Author(s): Mojtaba Saboori; Epari Ritesh Patro; Alireza Gohari; Ali Torabi Haghighi
Linked Author(s):
Keywords: No Keywords
Abstract: Climate extremes increasingly threaten crop production, yet the nonlinear and interacting nature of weather drivers makes it difficult to define actionable thresholds at which yields decline. This study presents a transferable, data-driven workflow that detects and evaluates agroclimatic stress thresholds (ST) and stress types using standard weather and yield records. Using 1 km gridded daily data and regional potato yields (1990–2022) from Finland and the Netherlands, we computed six ETCCDI-style indicators for the potato growing season: RX5day (maximum 5-day precipitation), RX1day (maximum 1-day precipitation), CDD (consecutive dry days < 1 mm), CWD (consecutive wet days ≥ 1 mm), CSU (consecutive hot days TX > 30 °C), and CFD (consecutive cool-night days TN < 5 °C). The workflow combines collinearity screening, Random Forest (RF) modeling, SHAP-based feature selection, and partial-dependence analysis to derive initial thresholds (STᵢ). These are refined using class-density diagnostics to obtain adjusted thresholds (STₐ) and integrated through a multivariate majority-vote system for classifying stressed (Sh) and non-stressed (NSh) seasons. Additionally, stress types (wet, dry, heat, cool) are identified for each shocked year. Results reveal that early-season wetness dominates yield shocks in Finland, particularly when May RX5day > ~27 mm or CWD > 6 days, while mid-season dryness and heat drive losses in the Netherlands (e.g., July CDD > 9 days, RX5day < 25 mm). The model achieved strong multivariate performance (F1 ≥ 0.8 for most regions) and accurately detected historical shock years (e.g., 2018). To enable operational use, we defined “screen–confirm–modify” rules by month: screen with key indicators (e.g., RX5day₅ or CDD₇), confirm with secondary cues (CWD or CSU), and modify management (drainage, irrigation, heat mitigation) when thresholds are crossed. Overall, the approach converts complex climate–yield interactions into a small set of month- and unit-based stress thresholds that can guide drainage, irrigation, and heat-risk management, and can be readily adapted to other crops and regions.
Year: 2026