Effects of Impervious Area Estimation Methods on Simulated Peak Discharges

In this paper, the variability of simulated flood discharges is evaluated using two methods for estimating impervious surfaces and common design methods in the Mission Creek watershed $\left(31{\mathrm{km}}^2\right)$ located in Santa Barbara, Calif. The two impervious estimation methods are characterized as high resolution from manually digitizing aerial photographs and medium resolution from automated interpretation of satellite data. In all cases, the high resolution fraction of impervious area is greater than the medium resolution. The difference between methods is greatest in urban regions having either small or large fractions of imperviousness (low- or high-intensity development). The largest difference observed at the modeling unit scale $(\sim 3.4{\mathrm{km}}^2)$ was 19 percentage points (26 versus 45%). To quantify discharge variability, four events (2- and $100\text{-}\mathrm{year}$ for 6- and $24\text{-}\mathrm{h}$ durations) and three models were used. The models used combinations of Green-Ampt and curve number (CN) losses and kinematic wave and unit hydrograph routing. The Green-Ampt model tended to overpredict peak discharge relative to the CN model, and kinematic wave routing and unit hydrograph agreed well except for the largest events, where the efficiency of the routing decreased travel time sufficiently to increase the peak discharge. The CN model was more sensitive to the impervious surface estimation method relative to the Green-Ampt model due to the mechanisms for incorporating imperviousness into the models. At the scale of the watershed used in this study, the effects of imperviousness estimation method on simulated peak discharge ranges from 16% for the $2\text{-}\mathrm{year}$ event to 9% for the $100\text{-}\mathrm{year}$ event. While the effects are present, they are not sufficient to shift the recurrence interval of the design discharges. At the model unit scale, the difference in peak discharge between impervious estimation methods exceeded 41 and 21% for the 2- and $100\text{-}\mathrm{year}$ events, respectively. Overall, the two impervious surface estimation methods result in different peak discharges but neither method consistently results in discharges less than flood frequency-based discharges. In terms of design risk, the results suggest that the design methods are sufficient to absorb the imperviousness variability from the two methods used here. For applications where a model is calibrated to a specific event, variability resulting from differences in estimation methods is likely to be further muted by adjusting model parameters in the calibration process.