Hydrologic responses of single land use urban and forested watersheds and their implications to improving urban drainage design

Abstract

The need for resilient stormwater infrastructure is increasingly critical as urbanization and climate change continue to threaten water resources. Engineers and practitioners require reliable methodologies to estimate rainfall-runoff responses to adequately size and design sewer pipes and inlets, flood controls, and stormwater control measures (SCMs). The National Resource Conservation Service Technical Release 55 (often referred to as curve number [CN] method), Simple, and Rational methods are methodologies commonly implemented for such designs by regulatory agencies due to the limited inputs needed to estimate runoff; however, uncertainty is present in each model since they simplify actual hydrological processes. In this study, 13 urban and two forested watersheds were monitored, and their observed hydrologic responses were compared to modeled hydrologic responses utilizing the aforementioned methods. Significant differences in observed normalized runoff volumes (i.e., runoff coefficients) and normalized peak flow rates were found between watersheds with similar watershed characteristics and rainfall patterns, demonstrating the meticulous model inputs required to differentiate hydrologic responses between similar watersheds. A suite of alternative predictive models, informed by feature selection algorithms, were formulated and compared to the performance of standard methods. Results suggested that composite CN methods were the best predictors of event runoff volume across all watersheds (Nash Sutcliffe [NSE] and Kling Gupta Efficiencies [KGE] of 0.74 and 0.52, respectively), but were outperformed by the Simple method for watersheds with more than 45% impervious cover (NSE and KGE scores of 0.85 and 0.76, respectively). However, composite CN methods underestimated runoff volume from every watershed, a limitation that was intended to be addressed through the creation of the distributed CN method. In the distributed approach, runoff volume estimations were improved compared to the composite CN approach only when directly connected impervious area in the watershed was extremely high or extremely low. The multi-linear regression runoff volume model created herein did not outperform traditional runoff models except when rainfall depth was less than 12.5 mm (i.e., the storms for which traditional runoff volume estimation methods performed the worst). Uncertainty in modeled peak flow rate was substantially greater than for runoff volume (NSE and KGE scores between 0.48 – 0.55 and 0.39 – 0.67, respectively) across all methodologies. There is a continued need to develop more dependable estimates of peak flow which are critical to the design of pipes, flood routing, and hydrograph prediction. Overall, these results suggest one model is not optimal in all scenarios. Municipalities, regulatory agencies, and stormwater engineers should consider the adoption of multiple methodologies and use guidance from the results herein to provide recommendations as to when each model is most applicable.