Development and Performance Evaluation of a WT-LSTM Hybrid Model for Global Land Meteorological Drought Prediction

Jinfeng Bu, Kebiao Mao, xueqi Xia, Jiancheng Shi, Sayed M. Bateni, Zijin Yuan

In recent years, droughts have become increasingly frequent worldwide, leading to issues such as reduced agricultural yields and ecological degradation in various regions. To mitigate the impact of drought on human survival and development, this study utilizes the Standardized Precipitation Evapotranspiration Index (SPEI) to analyze the spatiotemporal variations of global droughts. The results reveal a clear trend of increasing drought in regions such as Africa, South America, and Asia. To improve the accuracy of global drought prediction, a hybrid model combining wavelet transform and long short-term memory neural networks (WT-LSTM) was developed. Based on this hybrid model, two forecasting schemes were designed: the first scheme leverages deep learning on historical SPEI data to predict drought trends, while the second scheme integrates multiple factors, including precipitation, land surface temperature, and potential evapotranspiration, to forecast drought variations. Analysis using multi-source data from 1979 to 2022 shows that the WT-LSTM model outperforms traditional time series models in capturing spatiotemporal patterns. Both schemes demonstrated strong performance in model training and testing, achieving high prediction accuracy with a one-month lead time (Scheme 1: mean absolute error < 0.50, Pearson correlation > 0.80; Scheme 2: mean absolute error < 0.35, Pearson correlation > 0.90). The comparison between the two schemes indicates a high spatial consistency, with Scheme 2 exhibiting a clear advantage in drought prediction. The model was further applied to identify and reconstruct drought characteristics from 2013 to 2022 in typical drought-prone regions, revealing that Scheme 2 significantly outperforms Scheme 1 in reproducing drought duration and intensity. This indicates strong capability in drought feature recognition and regional adaptability. Overall, the proposed method provides effective technical support for meteorological drought forecasting, helping policymakers and the public to take early emergency measures, reduce agricultural losses, and raise awareness of drought issues. Especially in the context of escalating climate change, interdisciplinary collaboration and research are crucial for developing comprehensive and effective strategies to address global sustainable development challenges.