A comparative analysis of machine learning-based methods for impervious surface mapping using SAR and optical data

doi:10.1080/10106049.2025.2521833

A comparative analysis of machine learning-based methods for impervious surface mapping using SAR and optical data

2025-07-01

0

OA

AI

DOI：10.1080/10106049.2025.2521833

Original

Original request for help

Save

Abstract

En 中文

Accurate and timely access to information about impervious layers is essential for urban development and ecological environment. This study employs the random forest (RF) and extreme gradient boosting algorithm to rank the significance of features, which include sentinel-1 polarization and sentinel-2 spectral information, texture features, and vegetation indices, and analyze the contribution of each feature using the SHAP method. The change analysis of the impervious layer in Changsha County from 2016 to 2024 was conducted based on the better machine learning algorithm and indicators for extracting the impervious layer. The outperformed RF algorithm had an overall classification accuracy of over 92% from 2016 to 2024. Interestingly, there was a notable rise of the area of impervious surfaces in 2019 and a substantial fall in 2023, whereas the other years saw slight changes. The suggested approach can serve as a helpful guide for extracting the impervious layer chosen via index optimization.

Keywords:

Impervious surface

random forest

XGBoost

Sentinel-1/2

feature selection

A comparative analysis of machine learning-based methods for impervious surface mapping using SAR and optical data

Abstract

Journal

Organization