Agricultural GDP exposure to drought and its machine learning-based prediction in the Jialing River Basin, China

Investigating agricultural exposure to drought and enabling its long-term predictions are critical for climate adaptation and cropland management. This study integrates hydrological modeling, machine learning methods, and long-term agricultural economic data from 1991 to 2020 in the Jialing River Basin (JRB) to detect and forecast meteorological and agricultural droughts, as well as their impact on cropland. Initially, a soil moisture dataset with 0.083-degree resolution was generated using the Variable Infiltration Capacity (VIC) model. Subsequently, the standardized precipitation evapotranspiration index (SPEI) and standardized soil moisture index (SSMI) were applied to analyze the spatial-temporal patterns of droughts. Additionally, cropland exposure to drought was evaluated using gridded agricultural GDP data derived from pixel interpolation. Finally, four machine learning methods (Bayesian, BiGRU, CLA, and MLP) were employed to predict hydrometeorological variables from 2021 to 2030, and the agricultural economic exposures to drought under five shared socioeconomic pathways (SSPs) were also predicted. The results indicate that: (1) The JRB experienced a decline in drought severity and an increase in drought frequency from 1991 to 2020, with the drought centroid highly overlapping with cropland in the central and southern regions. (2) Over the past three decades, the proportion of high-exposure grids for agricultural GDP has increased, whereas the exposure of cropland area to high risks has decreased. Cropland has shifted from higher exposure to long-term drought to higher exposure to short-term, frequency drought. (3) Among the four machine learning models, the Bayesian model demonstrated superior performance in precipitation and temperature predictions, respectively, while the BiGRU model exhibited the best performance in long-term predictions of evaporation and soil moisture. (4) The central and southern regions will further increase in agricultural GDP exposure to both meteorological and agricultural droughts from 2021 to 2030, with exposures anticipated to increase by 20.2–34.8 % compared to the period from 2011 to 2020. Comprehensively, these findings underscore the necessity for precise drought monitoring and agricultural water management in the south-central JRB, providing vital scientific support for addressing drought management in the region.