Spectral Diversity as a Predictor of Tree Diversity: Exploring Challenges and Opportunities Across Forest Ecosystems

Abstract

Forests are crucial for ecosystem health, climate regulation, and biodiversity. However, many tree species face extinction threats, requiring active monitoring for conservation. The spectral variation hypothesis (SVH) suggests that spectral diversity can serve as a proxy for ground-measured biodiversity. Despite its promise, SVH’s application has shown inconsistent results, complicating its use in biodiversity monitoring. This study examines the relationship between tree diversity and Sentinel-2-derived spectral diversity across Quebec’s forests, analyzing 2531 inventory plots using a combination of spectral analysis, cluster analysis and random forest (RF) regressions. We evaluate four biodiversity indices: species richness, Shannon diversity, functional dispersion, and percent conifer. Our analysis reveals overlapping spectral signatures that make it challenging to differentiate between varying levels of species richness, Shannon diversity, and functional dispersion. However, percent conifer shows spectral separability and can be stratified using unsupervised k-means clustering. Using RF regression models, only the percent conifer models demonstrated strong performance (R2 = 0.77), while models for the other biodiversity indices did not exceed an R2 of 0.46. This study highlights the complex relationship between spectral diversity and tree diversity, and suggests that future research should aim to improve the understanding of the relationship, or lack thereof, between ground-measured biodiversity indices and relatable spectral metrics.