Answers are grounded in this paper and cite their sources.
Estimating leaf chlorophyll under deficit irrigation can support irrigation management. In this study, the CatBoost, XGBoost, and random forest (RF) algorithms were compared with support vector regression (SVR). The analyses were based on daily data (n = 1460) from eggplant plants in northern Turkey between 2015 and 2017, covering five different irrigation levels and three growing seasons, as well as nine different datasets (M1–M9) created using five key variables. Models were validated using stratified hold-out testing (n = 452), leave-one-series-out (LOSO; 15 series), and leave-one-year-out (LOYO; 3 seasons) cross-validation with nested hyperparameter tuning. Parsimonious CatBoost-M5 (irrigation level, cumulative growing degree days, cumulative solar radiation, plant height) achieved RMSE = 0.044 mg g−1 FW on hold-out testing and 0.111 mg g−1 FW (NSE = 0.87) under LOSO, outperforming the best simple benchmark (0.246 mg g−1 FW). Under LOYO cross-season extrapolation, performance degraded to mean RMSE = 0.319 mg g−1 FW (mean NSE = −0.48), matching baseline limits. Linear mixed-effects modeling of hold-out errors confirmed CatBoost achieved the lowest mean error (0.070 vs. 0.082–0.094 mg g−1 FW; p < 0.001). SHAP and LIME identified irrigation level and thermal time as the main contributors; these are model attributions, not causal effects. Transfer to unseen seasons remains limited.
No takes yet. Share an insight, caveat, or question.
Taşan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: