Drought prediction: Insights from the fusion of LSTM and multi-source factors

Tian Wang; Xinjun Tu; Vijay P Singh; Xiaohong Chen; Kairong Lin; Zonglin Zhou

doi:10.1016/j.scitotenv.2023.166361

Drought prediction: Insights from the fusion of LSTM and multi-source factors

Sci Total Environ. 2023 Dec 1:902:166361. doi: 10.1016/j.scitotenv.2023.166361. Epub 2023 Aug 16.

Authors

Tian Wang¹, Xinjun Tu², Vijay P Singh³, Xiaohong Chen¹, Kairong Lin¹, Zonglin Zhou¹

Affiliations

¹ Center of Water Resources and Environment, School of Civil Engineering, Sun Yat-Sen University, Guangzhou 510275, China.
² Center of Water Resources and Environment, School of Civil Engineering, Sun Yat-Sen University, Guangzhou 510275, China; Center of Water Security Engineering and Technology in Southern China of Guangdong, Guangzhou 510275, China; Guangdong Laboratory of Southern Ocean Science and Engineering, Zhuhai 519000, China. Electronic address: eestxj@mail.sysu.edu.cn.
³ Department of Biological and Agricultural Engineering, Texas A&M University, 2117, College Station, TX 77843, USA; Zachry Department of Civil & Environmental Engineering, Texas A&M University, 2117, College Station, TX 77843, USA; National Water & Energy Center, UAE University, Al Ain, United Arab Emirates.

PMID: 37595901
DOI: 10.1016/j.scitotenv.2023.166361

Abstract

In the context of current global climate change, accurate drought prediction is crucial for water resources management and agricultural production. Although traditional drought forecasting methods largely rely on historical climatic data, these methods cannot fully consider the long-term effects of factors, such as climate change, and the evaluation of prediction results is limited. Therefore, this study proposed a drought prediction and evaluation framework based on Long Short-Term Memory (LSTM), integrating multi-source factors to significantly enhance the accuracy and reliability of drought prediction models. This framework applied two distinct forecasting schemes. The first scheme utilized ten diverse factors, including precipitation, evaporation, bare soil percentage area coverage, percentage crop cover, leaf area index, runoff, surface runoff, soil moisture, temperature, and total vegetated percentage cover, to predict future precipitation and evaporation, which was then used to calculate the Standardized Precipitation-Evaporation Index (SPEI) to evaluate drought characteristics. The second scheme directly used these ten factors and historical SPEI to predict future SPEI, further assessing future drought characteristics. By comparing the drought prediction results of the two schemes in terms of data statistics, drought characteristics, and spatial patterns, it was found that the LSTM model significantly improved accuracy when handling high-dimensional complex data and predicting key factors such as precipitation, evaporation, temperature, and soil moisture. The first scheme was more accurate when predicting severe and extreme droughts, whereas the second scheme was more sensitive to predicting moderate and mild droughts and exhibited higher stability and regularity in predicting the spatial variability of drought. In summary, LSTM has made significant improvements in the accuracy, stability, and reliability of drought prediction, providing stronger support for practical applications, such as agriculture and water resources management, and offering a new research tool for further climate change research.

Keywords: CMIP6; Dongjiang basin; Drought prediction; LSTM; Multi-source factors.