R scripts for radar- and weather-informed forecasting of bird banding totals at Appledore Island

1_appledore_forecast_07_01_26.R — Main analysis and forecasts Compiles the daily dataset (bird-banding totals, weather-surveillance-radar measures from KBOX and KGYX, atmospheric conditions, and net-hours of effort) and fits the seasonal negative-binomial generalized additive models. Evaluates daily banding forecasts with leave-one-year-out cross-validation (observed vs. predicted), runs the training-length experiment estimating how forecast skill scales with the number of training years (including the segmented change-point analysis), generates per-species leave-one-year-out predictions, and builds the Magnolia Warbler example forecast time series. Input: daily_total_radar_wind.csv. Outputs: main-text performance figures, *_years_needed.csv, and the per-species prediction files in species_validation/ used by later scripts. 2_skill_scores_07_01_26.R — Forecast skill scores and validation robustness For each season, fits three nested models — a baseline (previous-day captures, effort, seasonal timing), a baseline + atmospheric model, and a full model adding radar measures — and evaluates them with leave-one-year-out cross-validation. Probabilistic forecasts are scored with the continuous ranked probability score (CRPS) via posterior-predictive simulation and point forecasts with RMSE, each expressed as a skill score relative to the baseline. Also includes an expanding-window (end-sample holdout) analysis that forecasts each year using only preceding years, as a robustness check on the cross-validation scheme. Input: daily_total_radar_wind.csv. Outputs: per-year skill tables (skill_by_year_*.csv), end-sample tables, and the skill figures. 3_figure4_species_lollipop_07_01_26.R — Species-level performance figure Reads the per-species leave-one-year-out prediction files produced by script 1, computes each species' forecast performance as CRPS normalized by mean daily count (alongside Pearson's r and RMSE), and plots the ten most-banded species per season as a ranked lollipop figure. Input: species_validation/*.csv (from script 1). Output: the species-level performance figure. 4_partial_dependence_helper_07_01_26.R — Predictor effects and importance (supplement) Refits the final full seasonal GAMs, generates partial-dependence (predicted-effect) plots for every predictor, and ranks predictor importance by leave-one-variable-out ΔAIC, ordering the figure panels and writing the importance tables accordingly. Input: daily_total_radar_wind.csv. Outputs: the partial-dependence supplement figures and ΔAIC importance tables (tableS_*_importance.csv).