How can we enable the full potential of spaceborne hyperspectral mineral mapping?

Observing and understanding Earth’s processes are more important than ever as the demand for resources and the human impact on the planet are skyrocketing. Hyperspectral mineral mapping is a crucial Earth observation (EO) tool with manifold applications, including understanding geological processes (e.g., for waste storage, geothermal energy), green- and brownfield mineral exploration, monitoring mining activities (e.g., grade control, monitoring of tailing dams), characterizing human-made mineral deposits (e.g. tailings, contaminations and mine drainage precipitates) and monitoring the local and global impact of mining on the environment.
EO enables digital archiving of mineralogic information and drives the transition from traditional maps to digital twins of the Earth’s surface. However, it also implies specific requirements that future spaceborne missions will need to meet in order to support the above applications:
Scale and orientation: Geological features of interest span a wide range and may require the simultaneous interpretation of cm-scale features (e.g. veins, fractures) and regional scale variations in mineral composition (e.g. alteration halos). At the same time, geological outcrops are often obliquely oriented and may be obscured if only nadir data are collected. The integration of data collected from different vantage points and at different scales (space-, air-, drone-borne, terrestrial) is critical for meaningful analysis. For a great part of applications (e.g. monitoring mining activities), additional temporal coverage is crucial. Coordinated acquisition of multi-mission data, established processing platforms, and careful corrections are required to enable this framework.
Spectral range: Mineralogically relevant information is contained (often exclusively) in confined spectral ranges, which are in the visible and shortwave, but also mid- and longwave infrared range. Especially the latter must not be forgotten if the full portfolio of hyperspectral mineral mapping is to be achieved. A careful selection of relevant spectral regions and adapted spectral resolution could help to reduce data load and improve spatial resolution.
Processing and validation: Tremendous progress has been made towards machine learning assisted processing of hyperspectral datasets. Nevertheless, developments too often rely on simple and small benchmark datasets. Large scale, mineralogically relevant datasets struggle with heterogeneous, scale dependent classes and often subjective geological interpretation. We recently established three reference sites in Europe that integrate reference data from different scales and technologies as part of the EU-funded INFACT project. We need to continue this effort and engage the community to provide both large-scale benchmarked datasets as well as architectures suitable for scalable machine learning approaches.