Soil temperature is a very important variable in agricultural meteorology and strongly influences agricultural activities and planning (e.g. date and depth of sowing crops, frost protection, etc.). There are many physically-based studies in the literature which model soil temperature, but few are easily applicable to use in the field. Simple and precise short-term forecasting of soil temperature with minimum data requirement is the main goal of this study. Soil temperature at 03, 09 and 15 GMT were forecasted based only on surface air temperatures using artificial neural network (ANN) and wavelet-artificial neural network (WANN) models. The hourly data were collected from the Mashhad synoptic station in Khorasan Razavi province in Iran between 2010 and 2013. The results of this study showed that using wavelet transform for preprocessing improved the accuracy of soil temperature forecasting. We also found that changing the temporal increment in forecasting time did not have a noticeable effect on errors in the WANN models. WANN models can be used as accurate tools to forecast soil temperature one to seven days ahead at depths of 5 to 30 cm.

Keywords