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

Summary

This paper presents a wavelet transform artificial neural network (WANN) approach for forecasting soil temperature using only surface air temperature observations, addressing the practical need for simple, field-applicable tools with minimal data requirements in agricultural meteorology. Analysis of hourly data from Mashhad, Iran (2010–2013) demonstrates that wavelet preprocessing significantly improves forecasting accuracy compared to standard ANN models, and that the method reliably forecasts soil temperature 1–7 days ahead at agriculturally relevant depths. The approach shows potential utility for agricultural decision-making including sowing timing and frost protection.

UK applicability

The methodology may be adaptable to UK conditions, though validation would be required for the distinct climate regimes and soil types encountered in British agriculture. UK synoptic stations with long-term hourly air and soil temperature records could provide datasets to test WANN model performance in cooler, more maritime conditions.

Key measures

Soil temperature forecasting accuracy (at 0300, 0900, and 1500 GMT); comparison of ANN versus WANN model performance; effect of temporal increment on forecast errors; forecast lead time (1–7 days ahead); soil depth (5–30 cm)

Outcomes reported

The study evaluated wavelet transform artificial neural network (WANN) models for forecasting soil temperature 1–7 days ahead at depths of 5–30 cm using only surface air temperature data. Forecasting accuracy was measured by comparing WANN performance against standard ANN models and assessing the effect of temporal increment changes on prediction errors.

Topic tags