Evaluation of GSSHA for Simulating Sediment Concentrations in Steep Hawaiian Watersheds

Abstract

Physics-based hydrological models, like GSSHA, potentially have an accuracy advantage over empirical or semi-empirical models, when simulating processes in steep Hawaiian watersheds, which can be very dissimilar to lands (or field plot experiments) where the empirical relationships were developed. GSSHA, therefore, could potentially be a valuable tool for hydrology projects in Hawaii, but is seldom used in Hawaii. In this study, GSSHA was evaluated and used to predict streamflow and sediment concentrations in the upper region of Halawa watershed on leeward Oahu—a watershed with an average overland basin slope of 0.65—during single-event storms. Single-event simulations were the focus here, because of the importance of single-event storms to Hawaii’s sediment issues (other studies claim that a large portion of annual sediment flux occurs during low-frequency, high-magnitude storms). Streamflow and sediment concentration data collected at 15-minute intervals were used to calibrate and validate the simulations. Several case studies were developed to test findings from past studies, including the finding that GSSHA tends to over predict sediment for low-frequency, high-magnitude storms, when calibrated with only high-frequency, low-magnitude storms (as stated in the GSSHA manual). It was found that GSSHA becomes less accurate in sediment concentration predictions for low-frequency storms (>2-year recurrence interval) when calibrated with only high-frequency storms (<2-year recurrence interval). It is shown that the degradation in accuracy was likely a result of over prediction in streamflow. It was also found that GSSHA made more accurate predictions of streamflow when the validation storm events were similar to calibration events in rainfall duration, accumulation, and peak intensity, indicating that one set of calibration parameters may not be sufficient for all storm events. Distinctively, however, it is shown that one set of calibration parameters may be sufficient for all storm events for predicting sediment concentrations. GSSHA predicted sediment load with an 8.8% error for a validation storm six times larger (based on streamflow) than calibration. Additionally, in response to studies that have shown that many sediment transport equations over predict sediment flux in steep terrain by orders of magnitude, two sediment transport equations—Engelund-Hansen and Kilinc-Richardson—were tested and compared to determine which equation is better for steep terrain. It was found that the Engelund-Hansen equation consistently outperformed the Kilinc-Richardson equation, indicating that the Engelund-Hansen equation is a better choice for use in steep Hawaiian watersheds.