Estimating Soil Temperature With Artificial Neural Networks Using Meteorological Parameters

The aim of this study is to develop a model which estimates monthly average soil temperature in the coming year by using some meteorological parameters that cover monthly average values measured by Turkish State Meteorological Service in 88 stations in Turkey between 1970 and 2011 years. Five different artificial neural network estimation models that are feed forward neural networks and algorithm of levenberg marquardt networks have been developed for soil temperature in different depths such as five, ten, twenty, fifty and a hundred centimeters. These models have been applied to lineer regression models and the productivity of artificial neural network models and regression models has been compared in regard to criteria like R-2, MSE and MAPE according to the criteria, it has been determined that estimations with artificial neural network models are much more better than the ones with regression models, and estimations with artificial neural network models are so close to the real soil temperatures.