Long-term streamflow forecasts by Adaptive Neuro-Fuzzy Inference System using satellite images and K-fold cross-validation (Case study: Dez, Iran)

Streamflow forecasting is an important issue in water resource management, as it determines some hydrological events. In recent decades, use of satellite images in water resources research has been considerably increased. In this research, obtained snow covered area by satellite images are used for a long-term seasonal streamflow forecasting model. For doing this, an Adaptive Neuro-Fuzzy Inference System (ANFIS) model is developed and its application is evaluated by using of K-fold cross-validation method. Results of this model are compared with those of the typical method (i.e., using 75% of data for training and the remaining 25% for testing the validity of the trained model). Case study is Dez basin which is located in the southwestern Iran. The used data in this research consists of 12 years (2000–2011) of monthly streamflow, precipitation, temperature and snow measurement records plus the Snow Covered Area (SCA) extracted from Moderate Resolution Imaging Spectroradiometer (MODIS) satellite images. To apply temperature and precipitation data in the model, the whole basin was divided into sub-basins and average values of each parameter for each sub-basin were allocated as model input. Finally, results are compared with those of the Artificial Neural Network (ANN) model. It was found that the K-fold validation method leads to better performance than the typical method in terms of statistical indices. Therefore, the ANFIS model that used K-fold is more reliable for forecasting especially in sophisticated models. Moreover, the application of areal versus point station data, in order to assess the Effect of regionalization on model performance, is compared. The Results showed that the use of basin data can improve the obtained results by the model considerably, in some cases results showed improvements more than 70% in coefficient of correlation (R). Moreover, it is shown that the use of satellite images could significantly improve the model performance and seems to be necessary in forecasting model. In sum, the ANFIS approach, including effective parameters on streamflow with the application of areal data, using appropriate cross-validation method demonstrates an acceptable performance on streamflow forecasting and is able to handle problems in forecasting related to shortcoming of data.