Investigating spatial and temporal variability in Whooping Crane habitat selection preference in Kansas and implications for wind energy development.

Abstract

The Whooping Crane (Grus americana) is a federally endangered species that relies on strategically located stopover habitats to complete its long migration between breeding and wintering grounds. Therefore, we need to identify suitable stopover habitats to aid their survival and recovery. Despite conservation efforts, we still don’t know how habitat preferences vary spatially and temporally, which hinders effective planning to balance species recovery with renewable energy development. To address this gap, we built upon recent research using machine learning algorithms to create species distribution models (SDM) for migratory avian species. We created an enhanced habitat suitability model for whooping cranes along the Kansas flyway using Random Forest algorithms, incorporating spatial (latitude/longitude) and temporal (date) variables to capture dynamic habitat use. Telemetry data from 2010–2016 (n = 1,253) were combined with pseudo-absences (n = 2,000) and environmental predictors (e.g., wetland cover, urban proximity) to train the model, which was validated using AUC and out-of-bag error rates. Areas with high wind potential and low conflict with suitable habitats were identified and ranked, while conflict zones with existing turbines were mapped by overlaying high-suitability habitats with current turbine locations from the U.S. Wind Turbine Database, with a 5 km buffer to assess displacement risks. Key results showed the improved model’s robustness (AUC = 0.98), outperforming previous static approaches by showing fine-scale seasonal shifts and spatial variability. Wetland cover was the top predictor in Kansas, while distance from road networks was the driver of habitat selection in Nebraska, reflecting regional landscape differences. Time-series analysis (2010–2016) showed corridor narrowing during drought years and seasonal divergence, with spring migrations using broader pathways than fall. The prediction map showed 75.07% of the study area was unsuitable and low conflict for energy development projects. 24.93% was mid-high conflict and should be designated as critical habitats. The wind-suitability overlay showed western Kansas’s agricultural zones as the optimal low-conflict areas for energy development.