forecasting of runoff and sediment using neural network and multi regression in aghajari marls

erosion and sediment movement phenomena are one of the most complex issues in management of rivers drainage areas that in water projects are very important. that its measurement wants high time and cost. issue of surface runoff in river basin is a complex issue that human knowledge and understanding about its physical laws a viewpoint of some mathematical formulas is limited. in this study to investigate modeling runoff and sediment production in different land uses of aaghajari formation deposits, part of margha watershed in izeh city with area 1609 hectares was selected. in this study, some soil physical and chemical characteristics such as percentage of sand very fine, sand, clay, silt, ph, electrical conductivity, moisture, calcium carbonate and soil salinity in different land uses of aghajari formation were used. then the rain simulator in 7 point and with three replicated in different intensities 0.75, 1 and 1.25 mm in minute in three land use range, residential areas and agricultural lands, were used the amount of runoff and sediment. and the same of number were sampled in 0-20 cm in soil layer. in totally, 126 times sampling runoff and sediment were done. and 189 soil experiments were done. in order to perform all statistical analysis were used 11.5 spss and excel and matlab 2008 software. the results showed that multi regression analysis in conditions with high input and little output data shows more favorable results than neural network. and in high intensities owing to data homogeny, neural network operation than to low precipitation intensities is better. but in multi regression in high and low precipitation intensities showed acceptable operation. the average of relative error in three land uses in sediment production in precipitation intensity 0.75 mm in minute were in multi regression 7.2 percent and root mean square error 0.06. and in neural network in same precipitation intensity the average of relative error 146/9 percent and root mean square error 0.41 were. the average of relative error in three land uses in sediment production in precipitation intensity 1 mm in minute were in multi regression 8.5 percent and root mean square error 0.19. and in neural network in same precipitation intensity the average of relative error 96.36 percent and root mean square error 0.85 were. the average of relative error in three land uses in sediment production in precipitation intensity 1.25 mm in minute were in multi regression 1.8 percent and root mean square error 0.38. and in neural network in same precipitation intensity were the average of relative error 37/6 percent and root mean square error 0.73.