Comparative Assessment of Classification Algorithms for Land Cover Mapping Using Multispectral and PCA Images of Landsat

The advancement of remote sensing technologies and the availability of free satellite data have significantly enhanced the precision of land use and land cover (LULC) mapping, facilitating the analysis of landscape transformations and ecosystem changes. However, selecting the most suitable classifier for LULC mapping remains a complex challenge. Therefore, it is essential to evaluate the accuracy of various LULC modeling algorithms to determine their effectiveness in different applications. This study conducted a comprehensive evaluation of both supervised machine learning algorithms and traditional classification methods applied to Landsat 8 imagery with a 30-meter spatial resolution, covering the Shangla and Battagram districts in Khyber Pakhtunkhwa (KPK), Pakistan. The study focused on three classification algorithms: Maximum Likelihood Classification (MLC), Support Vector Machines (SVM), and Random Forest (RF). The performance of these algorithms was assessed on both multispectral images and composite images derived from Principal Component Analysis (PCA) and Band Ratioing and/or Normalized Indices. Additionally, the accuracy of these algorithms, when applied to different datasets, was compared with the recently released World Cover LULC product by the European Space Agency (ESA). The results indicated that the SVM algorithm outperformed the others, achieving an overall accuracy of 90.43% and a kappa coefficient of 0.8792. The MLC and RF algorithms also produced promising results, with overall accuracies of 85.58% and 88.46%, respectively. Furthermore, the study found that the overall accuracy of ESA’s World Cover LULC product was 70.67% in the study area, based on similar validation samples. These findings underscore the strengths and limitations of each algorithm, providing valuable insights into their suitability for LULC classification and the applicability of existing global LULC maps.