The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences (Dec 2019)

LINEAR SPECTRAL UNMIXING OF SENTINEL-3 IMAGERY FOR URBAN LAND COVER - LAND SURFACE TEMPERATURE (LST) ANALYSIS: A CASE STUDY OF METRO MANILA, PHILIPPINES

  • C. Cruz,
  • A. C. Blanco,
  • A. C. Blanco,
  • J. Babaan,
  • J. A. Cruz,
  • R. R. Sta. Ana,
  • E. Paringit

DOI
https://doi.org/10.5194/isprs-archives-XLII-4-W19-141-2019
Journal volume & issue
Vol. XLII-4-W19
pp. 141 – 148

Abstract

Read online

The advancement of remote sensing technologies is a huge advantage in various environmental applications including the monitoring of the rapid development in an urban area. This study aims to estimate the composition of the different classes (vegetation, impervious surfaces, soil) in Metro Manila, Philippines using a 300-meter spatial resolution Sentinel-3 Ocean and Land Colour Instrument image. The relationship between these land cover fractions with the spatial distribution of land surface temperature at this scale is evaluated. Sentinel-3 image has a higher spectral resolution (i.e. 21 bands), as compared with other Landsat and Sentinel missions, which is a requirement for an accurate cover mapping. Linear Spectral Unmixing (LSU), a sub-pixel classification method, was employed in identifying the fractional components in the image based on their spectral characteristics. Field survey using spectroradiometer was conducted to acquire spectral signatures of an impervious surface, vegetation, and soil which were used as the endmembers in the unmixing process. To assess the accuracy of the resulting vegetation fractional image, this was compared with a separate land cover pixel-based classification result using a 3-meter high spatial resolution PlanetScope image and with another vegetation index product of Sentinel-3. The results indicate that the recently available Sentinel-3 image can accurately estimate vegetation fraction with R2 = 0.84 and 0.99, respectively. In addition, the land surface temperature (LST) retrieved from Climate Engine is negatively correlated with the vegetation fraction cover (R2 = 0.81) and positively correlated with the impervious surface fraction cover (R2 = 0.66). Soil, on the other hand, has no correlation with the LST.