Highly mobile species, such as migratory birds, respond dynamically and non-linearly to seasonal and inter-annual variability in resource availability. Despite the importance of resource thresholds in variable landscapes, species distribution models typically rely on long-term averages because of the difficulty in obtaining temporally resolved environmental data over large spatial extents. Earth observation data provide frequent measurements of changing landscapes that can be incorporated into distribution modeling. Using frequent surface water, land cover, and crop yield information in the California Central Valley, we explore how real-time environmental data influences waterbird distribution and abundance predictions. With its Mediterranean climate and seasonally irrigated agricultural, the Central Valley of California is an ideal study system for understanding the importance of temporal variability in habitat availability on birds. With 90% of wetlands having been lost to intensive agriculture and urbanization, considerable conservation resources are being devoted to identifying the location and timing of wetland and agriculture flooding that will maximize benefits for bird species.
Results/Conclusions
Using boosted regression trees and a modified hurdle technique to handle zero-inflation, we model monthly waterbird presence and abundance as a function of surface water availability, temperature, crop type, crop yield, and road density. We used data on four species of waterfowl and four species of shorebirds derived from both eBird and structured research projects in the Central Valley. Real-time surface water predictions obtained from Landsat data and crop type data from Cropscape (USDA National Agricultural Statistics Service) provided spatially explicit estimates of the distribution of suitable flooded agriculture. We compared (i) distribution models that use surface water availability within a 16-day interval around a given bird observation (i.e. “real-time” observations) to (ii) models that use long-term averages of surface water. Both real-time and long-term average models provided good fit to abundance data and distribution predictions will help managers prioritize seasonal flooding in the Central Valley. Predictions provide landscape managers and conservation partners with decision support for optimizing co-benefits to farmers and wildlife.