Changes in habitat suitability for wintering dabbling ducks during dry conditions in the Central Valley of California

Abstract

AbstractIn arid and Mediterranean regions, landscape‐scale wetland conservation requires understanding how wildlife responds to dynamic freshwater availability and conservation actions to enhance wetland habitat. Taking advantage of Landsat satellite data and structured and community science bird survey data, we built species distribution models to describe how three duck species, the Northern Pintail (Anas acuta), Green‐winged Teal (Anas crecca), and Northern Shoveler (Anas clypeata), respond to freshwater supply and food resources on different flooded land cover types in the Central Valley of California. Specifically, our models compared duck habitat suitability between the wettest and driest conditions in each month from September through April. Using abundance‐weighted boosted regression trees, we created three sets of species occurrence models based on different covariates: (1) near real‐time (hereafter “real‐time”) covariates in which duck observations were matched to the water availability within the 16‐day window of a Landsat observation, (2) a combination of real‐time covariates and waterfowl food resource covariates describing annual corn and rice biomass and managed wetland moist soil seed yield estimates derived from Landsat data, and (3) long‐term average covariates—the most common approach to species distribution modeling—in which long‐term average surface water availability was used. We modeled the monthly occurrence of three duck species as a function of surface water availability, land cover type, road density, temperature, and bird data source. We found that dry conditions result in reduced habitat suitability, with the biggest reductions in November through January and in agricultural fields; in contrast, suitability of flooded wetland habitat was relatively robust to surface water availability. When models of habitat suitability based on long‐term average climate conditions were compared to models based on real‐time conditions, the highest long‐term suitability values occurred in areas where suitability was high regardless of whether it was a wet or a dry year. While all models performed well, the inclusion of crop and wetland plant yield covariates resulted in slightly higher model performance. Overall, species distribution models created using data on the environmental conditions present at the time of bird observations can aid conservation efforts under extreme conditions over large spatial scales.