Exploring Contextual Appropriateness of Spectral Indices for Remote Sensing-Based Land Coverage Classification Using Machine Learning

Spectral indices obtained from satellite images are helpful for remote sensing, particularly for land cover classification (LCC). Therefore, the purpose of this paper will be to evaluate the performance of the three regularly used spectral indices, including Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Normalized Difference Built-up Index (NDBI), in disaggregating three classes of land cover, i.e., vegetation, water, and built-up areas. The work uses sentinel-2 generated satellite data from Google Earth Engine and Machine Learning classifiers to consider the contextual appropriateness of Spectral Indices for LCC of Remote sensing. The research proves that all indices are beneficial and non-beneficial in recognizing multiple forms of land covering; in other words, NDVI is more precise in identifying the vegetation-based land types than the other two indices. Contrary to this, NDWI is best suited to water-based land coverage, while NDBI is best suited to built-up areas-based land coverage. The study builds on the remote sensing domain by reviewing several spectral indices and determining their suitability for use by the LCC sub-domain.