A new tool for variable multiple endmember spectral mixture analysis (VMESMA)

Abstract

Spectral mixture analysis is a widely used method to determine the sub-pixel abundance of vegetation, soils and other spectrally distinct materials that fundamentally contribute to the spectral signal of mixed pixels. In this paper we present a computing and environmental analysis tool, named VMESMA, which extends the possibilities of conventional spectral unmixing. The basis is the categorization of the scene into different units or scene sub-areas and software guidance for endmember selection, allowing for a better adaptation of the model to the conditions of the main cover types. For each pixel an individual combination of endmembers may be selected by automated matching to model quality criteria. This hierarchical assessment can incorporate a priori knowledge from different data sources, including information derived from the unmixing results. Based on an iterative feedback process, the unmixing performance may be improved at each stage until an optimum level is reached. VMESMA allows an immediate estimate of the proportions, which is very robust against external factors (e.g. illumination) and canopy shade. An application of VMESMA on hyperspectral data has been conducted to evaluate the possibilities to map residual sludge and sludge derivatives for two consecutive years with changing land surface conditions. The method offered greater flexibility and new possibilities to improve the understanding and modelling of the scene characteristics.