University of Nebraska - Lincoln

DigitalCommons@University of Nebraska - Lincoln

USGS Staff -- Published Research US Geological Survey

2004

Use of Radar Remote Sensing (RADARSAT) to

Map Winter Wetland Habitat for Shorebirds in an
Agricultural Landscape
Oriane W. Taft
U.S. Geological Survey, [email protected]

Susan M. Haig
U.S. Geological Survey

Chris Kiilsgaard
Northwest Habitat Institute

Follow this and additional works at: http://digitalcommons.unl.edu/usgsstaffpub

Taft, Oriane W.; Haig, Susan M.; and Kiilsgaard, Chris, "Use of Radar Remote Sensing (RADARSAT) to Map Winter Wetland Habitat
for Shorebirds in an Agricultural Landscape" (2004). USGS Staff -- Published Research. 676.
http://digitalcommons.unl.edu/usgsstaffpub/676

This Article is brought to you for free and open access by the US Geological Survey at DigitalCommons@University of Nebraska - Lincoln. It has been
accepted for inclusion in USGS Staff -- Published Research by an authorized administrator of DigitalCommons@University of Nebraska - Lincoln.
 DOI: 10.1007/s00267-004-8920-8

Use of Radar Remote Sensing (RADARSAT) to Map

Winter Wetland Habitat for Shorebirds in an
Agricultural Landscape
ORIANE W. TAFT*,† cember 1999, 27 January 2000, and 15 March 2000) and
SUSAN M. HAIG simultaneously collected ground reference data to classify ra-
USGS Forest and Rangeland Ecosystem Science Center dar signatures and evaluate map accuracy of four habitat
3200 SW Jefferson Way classes: (1) wet with ⱕ 50% vegetation (considered optimal
Corvallis, Oregon 97331, USA shorebird habitat), (2) wet with ⬎ 50% vegetation, (3) dry with
ⱕ 50% vegetation, and (4) dry with ⬎ 50% vegetation. Overall
CHRIS KIILSGAARD accuracy varied from 45 to 60% among the three images, but
Northwest Habitat Institute the accuracy of focal class 1 was greater, ranging from 72 to
P.O. Box 855 80%. Class 4 coverage was stable and dominated maps
Corvallis, Oregon 97339, USA (40% of mapped study area) for all three dates, while cover-
age of class 3 decreased slightly throughout the study period.
ABSTRACT / Many of today’s agricultural landscapes once Among wet classes, class 1 was most abundant (about 30%
held vast amounts of wetland habitat for waterbirds and other coverage) in December and January, decreasing in March to
wildlife. Successful restoration of these landscapes relies on approximately 15%. Conversely, class 2 increased dramati-
access to accurate maps of the wetlands that remain. We cally from January to March, likely due to transition from class
used C-band (5.6-cm-wavelength), HH-polarized radar re- 1 as vegetation grew. This approach was successful in de-
mote sensing (RADARSAT) at a 38° incidence angle (8-m res- tecting optimal habitat for shorebirds on agricultural lands. For
olution) to map the distribution of winter shorebird (Charadri- modest classification schemes, radar remote sensing is a
iformes) habitat on agricultural lands in the Willamette Valley of valuable option for wetland mapping in areas where cloud
western Oregon. We acquired imagery on three dates (10 De- cover is persistent.

Wetlands have received worldwide recognition in modification of river channel morphology and drain-
the last few decades, especially in light of their alarming age of mesic prairies, the majority of native winter
loss and significant value to society. Western settlement riverine wetlands and wetland prairies that once pro-
and agricultural conversion are responsible for the vast vided habitat to countless wintering and migrating
majority of the estimated 53% loss of wetlands in the waterbirds (e.g., waterfowl, wading birds, shorebirds,
lower 48 US states (Dahl 1990, Mitsch and Gosselink cranes) has been lost to agriculture. Today, remaining
2000). In the Pacific West, statewide losses are esti- wetlands in the Willamette Valley (Valley) are primarily
mated at 31% for Washington, 38% for Oregon, and dispersed among small urban remnant wetlands, a few
91% for California (Dahl 1990) and local losses for duck hunting reserves, four larger state and federally
estuaries and river basins range from 50 to 95% (Akins protected wildlife refuges, and hundreds of scattered
1970, Dennis and Marcus 1984, Boulé and Bierly 1987). unprotected agricultural wetlands (“palustrine emer-
Wetland loss has been no less severe for one wetland gent-farmed wetlands” [Cowardin et al. 1979]). The
latter represent the legacy of a once vast wetland prai-
region in the Pacific Northwest—the Willamette Valley
rie. With estimates for total native wetland loss/conver-
of western Oregon (Taft and Haig 2003). Through
sion as high as 67% (Taft and Haig 2003) and contin-
ued yearly loss of habitat (Bernert and others 1999), a
KEY WORDS: Agricultural landscape; RADARSAT; Radar remote sens- number of local and federal agencies and coalitions
ing; Shorebirds; Wetland habitat mapping; Willamette Valley
now recognize the urgent need to identify, protect, and
restore extant Valley wetlands on a regionwide scale
Published online June 3, 2003.
(e.g., Good and Sawyer 1998, Drut and Buchanan 2000,
*Author to whom correspondence should be addressed, email:
[email protected]
Morlan 2000).
†
Also, Department of Fisheries and Wildlife, Oregon State University, Developing strategic wetland restoration plans on a
Corvallis, Oregon 97331, USA landscape scale requires a series of maps depicting the

Environmental Management Vol. 33, No. 5, pp. 750 –763 © 2004 Springer-Verlag New York, LLC

This article is a U.S. government work, and is not subject to copyright in the United States.
 Radar Remote Sensing and Agricultural Wetlands 751

spatial and temporal distribution of wetland habitats. ability to map land surface during periods of cloud
While the US Fish and Wildlife Service (1991) has cover (Ramsey 1999). Unlike optical sensors, radar sen-
completed National Wetlands Inventory (NWI) cover- sors (e.g., satellites ERS-1/2, JERS-1, and RADARSAT)
age for the Willamette Valley, these maps portray only do not rely on cloud-free conditions (Metternicht
Valley wetlands that typically occur during the spring 1999). These actively transmit energy at microwave fre-
growing season, not during the rainy winter months quencies (radar) to produce a black-and-white image
when wetlands are most prevalent. Other comprehen- from the energy returning to the sensor after interact-
sive Valley mapping efforts were also based on imagery ing with the Earth’s surface. Radar sensors can pene-
acquired in fall or spring (e.g., Kiilsgaard 1999, Oetter trate clouds, rain, and haze commonly encountered
and others 2000). Consequently, the principal type of during a rainy season. Moreover, radar is particularly
wetland habitat in the Valley— ephemeral agricultural effective at detecting boundaries between water and
wetlands— has been entirely overlooked. Because these land, flooding, surface roughness, and moisture con-
wetlands hold great potential for restoration, it is essen- tent of vegetation and soil, depending on parameter
tial to understand their spatial distribution and tempo- settings (Kasischke and others 1997, Elachi 1988, Met-
ral dynamics. ternicht 1999, Ramsey 1999). Wetlands used by water-
Wetland mapping has undergone considerable ad- birds tend to be shallow with open water unobscured by
vancement in the last two decades as researchers have vegetation, and any vegetation present is typically dif-
realized the application potential of various satellite ferent from that found within other land cover types
remote sensing techniques. While many sensoring plat- (Mitsch and Gosselink 2000). Such features structurally
forms have been used to inventory wetlands for moni- contrast them with the surrounding landscape, making
toring and regulatory purposes (see Lee and Lunetta them particularly favorable for radar mapping (Ramsey
1995), wetland biologists have only recently recognized 1999). In areas with a high incidence of rainfall, radar
the value of satellite remote sensing in wetland wildlife technologies may be useful in mapping standing shal-
conservation. The most prominent application of re- low sheetwater or saturated soil with differing vegeta-
mote sensing for wetland wildlife conservation has been tion coverage, conditions common on agricultural wet-
to map specific habitats important to wetland species lands of the Valley.
on a regional scale (e.g., Wickware and others 1980, We recognized the need for a series of maps depict-
Jacobson 1991, Kempka and others 1992, Gratto-Trevor ing the typical extent and dynamics of Valley wetlands
1996). With auxiliary data on the spatial distribution of to aid future restoration efforts. In addition, we had a
individuals and their preferences for different habitats, need for such maps to complete a related project ad-
imagery depicting habitat types can also be used to dressing determinants of habitat use by the 40,000 or
predict regional distribution or density of a species more Valley wintering shorebirds (order Charadriiformes
(e.g., Avery and Haines-Young 1990, Goss-Custard and [Johnson 1993, Nehls 1994, Gilligan and others 1994,
Yates 1992, Aspinall and Veitch 1993, Morrison 1997, Sanzenbacher and Haig 2002a, 2002b, Taft and Haig
Scott and others 2002). Specialized projects to map unpublished data]). Finally, winter cloud cover in the
particular wetland habitat types are becoming more Willamette Valley can be persistent. Thus, using re-
common as the spatial, temporal, or informational res- motely sensed radar (RADARSAT) data, we set out to
olution of already existing wetland maps often does not map winter wetland habitat in the Valley, specifically
match the needs of wetland wildlife biologists. focusing on agricultural wetland sites frequented by
Most satellite remote sensors are optical (e.g., Land- shorebirds. Our specific objectives were (1) to evaluate
sat Multispectral Scanner [MSS] and Thematic Mapper the utility and accuracy of using C-band HH polarized
[TM], Systeme Pour l’Observation de la Terre radar remote sensing to develop maps of ephemeral
[SPOT]), reliant on collecting reflected energy from wetland habitat (primarily agricultural) important to
the Earth’s surface at wavelengths in the visible portion shorebirds and (2) to produce three winter cover maps
of the electromagnetic spectrum (Avery and Berlin (beginning, middle, and end of winter) identifying wet
1992, Lee and Lunetta 1995). While these have been (either impounded water or shallow sheetwater/satu-
used to identify and monitor a variety of wetland types, rated soil), unvegetated (ⱕ 50% cover) wetland areas
they exhibit a number of limitations including spectral used by shorebirds and other waterbirds. In addition,
confusion of wetland and nonwetland categories this paper provides examples of potential uses of our
(Jensen and others 1987, Henderson and others 1999, maps and suggests possible approaches for improving
Bourgeau and others 2001) and, most notably, an in- radar performance for this particular application.
752 O. W. Taft and others

Figure 1. The Willamette Valley of western Oregon, including the area covered by radar imagery (large rectangle) and 10 areas
of focus for ground reference data collection (small dashed rectangles). The two lowland subecoregions are highlighted in light
gray (Prairie Terraces) and dark gray (Willamette River and Tributaries Gallery Forest).

Study Area (Jackson and Kimerling 1993), with an average annual

rainfall of 100 –125 cm and average temperatures rang-
Within the Willamette Basin of western Oregon, the ing from 1°C in January to 30°C in July (Oetter and
Valley encompasses a 9100-km2 area of lowland plains others 2000).
(including the “Prairie Terraces” and “Willamette River Our map study area included the central and south-
and Tributaries Gallery Forest” subecoregions but ex- ern regions of the Valley but excluded the Tualatin
cluding the “Valley Foothills” [Clark and others 1991, basin (Figure 1). General landforms in the Valley in-
Pater and others 1997]) varying in width from about 20 clude alluvial terraces and floodplains interrupted by
to 60 km and covering a north–south length from basalt outcrops and gently sloping hills of both volcanic
Portland to Eugene of roughly 290 km (Figure 1) (Ben- and sedimentary origin (Oetter and others 2000). The
ner and Sedell 1997, Hulse and others 1998). The majority of the Valley is dominated by agriculture, pri-
prominent hydrologic feature of the Valley is the north- marily grass seed fields on the alluvial terraces. Other
erly flowing Willamette River and its 13 major tributar- common lowland crops include vegetable crops, fruit
ies, which together drain the Willamette Basin, a orchards, nursery and greenhouse stock, and pepper-
29,000-km2 watershed between the Cascade and the mint (Hulse and others 1998, Oetter and others 2000).
Coast ranges of Oregon (Oetter and others 2000). The Grass seed crops and peppermint are planted in the fall
climate of this region is considered cool Mediterranean on plowed, leveled barren fields. While annual ryegrass
 Radar Remote Sensing and Agricultural Wetlands 753

Figure 2. Photographs exemplifying typical habitat in the four classes mapped with radar (RADARSAT) remote sensing in the
Willamette Valley of Oregon during winter 1999 –2000: (A) class 1, optimal shorebird habitat—wet, ⱕ 50% vegetation; (B) class
2—wet, ⬎ 50% vegetation; (C) class 3— dry, ⱕ 50% vegetation; (D) class 4 — dry, ⬎ 50% vegetation. Additional photos of class
1 are shown in (E) a flooded fallow cornfield with foraging long-billed dowitchers (Limnodromus scolopaceus) and (F) a linear
agricultural wetland/remnant slough in a newly planted grass field; many of these were not mapped if less than about 3 8-m pixels
(25 m) wide.

is replanted every year, perennial ryegrass is generally dominate (Figure 2). Common wintering shorebirds in
replanted on a 3-year cycle. Vegetable crops are annu- the Valley include dunlin (Calidris alpina), killdeer
ally plowed in fall and left fallow through the winter. By (Charadrius vociferus), common snipe (Gallinago galli-
spring (March), annual grass fields are fully covered nago), long-billed dowitcher (Limnodromus scolopaceus),
with vegetation, while perennial grass fields still provide black-bellied plover (Pluvialis squatarola), greater yel-
exposed soil between plants. These agricultural fields lowlegs (Tringa melanoleuca), and least sandpiper (Calid-
potentially hold water in winter where hydric soils pre- ris minutilla).
754 O. W. Taft and others

Methods F1F imagery because we were most interested in detect-

ing standing water in open habitat and because we
Classification Scheme needed the finest resolution possible to detect small
During the winter of 1998 –1999, we conducted a patches of standing water. Scenes at F1F are approxi-
preliminary study on the agricultural habitat associa- mately 50 ⫻ 50 km, thus it took three scenes to cover
tions of Dunlin and Killdeer, the two most abundant our focal study area (central and southern Valley) on
wintering shorebirds in the Valley. Roughly 90% of each acquisition date. These images were aligned and
fields used by dunlin were flooded with shallow (⬍ acquired north–south so that each final map, once
5-cm-deep) water and/or had saturated soil. About mosaicked, would continuously cover an area of
70% were covered with less than 50% vegetation. In roughly 50 ⫻ 150 km (4500 km2), encompassing Valley
contrast, killdeer were more of a wetland generalist; lowlands in the two subecoregions of interest (Prairie
only 50% of used fields held standing water/saturated Terraces and Willamette River and Tributaries Gallery
soil and 55% had less than 50% vegetative cover. Thus, Forest [Pater and others 1997]) from Eugene to New-
in mapping shorebird habitat on agricultural lands, we berg, Oregon (Figure 1). We used Pater and others
focused on identifying flooded fields with exposed soil (1997) to mask the Valley Foothills subecoregion from
(less than 50% cover; Figure 2E); these conditions are radar imagery acquired in the 50 ⫻ 150-km swath.
preferred by dunlin and ecological allies (species in the Thus, the size of the study area portrayed by final maps
family Scolopacidae). We designed a classification was 4209 km2.
scheme of four classes based on the presence of water Before rectifying scenes, we used Sigma and Lee
and vegetation (Figure 2): (1) wet (either impounded filters to reduce speckle noise (Rio and others 2000).
water or shallowly flooded) with ⱕ 50% vegetation, (2) Among nine scenes, we were able to reduce coefficients
wet with ⬎ 50% vegetation, (3) dry (without visible of variation for noise from 0.387– 0.399 at raw imagery
standing water) with ⱕ 50% vegetation, and (4) dry to 0.213– 0.223 at the second pass. We then rectified all
with ⬎ 50% vegetation. We considered habitat to in- December scenes to a 1996 Landsat Thematic Mapper
crease in suitability for shorebirds from the highest reference image of the Valley. Among the three Decem-
class number to the lowest, with class 1 regarded as ber scenes, we selected between 18 and 34 ground
optimal. control points (GCPs) to calculate third-order polyno-
mial transformations used to georegister input images
Imagery and Preprocessing to the reference scene. Among the three scenes, regis-
We acquired imagery on 3 selected days when the tration was achieved with an error of 22–28 m, approx-
Canadian RADARSAT satellite passed over the Valley imately the size of a Landsat TM pixel (30 m). Scenes
study area: 10 December 1999, 27 January 2000, and 15 for January (using 41–52 GCPs) and March (38 – 45
March 2000. Scene acquisition dates were 48 days apart, GCPs) were georegistered to the rectified December
or on every other 24-day orbit pass. With these dates we scenes, also using third-order polynomial transforma-
aimed to capture the extent of shorebird wetland hab- tions. Registration errors were lower among these six
itat at three time periods: at the onset of early winter scenes, ranging from 4.8 to 16.7 m. We used a cubic
rainfall (early December), at the peak of the midwinter convolution resampling procedure to assign new coor-
rainy season (January), and during waning rain activity dinate grids to scenes. After georegistration, the three
in late winter/early spring (March). scenes for each date were mosaicked into composite
The RADARSAT satellite carries a C-band (5.6-cm- images used in the classification effort.
wavelength) synthetic aperture radar (SAR) with HH
(horizontally transmitted/horizontally received) wave- Ground Reference Data Collection
length polarization (Corbley 1995). All scenes were For classification and verification, we collected
taken in fine 1 far (F1F) beam mode (8-meter pixel ground reference data from roadsides and aerial flights
resolution, 38.78° incidence angle) on descending or- within 1–2 days of each image acquisition date. Topog-
bits at 1418 hr. There is a trade-off between pixel raphy of the Valley alluvial terraces is strikingly flat (⬍
resolution and features one can detect with particular 1° slope), with maximum elevation fluctuations of the
incidence angles. Soil moisture and water under grass order of only meters over a horizontal distance of 10 –
canopies are best detected at steep incidence angles (⬍ 100 km (Hulse and others 1998). As a result, within
30°) to nadir (Elachi 1988, Ramsey 1995, Biftu and Gan individual agricultural fields, surface water tends to
1999), but RADARSAT imagery at these angles is avail- accumulate into numerous discrete small (of the order
able only at resolutions of 25 m or greater. Images at of decameters) shallow ponds (less than 25 cm deep) or
8-m resolution are acquired at larger angles. We chose as diffuse and widespread “sheetwater” (shallow, 1-cm-
 Radar Remote Sensing and Agricultural Wetlands 755

deep water spread across a flat surface; Figure 2). Such little (⬍ 10%) vegetation (class 1). Wet fields were
spatial patterning logistically precluded determining obvious because sheetwater/saturated soil appeared
the exact coordinate locations of the boundaries of all glossy and dark from the air.
individual water bodies. Instead, we considered refer-
ence “sites” as entire agricultural fields if they were Image Analysis and Final Maps
homogeneous in wetness and cover, or as wet areas We used both supervised and unsupervised classifi-
within otherwise dry fields (e.g., a large pond, a rem- cation techniques to associate spectral signature with
nant slough). We could not feasibly quantify soil mois- habitat classes. Using our field data, we were able to
ture, water depth, surface roughness, or percentage distinguish a specific signature to supervise the classifi-
cover of vegetation. Thus, we categorized sites into cation of class 1 habitat only; we used unsupervised
habitat classes by qualitatively noting the presence and
training to produce signatures for the remaining
prominence of surface water and visually estimating the
classes. For supervised classification of habitat in class 1,
percentage cover of vegetation as ⱕ 50% or ⬎ 50%. We
we delineated polygons around pixel clusters of homo-
did not separately distinguish habitat with saturated soil
geneous spectral returns within class 1 reference sites
(showing a glossy sheen but without standing water)
and calculated mean, standard deviation, minimum
from habitat with sheetwater because they almost always
and maximum digital values for all pixels within each
cooccur on a fine spatial scale (meters) within fields.
polygon. We then used a maximum likelihood classifi-
We selected ground reference sites opportunistically
cation algorithm to delineate class 1 habitat within
by traversing roads and taking aerial photos in areas
mosaicked scenes. To classify habitat in classes 2– 4, we
with a high incidence of hydric soils (Daggett and
separated returns remaining in the class 1-masked
others 1998). For both modes of data collection, we
scene into 50 distinct spectral clusters. We chose the
focused on finding wet, unvegetated sites (class 1, op-
number 50 to accommodate the variability of signatures
timal shorebird habitat). Sites in all other classes were
we expected from the myriad of permutations of wet
interspersed within these areas. We documented a total
and dry soils with varying amounts of vegetation. Using
of 689 reference sites for the December image, 731 for
roughly half of our ground reference data set (415 sites
January, and 592 for March. Sites were fairly well inter-
from December, 448 sites from January, and 308 sites
spersed among classes, but wet classes were better rep-
from March) as training sites (randomly chosen), we
resented. We selected sites in the north and south
then visually determined which of the three categories
Valley interspersed among 10 focal areas (Figure 1),
each of the 50 return clusters represented and recoded
but due to the concentrated nature of hydric soils, sites
the 50 clusters into three remaining habitat classes.
within these areas were locally clustered. For January
Finally, we overlaid each of the final classified mosa-
and March referencing, we revisited a large proportion
icked images with urban, forest, and permanent open
of sites we had visited in December to assess temporal
water (reservoirs, rivers, etc.) land cover using an an-
changes.
cillary map developed from 1996 Landsat TM imagery
For roadside-collected data, we located reference
(Kiilsgaard 1999).
sites by estimating the distance and direction to the
focal site of interest from a GPS- or landmark-located
Accuracy Assessment
roadside position (e.g., crossroad). During January and
March, we used a laptop computer displaying the De- We assessed the accuracy of final maps using the
cember raw imagery and a road layer overlay to mark remaining ground reference sites not used for training;
location of sites concurrently while in the field. We this resulted in 274 verification sites for December, 283
digitized sites by looking for pixels of uniform radar for January, and 284 for March. For each site, we com-
returns in the close vicinity of estimated locations and pared its predicted class from final maps to the site’s
then assigning the site’s positional coordinate to a cen- true class and constructed an error matrix of these
tral pixel for each cluster. To aid analyses and temporal comparisons for each date. The error assessment char-
comparisons, one or more photos were taken of each acterized the accuracy of each map, with commission
roadside site during each of the three dates. error or “user’s accuracy” (proportion of sites assigned
For data collected aerially (in a Cessna 182RG), we to a class that are correctly assigned) and omission
took oblique photos of the landscape at an elevation of error or “producer’s accuracy” (the proportion of sites
approximately 450 m and used visible roads and land- of a certain class that are correctly assigned) included.
marks to find the coordinate location of focal sites. For each map, we calculated overall accuracy as the
Because vegetative cover was often difficult to estimate number of correctly classified sites divided by the total
by air, we used aerial photos only to find sites with very number of testing sites.
756 O. W. Taft and others

Results habitat. Radar returns from dry, unvegetated sites were

darker relative to vegetated sites but brighter (lighter in
Spatial and Temporal Patterns tone) relative to wet sites.
Among all three maps, the majority of the 4209-km2 The error matrix for each map indicates an overall
study area was categorized as class 4 (Figure 3), with accuracy of 60% for December, 59% for January, and
roughly 40% coverage (Figure 4). For the December 45% for March (Table 1). By pooling cover types into
and January maps, class 1 was the next most abundant only two classes for wetness—wet and dry—the overall
class predicted (⬃30% of the study area) and class 2 map accuracy increased to 78% for December, 78% for
was the least prevalent, at about 10% coverage. More- January, and 75% for March. Pooling cover types into
over, optimal shorebird habitat (class 1) was generally two classes for vegetation—ⱕ 50% vegetation and ⬎
dispersed throughout the entire study area. However, 50% vegetation— overall map accuracies were 72% for
concentrations in the southern Valley occurred be- December, 69% for January, and 57% for March. For
tween Albany and Lebanon and southward, and west of all classification schemes, map accuracy diminished be-
the Willamette River between Corvallis and Eugene tween January and March.
(Figure 3). There were also coalesced patches of class 1 Among the four classes, class 1 (wet, ⱕ 50% vegeta-
habitat around Independence. In the central Valley, tion) was most accurately mapped, with user’s accuracy
class 1 habitat was most prevalent west and northeast of for all three maps between 72 and 80% (Table 1).
Salem and west of Woodburn. Producer’s accuracy was also fairly high in December
Overall, winter 1999 –2000 was an average season in (83% accuracy) and January (76%). The low produc-
rainfall, but precipitation varied across months. er’s accuracy (47%) in March was due primarily to class
Monthly precipitation totals for Corvallis, Oregon, were 1 sites being confused for class 2 (57/175 sites) and
as follows: November (23 cm; 6-cm departure from the class 4 (25/175 sites).
1961–1990 average), December (15 cm; ⫺5-cm depar- Classes 2– 4 did not map as accurately as class 1. For
ture), January (20 cm; 3-cm departure), February (16 all three maps, sites of class 2 (wet, ⬎ 50% vegetation)
cm; 3-cm departure), and March (9 cm; ⫺2-cm depar- were highly misclassified, commonly confused for class
ture [Oregon Climate Service 1999 –2000]). Covering 1 or class 4 (Table 1). Class 3 (dry, ⱕ 50% vegetation)
roughly 30% of the study area, optimal shorebird hab- was regularly misidentified as class 1 in December and
itat (class 1) was more prevalent in December and as class 4 in January and March. In general, class 4 (dry,
January (slightly more in January) than in March (Fig- ⬎ 50% vegetation) was moderately confused with each
ure 4). Although optimal habitat decreased in March, of the other three classes, but especially class 1. The
class 2 habitat increased, retaining the same proportion lower user’s than producer’s accuracy for most classes
(40%) of wet habitat (classes 1 and 2 combined) across (Table 1) indicates that commission errors tended to
dates despite monthly variation in rainfall. Likewise, be more prevalent than omission errors.
slight declines in coverage of class 3 were mirrored by
small increases in class 4 across dates. Estimates for total
Discussion
coverage of wet classes should be viewed with caution,
however, as class 2 accuracies were fairly low (see be- Radar remote sensing proved to be an effective and
low). The declines in classes 1 and 3 and corresponding valuable tool for mapping habitat important to shore-
increases in classes 2 and 4 may be partially explained birds on agricultural lands in winter. Final maps re-
by the steady growth of agricultural vegetation. Be- vealed far more shorebird/wetland habitat than was
tween January and March, many sites transitioned from previously thought to exist in the Valley during winter,
class 1 to class 2 or from class 3 to class 4. pointing to the importance of including agricultural
habitat in regional restoration and conservation plans.
Radar Signatures and Map Accuracy Our error assessments support the notion that single-
Among flooded habitat, radar returns for open wa- parameter radar imagery is useful for modest wetland
ter found on refuge impoundments, riceponds, and classification schemes (Lee and Lunetta 1995). More-
shallow reservoirs (weak return, black image tones) over, success depends on the number of classes that
were similar to returns from shallow sheetwater/satu- one hopes to depict accurately. Among classes, we were
rated soil on unvegetated fields (weak return, dark-gray most interested in mapping the distribution of optimal
image tones). We considered both signatures to display shorebird habitat, or land in class 1 (wet, ⱕ 50% vege-
“wet” habitat. Vegetated sites with either a wet or a dry tation). While the overall map accuracy was fairly low
substrate under grass canopies gave a brighter return (45– 60%) for each date, the user’s accuracy of class 1
(strong return, lighter-gray tones) than unvegetated was appreciably higher (72– 80%). The accuracy of wet
 Radar Remote Sensing and Agricultural Wetlands

Figure 3. Final cover maps of Willamette Valley wetlands for three dates in winter 1999 –2000.
757
758 O. W. Taft and others

Figure 4. Temporal changes in percent-

age cover (of the entire Willamette Valley
map study area) of four land cover classes
across the three image dates during the
winter of 1999 –2000. Class 1, optimal
shorebird habitat—wet, ⱕ 50% vegeta-
tion; class 2—wet, ⬎ 50% vegetation; class
3— dry, ⱕ 50% vegetation; class 4 — dry,
⬎ 50% vegetation.

Table 1. Error matrix for final wetland maps of the Willamette Valley, Oregona
Reference
Map prediction Class 1 Class 2 Class 3 Class 4 Total User’s accuracy (%)
Class 1: wet, ⱕ50% vegetation
10 Dec. 135 33 8 11 187 72
27 Jan. 130 33 4 14 181 72
15 Mar. 82 14 3 4 103 80
Class 2: wet, ⬎50% vegetation
10 Dec. 6 9 1 2 18 50
27 Jan. 11 10 0 1 22 45
15 Mar. 57 22 6 4 89 25
Class 3: dry, ⱕ50% vegetation
10 Dec. 9 3 11 7 30 37
27 Jan. 10 2 12 0 24 50
15 Mar. 11 4 9 6 30 30
Class 4: dry, ⬎50% vegetation
10 Dec. 13 13 3 10 39 26
27 Jan. 21 11 8 16 56 29
15 Mar. 25 15 7 15 62 24
Total
10 Dec. 163 58 23 30 274
27 Jan. 172 56 24 31 283
15 Mar. 175 55 25 29 284
Producer’s accuracy (%)
10 Dec. 83 16 48 33 60
27 Jan. 76 18 50 52 59
15 Mar. 47 40 36 52 45
a
Numbers on the diagonal (boldface) are reference sites correctly classified, whereas those off the diagonal signify reference sites (columns)
incorrectly predicted in each class (rows). Row totals signify the total number of sites predicted in each class, whereas column totals signify the
true total number of sites in each class used in the analysis.

habitat was further supported by visually assessing the face roughness, moisture content) and radar parameter
spatial correspondence of hydric soils (Daggett and configurations (incidence angle, wavelength, and po-
others 1998) with the distribution of classes 1 and 2. larization). These interactions are highly complex and
Considering class 1 as optimal shorebird habitat inter- the subject of much research (Lee and Lunetta 1995,
spersed in a matrix of suboptimal habitat (classes 2– 4), Metternicht 1999, Ramsey 1999). First, a given surface
the accuracy of class 1 habitat was the most meaningful will increase in specular reflectance (appear smoother)
measure for map accuracy. with larger incidence angles (Avery and Berlin 1992,
Some of the map error likely stems from interactions Sokol and others 2000). Thus, our relatively large inci-
between ground features (land/water boundaries, sur- dence angle may have been the source of some error in
 Radar Remote Sensing and Agricultural Wetlands 759

the confusion of vegetated (class 2 and 4) land cover Chavez and others 2001) may have provided a greater
for wet, unvegetated (class 1) cover, especially habitat breadth of information for image interpretation (Hess
with close to 50% vegetation (i.e., 40 – 60%). Second, and Melack 1994, Metternicht 1999). For example, by
detecting water under a grass canopy requires a radar cross-referencing RADARSAT data at multiple inci-
beam emitted at a steep incidence angle (⬍ 30°) or dence angles (e.g., steep 25-m resolution and shallow
with a long wavelength (Ormsby and others 1985, Ram- 8-m resolution), one may be able to produce a map
sey 1995, Wang and others 1995). RADARSAT’s shallow depicting small bodies of standing water while also
incidence angle combined with the short wavelength portraying more information about soil moisture. Mul-
suggests lessened penetration through grass canopies tiwavelength and multipolarization imagery would re-
and thus less of a return from flooded vegetated quire use of two different radar sensors.
ground (Ramsey 1995, 1999). This may account for Compared to other radar studies, our SAR data per-
confusion among vegetated wet (class 2) and dry (class formed well in mapping Valley wetlands. Radar satellite
4) sites. Third, RADARSAT’s polarization may have sensors have been used to detect coastal wetlands (e.g.,
been another source of error. Sensors with HH polar- (Henderson and others 1998, Dwivedi and others
ization are known to be less sensitive to changes in 1999), tidal flooding (e.g., Ramsey 1995), freshwater
vegetation moisture, content than are cross-polarized wetlands (e.g., Kasischke and Bourgeau-Chavez 1997),
sensors (HV or VH [Avery and Berlin 1992]). This may forested wetlands (e.g., Hess and Melack 1994,
provide an additional explanation for confusion be- Bourgeau-Chavez and others 2001), forested and un-
tween class 1 and class 2, as moisture in the vegetation forested peat bogs and marshes (e.g., Baghdadi and
of class 2 sites may not have been detected. others 2001), and soil moisture (e.g., Dobson and oth-
Other errors may have been caused simply by simi- ers 1995, Biftu and Gan 1999). However, much of this
larities among signatures. For example, grass crops of
work has focused on the relationships between radar
high biomass and homogeneous in horizontal cover
backscatter returns and particular wetland properties
and vertical height (e.g., rice, annual ryegrass) tend to
(e.g., flooding, soil moisture), rather than on docu-
act like a smooth surface (similar to calm standing
menting the accuracy of maps created using these re-
water), decreasing radar returns (Durden and others
lationships. Of the few studies that have documented
1995, Dobson and others 1996). Many of the annual
accuracy, success has been variable. For example, Bagh-
ryegrass sites in class 4 (dry, vegetated) may have been
dadi and others (2001) defined six cover types (for-
confused for class 1 (wet, unvegetated) for this reason.
ested and nonforested peat bog, marsh, open water,
Moreover, the confusion of class 1 sites for class 2 in
clearing, and forest) from variable radar signatures
March may be explained by the fact that many class 1
using ERS-1 C-band SAR at different polarizations. For
sites had grown to nearly 50% cover by the third date of
imagery. one season, they reported a high overall accuracy of
Had we not chosen to use imagery with the finest 86% for cross-polarized data and 76% for HH polariza-
resolution from one sensor (RADARSAT), different tion. In contrast, Henderson and others (1998) exam-
parameter settings or the use of setting combinations ined variation in classification accuracy among ERS and
from multiple sensors may have improved mapping RADARSAT images acquired with various incidence
performance for nonfocal habitats (classes 2– 4). While angles and polarizations. They classified coastal wet-
some settings are predetermined by the RADARSAT lands into four classes: estuarine emergent, palustrine
sensor (i.e., wavelength and wavelength polarization), emergent, palustrine scrub shrub, and palustrine for-
others are adaptable to user needs (i.e., incidence an- ested wetland. Accuracy was extremely poor for both
gle and associated resolution). Choice of a smaller sensors and all settings, with 17% the highest accuracy
incidence angle may have lessened confusion between achieved for any one class. In comparison to these, our
vegetated and unvegetated classes and/or enabled pen- results are encouraging. The high accuracies found by
etration through grass canopies to distinguish vege- Baghdadi and others (2001) are partly a function of
tated areas with understory flooding. Similarly, a extreme structural differences among the classes they
steeper incidence angle could have been used had set out to classify, while high error rates found by
detecting differences in soil moisture been our main Henderson and others (1998) stem from complex, eas-
objective (Dobson and others 1995). Additionally, be- ily confused wetland classes (i.e., all of their classes have
cause backscatter returns vary with all three radar pa- some kind of flooding and vegetative cover). In con-
rameters, the use of multi-incidence angle, multi-wave- trast to these two studies, our classification scheme was
length, or multipolarization imagery (e.g., (Wang and modest, with classes distinct enough to yield sufficient
others 1995, Dobson and others 1996, Bourgeau- accuracy with the parameter settings we chose.
760 O. W. Taft and others

Among wetland mapping projects that depict water- practical aspects that should be considered in using
bird habitat, most have been able to use optical remote these maps, however. The first is that farming practices
sensors (e.g., Avery and Haines-Young 1990, Jacobsen (affecting patterns of vegetative cover) and rainfall pat-
1991, Gratto-Trevor 1996, Morrison 1997). For single- terns vary annually. Thus within the subset of lands with
pass imagery, optical sensors (particularly Landsat TM) hydric soils in the Valley, the extent and distribution of
generally still outperform radar sensors in classification shorebird habitat will vary from year to year. Moreover,
accuracy (e.g., Jensen and others 1993, Gratto-Trevor although waterfowl use agricultural wetlands (Taft and
1996, Morrison 1997, Henderson and others 1998, Haig, unpublished data) and impounded wetlands, the
Fuller and others 1998, Lunetta and Balogh 1999). maps fail to accurately map emergent ponds and shal-
Moreover, augmenting radar data with optical imagery low riparian areas also used by waterfowl and other
will result in the greatest accuracies for wetland maps wetland-dependent species (e.g., amphibians). This is
(Place 1985, Ramsey and others 1998, Pietroniro and primarily because our focus was on referencing and
others 2000, Töyrä and others 2001). However, for verifying shorebird habitat in particular. Therefore, fi-
studies where cloud cover can be prevalent, using an nal maps can be regarded as depicting the majority of
optical sensor is seldom an option. Radar imagery en- habitat important to shorebirds but only some of the
ables mapping important waterbird habitat when it habitat important to other wetland-dependent species.
otherwise would not be possible, particularly in regions Finally, radar was unable to detect linear wetlands (e.g.,
with persistent cloud cover such as Oregon’s Wil- remnant sloughs, shallow streams) less than 3 pixels
lamette Valley in winter. wide (⬃25 m), habitat that is frequented by both wa-
Our Valley wetland maps hold promise for a number terfowl and shorebirds (Figure 2F). Consequently,
of local applications. First, the spatial distribution of these maps should be regarded as “snapshot” views of a
extensive drainage tiling on hydric soil farmlands in the certain kind of wetland habitat (i.e., agricultural habitat
Valley is poorly documented. Hydric soils maps are not important to wintering shorebirds) in a given year of
average rainfall.
always indicative of where ponding will occur. Using
our maps, restorationists can now evaluate which gen-
eral areas of the Valley would be most optimal for
restoration efforts. Additionally, information on the Summary
location and dynamics of class 1 habitat at local spatial The specific radar parameters we used (C-band, HH
scales may be useful to resource agencies or watershed polarization, 38° incidence angle) and our approach
councils involved in local restoration in the Valley. (single sensor, single settings for each parameter) per-
Finally, these maps enable examining important ques- formed well for the modest goal of mapping shorebird
tions in landscape ecology that are relevant to restora- habitat at a relatively fine resolution on agricultural
tion and management (e.g., Turner 1989, Scott and lands in winter. When cloud cover restricts the use of
others 2002), namely, investigating the influence of the Landsat TM data, radar imagery may be worth acquir-
spatial and temporal patterning of habitat on the dis- ing for projects with objectives similar to ours, espe-
tribution, abundance, and movements of wetland spe- cially as the availability of imagery increases commer-
cies in the Valley. With spatially continuous data, one cially and financially. Moreover, we may find increased
can ask these questions at multiple spatial scales perti- value for radar in mapping wetland habitat for wildlife
nent to the species of interest. as future research refines our knowledge of radar–
Although class 1 habitat invariably includes many ground feature interactions and as the accuracies of
wetlands already identified by the National Wetlands maps created with radar data are further evaluated.
Inventory, much of it depicts agricultural wetlands that
were unidentified prior to this study. Thus, by illustrat-
ing the great extent of winter wetlands on agricultural Acknowledgements
land, these maps represent a significant addition to This work benefited greatly from the technical assis-
other available wetland maps for the Valley. They also tance and advice of C. Barrett, K. Bierly, G. Bonser, W.
provide a much different picture of the distribution Cohen, W. Jensen, G. Lienkaemper, J. Morlan, D. Oet-
and abundance of potential waterbird habitat in the ter, D. Peters, E. Ramsey III, P. Sanzenbacher, and T.
Valley. In general, radar remote sensing has the poten- Zarriello. R. Kennedy contributed helpful comments
tial to identify both intermittent and permanent flood- on early drafts of the manuscript. J. Dhundale, M.
ing not necessarily extractable from National Wetlands Henschen, W. Jensen, H. Packard, and P. Sanzen-
Inventory maps with NWI categories. There are a few bacher provided invaluable assistance in collecting
 Radar Remote Sensing and Agricultural Wetlands 761

ground reference data. This study was funded by the California wetlands. Environmental Science Associates, San
Oregon Watershed Enhancement Board, USGS Forest Francisco, California, 125 pp.
and Rangeland Ecosystem Science Center, US Fish and Dobson, M. C., L. E. Pierce, and F. T. Ulaby. 1996. Knowledge-
Wildlife Service, Oregon Department of Fish and Wild- based land-cover classification using ERS-1/JERS-1 SAR
composites. Institute of Electrical and Electronics Engineers
life, and US Bureau of Land Management.
Transactions on Geoscience and Remote Sensing 34:83–99.
Dobson, M. C., F. T. Ulaby, and L. E. Pierce. 1995. Land-cover
References classification and estimation of terrain attributes using syn-
thetic aperture radar. Remote Sensing of Environment
Akins, G. J. 1970. The effects of land use and land manage- 51:199 –214.
ment on the wetlands of the upper Klamath Basin. M.S.
thesis. Western Washington State College, Bellingham, Drut, M. S., and J. B. Buchanan. 2000. The U.S. shorebird
Washington. conservation plan: Northern pacific coast regional shore-
bird management plan. Manomet Center for Conservation
Aspinall, R., and N. Veitch. 1993. Habitat mapping from Sciences, Manomet, Massachusetts, 32 pp.
satellite imagery and wildlife survey data using a Bayesian
modeling procedure in a GIS. Photogrammetric Engineering Durden, S. L., L. A. Morrissey, and G. P. Livingston. 1995.
and Remote Sensing 59:537–543. Microwave backscatter and attenuation dependence on leaf
area index for flooded rice fields. Institute of Electrical and
Avery, T. E., and G. L. Berlin. 1992. Fundamentals of remote
Electronics Engineers Transactions on Geoscience and Remote Sens-
sensing and airphoto interpretation. 5th editionMacmillan,
ing 33:807– 810.
New York, 472 pp.
Dwivedi, R. S., B. R. M. Rao, and S. Bhattacharya. 1999.
Avery, M. I., and R. H. Haines-Young. 1990. Population esti-
Mapping wetlands of the Sundaban Delta and its environs
mates for the Dunlin Calidris alpina derived from remotely
using ERS-1 SAR data. International Journal of Remote Sensing
sensed satellite imagery of the Flow Country of northern
20:2235–2247.
Scotland. Nature 344:860 – 862.
Elachi, C. 1988. Spaceborne radar remote sensing: Applica-
Baghdadi, N., M. Bernier, R. Gauthier, and I. Neeson. 2001.
tions and techniques. Institute of Electrical and Electronics
Evaluation of C-band SAR data for wetlands mapping. In-
Engineers Press, New York, 255 pp.
ternational Journal of Remote Sensing 22:71– 88.
Fuller, R. M., P. Ipulet, D. Pomeroy, A. Katende, R. Bailey, R.
Benner, P. A., and J. R. Sedell. 1997. Upper Willamette River
Ogutu-Ohwayo, G. B. Groom, and S. Mugisha. 1998. The
landscape: A historic perspective. Pages 23– 47 in A. Lae-
integration of field survey and remote sensing for biodiver-
nen, and D. A. Dunnette. Eds, River quality: Dynamics and
sity assessment: A case study in the tropical forests and
restoration. Lewis, New York.
wetlands of Sango Bay, Uganda. Biological Conservation
Bernert, J. A., Eilers, J. M., Eilers, B. J., Blok, E., Daggett, S. G., 86:379 –391.
Bierly, K. F. (1999) ⬙Recent wetlands trends (1981/82–
Gilligan J., D. Rogers, M. Smith, and A. Contreras. Eds, 1994.
1994) in the Williamette Valley, Oregon, USA.⬙ Wetlands
Birds of Oregon: Status and distribution. Cinclus, McMin-
19:545–559.
nville, Oregon, 330 pp.
Biftu, G. F., and T. Y. Gan. 1999. Retrieving near-surface soil
Good, J. W., and C. B. Sawyer. 1998. Recommendations for a
moisture from Radarsat SAR data. Water Resources Research
nonregulatory wetland restoration program for Oregon.
35:1569 –1579.
Oregon Sea Grant, Corvallis, Oregon, 96 pp.
Boulé, M. E., and K. F. Bierly. 1987. History of estuarine
Goss-Custard, J. D., and M. G. Yates. 1992. Towards predicting
wetland development and alteration: What have we
the effect of salt-marsh reclamation on feeding bird num-
wrought?. Northwest Environmental Journal 3:43– 61.
bers on the Wash. Journal of Applied Ecology 29:330 –340.
Bourgeau-Chavez, L. L., E. S. Kasischke, S. M. Brunzell, J. P.
Gratto-Trevor, C. L. 1996. Use of Landsat TM imagery in
Mudd, K. B. Smith, and A. L. Frick. 2001. Analysis of space-
determining important shorebird habitat in the Outer
borne SAR data for wetland mapping in Virginia riparian
Mackenzie Delta, Northwest Territories. Arctic 49:11–22.
ecosystems. International Journal of Remote Sensing
22:3665–3687. Henderson, F. M., R. Chasan, T. F. Hart Jr., and J. Portolese.
1998. Variation in SAR signal response from coastal wet-
Clark, S. E., D. White, and A. L. Schaedel. 1991. Oregon, USA,
lands in a complex urban setting. Pages 487– 492 in Pro-
ecological regions and subregions for water quality manage-
ceedings of the 2nd international workshop on retrieval of
ment. Environmental Management 15:847– 856.
bio- and geo-physical parameters from SAR data for land
Corbley, K. P. 1995. RADARSAT: Meeting the needs of users applications, 21–23 October. ESTEC, Noordwijk, The
worldwide. Geocarto International 10:87–91. Netherlands.
Daggett, S. G., M. E. Boulé, J. A. Bernert, J. M. Eilers, E. Blok, Henderson, F. M., T. F. Hart Jr, B. P. Heaton, and J. E.
D. Peters, and J. Morlan. 1998. Wetland and land use Portolese. 1999. Mapping coastal ecosystems over a steep
changes in the Williamette Valley, Oregon: 1982 to 1994. development gradient using C-CAP protocols. International
Report to the Oregon Division of State Lands. Shapiro and Journal of Remote Sensing 20:727–744.
Associates, Salem, Oregon, 38 pp.
Hess, L. L., and J. M. Melack. 1994. Mapping wetland hydrol-
Dahl, T. E. 1990. Wetlands: Losses in the United States 1780s ogy and vegetation with synthetic aperture radar. Interna-
to 1980s. Fish and Wildlife Service, Washington, DC, 21 pp. tional Journal of Ecology and Environmental Sciences
Dennis, N. B., and M. L. Marcus. 1984. Status and trends of 20:197–205.
762 O. W. Taft and others

Hulse, D., Branscomb, A., Giocochea-Duclos, J., Gregory, S., sperger, and R. E. Kennedy. 2000. Land cover mapping in
Ashkenas, L., Minear, P., Payne, S., Richey, D., Dearborn, an agricultural setting using multiseasonal Thematic Map-
H., Christy, J., Alverson, E., Richmond, M. 1998. Willamette per data. Remote Sensing of Environment 76:139 –155.
River Basin planning atlas, 1st ed. Pacific Northwest Ecosys- Ormsby, J. P., B. J. Blanchard, and A. J. Blanchard. 1985.
tem Research Consortium, Institute for a Sustainable Envi- Detection of lowland flooding using active microwave sys-
ronment, University of Oregon, Eugene, Oregon, 72 pp. tems. Photogrammetric Engineering and Remote Sensing
Jackson, P. L., and A. J. Kimerling. 1993. Atlas of the Pacific 51:317–328.
Northwest. Oregon State University Press, Corvallis, Ore- Pater, D. E., S. Bryce, T. D. Thorson, J. Kagan, C. Chappell, J.
gon, 152 pp. Omernik, S. Azevedo, and A. J. Woods. 1997. Ecoregions of
Jacobson, J. E. 1991. An operational program for the inven- western Washington and Oregon. US Environmental Pro-
tory of waterfowl habitat. American Congress on Surveying and tection Agency, Corvallis, Oregon.
Mapping/American Society for Photogrammetry and Remote Sens- Pietroniro, A., Leconte, R., Peters, D. L., Prowse, T. D. 2000.
ing 3:215–223. Application of a hydrodynamic model in a freshwater delta
Jensen, J. R., H. E. Mackey Jr, E. J. Christensen, and R. R. using remote sensing. Pages 519 –525 in IAHS–AISH Publi-
Sharitz. 1987. Inland wetland change detection using air- cation No. 267, Remote sensing and hydrology 2000 —Proceedings
craft MSS data. Photogrammetric Engineering and Remote Sens- of a symposium, April. Santa Fe, New Mexico.
ing 35:521–529. Place, J. L. 1985. Mapping of forested wetland: Use of Seasat
Jensen, J. R., D. J. Cowen, J. D. Althausen, and S. Narumalani. RADAR images to complement conventional sources. Pro-
1993. An evaluation of the CoastWatch change detection fessional Geographer 37:463– 469.
protocol in South Carolina. Photogrammetric Engineering and Ramsey III, E. W. 1995. Monitoring flooding in coastal wet-
Remote Sensing 59:1039 –1046. lands by using radar imagery and ground-based measure-
Johnson, J. 1993. Fieldnotes: Western Oregon, winter 1992– ments. International Journal of Remote Sensing 16:2495–2502.
93. Oregon Birds 19:86 – 88. Ramsey III, E. W. 1999. Radar remote sensing of wetlands.
Kasischke, E. S., and L. L. Bourgeau-Chavez. 1997. Monitoring Pages 211–243 in R. S. Lunetta, and C. D. Elvidge. Eds,
south Florida wetlands using ERS-1 SAR Imagery. Photogram- Remote sensing change detection. Ann Arbor Press,
metric Engineering and Remote Sensing 63:281–291. Chelsea, Michigan.
Kasischke, E. S., J. M. Melack, and M. C. Dobson. 1997. The Ramsey III, E. W., G. A. Nelson, and S. K. Sapkota. 1998.
use of imaging radars for ecological applications—A review. Classifying coastal resources by integrating optical and ra-
Remote Sensing of Environment 59:141–156. dar imagery and color infrared photography. Mangroves and
Kempka, R. G., R. P. Kollasch, and G. T. Koeln. 1992. Ducks Salt Marshes 2:109 –119.
Unlimited: Using GIS to preserve the Pacific Flyway’s wet- Rio, J. N. R., and D. F. Lozano-Garcia. 2000. Spatial filtering of
land resource. GIS World 5:46 –52. radar data (RADARSAT) for wetlands (brackish marshes)
Kiilsgaard, C. 1999. Oregon vegetation: Mapping and classifi- classification. Remote Sensing of Environment 73:143–151.
cation of landscape-level cover types. Final report. USGS- Sanzenbacher, P. M., and S. M. Haig. 2002a. Residency and
BRD GAP Analysis Program, Moscow, Idaho, 22 pp. movement patterns of wintering Dunlin in the Willamette
Lee, K. H., and R. S. Lunetta. 1995. Wetland detection meth- Valley of Oregon. Condor 104:271–280.
ods. Pages 249 –284 in J. G. Lyon, and J. McCarthy. Eds, Sanzenbacher, P. M., and S. M. Haig. 2002b. Regional fidelity
Wetland and environmental applications in GIS. Lewis, and movement patterns of wintering Killdeer in an agricul-
Boca Raton, Florida. tural landscape. Waterbirds 25:16 –25.
Lunetta, R. S., and M. E. Balogh. 1999. Application of multi- Scott, J. M., P. J. Heglund, and M. L. Morrison. 2002. Predict-
temporal Landsat 5 TM imagery for wetland identification. ing species occurrences: Issues of accuracy and scale. Island
Photogrammetric Engineering and Remote Sensing 65:1303–1310. Press, Washington, DC, 840 pp.
Metternicht, G. I. 1999. Current status and future prospects of Sokol, J., Pultz, T. J., Bulzgis, V. 2000. Monitoring wetland
radar remote sensing for cartographic applications. Cartog- hydrology in Atlantic Canada using multi-temporal and
raphy 28:1–17. multi-beam Radarsat data. Pages 536 –538 in IAHS–AISH
Mitsch, W. J., and J. G. Gosselink. 2000. Wetlands, 3rd ed. Publication No. 267, Remote sensing and hydrology 2000 —Pro-
John Wiley and Sons, New York, 920 pp. ceedings of a symposium, April. Santa Fe, New Mexico.
Morlan, J. C. 2000. Summary of current status and health of Taft, O. W., and S. M. Haig. 2003. Historical wetlands in
Oregon’s freshwater wetlands. Pages 45–52 in Oregon state Oregon’s Willamette Valley: Implications for restoration of
of environment report 2000. Division of State Lands, Salem, winter waterbird habitat. Wetlands 23:51– 64.
Oregon. Töyrä, J., A. Pietroniro, and L. W. Martz. 2001. Multisensor
Morrison, R. I. G. 1997. The use of remote sensing to evaluate hydrologic assessment of a freshwater wetland. Remote Sens-
shorebird habitats and populations on Prince Charles Is- ing of Environment 75:162–173.
land, Foxe Basin, Canada. Arctic 50:55–75. Turner, M. G. 1989. Landscape ecology: The effect of pattern
Nehls, H. B. 1994. Oregon shorebirds: Their status and move- on process. Annual Review of Ecology and Systematics
ments. Technical Report 94-1-02. Oregon Department of 20:171–197.
Fish and Wildlife, Portland, Oregon, 58 pp. US Fish and Wildlife Service 1991. Status of National Wetlands
Oetter, D. R., W. B. Cohen, M. Berterretche, T. K. Maier- Inventory: Region 1. USFWS, Portland, Oregon.
 Radar Remote Sensing and Agricultural Wetlands 763

Wang, Y., L. L. Hess, S. Filoso, and J. M. Melack. 1995. Un- The application of remote sensing techniques for an eco-
derstanding the radar backscattering from flooded and logical land survey of the Snow Goose colony at Cape
nonflooded Amazonian forests: Results from canopy back- Henrietta Maria, Hudson Bay. Pages 387–395 in Proceedings
scatter modeling. Remote Sensing of Environment 54:324 –332. of the Sixth Canadian Symposium on Remote Sensing. Halifax,
Wickware, G. M., Sims, R. A., Ross, R. K., Cowell, D. W. 1980. Nova Scotia.