The urbisphere Spectral Library v1.0: Enabling Urban Material Identification Across Cities Using Multi-Sensor Satellite Imaging

Being able to identify urban surface materials using Earth Observation data has the potential to support urban weather and climate modelling, urban planning and to help assess climate resilience strategies. However, the varied composition of these materials, the three-dimensionality of the urban canopy and the spatial scales cause high likelihood of mixed pixels, each pose numerous challenges. Existing spectral libraries do not cover the diversity of urban materials needed for image-based surface cover classification. Here, we develop an urban hyperspectral library in the frame of the European Research Council project urbisphere with high-resolution spectral data collected from a wide range of urban materials across Europe.

The urbisphere spectral library v1.0 contains more than 10,000 in-situ hyperspectral measurements from various natural and artificial materials collected from several European cities, such as Heraklion, Paris, and Berlin, and is set to be expanded in the coming years. These measurements were captured under varying conditions of shading, weathering, and viewing angles, using the HySpex Mjolnir VS-620 hyperspectral camera harmonized with the RS-3500 spectroradiometer.

In an exploratory study, hyperspectral signatures from the urbisphere library were adjusted to align with the multispectral bands of the WorldView-3 and Sentinel-2 satellites, as well as the hyperspectral bands of PRISMA and EnMap. These adjusted signatures were used to train separate X-SVM classifiers for each satellite. The trained models were then applied to classify the respective satellite imagery acquired over Heraklion, Greece. The results highlight the library's capability to detect various natural and artificial materials in urban environments and reveal the limitations associated with differing spatial and spectral resolutions. This methodology demonstrates precise identification of urban surface materials while reducing reliance on labour-intensive, image-based end-member extraction.