Comparing statistical methods for detecting climatic drivers of mast seeding

Understanding the drivers of mast seeding is critical for predicting reproductive dynamics in perennial plants. Here, we evaluate the performance of four statistical methods for identifying weather-associated drivers of annual seed production, i.e, weather cues: climate sensitivity profile, P-spline regression, sliding window analysis, and peak signal detection. Using long-term seed production data from 50 European beech (\emph{Fagus sylvatica}) populations and temperature records, we assessed each method’s ability to detect a benchmark window around the summer solstice. All methods successfully identified biologically meaningful windows, but their performance varied with data quality, signal strength, and sample size. Sliding window and climate sensitivity profile methods showed the best balance of accuracy and robustness, while peak signal detection had lower consistency. Cue identification was more reliable with at least 20 years of data, and predictive accuracy was highest when models were based on seed trap data. A simulation study showed method-specific sensitivity to signal strength, with the sliding window performing best. Our findings provide a means to improve masting forecasts through a practical guide for selecting appropriate cue identification methods under varying data constraints.