Separate pond gain and loss slopes gave no clear mean gain in spatial predictions for ten prairie duck taxa in 2000–2019

We tested whether separate responses to pond gains and losses improve predictions of regional duck shares. We used ten taxa and 24 fixed prairie regions from the public waterfowl breeding survey. Each model used previous bird shares, previous May pond levels, and current pond change. We fitted the models on 23 years before 2000 and tested them on 18 complete years in 2000–2019. Separate gain and loss slopes decreased mean squared Hellinger error by 0.013%. The nominal 95% test-year block range was −0.372% to +0.365%. A check with 300 training refits and test-year draws gave a range of −1.861% to +0.701%. Thus separate slopes gave no clear mean gain. In contrast, the symmetric pond-history model decreased error by 20.54% against bird history alone. Taxon, country, and time-split results had mixed signs. A smooth curved response also gave no clear mean gain. This is a retrospective model test on previously studied survey data. It does not establish equal ecological responses or a causal effect of ponds. The package preserves source data, code, provenance, predictions, uncertainty draws, figures, and reproduction instructions.

waterfowl · species distribution · wetlands · pond change · temporal validation · ecological prediction · directional response