Estimating Soil Temperature Using Hybrid Models Based on Discrete Wavelet Transform

Soil temperature is one of the most important variables in agricultural meteorology and has various effects on physical, chemical and biological processes in soil. In this study, daily soil temperature at depths of 5, 10, 20 and 30 cm were estimated using artificial neural network, adaptive network-based fuzzy inference system, multiple linear regression and also hybrid models based on the mentioned methods and discrete wavelet transform. To develop the models, daily minimum and maximum air temperature and daily soil temperature at depths of 5 to 30 cm in Ghoochan and Gonabad synoptic stations during 2010 to 2013 were employed, and for testing the performance of models, data of 2004 (wet year) and 2008 (dry year) were employed. Results of this study showed that hybrid models based on discrete wavelet transform could produce more accurate results compared to non-hybrid models in estimating daily soil temperature. Also, in most of the cases, the multiple linear regression (MLR) model had the weakest performance.