Noise-Augmented Boosting Framework for Accurate Soil Temperature Prediction

Abstract

Soil temperature prediction is important for farming, climate research, and environmental modeling. This research proposes an ensemble prediction method for soil temperature prediction on a daily basis using lag feature and Gaussian noise. In the proposed framework, the ensemble algorithms Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), and Gradient Boosting Regressor (GradientBoosting) are compared. The models are trained on a dataset enriched with lagged temperature values and synthetic noise and evaluated using metrics such as Root Mean Square Error (RMSE), Coefficient of Determination (R²) and Mean Absolute Percentage Error (MAPE). The results show that all models exhibited high prediction performance. However, the GradientBoosting model has the best performance with an R² value of approximately 0.96. To further assess the time series structure and model behavior in detail, diagnostic analyses such as autocorrelation analysis, box plots and residual histograms are also conducted. The proposed strategy works well for reliable short-term soil temperature prediction.