Modelling Capabilities of Two Physically Based Hydrologic Models for Streamflow Simulations

Hydrologic processes in a watershed are typically simulated through hydrologic models due to their availability in the public domain and improved computational capacities. However, choosing a suitable model among the many available for a region of interest is challenging. In our work, we compared streamflow generated by the Soil and Water Assessment Tool (SWAT) and the Hydrological Engineering Centre-Hydrologic Modeling System (HEC-HMS) in the Kalu River Basin (KRB), Sri Lanka, frequently impacted by floods. Meteorological data including rainfall and temperature from 1990 to 2000 were used to force the hydrologic models. In addition, we used soil, land use data and a digital elevation model (DEM) for model development. During the calibration phase (1993-1996) of the SWAT model we achieved a coefficient of determination (R²) of 0.93 and a Nash-Sutcliffe Efficiency (NSE) of 0.87. In the validation phase (1997–2000), these indices yielded values of 0.87 and 0.66, respectively. In the HEC-HMS model, during the calibration phase, R2 and NSE yielded values of 0.89 and 0.91 while in the validation phase, these indices yielded values of 0.77 and 0.56, respectively. The exceedance probabilities at 10%, 50%, and 90% derived from flow duration curves (FDCs) from HEC-HMS and SWAT models were 395, 159, 54.5 and 400.5, 148, 29.11 (all in m3/s), respectively. Similarly, for observed flow, these values were 344.40, 138.98, and 65.35 m3/s, respectively. Thus, the FDCs suggest that the HEC-HMS model captures low flows reasonably. Neither model accurately resembled high flows. During the first inter-monsoon season (March-April) the HEC-HMS and SWAT underpredicted 3%, and 4% while during the northeast monsoon season (December-February) the models underpredicted 9%, and 2%, respectively. Similarly, during the second inter-monsoon season (October-November) and the southwest monsoon season (May-September), HECHMS and SWAT models overestimated observed flow by 11%, 5%, and 8%, 17%, respectively. Both models performed reasonably well on a seasonal basis with slight over-predictions and under-predictions. Overall, it is clear that both models can generally capture the hydrology of the KRB.