Remotely sensed mapping of plant diversity in Earth's largest mangrove forests: Developing a spectral diversity metric with DESIS hyperspectral data and the ‘spectral species’ concept

Global biodiversity monitoring faces significant challenges, yet recent advancements in spaceborne remote sensing, particularly through hyperspectral sensors, are opening new avenues for cost-effective and scalable plant diversity mapping. The high spectral resolution of these sensors enables precise identification of plant traits and community compositions. Employing the “Spectral Species” concept, which categorizes spectral imagery pixels into distinct spectral types, we have developed a novel semi-discrete “Spectral Species Diversity” (SSD) metric. This metric has proven effective in modeling plant diversity, as demonstrated by our study in the mangrove forests of Bangladesh's Sundarbans using DESIS (DLR Earth Sensing Imaging Spectrometer) hyperspectral imagery.

In this study, we analyzed data from 110 Permanent Sampling Plots in the Sundarbans, calculated traditional plant diversity indices (Species Richness, Shannon and Simpson Diversity), and compared these with our spectral diversity metric. The comparison revealed robust correlations between field-measured plant diversity and our DESIS-derived SSD (R² = 0.473 for Shannon diversity and R² = 0.468 for Simpson diversity). However, species richness showed poor correlation with the newly developed SSD metric. Conversely, the continuous conventional Coefficient of Variation (CV) spectral diversity metric, also computed using the same hyperspectral dataset, underperformed relative to the SSD metric. Furthermore, when assessing the performance of our SSD metric using multispectral imagery from Sentinel-2 and Landsat 8, the metrics derived from Sentinel-2 exhibited weaker relationships with plant diversity (R² = 0.152 for Shannon Diversity and R² = 0.144 for Simpson Diversity), and those from Landsat 8 were less effective.

Upon examining different spectral space sizes, we determined that the optimal size for computing spectral diversity metrics was 150 m × 150 m. This size most effectively captured plant diversity in the vegetation survey plots. While the SSD metric within these spectral spaces mirrored the plant diversity trend across the three salinity zones of the Sundarbans, the observed differences were not statistically significant. Nonetheless, the alignment in pattern highlights the ecological relevance of the SSD metric.

This study underscores that the newly developed SSD metric, utilizing hyperspectral imaging and adapting the Spectral Species concept, can accurately map plant diversity in ecologically diverse ecosystems like the Sundarbans mangroves. Future enhancements, such as aligning spectral space with vegetation survey plot dimensions and incorporating data from SWIR sensors, SAR, or LiDAR, could further refine the metric's robustness and global applicability. These improvements will provide crucial insights for biodiversity conservation and ecological research through precise plant diversity and distribution mapping.