Abstract
Accurate estimation of drought impacts is critical for drought relief or management, which relies on suitable modelling of the relationship between drought indicators and drought losses. However, drought impacts generally result from a multitude of influencing factors, which present myriad challenges for accurate estimation of impacts. Here, we leverage the Extreme Gradient Boosting (XGB) model to build the relationship between influencing factors (e.g., drought characteristics, socioeconomic vulnerabilities) and drought impact data, including affected crop areas (AA), affected populations (AP), and economic losses (EL), at the provincial level for the period from 2003 to 2020 in China. The model performance is assessed based on drought impact data from six provinces in the southwestern and northern regions, which are also compared with the baseline approach of similarity analysis. Results show that the XGB model performs well in capturing drought impacts across northern and southwestern China (depending on the provinces or impact data). In addition, the XGB model overall outperforms the similarity analysis for most cases of the impacts of AA and AP. Specifically, the average correlation coefficients between observations and estimation for AA, AP, and EL are 0.62, 0.52, and 0.30 based on the XGB model and 0.33, 0.41, and 0.34 based on the similarity analysis. Results from this study can be useful for managing drought risk and enhancing drought mitigation strategies in China.