Improving wheat phenology and yield forecasting with a deep learning-enhanced WOFOST model under extreme weather conditions

Abstract. Extreme weather events pose significant challenges to crop production, making their assessment essential for developing effective climate adaptation strategies. Process-based crop models are valuable for evaluating climate change impacts on crop yields but often struggle to simulate the effects of extreme weather accurately. To fill this knowledge gap, this study introduces WOFOST-EW model, an enhanced version of the World Food Studies Simulation Model (WOFOST), which integrates extreme weather indices and deep learning algorithm to improve simulations of winter wheat growth under extreme conditions. We validate WOFOST-EW using phenological, yield, and extreme weather data from agricultural meteorological stations in the North China Plain. The results show that WOFOST-EW improves simulation accuracy, with heading and maturity dates predicted more accurately by 10.64 % and 12.86 %, respectively. The R² value for yield simulations increases from 0.67 to 0.76. Validation during extreme weather years (2008 and 2018) further highlights the model's improved performance, with the R² increasing from 0.69 to 0.79 in 2008 and from 0.61 to 0.80 in 2018, respectively. WOFOST-EW effectively captures the impacts of extreme weather, offering a reliable tool for agricultural planning and climate adaptation. As extreme weather events become increasingly frequent, WOFOST-EW can assist decision-makers in more accurately evaluating crop yields, providing technical support for agricultural systems in the context of global climate change.