Assessing the Joint Impact of Climatic Variables on Meteorological Drought Using Machine Learning

Abstract

With the intensification of climate change, the coupling effect between climate variables plays an important role in meteorological drought identification. However, little is known about the contribution of climate variables to drought development. This study constructed four scenarios using the random forest model during 1981–2016 in the Luanhe River Basin (LRB) and quantitatively revealed the contribution of climate variables (precipitation; temperature; wind speed; solar radiation; relative humidity; and evaporative demand) to drought indices and drought characteristics, that is, the Standard Precipitation Evapotranspiration Index (SPEI), Standard Precipitation Index (SPI), and Evaporative Demand Drought Index (EDDI). The result showed that the R2 of the model is above 0.88, and the performance of the model is good. The coupling between climate variables can not only amplify drought characteristics but also lead to the SPEI, SPI, and EDDI showing different drought states when identifying drought. With the decrease in timescale, the drought intensity of the three drought indices became stronger and the drought duration shortened, but the drought frequency increased. For short-term drought (1 mon), four scenarios displayed that the SPEI and SPI can identify more drought events. On the contrary, compared with the SPEI and SPI, the EDDI can identify long and serious drought events. This is mainly due to the coupling of evaporative demand, solar radiation, and wind speed. Evaporation demand also contributed to the SPEI, but the contribution (6–13%) was much less than the EDDI (45–85%). For SPEI-1, SPEI-3, and SPEI-6, the effect of temperature cannot be ignored. These results are helpful to understand and describe drought events for drought risk management under the condition of global warming.