In this paper we apply the usage of thermal weights, a new variable for geostatistical analysis and we present the method for their determination. In the case study we tested a data fusion between Sentinel-2 and Landsat 7/8 data, to incorporate also the thermal factor in the detection of land cover changes. The process distinguishes grasslands from other crops with similar vegetative appearance and offers us the possibility to create a new statistical sample with just grasslands. The data fusion is incorporated in the calculation of Land Surface Temperature (LSTFU) by combining the Sentinel-2 derived Normalized Difference Vegetation Index (NDVI), and from it derived land surface emissivity, with the Landsat 7/8 derived Top of Atmosphere Brightness Temperature (TOABT). The experimental LSTFU is modified into a normalized assessment variable by a time-series analysis. The result is a thermal weight layer which can help us in further object-based image analyses and classification. The thermal weight is calculated from Sentinel-2 and Landsat 7/8 datasets that has small acquisition time gaps between them. The accuracy assessment due to time gaps and sensor differences was evaluated with Cohens’s kappa (κ) and correlation matrix validation. The data fusion is made to test if a Sentinel-2 fusion approach could improve the Thermal Weight created just by Landsat imagery. The purpose was to evaluate the importance of thermal bands for LU/LC cover.
Mangafić, Alen; Mesner, Nika; and Triglav Čekada, Mihaela
"Grassland Recognition with the Usage of Thermal Weights,"
Free and Open Source Software for Geospatial (FOSS4G) Conference Proceedings: Vol. 18
, Article 3.
Available at: https://scholarworks.umass.edu/foss4g/vol18/iss1/3