Modelling Soil Temperature by Tree-Based Machine Learning Methods in Different Climatic Regions of China

Accurate estimation of soil temperature (Ts) at a national scale under different climatic conditions is important for soil–plant–atmosphere interactions. This study estimated daily Ts the 0 cm depth 689 meteorological stations in seven climate zones China period 1966–2015 with M5P model tree (M5P), random forests (RF), and extreme gradient boosting (XGBoost). The results showed that XGBoost (averaged coefficient determination (R2) = 0.964 root mean square error (RMSE) 2.066 °C) overall performed better than RF R2 0.959 RMSE 2.130 0.954 2.280 models estimating higher computational efficiency. With combination air (Tmean) global solar radiation (Rs) as inputs, accuracy was considerably high 0.96–0.97 1.73–1.99 °C). On basis Tmean, adding Rs to input had greater degree influence on other factors input. Principal component analysis indicated organic matter, water content, relative humidity (RH), Rs, wind speed (U2) are main cause errors Ts, total interpretation rate 97.9%. Overall, would be suitable algorithm China, Tmean inputs more practical combinations.