Predicting Spatial Distribution of Snow Water Equivalent Using Multivariate Non-linear Regression and Computational Intelligence Methods

Abstract

The evaluation of water resources given by snowfall is very important in the mountainous basins. In this study, the snow depth (SD) and snow water equivalent (SWE) were investigated to quantify the water resources stored in the snow. Multivariate non-linear regression (MNLR) method, four types of artificial neural network (ANN) and neural network-genetic algorithm (NNGA) model were initially evaluated to predict SWE in the Samsami basin of Iran. Afterwards, ordinary kriging (OK) technique was applied to interpolate the SWE values estimated by the best-performed model. For this regard, seven different MNLR, ANN and NNGA models comprising various combinations of climatic and topographic parameters including elevation (El.), slope (S), north–south (N-S) and east–west (E-W) aspects, maximum upwind slope (Sx), longitude (X) and latitude (Y) were developed to evaluate degree of effect of each of these parameters on SWE. The different experiment results showed that the NNGA5 model characterized by Delta-Bar-Delta learning algorithm and Sigmoid activation function with inputs of El., Sx, N-S aspects, S and X performed best in estimating SWE. In general, the results indicated that the NNGA technique was the most suitable method for estimation of SWE in the study area. The ANN and MNLR models were identified as the next categories, respectively. The sensitivity analysis revealed that El. and Sx were more important parameters influencing SWE than the other input parameters.