Abstract
Soil temperature (ST) is one of the critical parameters in agricultural meteorology and significantly influences physical, chemical, and biological activities in the soil environment. One of the major challenges in agricultural studies is the limited number of synoptic stations for measuring ST. Novel data mining methods offer effective solutions for obtaining reliable estimations while reducing costs and improving accuracy. This study estimated daily ST at depths of 5 cm, 10 cm, 20 cm, 50 cm, and 100 cm using meteorological parameters from an American synoptic station over three years (2020-2022). Seasonal ARIMA (SARIMA), Multiple Linear Regression (MLR), and Artificial Neural Network (ANN) models were applied for ST prediction. A bivariate correlation test with a p-value less than 0.05 was conducted to determine the relationship between meteorological variables and ST. The results indicate that the ANN model outperforms SARIMA and MLR in predicting ST at all depths. For instance, at 5 cm depth, the ANN model achieved RMSE = 0.85, r = 0.98, MAE = 1, and PBIAS = 1.5%, compared to SARIMA (RMSE = 1.5, r = 0.96, MAE = 1.16, PBIAS = 2.5%) and MLR (RMSE = 1.35, r = 0.97, MAE = 1.3, PBIAS = 3%). Similarly, at 100 cm depth, the ANN model achieved RMSE = 0.65, r = 0.91, MAE = 0.55, and PBIAS = 2.2%, compared to SARIMA (RMSE = 1.3, r = 0.96, MAE = 0.58, PBIAS = 3.5%) and MLR (RMSE = 1.15, r = 0.91, MAE = 0.68, PBIAS = 3.8%). The analysis also revealed that surface temperature (Avg. Infrared) and air temperature (Avg. T) were the most influential parameters in ST prediction. Additionally, the ANN model exhibited the lowest error rates across all depths, highlighting its superior capability for estimating daily ST in agricultural soils. This study provides a comprehensive framework for accurately estimating daily ST at different depths, offering valuable insights for agricultural and environmental applications.