Evaluating the association between morphological characteristics of urban land and pluvial floods using machine learning methods

Abstract

Spatial pattern of urban land plays a significant role in the occurrence of pluvial floods. While landscape metrics have been widely used to reflect land use spatial patterns, the association between the morphological spatial pattern of urban land and pluvial floods is much less understood. Unlike landscape metrics, morphological spatial pattern analysis could discover more spatial information on the formation and layout of land use. To fill such knowledge gap, this research evaluated whether the morphological characteristics of urban land matter to pluvial floods based on machine learning techniques. Pearson's correlation test and random forest algorithm were employed to discover the linear and non-linear associations between the density of flood hotspots and a set of underlying influencing factors, respectively. A case study in a low-lying coastal city has highlighted the strong influence of land use morphological characteristics on pluvial floods with baseline influencing factors considered. The density of flood hotspots was positively associated with the core, loop, edge, and bridge of urban land at sub-watershed level, but negatively associated with islet urban land. While it is unfeasible to restrain rapid urban sprawl and population growth in many developing countries, this research could give specific guidance for upgrading the morphological spatial pattern of urban land. A proper design of key morphological elements may overcome the negative influence of urban expansion on pluvial floods. These findings could help urban planners and policy makers to highlight the importance of land use morphological spatial patterns.