Response of streamflow to groundwater pumping in a tropical watershed

Hawaii streams are known for generating intense and short-duration runoffs that often cause flash floods. In the past, many of these streams have been reported for decreasing trends in base flow. Trend in precipitation is believed to be correlated with trend in base flow but its relationship with other factors (e.g., Groundwater pumping) is unknown. The distributed hydrology soil vegetation model (DHSVM) was applied in the mountainous 13 km2 Makaha watershed, Hawaii to: 1) assess model applicability in predicting hydrological processes of this flashy tropical watershed and ii) evaluate model suitability as a tool to assess the link between streamflow and groundwater pumping. The model was calibrated (1971-1980) and validated (1981-1990) for the pre-pumping conditions against measured daily streamflow. Simulations were performed at 3 h time step and at 30 m grid resolution. Model initial states were generated by running it for a warm-up period (1966-1970). Performance of the model was evaluated by comparing the daily measured and simulated streamflow using Nash-Sutcliffe efficiency (NE), correlation coefficient (R), and root mean squared error (RMSE) for each year of simulation. Monthly measured and simulated streamflow for pre-pumping (before 1991) and during pumping (after 1990) periods were used to quantify the decline in streamflow as a result of pumping. During calibration and validation NE were >0 except for the years 1975, 1976, 1980, 1987, and 1989. Negative NE values during these years are mostly attributed to large events for which streamflow was over predicted. Correlation coefficient and RMSE significantly varied between years. During calibration and validation correlation coefficient and RMSE were between 0.27-0.85 and 0.03-0.81 m3 s-1 , respectively. Initial results indicate 15% decline in streamflow compared to pre-pumping conditions which represent 38% of mean annual streamflow decline during the same period. This decline in streamflow can be partly attributed to groundwater pumping. Overall, the model performance is reasonably good despite its use in a relatively small flashy watershed with an average slope of 66% and high precipitation gradient (22.4 cm km-1). In small watersheds, spatial variability of soil and vegetation data can lead to poor model performance. Therefore, further efforts are needed in quantifying the spatial distribution of soil and vegetation parameters to improve the model performance.