Forecasting soil temperature based on surface air temperature using a wavelet artificial neural network

Summary

This study presents a practical machine-learning approach to short-term soil temperature forecasting using only surface air temperature as input, addressing a significant gap between physically rigorous soil temperature models and field-applicable tools. Data from a four-year meteorological station record in Iran were used to develop and validate wavelet-enhanced artificial neural network models. The findings indicate that wavelet preprocessing substantially improves forecasting accuracy and that WANN models are viable for agronomic decision-support applications requiring soil temperature predictions up to seven days in advance.

UK applicability

Whilst the model was calibrated on Iranian climate data from Mashhad, the methodological approach of wavelet-enhanced ANN for soil temperature forecasting is transferable to UK conditions. UK users would need to retrain models using local air and soil temperature data; the minimal data requirements (surface air temperature only) make adoption feasible for UK meteorological networks and agricultural advisory services.

Key measures

Soil temperature forecasting accuracy; comparison of ANN versus WANN model performance; temporal forecasting increments (1–7 days ahead); soil depths (5–30 cm); forecast times at 0300, 0900, and 1500 GMT

Outcomes reported

The study developed wavelet transform artificial neural network (WANN) models to forecast soil temperature at depths of 5–30 cm using only surface air temperature data. WANN models demonstrated improved accuracy compared to standard ANN approaches and provided reliable forecasts 1–7 days ahead.

Topic tags