A comprehensive management framework integrating environmental and agronomic factors is critical for stable and resource-efficient rice production. The primary objective of this study was to develop an optimization framework for transplanted rice in Jiangsu Province, China, using a Generalized Additive Model (GAM). The framework was used to quantify the inter-annual stability of optimized management schemes and assess their sensitivity to future climate scenarios. The study evaluated the model’s generalization capability using two cross-validation strategies: Leave-One-Year-Out (LOYO) and Leave-One-Site-Out (LOSO). By predicting the yield of each candidate, the scheme maximizing yield was selected as the annual optimal management practice. Validation results demonstrated robust generalization capabilities across both spatial and temporal dimensions, with the model achieving an R2 of 0.66 and an RMSE of 836 kg ha−1 in LOSO validation, and an R2 of 0.61 and an RMSE of 848 kg ha−1 in LOYO validation. Analysis of the optimized schemes revealed that transplanting date and seedling age functioned as relatively stable planning benchmarks across years, whereas inter-annual adaptation was achieved primarily through adjustments in planting density and nitrogen inputs. Beyond yield prediction alone, this framework translates interpretable GAM response surfaces into spatially differentiated management prescriptions and highlights both soil-conditioned variable-rate strategies and the distinction between stable and adaptive management components under future climate scenarios.
Qiu et al. (Tue,) studied this question.