COS 290-2 - Detecting grassland plant diversity using airborne and spaceborne imaging spectroscopy

Abstract: Grasslands are among the most endangered ecosystems, primarily due to land-use/land-cover change, disturbances, and invasive species. The increasing threat to grasslands underlines the need to monitor their biodiversity. Remote sensing is a promising approach to capture plant diversity in these threatened ecosystems. While the majority of previous remote sensing studies have focused on quantifying α-diversity (diversity within communities), our goal was to use remote sensing to estimate β-diversity (diversity among communities) in grasslands subject to common management practices.

To address our goal, we collected airborne imaging spectroscopic data with spatial resolution of 1m and spaceborne imaging spectroscopic data from DLR’s DESIS sensor with spatial resolution of 30m. We also collected species composition at 100 240×240m plots from the Joseph H. Williams Tallgrass Prairie Preserve, OK, USA, which is managed using prescribed fire and grazing. To estimate β-diversity remotely, we used an approach based on spectral species, where we grouped pixels into clusters based on their spectral similarity. Using the Bray-Curtis β-diversity metric, we estimated species turnover (from field data) and spectral species turnover (from remote sensing datasets). We calculated Mantel r-statistics to assess the agreement between in-situ and remote sensing-based β-diversity. To understand the degree to which management practices affect β-diversity and our ability to estimate it remotely, we also grouped our plots into three categories based on time since fire, including recently-burned (burn age < 1 year), transitional (burn age of 1-2 years), and unburned (burn age >2 years) and assessed the relationship between field-based and remote sensing-based β-diversity.

Results showed that remote sensing data were capable of estimating β-diversity, although the agreement between field-based and remote sensing-based β-diversity was not the same for airborne and spaceborne data. Results also suggested that the association between field-based and remote sensing-based β-diversity was highly affected by time since fire. Mantel r correlations between airborne and in-situ β-diversity for recently burned, transitional, and unburned plots were 0.020, 0.136^*, and 0.279^*, respectively, whereas the same analysis showed Mantel r correlations of 0.009, 0.182^*, and 0.023 for DESIS (*denotes significant relationship at p ≤0.05). We also found that soil exposure confounds our ability to estimate β-diversity remotely. After removing plots with soil percent cover >50%, the capability of remote sensing data to estimate β-diversity increased. Overall, this study suggests that regardless of spatial resolution of remotely sensed data, spectral information from imaging spectroscopy can be used to estimate plant β-diversity in naturally-assembled grasslands.