Accurate yield estimation is crucial for ensuring national food security and balancing supply and demand. Remote sensing (RS) process models are commonly used for regional-scale crop yield estimation, with the maximum carboxylation rate at 25 C-degrees ( V-m25 ) being a key parameter influenced by genetic varieties, environmental conditions, and spatio-temporal variations. However, most RS process models use a fixed Vm25 value to simulate maize yield at regional scales. These models ignore variations in V-m25 across time, space, and environmental conditions, leading to uncertainties in simulation. To address this issue, we developed a convolutional neural network (CNN) model combined with V-m25 -related variables to estimate dynamic V-m25 values for maize in Ningxia (NX), a typical semi-arid region of China. By integrating the normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), and leaf area index (LAI) with three drought-related hydrometeorological factors-evapotranspiration (ET), vapor pressure deficit (VPD), and soil water content (SWC)-the model's prediction accuracy of V-m25 was significantly improved, achieving an R2 of 0.88 and a root mean square error (RMSE) of 2.21 mu mol & sdot; m -2 & sdot; s(-1)on the test set. These dynamic V-m25 values were then integrated into the RS process model (PRYM-Maize-Dr) to improve maize yield simulations under drought conditions in NX. Validation using data from 2010 to 2021 showed that, at the city level, the R2 increased from 0.67 to 0.78 and RMSE decreased from 0.57 to 0.43 t & sdot; ha(-1), while at the county level, the R-2 increased from 0.59 to 0.71 and RMSE decreased from 0.60 to 0.53 t & sdot; ha(-1). These results highlight the potential of integrating RS process models with optimized V-m25 for accurate spatio-temporal crop yield estimation at the regional scale.