Spatiotemporal analysis of meteorological drought across China based on the high-spatial-resolution multiscale SPI generated by machine learning

Abstract

An accurate gridded Standardized Precipitation Index (SPI) at high spatial resolution is important for meteorological drought monitoring and assessment. However, the spatial estimation of the SPI solely derived from gridded precipitation products or interpolation of meteorological station-based data is subject to low accuracy and large uncertainty. Here, we developed a Gaussian process regression-based method by blending multiple precipitation products (GPR-BMP) with information from meteorological stations and topographical characteristics to improve the resolution and accuracy of spatial SPI. GPR-BMP was applied over China in this study, using five widely used precipitation products and meteorological station data, as well as elevation, slope, longitude and latitude. The gridded multiscale (i.e., 1-month, 3-month, 6-month, 9-month and 12-month) SPI datasets over China for 1984–2020 (GPR-BMP-SPI) were generated. Results showed that GPR-BMI-SPIs have higher accuracy than the SPIs derived from single precipitation products at all time scales. Additionally, GPR-BMP-SPIs can relatively accurately identify historical drought events. By analysing the drought spatiotemporal variation over China during 1984–2020, it was found that the annual drought frequency, severity and duration were relatively high in southeast China during 1984–2020. The annual trends of drought frequency, severity and duration showed a significant decrease on the Qinghai-Tibet Plateau. Winter drought in 1984–2020 exhibited the most distinctive features between the east and the west. Additionally, the drought area percentage showed an overall decreasing trend over the whole region of China on both annual and seasonal scales. It was also found that drought frequency contributed more to drought severity than drought duration. The GPR-BMP-SPI datasets generated in this study can serve as fundamental data support for future studies. The proposed model can be applied to other regions globally for drought monitoring and mitigation.