Prediction of Drainage Water and Soil Salinity in Rainfed Farms of Behshahr Ran Subsurface Drainage Network Using Time-series Modeling

Stochastic drainage of rainfed lands (due to its dependence on rainfall) led to the application of random variables and time series modeling in predicting the performance of drainage systems. The aim of this study was to investigate the potential of time-series models in predicting drainage water and soil salinity in rainfed farms of subsurface drainage in Ran Behshahr, Iran. First, Drainmod-S model was calibrated using measured data. Then, drainage water and soil salinity were simulated via the calibrated Drainmod-S model. The simulated outputs were used for evaluation of the results of the time-series models including AR, ARX, ARMA and ARMAX. The results showed that the ARMAX model with exogenous variables including daily value, precipitation during the previous days and average desired variables in the last two days was efficient in predicting soil and drainage water salinity, so that the absolute mean modeling error for soil surface salinity (0-50cm), soil subsurface salinity (50-100cm) and drainage water salinity was 4%, 0.4% and 5%, respectively. Comparison between the selected times-series models and the calibrated Drainmod –S model results indicated that the application of time-series models in predicting the performance of the subsurface drainage system was satisfactory. The coefficients of determination were 0.75, 0.63 and 0.57 for for salinity of soil surface and subsurface layers and drainage water, respectively. The root mean squared errors for these variables were 2412.6, 331.8 and 1724.6 mg/ L, respectively. According to the evaluation indicies, time series models were efficient in predicting soil and drainage water salinity.