Northern Prairie Wildlife Research Center
For each of the 335 sample plots, we identified and mapped upland and wetland habitats according to the classification described by Cowardin et al. (1988). Upland classes were grassland (i.e., pastureland), idle grassland, hayland, cropland, woodland, scrubland, planted cover, road and railroad rights-of-way, barren land, and other (e.g., rockpiles, shelter belts, etc.). For use in models, we combined woodland, scrubland, and other classes into a single odd area class (Klett et al. 1988). Planted cover was separated further into CRP, WPA, and Waterbank Program, all of which had similar vegetative characteristics, but were established through different programs. Upland habitat areas were delineated at the scale of 12.5 cm/km on hard-copy maps, which were georeferenced and digitized using ARC/INFO (Environmental Systems Research Institute, Redlands, California, USA) Geographic Information Systems (GIS) software. We obtained digital wetland data for each plot from the USFWS, National Wetlands Inventory, St. Petersburg, Florida, USA. These data were derived from high-altitude (1:63,360), color-infrared photography acquired during the late 1970s and early 1980s. We combined this wetland data layer with our digital upland data layer to form a complete land classification coverage for each of the 335 sample plots. We also calculated the Universal Transverse Mercator easting and northing coordinates, projected in zone 14, for the center of each plot.
Each year during May, we measured the number and wet area of all wetland basins that contained water on all sample plots. We derived wetland information from aerial videography taken vertically at an altitude of approximately 4,100 m above ground from small, fixed-wing aircraft. Video imagery was later replayed, and a fixed scene was captured into a computer (equipped with a video capture card) and then saved as a Raster Vector CAD file into Map and Image Processing System (MIPS; MicroImages, Lincoln, Nebraska, USA) GIS software. Captured video scenes were then overlaid with digital wetland polygon data obtained from the USFWS, National Wetlands Inventory, St. Petersburg, Florida, USA. Using the original video as a reference, we delineated wet areas ≥0.008 ha for all wetland basins on our sample plots.
Estimation of Numbers of Breeding Pairs
We annually estimated numbers of breeding mallard, gadwall, blue-winged teal, northern shoveler (A. clypeata), and northern pintail pairs on each sample plot. We derived these estimates from a survey conducted annually on the 335 sample plots to estimate breeding duck pairs and production for the USFWS, Region 6 portion of the Prairie Pothole Joint Venture, North American Waterfowl Management Plan (Cowardin et al. 1995). We used aerial videography, as described above, to determine the wet area of all wetland basins on each sample plot, and we conducted ground counts of breeding duck pairs on a sub-sample of those wetland basins. These data were used in regression-ratio models (Cowardin et al. 1995) to estimate breeding duck pairs on each sample plot in our study area. The regression-ratio estimator was
where γ corrected for annual and geographic variation, ap was the area of pond p, N was the number of ponds on a sample plot, and (ap) was the uncorrected estimate of breeding population, (ap) = A × (ap) + B × . For A and B, we used regression coefficients provided by Cowardin et al. (1995:7). We computed the correction factor "γ" for each USFWS, Wetland Management District (Cowardin et al. 1995) in our study area each year as
where yp was the number of breeding ducks counted on pond p, and n was the number of ponds surveyed.
Selection of Plots for Nest Data Collection
Two-hundred-fifteen of the 335 sample plots contained ≥1 CRP field. We considered in final selection only those sample plots that had ≥16.2 ha of CRP and contained sufficient wetland basins to support ≥20 mallard breeding pairs during average wet conditions, as determined from the pair-wetland regression model described by Cowardin et al. (1988). These criteria were established to increase the chance of locating adequate numbers of nests to estimate nest success on individual plots. One-hundred-thirty-eight plots met our criteria (hereafter study plots). For each of the 138 study plots, we identified the nearest WPA containing ≥16.2 ha of planted cover that had plant species composition and structure similar to CRP cover.
We generated cover maps for the 138 study plots using MIPS GIS software. Each year, we selected a sample of plots to be studied from the 138 study plots. We refer to a study plot and its neighboring WPA as a replicate. To maximize the number of replicates that could be searched for duck nests, we limited the amount of cover to be searched on each replicate. Each year, we selected replicates from the pool and randomly chose fields of CRP cover and WPA cover until the last field selected reached or exceeded 81 ha for each cover type on that replicate. On replicates with ≤81 ha of a specific cover, we selected all fields of that cover type.
During 1992, we selected 14 replicates (7 in North Dakota and 7 in South Dakota) to be searched by field crews. Our goal for the first year was to gather data to determine the number of replicates needed to meet our first objective. We assumed if we had enough replicates to address objective 1, other objectives would also be met. For objective 1, we set α = 0.10, β = 0.20 and δ = 0.03 for assessing differences (2-tailed test) in DSR between CRP and WPA cover. Variance estimates for the species examined (mallard, gadwall, blue-winged teal, northern shoveler, and northern pintail) in 1992 ranged from 0.05-0.15. Because drought conditions prevailed in our study area during the pilot year that resulted in reduced nesting effort, we believed variances of ≤0.08 were attainable under more favorable conditions. With a variance of 0.08, about 100 replicates would be needed to meet our study objectives. Therefore, we developed a minimum goal of 35 replicates per year for 1993-95.
Ideally, we would have selected replicates randomly from our pool of study plots each year. However, due to the distribution of replicates and logistics, this was not feasible. Instead, we first delineated crew-areas, defined as geographic areas with ≥7 replicates that were ≤161 km apart. The 7 replicates and distance criterion were based on what we judged as logistically feasible for a field crew to collect data under the study design. Eleven crew-areas covered our entire study area, and we were able to work in 7-8 crew-areas each year (1993-1995). Klett et al. (1988) found regional differences in duck nest success among eastern and central North Dakota and eastern and central South Dakota. Thus, to optimize spatio-temporal variation, each year (except 1992) we included ≥1 crew area in each of the regions described by Klett et al. (1988). We selected replicates ≥2 times from each of the 11 crew-areas during the period of study. Each year, 3-8 replicates were chosen at random from those available in selected crew-areas.
We did not collect predator abundance data on study plots. Instead, we obtained annual indices of coyote and red fox abundance for counties in North Dakota where we had study plots (S. H. Allen, North Dakota Game and Fish Department, unpublished data). We had no indices of coyote or red fox abundance in South Dakota or Montana.
Nest Searching.In spring-summer 1992-1995, we located duck nests in CRP and WPA cover following methods of Klett et al. (1986). A nest was defined as ≥1 egg tended by a hen when found (Klett et al. 1986). Each field selected for study was searched 3 times at approximate 21-day intervals between 1 May and about 2 July of a particular year. Standard procedures were followed for marking nests and recording location and nest site data (Klett et al. 1986), and stage of incubation was determined for each nest following Weller (1956). Nests were revisited on subsequent searches or more frequently to determine fate (hatched, destroyed, or abandoned). Nests that appeared to have been abandoned on the day of discovery were considered failed due to investigator influence.
Duck Nesting Study.We estimated DSR using the methods of Mayfield (1961, 1975) and Johnson (1979). We excluded nests that were terminated when found, those that showed evidence of egg depredation or that contained eggs laid by nest parasites when found, and all nests that likely were abandoned due to investigator influence or that contained eggs broken by investigators. DSR were calculated for CRP and WPA cover on each replicate for each of 5 principal duck species: mallard, gadwall, blue-winged teal, northern shoveler, and northern pintail. Statistical analyses were conducted on DSR. Nest success (probability that ≥1 egg in a clutch hatched) was derived exponentially from DSR and laying and incubation periods (Klett et al. 1986) for presentation in portions of the results and discussion.
We developed models of DSR to identify sources of variability in observed DSR and to improve estimates of DSR in CRP and WPA cover on our study plots and to extrapolate to sample plots not studied. We used correlation analysis and stepwise regression (SAS Institute 1989) to identify variables that best explained variation in observed DSR. Explanatory variables considered for inclusion in our models were: (1) indicated breeding pairs (BPOP); (2) number of wet ponds (WETPOND); (3) area of wet ponds (WETAREA); (4) percent perennial grass cover (PGRASS); (5) indices to coyote (COYINDX) and red fox (FOXINDX) abundance in the spring, measured at the county level in North Dakota; and (6) location corresponding to Universal Transverse Mercator coordinates for the center of each study plot or sample plot (i.e., easting [EAST], northing [NORTH] and their product [E × N]). The last 3 variables were treated as a variable subset (LOC), meaning that all 3 variables were either included in or excluded from a particular model. Because of large numbers of missing values for FOXINDX and COYINDX (Montana and South Dakota study plots), we fitted a second suite of regression models after excluding those 2 variables from the group of predictors. Significance levels for adding new variables or variable subsets and retaining existing ones in our models were both P ≤ 0.15. We used weighted least squares (Snedecor and Cochran 1980), weighted by exposure days (Johnson 1979) to fit the stepwise models.
Prior to conducting stepwise analyses, we examined correlation coefficients among the explanatory variables to check for multicollinearity. Residuals from regression models were plotted against predicted values of DSR and against each explanatory variable and examined for evidence of nonconstant variance or nonlinear relations between DSR and explanatory variables.
Based on results of initial analyses, we fitted regression models for DSR in CRP cover as a function of PGRASS and LOC, by species to increase accuracy of estimating DSR in CRP fields on sample plots not studied. We then used analysis of covariance (ANCOVA), with DSR as the response variable and PGRASS, LOC, and species as explanatory variables, to identify and combine regression coefficients showing nonsignificant species effects. This approach allowed us to preserve species effects that were supported by our data and pool those that could not be statistically separated. We also considered models allowing for effects of nest age and initiation date on DSR (Klett and Johnson 1982, Grand 1995).
Duck Production Models.We used models presented by Cowardin et al. (1995: equations 3-7) and Krapu et al. (2000) to estimate production parameters for the principal species for years 1992-1997 (peak-CRP period) on each of the 335 sample plots. These production models use input data for breeding population size, availability of various nesting habitats, nesting habitat preference, nest success by habitat, wetland condition, brood survival, and brood size at fledging to estimate duck production from 10.4-km² landscapes (size of our sample plots; Table 1). Except for brood survival of gadwall and brood size at fledging for all species, inputs to production models were derived from this study, and analyses of nesting data that were collected during the period of our study. Brood survival for species other than gadwall was estimated for each sample plot using a proportional hazards model for mallard brood survival presented by Krapu et al. (2000) in which brood survival is a function of (1) percent seasonal wetlands with water, (2) hatch date, and (3) precipitation events. We assumed this model was appropriate for blue-winged teal, northern shoveler, and northern pintail. For gadwall, we treated brood survival as a constant (0.84) based on unpublished data collected in our study area (P. J. Pietz, U.S. Geological Survey, personal communication). Brood size at fledging was taken from Cowardin et al. (1995). Principal production parameters estimated for each plot were (1) overall nest success, (2) recruitment rate (number of females fledged/adult female in the breeding population), and (3) recruits (total males and females fledged). We expanded estimates from the sample plots to our entire study area following the methods of Cowardin et al. (1995) and calculated weighted means for some parameters, using weights equal to the breeding populations estimated on sample plots.
|Table 1. Input parameters for duck production models presented by Cowardin et al. (1995), and data sources used in analyses to estimate production for 5 principal duck species in the U.S. Prairie Pothole Region for 1992-1997 under 2 scenarios: (1) actual landscape configuration, and (2) cropland in place of Conservation Reserve Program cover.|
|Input parameter||Data source|
|Breeding duck pair estimates||This study, following methods of Cowardin et al. (1995)|
|W, percentage of wetland basins containing water||This study, using aerial videography|
|α, index to nesting intensity derived from W||Cowardin et al. (1995): equation 6|
|Nesting habitat preference of female ducks||NPWRCa files analyzed for this study following methods of Klett et al. (1988)|
|Area of available nesting habitat||This study|
|Duck nest success|
|CRP and planted cover||This study|
| Other nesting cover types
(grassland, hayland, cropland, etc.)
|NPWRC nest files analyzed for this study following methods of Klett et al. (1998)|
|Z, survival rate of broods||This study, using model from Krapu et al. (2000)|
|B, average brood size at fledging||Cowardin et al. (1995)|
|a U.S. Geological Survey, Northern Prairie Wildlife Research Center.|
We estimated duck production during 1992-1997 under 2 scenarios: (1) assuming actual landscape configuration (CRP present), and (2) assuming that cropland had never been converted to CRP cover. Northern Prairie Wildlife Research Center (NPWRC) maintains a repository of waterfowl nest records submitted by researchers and managers from numerous independent studies conducted throughout our study area. We used DSR estimates from nest data collected during 1990-1994 and submitted to the NPWRC Waterfowl Nest File for all habitats except CRP and WPA cover to estimate duck production under actual landscape configuration during 1992-1997. The 1990-1994 period is the most recent for which data are available that coincided with the CRP period. Because the nest file did not contain sufficient data from northeast Montana, we used DSR estimates from central North Dakota (see Klett et al. 1988) for sample plots in Montana. We used data collected on our replicates during 1992-1995 to estimate DSR in CRP and WPA cover.
To simulate duck production under the scenario in which cropland had never been converted to CRP cover, we used DSR estimates from the NPWRC Waterfowl Nest File for 1980-1984, the latest 5-year pre-CRP period. We also used the nest file to determine the preference that hens of each species display for different nesting habitats (probability that a hen will select a particular habitat for nesting, given all habitats are equally available). This analysis followed the methods of Klett et al. (1988), except that, in addition to their 1966-1984 data, we included 1985-1994 data. Preference values were derived from data for central North Dakota and were assumed to apply elsewhere in our study area. Preference for CRP, WPA, and Waterbank Program covers were assumed to be the same. Daily survival rates for nests initiated in planted cover enrolled in the USDA Waterbank Program were assumed to be the same as CRP. We used these preference values and availability of habitat types on each sample plot as inputs to our production models.