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

Summary

This paper presents a practical forecasting tool for soil temperature based on surface air temperature data using wavelet-preprocessed artificial neural networks. The authors demonstrate that wavelet transform preprocessing improves forecast accuracy compared to standard ANN models, and that WANN models can reliably forecast soil temperature 1–7 days in advance at agriculturally relevant soil depths with minimal input data requirements. The approach addresses a gap in accessible, field-applicable soil temperature prediction methods important for agronomic decision-making.

UK applicability

The methodology is potentially transferable to UK conditions, as soil temperature forecasting is equally relevant for UK crop management (sowing timing, frost risk). However, the model was trained on Iranian climate data; recalibration using UK meteorological records would be necessary to ensure accuracy under UK weather regimes and soil types.

Key measures

Soil temperature forecasting accuracy (presumably mean absolute error, root mean square error, or similar metrics comparing predicted to observed soil temperatures at 0300, 0900, and 1500 GMT)

Outcomes reported

The study developed and compared artificial neural network (ANN) and wavelet transform artificial neural network (WANN) models to forecast soil temperature at multiple depths (5–30 cm) and times of day based solely on surface air temperature data. The models were validated using hourly meteorological data from Mashhad, Iran (2010–2013) and evaluated for forecasting accuracy across 1–7 day lead times.

Topic tags