The pond gain depended on northern shoveler in fixed one-year forecasts for ten prairie duck taxa in 2000–2019

We tested whether known pond estimates improve forecasts of duck numbers one year later. We used ten duck taxa and 24 fixed prairie areas from the USFWS waterfowl survey. We fitted separate log-abundance models to 23 annual transitions with targets in 1975–1999. We kept the fits fixed and tested 18 complete target years in 2000–2019. Adding origin-year log ponds to autoregression decreased mean squared log error from 0.064998 to 0.061345, or 5.62%. A conditional 95% test-year block interval was 2.38% to 8.84%. Seven taxa had higher error. Northern shoveler had a 41.32% gain and supplied 168.54% of the net pooled reduction. A later check without that taxon gave a gain of −5.00% (−10.65% to +0.66%). A no-change forecast had lower mean error, 0.049306. Against no change, the pond forecast gain was −24.42% (−59.06% to +0.89%). Annual refitting gave a smaller pond gain of 1.90%, with an interval that included zero. The benchmark shows that a pooled gain can depend on one taxon and a weak baseline. It does not identify a causal pond effect or test all habitat models. The intervals omit source sampling error and condition on fitted models. The package has unchanged public source data, code, provenance, all predictions, block draws, figures and offline instructions. Separate arithmetic checks and a clean repeat passed.

waterfowl · ecological forecasting · wetlands · autoregression · prediction benchmarks · northern shoveler · open data