Land use land cover classification in Japanese wetland and agricultural landscapes via machine learning and multi-source sentinel data

Abstract

This study assessed land use and land cover (LULC) in the Oze wetland and Hatase agricultural fields using Random Forest (RF) and Support Vector Machine (SVM). Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 optical data from 2023–2024 were processed into seasonal median composites. Input features included SAR backscatter coefficients (vertical transmit–vertical receive, vertical transmit–horizontal receive, and their ratio), Sentinel-2 bands (10 m resolution), and vegetation indices: Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), Bare Soil Index (BSI), and Modified Normalized Difference Water Index (MNDWI). Training and testing data were derived from high-resolution PlanetScope and drone imagery. Models were implemented in Python (Google Colab). Results showed RF consistently outperformed SVM, achieving kappa scores of 81%–83% in Oze and 79%–81% in Hatase, while SVM failed to exceed 80%. RF’s robustness for seasonal LULC mapping highlights its potential to support monitoring and sustainable land management in cloud-prone wetland–agriculture systems.