A novel Approach For Automatically Identifying and Evaluating Climate Stress Thresholds in Potato Yields

The stability of agricultural production, critical for global food security, is increasingly threatened by climate variability and extreme weather events. This study focuses on identifying and evaluating climate-induced stress thresholds for potato yields in Finland and the Netherlands, two regions with contrasting climatic and agronomic conditions. A comprehensive dataset spanning multiple decades was analyzed using advanced machine learning techniques, including Random Forest modeling, SHAP (SHapley Additive exPlanations) values for feature importance, and Partial Dependence Plots (PDPs) to detect key climate indicators and their thresholds. By classifying yields into shocked, normal, and boosted categories based on detrended yield percentiles, the study pinpoints the specific climatic conditions that transition potato yields into stress states. District-level analyses highlight spatial variations, with northern Finland and southern Netherlands particularly sensitive to compound climatic extremes, emphasizing the need for localized adaptation strategies. Findings reveal distinct regional stressors: in Finland, excessive June precipitation (>69 mm) consistently emerged as a critical driver of yield reductions, while in the Netherlands, extreme July temperatures (>31.5°C) and deviations in warm-day counts were the dominant stressors. This research is the first to identify climate-induced stress thresholds by accounting for the nonlinear and interactive effects of multiple climate factors. The findings provide actionable thresholds for policymakers and farmers, enhancing climate resilience and ensuring sustainable agricultural practices under future climate scenarios.