Comparison of Artificial Neural Network with Multiple Linear Regression in Predicting Soil Pore Size Distributions

Assessment of the spatial distribution of soil properties has achieved considerable interest among soil scientists, both for testing hypotheses about soil formation processes and for predicting the properties of soils at non- sampled locations (mapping). In this paper, a discussion of two various procedures for modeling of spatial varieties were provided. Two modeling framework that were able to incorporate the most important effects usually found in spatial varieties, including fixed and random spatial effects, spatial trends, and heteroscedasticity were proposed. Artificial natural network (ANN) and multiple linear regressions(MLR) were used and compared for relating various soil pore distributions as outputs with 8 soil physicochemical properties, 22 topographic attributes and NDVI as inputs in a mountainous catchment in the Lordegan basin (IRAN). The results showed that MLR models could explain 31 to 60 percent of the variability in soil pore size distributions, However, ANN developed models explained the prediction of 50 to 89% of them in the study area. In ANN models RMSE and MAE values were less than them in MLR model which show higher accuracy of ANN models. So, ANN could explain the variability of the soil properties with more efficiency. Keyword: Microspores, Mesopores, Macropores, soil pores size distribution, artificial neural network development, Multivariate statistical regression