Mapping The Spatio-Temporal Dynamics of Drought in Northeast Thailand

Abstract

Drought, a globally significant natural disaster, imposes considerable economic and environmental impacts, severely impacting agriculture and socio-economic annually. This study aims to investigate the spatiotemporal dynamics of drought in Northeast Thailand by integrating remote sensing (RS) and ground observations with machine learning models. Machine learning (ML) algorithms were employed to combine these data. The study utilized five RS parameters: Vegetation Condition Index (VCI), Enhanced Vegetation Index (EVI), Temperature Condition Index (TCI), topography, and precipitation, along with ground-based data such as the Standardized Precipitation Evapotranspiration Index (SPEI). Various ML techniques, including XGBoost, Random Forest, and Extra Trees, were applied to assess the relationships among the variables. The results indicate that the Extra Trees model outperforms others in predicting drought indices. For short-term predictions, the model achieved an R² ranging from 65.26% to 94.28%, an RMSE between 1.58% and 33.28%, and an MAE ranging from 0.09% to 18.55%. For long-term predictions, the R² ranged from 78.73% to 94.8%, the RMSE from 4.55% to 31.93%, and the MAE from 0.45% to 18.14%. Key variables contributing to the model’s accuracy include precipitation (27%–67%), topography (19%–37%), and land surface temperature (6%–21%). The study examined both short-term and long-term drought conditions using the Standardized Precipitation Evapotranspiration Index (SPEI). The short-term analysis revealed significant drought events in June 2015 and April 2016, as well as recurrent droughts from late 2018 through 2019, and the early months of 2020 and 2021. In the long-term analysis, sustained negative SPEI values from mid-2015 to 2016 signaled the onset of drought, while a prolonged negative trend from mid-2018 to 2020 marked an extended drought period lasting several months, emphasizing the severity and duration of the event. In conclusion, the study provides a framework for strategic planning in drought management by integrating RS and ground observation data.