Random Forest-Based Model for Estimating Weighted Mean Temperature in Mainland China

Abstract

The weighted mean temperature (Tm) is a vital parameter for converting zenith wet delay (ZWD) into precipitation water vapor (PWV) and plays an essential part in the Global Navigation Satellite System (GNSS) inversion of PWV. To address the inability of current mainstream models to fit the nonlinear relationship between Tm and meteorological and spatiotemporal factors, whose accuracy is limited, a weighted mean temperature model using the random forest (named RFTm) was proposed to enhance the accuracy of the Tm predictions in mainland China. The validation with the Tm from 84 radiosonde stations in 2018 showed that the root mean square (RMS) of the RFTm model was reduced by 38.8%, 44.7%, and 35.5% relative to the widely used Global Pressure and Temperature 3 (GPT3) with 1° × 1°/5° × 5° versions and Bevis, respectively. The Bias and RMS of the new model in different latitude bands, various height intervals, and different times were significantly better than those of the other three comparative models. The accuracy of the new model presented a more stable adaptability. Therefore, this study provides a new idea for estimating Tm and can provide a more accurate Tm for GNSS meteorology.