A hybrid hydrologic modeling framework-role of spatial resolution, calibration approaches and error modeling

Narmada River Basin, India. Spatial resolution and calibration strategies are two key aspects of semi-distributed streamflow modeling. Further, to enhance the skill, post-processing of model estimates is crucial to correct biases and potential model inadequacies. To this end, this study assesses these aspects: (i) four spatial resolutions to identify the best resolution, (ii) two multi-site calibration approaches for flow estimations - Basin specific (BS) and Sequential ungauged basin (SQ), and (iii) Random Forest (RF), a machine learning technique, to model the errors from the hydrological model for post-processing. To address these, a framework combining a hydrological model, Sacramento Soil Moisture Accounting (SAC-SMA), and a hydraulic routing scheme, Kinematic wave routing, was developed and employed for six diverse sub-basins. The medium resolution was chosen due to its performance and lower computational requirements, and streamflows were estimated using two calibration approaches. Of these, SQ demonstrated NSE improvements ranging from 0.05 to 0.37 for all flows and 0.08–0.69 for high flows compared to BS. Error modeling improved overall model performance, with BS showing relatively greater improvements. The results indicate that a modeling framework using medium resolution combined with BS calibration and error-correction is both computationally efficient and skillful. Further, validation of the framework on two locations indicates its strong potential to estimate streamflows in ungauged basins. • Medium resolution configuration (29 sub-basins) is chosen for streamflow estimation. • A novel calibration technique is developed to account for routing in calibration. • Modeling of error improved original streamflow estimates. • Improvement via error modeling is high for the poorly performed calibration approach. • The proposed framework has potential for streamflow estimation in data-poor regions.