Scholarlink

<p>Urban land use/land cover (LULC) classification remains challenging due to high intra-class variability, spectral ambiguity, and mixed pixels in heterogeneous urban environments. For example, concrete and asphalt often exhibit spectral responses similar to barren leading to frequent misclassification in peri-urban transition zones. This study presents a comparative assessment of four widely used machine learning classifiers- Classification and Regression Tree (CART), Support Vector Machine (SVM), Random Forest (RF), and Gradient Tree Boosting (GTB) implemented on the Google Earth Engine platform for LULC mapping. Freely available multi-sensor satellite datasets from LISS-IV (5.8&#xa0;m), Sentinel-2 (10&#xa0;m), and Landsat-9 (30&#xa0;m) were selected for their operational relevance and complementary resolutions, enabling systematic evaluation of resolution and spectral trade-offs in classification. Classifiers were trained using raw spectral bands and enhanced feature sets incorporating NDVI, NDWI, and sensor-specific band ratios (Red/Green and Green/Blue for Landsat-9 and Sentinel-2, Red/NIR and Green/NIR for LISS-IV). A standardized dataset of 100 training and 50 testing samples per class was used. Higher training proportions yielded only marginal accuracy gains, while lower proportions reduced classification stability. Results indicate that feature enhancement improved overall classification accuracy by approximately 2.4%-8.8% across sensors and classifiers. SVM, RF, and GTB consistently outperformed CART, with SVM applied to Landsat-9 achieving the highest accuracy (93.2%, Kappa = 0.915). The findings demonstrate the importance of integrating optimized feature engineering with robust classifiers to improve urban LULC mapping reliability. This study provides practical guidance for remote sensing analysis, urban planners and supports data-driven, sustainable urban development.<?pagebreak??></p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Urban Land Use Land Cover Mapping Using Multi-Sensor Satellite Data and Machine Learning in Google Earth Engine: Toward Sustainable Planning

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Adarsh Agrawal,
  • Priyamitra Munoth,
  • Vikas Poonia,
  • Lixin Wang

摘要

Urban land use/land cover (LULC) classification remains challenging due to high intra-class variability, spectral ambiguity, and mixed pixels in heterogeneous urban environments. For example, concrete and asphalt often exhibit spectral responses similar to barren leading to frequent misclassification in peri-urban transition zones. This study presents a comparative assessment of four widely used machine learning classifiers- Classification and Regression Tree (CART), Support Vector Machine (SVM), Random Forest (RF), and Gradient Tree Boosting (GTB) implemented on the Google Earth Engine platform for LULC mapping. Freely available multi-sensor satellite datasets from LISS-IV (5.8 m), Sentinel-2 (10 m), and Landsat-9 (30 m) were selected for their operational relevance and complementary resolutions, enabling systematic evaluation of resolution and spectral trade-offs in classification. Classifiers were trained using raw spectral bands and enhanced feature sets incorporating NDVI, NDWI, and sensor-specific band ratios (Red/Green and Green/Blue for Landsat-9 and Sentinel-2, Red/NIR and Green/NIR for LISS-IV). A standardized dataset of 100 training and 50 testing samples per class was used. Higher training proportions yielded only marginal accuracy gains, while lower proportions reduced classification stability. Results indicate that feature enhancement improved overall classification accuracy by approximately 2.4%-8.8% across sensors and classifiers. SVM, RF, and GTB consistently outperformed CART, with SVM applied to Landsat-9 achieving the highest accuracy (93.2%, Kappa = 0.915). The findings demonstrate the importance of integrating optimized feature engineering with robust classifiers to improve urban LULC mapping reliability. This study provides practical guidance for remote sensing analysis, urban planners and supports data-driven, sustainable urban development.

Graphical Abstract