Fig 1.
Location of the Prairie Pothole Region (PPR) of the U.S. and Canada.
Ecoregions within the PPR are shown at their coarsest delineations to provide context for the settling patterns of 5 species of dabbling ducks across the traditional Waterfowl Breeding Population and Habitat Survey (BPOP) areas in the Prairie Pothole Region during 2002–2010. These species included; blue-winged teal (Anas discors), gadwall (A. strepera), mallard (A. platyrhynchos), northern pintail (A. acuta), and northern shoveler (A. clypeata). In this pilot effort, we only modeled areas within the traditional BPOP survey area within the PPJV & PHJV boundaries. Stratum boundaries and transect centroids show the spatial distribution of the BPOP survey population data which was linked to GIS based habitat variables.
Table 1.
Description of the explanatory variables used to predict the abundance of count of 5 species of dabbling ducks within a ∼ 11 km2 scale within the Prairie Pothole Region of the U.S. and Canada during 2002–2010.
Fig 2.
Abundance and distribution of 5 species of dabbling ducks across the traditional BPOP survey areas in the Prairie Pothole Region during 2002–2010.
These species included; blue-winged teal (Anas discors), gadwall (A. strepera), mallard (A. platyrhynchos), northern pintail (A. acuta), and northern shoveler (A. clypeata). Population estimates derived from our spatially explicit models were summed across the entire landscape and grouped into 10 percent bins, such that a value of 10 represents the smallest area in which 10% of the population is contained relative to each year. Our spatially explicit population estimates show large variation in both population estimates and settling patterns across the years we modeled. Models explained between 64% and 79% of the variation in population counts.
Fig 3.
Abundance and distribution of 5 species of dabbling ducks across the U.S. and Canadian Prairie Pothole Region.
These species included; blue-winged teal (Anas discors), gadwall (A. strepera), mallard (A. platyrhynchos), northern pintail (A. acuta), and northern shoveler (A. clypeata). Maps depict the mean and standard deviation of our yearly predictions from 2002–2010. For the mean population estimate (left inset) estimates were summed across the entire landscape and grouped into 10 percent bins, such that a value of 10 represents the smallest area in which 10% of the population is contained relative to each year.
Table 2.
Variation explained by year and the number of predictor variables selected by Random Forest model selection techniques.
Table 3.
Top 5 variables selected for each year from 2002–2010.
Table 4.
Goodness of fit statistics generated from comparing model predictions versus the out of bag test data.
Fig 4.
Linear regression of mean year and stratum level BPOP estimates as predicted by compared random forest stratum level population estimates from 2002–2010.
Random forest estimates predicted BPOP estimates well with an r2 = 0.977 and a regression coefficient of 1.005. Plots of BPOP estimates versus random forest predictions highlight a good model fit, but also show variation for certain transect and year combinations.
Fig 5.
BPOP population estimates and random forest population estimates track each other well in most population strata and in the U.S. (45) and Canada (32) strata that have the highest ducks during 2000–2010.
However, strata 34 & 47 were the two strata that consistently had highest standardized residual < -2. Post hoc inspection showed that these are two of the most intensively cropped transects within the Canadian and US PPR respectively.
Fig 6.
Comparison of design based BPOP estimates compared to population estimates generated by summing Random Forest predictions 2000–2002.
We computed yearly 95% CI’s from transect and species-specific SE’s. We only compared BPOP versus Random Forest spatial methods for strata that had almost complete overlap (strata 26–28, 30, 32–35, 38–41, 45–47). For the years we modeled, summation of random forest spatial models across all overlapping strata predicted higher population estimates than the designed based BPOP estimates (Mean = 10.6% increase (range -1.6% [2002] to 15.3% [2007]), however estimates were within the 95% confidence intervals and population trends tracked each other.
Fig 7.
The functional response of waterfowl abundance to wetlands density varied with changing population sizes within the Prairie Pothole Region of the U.S. and Canada during 2002–2010.
Waterfowl abundance was positively associated with wetlands density regardless of time lags tested, wetland density, or overall population size within the PPR. For each panel in the figure, the x-axis is the count of wetlands (0 to 100) and the y-axis is the count of 5 species of dabbling ducks (0 to 300) within a ∼ 11 km2 scale. Functional responses were generated using Loess smoothing functions in R.