A Multi-Model Approach for Improving Streamflow Prediction with Statistical Post-Processing in the Upper Narmada River Basin

In the last few decades, many hydrological models are developed for simulating various hydrological processes. The entire hydrological model has its algorithm, structure, and equations. Based on these structures, these hydrological models are divided as lumped and semi-distributed models. Differently structured models have their advantages and limitations that reduce the reliability of a particular hydrological model. To overcome the limitations of a specific model, multi-modeling was introduced in the last decade.By combining two or more hydrological models, we can reduce the uncertainty propagation in the results by a single hydrological model. Earlier Linear Regression (LR) and Quantile Mapping (QM) methods are used as multimodality methods; these methods have limitations.Bayesian Model Average (BMA) and Quantile Model Average (QMA) post-processing techniques are developed and widely used to overcome these limitations. For this multi-modeling approach, we use three hydrological models, namely; Soil and Water Assessment Tool (SWAT), Hydrologic Engineering Centre-Hydrologic Modeling System (HEC-HMS), Variable Infiltration Capacity (VIC). This study applies different multi-modeling methods for streamflow forecast in the Manot watershed part of the Narmada River basin for the monsoon (June to September) period of 2018 with multiple lead times, i.e., 1 to 5 days. Three different numerical weather prediction models, namely Japan Metrological Agency (JMA), National Center for Medium Range Weather Forecasting (NCMRWF), and European Center for Medium-Range Forecast (ECMWF), have been used for forecasting the streamflow. So far, we have SWAT and VIC model simulated results. For the Calibration period and validation period, SWAT and VIC has an R2 value of 0.6 and 0.59 for calibration and 0.59 and 0.44 for validation, respectively. The detailed results of ongoing work will be presented at the conference.