Optimizing mixed spectra generation for regression-based unmixing of land cover in urban areas

Regression-based unmixing for quantifying urban land cover at the subpixel scale requires mixed training spectra for model calibration. In this paper optimization and synthetic mixing of hyperspectral image endmember libraries for the calibration of unmixing models are investigated. APEX and HyMap airborne hyperspectral transects respectively covering Brussels and Berlin are used to produce an endmember library for unmixing, as well as reference land cover fractions for validation. The library is spectrally resampled, optimized and synthetically mixed to produce quantitative training data for unmixing of a Sentinel-2 surface reflectance image of Brussels (Belgium). Support Vector Regression models are developed for vegetation-impervious-soil land cover mapping. Findings may contribute to the use of multi-sensor data and to the demonstration of Sentinel-2's added value for quantitative urban land cover assessment.

ISBN:978-1-5090-5809-9

Publication year:2017

Keywords:endmember libraries, land cover, Sentinel-2, sub-pixel mapping, Support Vector Regression, synthetic mixing, urban remote sensing

Authors:International

Accessibility:Closed