Advancing soil temperature forecasts: an integrated evaluation of input variable selection techniques and their synergistic potential in predictive modelling

Soil temperature is a critical factor influencing plant growth, crop yield, and ecological processes. This study evaluates feature selection techniques to improve soil temperature forecasting. We applied these techniques to 39 weather stations across Florida, using meteorological data spanning 2000 to 2022, with 13 input variables, including evapotranspiration and minimum temperature. Three models, namely Multi-Layer Perceptron (MLP) Long Short-Term Memory (LSTM), and Neural Basis Expansion Analysis for Time series (N-BEATS), are used. Moreover, three optimization algorithms are applied to improve the MLP model’s accuracy: Adam, RangerAdaBelief, and AdaBelief. When integrated with the innovative SS_MLP_AdaBelief model, the standout method, Stability Selection demonstrated significant predictive accuracy, underscoring the importance of evapotranspiration and minimum temperature as key variables. The model achieved an RMSE of 0.328, an NSE of 0.873, and a CC of 0.95 at the Alachua station, demonstrating strong predictive performance. Similar trends were observed across multiple locations, indicating the model’s consistency and reliability in soil temperature forecasting. Despite the N-Beats model’s limitations, our comparative analysis, visualized through Taylor diagrams, emphasizes the necessity for precise feature selection and the synergistic application of variables and models. This research not only advances the field of soil temperature prediction but also offers valuable insights for future applications, highlighting the potential of methodical feature selection and model integration in overcoming the challenges of traditional deep learning approaches. Future research should explore hybrid deep learning architectures, larger datasets, and real-time predictive applications. This study advances soil temperature forecasting by demonstrating the synergistic impact of feature selection and optimization techniques, contributing to precision agriculture, climate change adaptation, and environmental sustainability.