FRANCA - Developing an automated processing chain to quantify soil, green and dry vegetation in spaceborne imaging spectroscopy data

Kurzfassung

In many previous studies linear spectral unmixing approaches have shown to be an accurate retrieval method for quantitative subpixel information. In addition, when the spectra of all endmembers (EM) are known, approaches based on multiple endmember spectral mixture analysis (MESMA) are highly accurate. For the retrieval of the EM spectra, spatial-spectral approaches such as the SSEE [ROGGE et al., 2007] were developed and successfully applied in many studies. But for many real-world applications with less-than-perfect EM knowledge, the retrieval accuracies are signifcantly lower. To overcome this problem, an automated MESMA methodology with an increased stability in case of inaccurate EM, or when the whole spectral variability cannot be completely represented by EMs, was developed [BACHMANN 2008, BACHMANN et al., 2009]. Also the combination of the SSEE endmember extraction and the uMESMA unmixing was already applied for soil-related application [MALEC et al., 2015; BAYER et al., 2016 ]. In preparation of the upcoming DESIS and EnMAP missions, an automated processing chain from including EM extraction, EM classifcation and MESMA unmixing is currently developed, which are based on the Python versions of SSSEE, uMESMA, and newly developed spectral classifers. The full chain is developed in Python3, and is using the Luigi workfow management in combination with a Docker-based distribution, therefore preparing for the improved availability of hyperspectral data in context of the upcoming hyperspectral spaceborne missions. Within this presentation, the overall processing chain design is presented, and details are given on the implementational and confgurational requirements for SSEE and uMESMA allowing for a fully automated and orchestrated processing.