Reproducible benchmark for forecasting U.S. honey‑bee colony loss
The project builds a state‑quarter panel from USDA NASS data and evaluates whether reported stressors improve next‑quarter loss forecasts. It includes scripts to fetch data, construct the panel, run gradient‑boosting and Lasso models, and generate reproducible figures and a leaderboard. Researchers and policymakers can use it to assess forecasting value of agricultural surveys and compare new models against a calibrated seasonal baseline. By providing a pre‑registered, open benchmark, it encourages transparent, out‑of‑sample validation in ecological forecasting.
View on GitHub →BenMalaga/bee-colony-loss-forecast