Abstract: The aim of this study was to investigate the relationship between soil temperature and meteorological factors in the Huaibei region and to make predictions about future soil temperature. To achieve this, authors analyzed 58 years of daily measured data from the Wudaogou Hydrological Experiment Station, spanning from 1986 to 2021. The data included soil temperature readings from 9 soil layers at depths of 0-320cm, as well as measurements of 8 meteorological factors: mean air temperature, water surface evaporation, sunshine hours, precipitation, mean wind speed at 1.5m height, mean air relative humidity, absolute humidity, and water vapor pressure difference. The study analyzed the correlations between soil temperature and meteorological factors. The data of the eight meteorological factors were downscaled using Principal Component Analysis. Finally, a PCA−BP soil temperature prediction model was constructed using the extracted principal components as input. The model was compared with a BP neural network model that used only mean air temperature as input. The results indicate that (1) soil temperature is influenced not only by mean air temperature but also by meteorological factors such as water surface evaporation, absolute humidity, and water vapor pressure difference. The influence of meteorological factors on soil temperature decreases with increasing soil depth. The response of deep soil temperature to meteorological factors exhibits a time lag, which increases with soil depth. (2) The accuracy of the PCA−BP soil temperature model in predicting soil temperature at depths of 0-80cm after the 1^st, 3^rd, and 7^th day in the future was high, with an R² value greater than 0.79. The accuracy of the PCA−BP soil temperature model in predicting deep soil temperature increased as the number of prediction days increased due to the lag effect of soil temperature response to meteorological factors. (3) Adding new meteorological factors as inputs can enhance the accuracy of the soil temperature prediction model. The PCA−BP model outperformed the BP model, which only used the mean air temperature as input, in predicting the soil temperature after the 3^rd and 7^th predicted days.

Key words: Soil temperature, Meteorological factors, Response relationship, Prediction model, Hysteresis