Assessing Performance of Modified Spectral Indices as Land Surface Temperature Indicators in Tropical Urban Areas

Abstract

Spectral indices derived from remote sensing data such as NDVI, SAVI, NDBI, NDWI, and MNDWI were modified using various mathematical operations to enhance the overall accuracy of the automated mapping built-up (BU) areas. Although some of the modified spectral indices succeed in improving the accuracy of mapping BU areas, their performance as land surface temperature (LST) indicators needs to be investigated. Hence, this paper aims 1) to assess the performance of various spectral indices as LST indicators and 2) to investigate the relationship between performance of spectral indices as LST indicators and their accuracy for mapping BU areas. Research locations were located in Yogyakarta and Surakarta City, Indonesia. Spectral indices and LST were derived from Landsat 8. Results showed that NDBI, NDBI*MNDWI, NDVI*NDBI, and NDBI*SAVI were good LST indicators both in Yogyakarta and Surakarta proven by moderate to strong coefficient of Pearson correlation. The relationship between the performance of various spectral indices as LST indicators and their accuracy for mapping BU areas showed strong positive Spearman correlation by 0.69 and 0.73 for Yogyakarta and Surakarta, respectively. In other words, spectral indices, for both original and modified indices, with good accuracy for automated mapping BU areas tended to be good LST indicators.