Modelling infiltration rates in permeable stormwater channels using soft computing techniques*

In the design of permeable stormwater channels, the ability to quantify infiltration rates accurately is important for assessing the capability of such channels to perform their required functions. Most of the available infiltration models neglect the effects of water level and channel section on the infiltration rate. In this study, physical channel models, with different channel sections, were developed in the laboratory and used to measure the infiltration rates. The performance of three soft computing techniques, including Gaussian process regression, M5P, and random forest (RF) models, were evaluated against measured values. Seven independent input variables, namely, channel side slope (m), base width (b), water level (y), sand (%), silt (%), clay (%), and time (T) and the output variable infiltration rate (f(t)), were considered in the model development and validation. The Gaussian progression–Pearson VII universal kernel function model approach was found to perform best for the data set considered, followed by the RF-based model. The sensitivity investigation showed that time, water level, and channel side slope were the most influential input variables in predicting infiltration rates for permeable stormwater channels and should be given primary consideration in designing such channels.