Comparison of landuse/landcover classifier for monitoring urban dynamics using spatially enhanced landsat dataset

Urban dynamics refers to a phenomenon wherein certain factors contribute to imparting changes to an urban area. These factors can aid in either growth or deterioration of the city. One important factor that acts as a threat to the urban environment is rapid urbanization. To monitor and control such urban expansion, prediction is a necessity. It will throw some light on how the city grows and how it will affect the environment and living conditions. A precise and accurate dataset on land use/land cover (LULC) is a must for such analysis. Several methods exist to classify satellite data based on spectral reflectance with advancements in satellite technology and the means to process it. One problem is that there is no single classifier that can produce accurate results on LULC predictions. Each classifier varies in its performance based on several factors. Hence this research was carried out by choosing three classifiers, namely maximum likelihood (MLC), random forest algorithm (RFA) as well as support vector machines (SVM), which are commonly used. Its performance is then evaluated for a pan-sharpened Landsat 8 data of the study area, i.e., Salem and its surrounding urban agglomerations. From the results, it was inferred that the overall accuracies of MLC, SVM and RFA are 0.885, 0.930 and 0.945, respectively. Though the MLC accuracy is acceptable, it had many misclassifications of other classes into built-up classes. SVM and RFA performances were found good overall, but SVM had fewer misclassifications compared to RFA. SVM produced results close to reality and was concluded as the best classification method for pan-sharpened Landsat 8 data for Salem and its surrounding area.