Rainfall-runoff modeling in a flashy tropical watershed using the distributed HL-RDHM model

Many watersheds in Hawai\u27i are flash flood prone due to their small contributing areas and frequent intense rainfall. Motivated by the possibility of developing an operational flood forecasting system, this study evaluated the performance of the National Weather Service (NWS) model, the Hydrology Laboratory Research Distributed Hydrologic Model (HL-RDHM) in simulating the hydrology of the flood-prone Hanalei watershed in Kaua\u27i, Hawai\u27i. This rural watershed is very wet and has strong spatial rainfall gradients. Application of HL-RDHM to Hanalei watershed required (i) modifying the Hydrologic Rainfall Analysis Project (HRAP) coordinate system; (ii) generating precipitation grids from rain gauge data, and (iii) generating parameters for Sacramento Soil Moisture Accounting Model (SAC-SMA) and routing parameter grids for the modified HRAP coordinate system. Results were obtained for several spatial resolutions. Hourly basin-average rainfall calculated from one HRAP resolution grid (4km×4km) was too low and inaccurate. More realistic rainfall and more accurate streamflow predictions were obtained with the 1/2 and 1/4 HRAP grids. For a one year period with the best precipitation data, the performance of HL-RDHM was satisfactory even without calibration for basin-averaged and distributed a priori parameter grids. Calibration and validation of HL-RDHM were conducted using four-year data set each. The model reasonably matched the observed peak discharges and time to peak during calibration and validation periods. The performance of model was assessed using the following three statistical measures: Root Mean Square Error (RMSE), Nash-Sutcliffe efficiency (NSE) and Percent bias (PBIAS). Overall, HL-RDHM\u27s performance was very good (NSE\u3e0.75, PBIAS\u3c±10) for the finer resolution grids (1/2 HRAP or 1/4 HRAP). The quality of flood forecasting capability of the model was accessed using four accuracy measures (probability of false detection, false alarm ratio, critical success index, and probability of detection) for three return periods 1.005, 1.05 and 1.25. HL-RDHM is more accurate at the lower threshold discharge (65m3s-1) than at the higher ones (140 and 248m3s-1) as indicated by lower values of probability of detection and critical success index, and a higher value of the false alarm ratio. These results suggest that HL-RDHM may be suitable for flood forecasting applications in watersheds with steep terrain and strong spatio-temporal variability of precipitation, e.g. US Pacific North-West, and tropical islands