ISSN: 2463-0861 (En línea)
XXX versión virtual
XXX - SIE 2021
Septiembre 21 - 24 de 2021
Local control of soil heterogeneity using records from precision agriculture
programs
Carlos Alberto Martínez
[email protected] Corporación Colombiana de Investigación Agropecuaria - Agrosavia. Sede Central. Km. 14, vía
Mosquera - Bogotá, Mosquera - Cundinamarca, Colombia
Oscar Orlando Melo
[email protected]Departamento de Estadística, Facultad de Ciencias, Universidad Nacional de Colombia, Sede Bogotá,
Colombia
Sandra Esperanza Melo
[email protected]Departamento de Agronomía, Facultad de Ciencias Agrarias, Universidad Nacional de Colombia, Sede
Bogotá, Colombia
Nelson Alirio Cruz
[email protected]Departamento de Estadística, Facultad de Ciencias, Universidad Nacional de Colombia, Sede Bogotá,
Colombia
Ruy Edeymar Vargas
[email protected] Corporación Colombiana de Investigación Agropecuaria – AGROSAVIA. Centro de Investigación
Tibaitatá. Km. 14, vía Mosquera - Bogotá, Mosquera - Cundinamarca, Colombia
Abstract
Under the presence of nuisance parameters like soil heterogeneity effects, statistical
models used to analyze data from agricultural experiments need to carry out local control
to correct for them. There is a question regarding the performance of two frequently used
approaches: blocking (A1) which seems to be the standard, and the use of covariates (A2).
The advent of precision agriculture permits the use of site-specific soil composition traits as
covariates. Preliminary research has attempted to compare these approaches empirically;
to the best of our knowledge, analytical approximations are not available in the literature.
In this ongoing study, we focus on linear fixed effects models with iid Gaussian errors;
under this setting, the precision to infer estimable functions of treatment means can be ad-
dressed through the mean square of error (MSE). Thus, simulation and analytical studies
are being carried out to compare the performance of these approximations. A dataset was
simulated in two steps; the first one involved the random generation of 4 soil composition
variables while in the second one, non-linear functions of these variables were added to
treatment means and errors to compute the response variable. The simulation was repli-
cated 1000 times. Mean absolute error of the response variable (MAER) and two sets of es-
timable functions (treatment means: MAEE1 and pairwise differences of treatment means:
MAEE2) along with MSE were used to evaluate the performance of the two models.
UNAL https://www.simposioestadistica.unal.edu.co/ SIE
XXX versión virtual
XXX - SIE 2021
Septiembre 21 - 24 de 2021
Results showed smaller MAER, MAEE1 and MSE for A2; differences for MAEE2
were negligible. Regarding the analytical comparison, this is unfinished work; however,
we believe that it is worth mentioning that some formal results concerning precision have
been established. So far, it seems that the performance of these models is highly dependent
on the conditions of the experiment, it is difficult to elucidate a general setup under which
one model outperforms the other.
Key words: Agronomy, blocking, covariates, design of experiments, linear models.
Acknowledgements
This study was supported by Ministerio de Agricultura y Desarrollo Rural – MADR and Uni-
versidad Nacional de Colombia.
UNAL https://www.simposioestadistica.unal.edu.co/ SIE
