Порівняльний аналіз спектральних індексів для виявлення забудови на основі супутникових знімків Sentinel-2

As a result of the accelerated processes of urbanization and the seasonal variations in the landscape cover, adaptive and precise methods for built-up area identification from space are necessary. The conventional spectral indices, majorly NDBI and NDVI, show inadequate stability over seasonal variations particularly when snow is present and where open soil exists. Within this study, a comparison between NDBI, NDWI, modified NDBIBlue, and EVI-S indices proposed by the authors is made. Seasonal multispectral Sentinel-2 images were chosen to test them. This gives an opportunity to evaluate these indices’ effectiveness in four seasonal periods. According to the three types of categories — building water areas and arable land each index gave out varied sensitivity levels. The optimum results for the discrimination of urbanized areas were obtained during summer, and winter proved to be an inappropriate season for classification since there is a very high number of false positives due to snow coverage. The last step of this work was classification with Maximum Likelihood using three index layers, building a vectorized model of the structures with great fidelity to what is seen in the satellite image. This study can initiate trust in a hybrid approach that takes seasonal sensitivity from spectral indices and hands over to automated methods of classification while it points out that there is constant need that methods should be advanced having errors within classes reduced particularly when open ground or shaded areas are present.