Prediction of Near-surface Air Temperature at Glacier Region using ERA5 Data based on Random Forest Regression method

Abstract

Near-surface air temperature is an essential physical parameter to estimating the Glacier ablation model. There is a significant difference between the temperature in glacier and non-glacier areas at the same altitude. The estimation model that uses low-altitude automatic weather stations (AWS) data can't be applied to the glaciers far away from the stations. This paper studies a scheme of air temperature prediction by using the random forest regression model (RFR) and ECMWF Reanalysis 5(ERA5). The results show that the average MAE and RMSE of ERA5 raw temperature product in three glaciers comes to $4.32^{\circ}\mathrm{C}$ and $5.72^{\circ}\mathrm{C}$. After the RFR model revision, the MAE and RMSE significantly decreased to $1.01^{\circ}\mathrm{C}$ and $1.41^{\circ}\mathrm{C}$. We also found that when using Linzhi weather station to calculate the temperature in Parlung No.4 glacier, the MAE of GB model just decreased $0.18^{\circ}\mathrm{C}$ compared to the lapse rate method, so it is better to use the method in this paper than the physical methods when we study a distant glacier.