干旱区地理2026，Vol.49Issue(5)：881-893,13.DOI:10.12118/j.issn.1000-6060.2025.445
中国气象干旱时空特征与混合模型预测
Spatiotemporal characteristics and hybrid model prediction of meteorological drought in China
摘要
Abstract
To enhance drought prediction accuracy,this study uses Chinese ground meteorological observations from 1980 to 2023 and selects the standardized precipitation evapotranspiration index(SPEI)as the drought indi-cator.Key predictors were identified through correlation analysis,and a hybrid wavelet transform-long short-term memory(WT-LSTM)model was developed using Theil-Sen Median trend analysis and related methods.Two pre-diction schemes—single-factor and multifactor—were designed to analyze the spatiotemporal evolution of meteo-rological drought in China.Results show(1)Spatial trends of factors are uneven;precipitation and potential evapotranspiration generally increase,with precipitation showing an"increase-decrease-increase-decrease"pat-tern from southeast to northwest,and potential evapotranspiration increasing from northwest to southeast.SPEI trends are negative in 88.87%of areas,indicating widespread drought intensification.(2)The average annual drought duration is mostly 1-2 months,with significant increases in average annual drought severity mainly in northwest,north,and northern northeast China.Trends in average annual drought characteristics exhibit a spatial pattern of higher values in the north and lower values in the south.(3)Regions with long seasonal drought dura-tions in each season do not correspond to high drought intensity;high-value areas of summer drought frequency are widely distributed,while winter drought frequency is lowest.(4)Compared with LSTM,the WT-LSTM mod-el performs better,and for single-factor predictions,the multi-factor approach enhances the ability of the model to represent complex drought patterns,significantly improving prediction performance.(5)Under the hybrid model,single-factor prediction is more suitable for regions with relatively stable climatic drought patterns,while multi-factor prediction better captures drought trends in climatically complex areas such as Xinjiang Uygur Autono-mous Region and the Qinghai-Xizang Plateau.关键词
干旱预测/标准化降水蒸散发指数/离散小波变换/长短期记忆神经网络Key words
drought forecasting/standardized precipitation evapotranspiration index/discrete wavelet transform/long short-term memory neural network引用本文复制引用
刘洋洋,毛克彪,郭中华,袁紫晋..中国气象干旱时空特征与混合模型预测[J].干旱区地理,2026,49(5):881-893,13.基金项目
中央级公益性科研院所基本科研业务费专项(Y2025YC86) （Y2025YC86）
宁夏回族自治区科技厅自然科学基金重点项目(2024AC02032)资助 （2024AC02032）
