Measuring biodiversity across spatial and spectral scales

The rapid expansion of imaging spectroscopy missions has opened new opportunities for remote sensing based biodiversity assessment. However, the intrinsic trade-off between spatial resolution, spectral resolution, and signal-to-noise ratio necessitates synergistic approaches that combine multispectral and hyperspectral data in order to preserve both spatial detail and rich spectral information. Many biodiversity indices require the estimation of variation in a moving window manner, where the resulting metric is an aggregation of several (hundreds of) pixels, thus often the resulting biodiversity maps are an order of magnitude lower in resolution compared to the input dataset.

Current spaceborne imaging spectroscopy missions such as EnMAP, PRISMA, and the forthcoming CHIME provide imagery at 30m resolution, yet data fusion techniques that integrate these products with higher resolution multispectral sensors like Sentinel-2 (10 m) offer a pathway to enhance spatial detail while retaining the spectral properties of an image. These techniques can help us create biodiversity products that have high spatial resolution and at the same time capture spectral variation in a way not possible with existing multispectral sensors. Here, we present a case study showing the application of a fusion method based on spectral unmixing to derive 10m EnMAP-like product over the Bavarian Forest National Park in Germany. Using the fused products, we compute several functional diversity metrics - functional richness, divergence, and evenness - and compare outputs derived from multispectral data, hyperspectral data, and fused datasets. Leveraging the continuous coverage of the study area by EnMAP, we are able to span our analysis over four years (2022-2025) allowing us to account for uncertainties on interannual scale.

We further emphasize the critical role of rigorous data harmonization - including advanced cloud and cloud shadow masking, precise co-registration, and image alignment - which extends beyond standard Level-2A processing and is essential for the effective integration of multisensor data in biodiversity applications.