Abstract Combining BP neural network with particle swarm optimization(PSO-BP), using the data of 88 radiosonds stations in China from 2015 to 2017, taking the surface temperature, surface water vapor pressure, latitude, elevation, and day of year as the model input factors, and taking the Tm value obtained by the integration method as the learning objective, we establish the Tm model which is applicable to the China region(PSOTM). The accuracy of the PSOTM model is evaluated using the sounding data of 2018 as reference values, and compared with Bevis, GPT3, traditional BP neural network (BPTM) and general regression neural network (GRNNTM) models. The results show that the average annual RMSE of the PSOTM model is 3.08 K, which is 26.84%, 35.97%, 15.38% and 4.94% lower than that of the Bevis, GPT3, BPTM and GRNNTM models, respectively. The average annual bias of the PSOTM model is 0.32 K, which is 68.93%, 82.42% and 72.41% lower than that of the Bevis, GPT3 and BPTM models, respectively, and 37.50% higher than that of the GRNNTM model. The PSOTM model has relatively better accuracy and stability than the Bevis, GPT3 and BPTM models at different latitudes and elevations in China, and has good applicability in China.
SHI Yifan,LIU Lilong,LAN Shengwei et al. An Atmospheric Weighted Mean Temperature Model of China Region Based on PSO-BP Neural Network[J]. jgg, 2023, 43(12): 1300-1306.
SHI Yifan,LIU Lilong,LAN Shengwei et al. An Atmospheric Weighted Mean Temperature Model of China Region Based on PSO-BP Neural Network[J]. jgg, 2023, 43(12): 1300-1306.