Quantifying urban land cover imperviousness as input for flood simulation using machine learning: South African case study.

The imperviousness of urban surfaces is an important parameter in simulating urban hydrological responses, but quantifying imperviousness can be challenging and time-consuming. In response, this study presents a new framework to efficiently estimate the imperviousness of urban surfaces, using satellite images with Red, Green and Blue bands and a land cover dataset with multiple built-up urban classes through remote sensing, machine learning and field verification. The methodology is adaptable to other regions with similar datasets. For a case study in Pretoria, South Africa, major differences in median total impervious area percentages (mTIA%) were identified when compared between land cover groups: residential areas had a lower imperviousness median (mTIA% = 38%) than commercial (mTIA% = 81%) and industrial (mTIA% = 89%) land cover. The mTIA% also varies between 17 and 61% for a range of different formally developed residential classes and between 14 and 43% for a range of different informally developed residential classes. These mTIA% are recommended for any urban area within the South African National Land Cover dataset. These values can be incorporated into hydraulic and hydrological models, which improve the efficiency of parameter estimation for modelling. The methodology successfully quantified temporal imperviousness changes in the study area.