Comparative Analysis of Land Use Classification Accuracy Using Maximum Likelihood Classification (MLC) and Spectral Angle Mapping (SAM) Methods

Abstract

This study examines the effectiveness of Maximum Likelihood Classification (MLC) and Spectral Angle Mapping (SAM) methods in classifying land use types across a mixed-use landscape. Using an error matrix assessment based on 791 ground-reference points, the study evaluates the classification accuracy for different land use classes, including Paddy Fields, Water Bodies, Forests, Sugarcane, Cassava, Eucalyptus Plantations, and Built-up Areas. Results reveal that MLC achieved higher accuracy in classifying Paddy Fields and Sugarcane, demonstrating a strong overall accuracy of 83.944% with a Kappa coefficient of 0.808, indicating robust classification reliability. SAM, with an overall accuracy of 79.267% and a Kappa coefficient of 0.752, showed strengths in specific classes but struggled with spectral overlap in mixed or similar land cover types, particularly Built-up Areas and certain plantation classes. The study’s findings suggest that MLC’s probabilistic approach is more suitable for complex land cover patterns, whereas SAM performs best in distinguishing classes with distinct spectral properties. The comparative insights inform the selection of classification methods based on landscape characteristics, offering guidance for improved land use mapping accuracy in remote sensing applications.