Evaluation the Preprocessing Effect of Satellite Images Input Parameters in to Artificial Neural Network for soil texture determination

Soil properties determination such as soil texture is an important tool for proper management, optimized and sustainable use of soil. The aim of this research is determination the soil texture, geometric mean and standard deviation of soil particles using images of MODIS sensor in the period of 2015-2016. After soil texture determination using hydrometer method, artificial neural network model have been used for soil properties determination using reflectance, thermal bands and indices of satellite images. The preprocessing is one the most important parts in the modeling process. In this research, the preprocessing of input parameters was based on the significance of correlation coefficient, using the constant number of input parameters and stepwise regression. Stepwise regression method has the minimum error which the RMSE decreasing rather to the significance correlation and constant input parameter methods for clay content determination was 22 and 18.6 percent, for sand determination 43.19 and 71.23 percent, for geometric mean determination 80.14 and 27.29 percent, for standard devotion determination 21.17 and 38.71 percent. Also, in the case of silt calculation, the RMSE decreasing rather to the constant input parameter method was 55.13 percent. The minimum of average criteria; RMSE, MAE and MRE for three different preprocessing methods was related to the sand particle, for example the average of MAE for clay, sand and silt was 1.74, 1,2 and 1,66 respectively. RMSE decreasing of sand, 27.77, using artificial neural networks indicted the better performance of method relative to the classical regression. Generally, the kind of input parameters and kind of modeling is important factors in the soil texture determination.