata (red) and D. rotundata (blue) selling price at the market of Glazoué (Benin)
depending on the size of the tuber (circle: tuber size between 0.8 and 2.5 kg; triangle: tuber
size over 2.5 kg). Error bars are standard deviation of mean price over 2007, 2008 and 2009.
........................................................................................................................................................... 126
Figure 41. Estimated against observed outputs of best model components: plant leaf area (a and
b) and plant yield (c and d) of the two main yam species: D. alata (in red) and D. rotundata (in
blue) as influenced by temperature and daylength from plant emergence. ........................... 129
Figure 42. Temperature (a), photoperiod (b) and estimated growth stage durations (c and d) as a
function of emergence date of the two main yam species (D. alata in red and D. rotundata in
blue). EM-TI is the vegetative phase, and TI-HA the tuberization phase. The density plot at
the top of the figure presents the density function of the observed emergence dates of both
yam species within the six experiments of the current study................................................... 130
Figure 43. Influence of the planting date on the estimated yield differences between improved
(solid line) and traditional (dashed line) cropping systems of the two main yam species (D.
alata in red and D. rotundata in blue). The black arrow indicates the traditional planting date
in West Africa. ................................................................................................................................ 131
Figure 44. Yield (t of tuber dry matter per ha) of the traditional cropping system (a and b) and
yield improvement linked to the use of improved planting material (c and d) for the two
main yam species, D. alata (in red) and D. rotundata (in blue) as a function of planting date
and seed-tuber weight. Contour lines represent level of equal yield (a and b) or yield
improvement (c and d). ................................................................................................................. 132
Figure 45. Profitability (k€ ha-1) of the traditional cropping system for D. alata (a) and D.
rotundata (b) and profitability improvement linked to the use of improved planting material
for D. alata (c) and D. rotundata (d) as a function of planting date and seed-tuber weight.
Contour lines represent level of equal profitability (a and b) and profitability improvement (c
and d). .............................................................................................................................................. 133
Page | 170
Page | 171
Liste des tableaux
Tableau 1. Principales caractéristiques pédologiques des deux sites d’expérimentation de 2007 à
2009. ................................................................................................................................................... 46
Table 4. Individual yield inequality statistics: number of samples (N), coefficient of variation
(CV), unbiased Gini coefficient (G’) and Lorenz asymmetry coefficient (S). Yield values
having different letters are significantly different at the 5% level ............................................. 62
Table 5. Summary of individual plant weight statistics over time. DAE is the number of days
after 50% emergence, N is the number of observations and CV is the coefficient of
variation ............................................................................................................................................. 63
Table 6. Beta growth function parameters fitted to emergence classes and yam species. P0-20 is
the first 20 percentile class, P20-40 is the 20th to 40th percentile class, P40-60 is the 40th to
60th percentile class, P60-80 is the 60th to 80th percentile class, P80-100 is last 20 percentile
class. Wmax is the maximum value of tuber weight, te is the time at maximum tuber weight
and tm is the time at maximum growth rate. Model parameters with different letters are
significantly different at the 5% level for a given species. .......................................................... 64
Table 7. Explanatory variables used to build the Bayesian model explaining plant yield. ............ 77
Table 8. Marginal posterior quantiles for each effect parameter (precision and intercept terms
not shown) in the final best model associating early growth variables with yield components
of D. alata and D. rotundata. ............................................................................................................. 88
Table 9. Experiment characteristics of each data set used in this study. ....................................... 101
Table 10. Summary (minimal, maximal and mean ± SD values) of the calibration and validation
datasets used to build predictive allometric models to assess vegetative growth of the two
main species of yam. ...................................................................................................................... 106
Page | 172
Table 11. Candidate allometric models ranked by performance metrics: Akaike Information
Criteria (AIC), mean percent standard error ( S % ), adjusted coefficient of determination (R²)
and, the model bias ( b% ); L is the maximum length, M is the midrib length, W is the
maximum width, a and b are model parameters. ....................................................................... 107
Table 12. Summary of the relative influence of the experimental factors on the allometric
models. Bootstrap confidence intervals for relative importance were based on 1000 samples.
Differences in model performance were calculated between the minimal adequate model
and the equivalent pooled regression (i.e. ΔR²: difference in adjusted coefficient of
determination, ΔAIC: difference in Akaike Information Criteria, Δ S % : difference in mean
percent standard error and, Δ b% : difference in the model bias). L is the maximum length, M
is the midrib length and W is the maximum width. .................................................................. 108
Table 13. Summary of final model parameter values (mean and confidence interval, CI) and
performance at calibration and validation stage. Performance statistics: the adjusted
coefficient of determination (R²), mean percent standard error ( S % ) and, the model bias (
b% ). .................................................................................................................................................. 110
Table 14. Characteristics of the plant populations generated for the simulation study ............... 125
Table 15. Comparison of global and species-specific model’s performances ............................... 127
Table 16. Parameter estimates for the selected species-specific models during vegetative and
tuberization phases......................................................................................................................... 128
Page | 173