Integrating hyperspectral imagery at different scales to estimate component surface temperatures

Different methods for classifying land cover and extracting temperatures of surface components from hyperspectral images at different scales were compared using airborne imagery (Reflective Optics System Imaging Spectrometer (ROSIS) at 1.2 m spatial resolution and Digital Airborne Imaging Spectrometer (DAIS 7915) at 3.3 m spatial resolution) for a ?montado/dehesa? landscape in the Alentejo, Portugal. For calibration purposes, surface temperatures and stomatal conductance of component vegetation types were also measured at ground level. Manual classification was compared with a range of automated classification methods to determine the most accurate method for the study area. The ?scale? for each cover type was characterized by analysing the frequency distribution of contiguous pixels of each cover type at 1.2 m. Temperatures of different surface components were estimated using different combinations of 1.2 m and 3.3 m data (using spectral angle mapper classification) as well as linear spectral unmixing and disaggregation approaches for extracting thermal information at sub?pixel resolution. The relative advantages of the different methods are discussed and a recommended strategy for integrating hyperspectral imagery at different scales to extract component surface temperatures in montado/dehesa?type systems is proposed.