A regression tree model to predict daily return flow of irrigation networks (Case study: Salimeh irrigation area of Dez irrigation network)

Return flow is one of the main factors in reduction of water efficiency within the irrigation networks. Hence, an accurate prediction of return flow and providing the managerial strategies to reduce it’s quantity or reuse it in an optimal way could improve the efficiency of irrigation systems. In the present study, a regression tree model is developed to predict daily return flow discharge for Salimeh Irrigation Area, SIA, in Dez Irrigation Network. The daily inflow to the irrigation area, effective rainfall, consumptive water demand, percolation loss and evaporation from the surface canals are taken as predictor variables and return flow is treated as the target variable. Four sub-models of regression tree are evaluated. Model training, validation and testing carried out based on the observed data of 1386 and 1388. The model performance shows a good match between the simulated and the field measured return flow values. Results of statistical analysis indicated that the correlation coefficients are high for all sub-models and it can be concluded that the model performs satisfactorily in simulating the return flow. Accordingly, all tree regression sub-models may be recommended in the monitoring process of the SIA. The sensitivity of the model to the predictor variables is evaluated as; daily inflow> consumptive water demand> percolation loss> effective rainfall> evaporation from the surface canals. The findings demonstrate that the model has also a desirable accuracy in spite of deleting effective rainfall and evaporation from the surface canals as predictor variables.