Assessing the role of Red-Edge and SWIR bands in urban land cover mapping using machine learning: spectral and spatial resolution trade-offs

Accurate urban landscape mapping is essential for sustainable urban planning, yet achieving high accuracy remains challenging due to the trade-off between spectral and spatial resolution. This study examines how spectral information from Red-Edge (RE) and Short-wave Infrared (SWIR) bands, even relatively lower spatial resolution enhances accuracy of urban land cover classification using machine learning techniques. The classification was performed using multispectral composites from Landsat-8 (Visible and Near-Infrared (VNIR) and SWIR bands at 30 m) and Sentinel-2 VNIR bands at 10 m; RE and SWIR bands at 20 m; and VNIR, Red-Edge and SWIR bands at 20 m. Five machine learning classifiers—Support Vector Machine (SVM), Gradient Boosting Trees (GTB), Random Forest (RF), K-Nearest Neighbors (KNN), and Classification and Regression Trees (CART)—were applied to map nine urban and non-urban land cover classes, including compact and dispersed urban areas, agriculture, grass, bare soil, roads, tree cover, water, and barren areas. The results showed that the Sentinel-2 spectral composite (VNIR, RE and SWIR bands at 20 m) achieved the highest accuracy (93.9%), followed by Landsat-8 (91.7%). The other spectral composites of Sentinel-2 produced lower accuracies of 90.4% and 87.39%, respectively. Water and compact urban areas were classified with the highest accuracy (F1 score: 0.97), while dispersed urban areas were better mapped using Landsat-8 (F1 score: 0.91). In contrast, bare soil had the lowest accuracy (F1 score: 0.63) due to spectral similarities with grass and agriculture. Among the classifiers, RF consistently performed best across all datasets. These findings highlight the importance of spectral information, particularly RE and SWIR bands, in improving urban mapping accuracy, even with lower spatial resolution. The integration of machine learning with Sentinel-2 and Landsat-8 data provides valuable insights into urban land cover dynamics, supporting data-driven planning for sustainable urban development.