Abstract:

Based on observational data from 9 meteorological stations in Shaanxi province and the CRU TS v4.09 monthly grid climate dataset released by the Climate Research Centre of the University of East Anglia in the United Kingdom from 1981 to 2019, the XGBoost machine learning algorithm was introduced. Taking 1981−2010 as the training period and 2011−2019 as the testing period, a machine learning correction model was established. A correction was applied to the hydrothermal meteorological data from the CRU TS v4.09 monthly grid climate data set for Shaanxi province from 1912 to 1949. Combined with the Standardized precipitation evapotranspiration index (SPEI), the meteorological drought index of Shaanxi from 1912 to 1949 was reconstructed, and the spatiotemporal evolution characteristics of meteorological droughts were analyzed to provide a reference for regional historical drought reconstruction. The results showed that: (1) the hydrothermal meteorological correction model constructed by the XGBoost algorithm exhibited satisfactory simulation performance, with precipitation R² of 0.77 and RMSE of 29.27mm, and mean temperature R² of 0.92 and RMSE of 2.08℃ during the testing period. (2) Based on the hydrothermal meteorological correction model, the monthly grid based hydrothermal meteorological data of CRU TS v4.09 in Shaanxi province from 1912 to 1949 were corrected. From 1912 to 1949, precipitation in Shaanxi showed a slightly fluctuating pattern of increasing in the north and decreasing in the south, with a nonsignificant trend. The mean temperature exhibited an extremely significant increasing trend (P<0.05). (3) The drought coverage and intensity in Shaanxi were relatively low from 1912 to 1921, but rose from 1922 to 1949 and showed a characteristic of multi peak oscillation. In 1941, the drought coverage and intensity reached the peak of the entire region. (4) From 1912 to 1949, drought frequency increased while drought intensity decreased from south to north across the province. High-frequency areas of severe drought were mainly distributed in Guanzhong and parts of southern Shaanxi. (5) The overall drought tendency of Shaanxi intensified from 1912 to 1949, and southern Shaanxi witnessed the most significant trend. In summary, the XGBoost algorithm can effectively correct historical gridded meteorological data, providing technical support and methodological reference for the reconstruction and evolution analysis of long−term historical meteorological droughts.

Key words: Meteorological data correction, Drought reconstruction, SPEI index, Spatiotemporal evolution