The data concept behind the data: From metadata models and labelling schemes towards a generic spectral library

Abstract

Spectral libraries play a major role in imaging spectroscopy. They are commonly used to store end-member and spectrally pure material spectra, which are primarily used for mapping or unmixing purposes. However, the development of spectral libraries is time consuming and usually sensor and site dependent. Spectral libraries are therefore often developed, used and tailored only for a specific case study and only for one sensor. Multi-sensor and multi-site use of spectral libraries is difficult and requires technical effort for adaptation, transformation, and data harmonization steps. Especially the huge amount of urban material specifications and its spectral variations hamper the setup of a complete spectral library consisting of all available urban material spectra. By a combined use of different urban spectral libraries, besides the improvement of spectral inter- and intra-class variability, missing material spectra could be considered with respect to a multi-sensor/ -site use. Publicly available spectral libraries mostly lack the metadata information that is essential for describing spectra acquisition and sampling background, and can serve to some extent as a measure of quality and reliability of the spectra and the entire library itself. In the GenLib project, a concept for a generic, multi-site and multi-sensor usable spectral library for image spectra on the urban focus was developed. This presentation will introduce a 1) unified, easy-to-understand hierarchical labeling scheme combined with 2) a comprehensive metadata concept that is 3) implemented in the SPECCHIO spectral information system to promote the setup and usability of a generic urban spectral library (GUSL). The labelling scheme was developed to ensure the translation of individual spectral libraries with their own labelling schemes and their usually varying level of details into the GUSL framework. It is based on a modified version of the EAGLE classification concept by combining land use, land cover, land characteristics and spectral characteristics. The metadata concept consists of 59 mandatory and optional attributes that are intended to specify the spatial context, spectral library information, references, accessibility, calibration, preprocessing steps, and spectra specific information describing library spectra implemented in the GUSL. It was developed on the basis of existing metadata concepts and was subject of an expert survey. The metadata concept and the labelling scheme are implemented in the spectral information system SPECCHIO, which is used for sharing and holding GUSL spectra. It allows easy implementation of spectra as well as their specification with the proposed metadata information to extend the GUSL. Therefore, the proposed data model represents a first fundamental step towards a generic usable and continuously expandable spectral library for urban areas. The metadata concept and the labelling scheme also build the basis for the necessary adaptation and transformation steps of the GUSL in order to use it entirely or in excerpts for further multi-site and multi-sensor applications.