Investigating Capabilities of Machine Learning Techniques in Forecasting Daily Streamflow Using Some Meteorological Data and Normalized Difference Snow Index (Case Study: Latian and Navroud Basins)

Accurate prediction of streamflow is crucial for water resources management, especially for the prediction of floods and soil erosion. In the current study, the capability of three machine learning (ML) methods, including Support Vector Regression (SVR), Artificial Neural Network with Backpropagation (ANN-BP), and Gradient Boosting Regression (GBR) was investigated using meteorological observations and MODIS snow cover data to forecast daily streamflow in two different basins, namely Latian and Navroud. For model development, four major predictors, including daily rainfall (P), maximum temperature (Tmax), minimum temperature (Tmin), and the Normalized Difference Snow Index (NDSI) from the MODIS satellite were used from 2000 to 2018. The performance of these models was evaluated using statistical indices. Simulation results revealed that all models presented satisfactory results in simulating daily streamflow using meteorological predictors, and the efficiency of all applied models was improved when the NDSI index was applied as an additional predictor. The best performance was observed in GBR among all studied models in both basins, whereas SVR revealed the lowest performance in forecasting streamflow for both validation and calibration steps in most cases. In general, the simulation results demonstrated higher accuracy in Latian basin than Navroud basin in both calibration and validation steps. The best performance among all models was observed using GBR with R = 0.85, NS=0.72, and RMSE = 3.43 m3/s using the NDSI index in Latian basin indicating the significant effect of snowmelt on streamflow generation in snowmelt-dominated regions.