基于XGBoost模型的广西区域PWV估算及极端降雨分析OA

Aiming at the limitation of real/near-real-time GNSS PWV retrieval under missing meteor-ological parameters,three PWV estimation models without the need for measured meteorological pa-rameters were established in the Guangxi region based on the XGBoost model.First,the XGBZ-PWV model was developed with inputs including station time(DOY and HOD),location(Longitude,Lati-tude,and Height),and GNSS ZTD,and the output feature being GNSS PWV.Then,based on the XGBZ-PWV model,two empirical PWV values were incorporated to establish the XGBZG-PWV and XGBZE-PWV models,respectively.For comparison,the GPT3 model was used to provide pressure and temperature for PWV retrieval based on GNSS ZTD(GPT3-PWV model).The accuracy of the established models was validated using GNSS PWV retrieved from GNSS ZTD,ERA5 surface pres-sure,and temperature in the Guangxi region in 2022 as the reference value.The results show that,compared to the GPT3-PWV model,the estimation accuracy of the XGBZ-PWV,XGBZG-PWV,and XGBZE-PWV models improved by 22.98％,29.03％,and 31.45％,respectively,with the XGBZE-PWV model performing the best.During two extreme rainfall events in 2022,the spatiotemporal evo-lution characteristics of PWV and rainfall were analyzed.The results demonstrate that the XGBZE-PWV model maintains good applicability even under extreme weather conditions.