Modeling of Subsurface Throughflow in Urban Pervious Areas

Abstract

Infiltration excess runoff, i.e., runoff as a result of the rainfall intensity exceeding the infiltration capacity of the soil surface, has traditionally been considered the only contributor to the surface runoff from urban pervious areas. However, recent studies show that subsurface throughflow also can be a significant contributor to urban stormwater runoff. Although rainfall-runoff from urban pervious areas can contribute with large quantities of runoff, only little knowledge exists on this topic. In this study, experimental field observations of subsurface throughflow from the literature are used to assess the capability of different models to simulate this type of runoff. It is investigated how well three new modeling approaches in urban drainage engineering (linear reservoir, regression, and shallow neural network models) performs in simulating subsurface throughflow compared to two commonly used models (the time-area and kinematic wave model). The models are compared with the measured runoff rate and evaluated by the root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE), and Bayesian likelihood (L). Generally, a neural network containing 60 neurons and using up to 180 min of data back in time produces the best results (RMSE=0.59 Lmin−1, NSE=0.91, and L=0.92). However, both the kinematic wave (RMSE=1.06 L min−1, NSE=0.71, and L=0.76) and linear reservoir model (RMSE=0.98 L min−1, NSE=0.75, and L=0.78) generate reasonable results despite their significantly simpler modeling approaches.