A Comparative Analysis of Land Use Classification Methods Using Landsat and Ancillary Data in Urban Mapping

Abstract

This study compares the performance of parametric (LDA) and non-parametric (CTA, RF, SVM) classification algorithms in mapping urban and surrounding land cover types in Balikesir, T & uuml;rkiye, using Landsat 8 OLI/TIRS imagery and ancillary data. Seven land cover classes-built-up areas, roads, water bodies, forests, meadows, agriculture, and barren land-were classified based on 2,480 ground truth points. The Random Forest (RF) classifier achieved the highest overall classification accuracy (Kappa = 0.90) and an F1-score of 0.99 for the built-up class, outperforming LDA (Kappa = 0.86), SVM (0.83), and CTA (0.78). The integration of the Digital Elevation Model (DEM) with spectral wavebands improved classification performance, particularly in distinguishing urban areas from spectrally similar classes such as barren land and roads. In contrast, additional indices like NDBI and SAVI provided only marginal improvements. Results suggest that incorporating DEM enhances model robustness and spatial accuracy, while the sole use of ancillary indices may introduce redundancy. The study underscores the importance of selecting appropriate classifier-data combinations and highlights the utility of the F1-score, alongside Kappa, for evaluating class-specific accuracy. This research contributes to urban land cover mapping by offering a comparative framework that integrates ancillary variables, helping to refine classification strategies in heterogeneous landscapes.