Extraction of impervious surface with Landsat based on machine learning in Chengdu urban, China

Abstract

In the process of urbanization, materials such as steel and concrete are widely used in urban construction such as road paving and house building, resulting in an increase in impervious surface area in cities. Rapid urbanization in China today has led to a huge increase in impervious surface area, which has increased the risk of urban waterlogging. Therefore, there are more reasons to conduct research on impervious surface and explore the relationship between impervious surface and urban waterlogging. This research has been done to obtain the optimal model for extracting impervious surface in Chengdu urban area and to explore the correlation between impervious surface changes and urban waterlogging in the same area. In order to obtain the optimal model, five different machine learning and deep learning algorithms such as Naive Bayes (NB), Convolutional Neural Network (CNN), Support Vector Machine (SVM), Capsules Network (CapsNet) and Random Forest (RF) have been used to extract the urban impervious surface in this study. The accuracy of the classification models is evaluated using the overall accuracy (OA), kappa coefficient (KC), and macro F1 (F1). It is found that CNN is the best option to extract the urban impervious surface of Chengdu with an OA of 0.7809. In addition, based on the optimal model to extract the impervious surface of Chengdu, a partial correlation analysis was conducted between urban waterlogging and urban impervious in the study area, it was concluded that there was a positive correlation between the two with a partial correlation coefficient of 0.979.