Climate-smart Optimization of Sowing Dates for Double-Cropping Rice by Combining Crop Model and Machine Learning

CONTEXT Adjusting sowing dates is a crucial strategy for enhancing crop yields, improving resource utilization efficiency, and adapting to climate change. However, traditional sowing date optimization primarily relies on field trials and empirical guidance. OBJECTIVE This study aims to develop a digital method to support climate-smart decisions on sowing dates, helping ensure sustainable and secure rice production. METHODS This study developed a rice yield prediction framework by integrating crop model, extreme climate indices, and machine learning algorithms. Using this framework, the study assessed the impact of sowing date shifts on double-cropping rice yields. RESULTS AND CONCLUSION The hybrid modelling approach effectively and accurately predicted rice yields, achieving a normalized root mean square error (NRMSE) of 10.6% in the test set. Furthermore, considering rotational constraints between early and late rice, this study identified the optimal sowing dates (OSDs) and the suitable sowing window to maximize the yield of double-cropping rice in southern China. Compared to actual sowing dates, projected yield improvements under optimal sowing dates were 0.60%–4.59% for early rice, 4.55%–10.86% for late rice, and 4.27%–6.46% for total double-cropping rice yield. The findings revealed a northward delay in suitable sowing periods (SSPs) for early rice, with the ideal period ranging from February 20 to April 25 across different provinces. The SSPs for late rice generally ranged from May 24 to August 5, with the latest suitable dates mainly occurring in lower-latitude areas where thermal resources were sufficient for safe maturity. SIGNIFICANCE This study provided valuable insights into optimizing thermal and light resource utilization for double-cropping rice systems and developing adaptive strategies under climate change.