Land-atmosphere and ocean–atmosphere couplings dominate the dynamics of agricultural drought predictability in the Loess Plateau, China

Accurate agricultural drought prediction is crucial for preparation for regional agricultural drought disasters. However, existing prediction models, while making some progress, have trade-offs between high accuracy and computational complexity and a poor understanding of prediction mechanisms. To bridge this gap, this study introduces the Meta-Gaussian model, a state-of-the-art statistical forecasting tool that requires no parameter adjustment for agricultural drought prediction. Its forecasting performance was used to characterize drought predictability. Four types of elements, including atmosphere elements (AT), ocean–atmosphere coupling (OA), land–atmosphere coupling (LA), and land surface elements (LD), were applied to the attribution of predictability on the Loess Plateau in China from both spatial and temporal perspectives, based on Geodetector and Random Forest, respectively. Results indicated that: (1) the spatial pattern of predictability was high in the northeast and southwest, while it was low in the middle. LD, such as soil moisture, were the most important factors dominating the spatial changes in predictability; (2) from a seasonal perspective, winter exhibited the highest predictability, while summer had the lowest; and (3) generally, most areas showed a significant downward trend at both annual and seasonal scales, except for summer. LA drove 48% of spring and 62% of autumn predictability decline areas. Meanwhile, OA drove 46% of summer predictability increase areas, and 44% of winter predictability decrease areas. Overall, the findings of this study provide valuable insights for regional drought prediction and further support the development of effective drought forecasting systems.