COS 245-1 - Remote sensing of biodiversity in grassland ecosystems: exploring context dependencies in the spectral variance hypothesis

Abstract: Grasslands are diverse ecosystems that provide a myriad of ecosystem services to humans. However, over half of the grasslands in the Great Plains have been converted or degraded, impacting grassland structure and functioning. As a result, conservation and promotion of grassland resilience are primary aims for many land management agencies. In order to make the best management decisions towards these aims, it is imperative that we understand the current state of grasslands and how they are changing over time. Nonetheless, monitoring grassland composition, diversity and productivity over space and time is resource intensive. Driven by limitations in traditional monitoring, remote sensing is increasingly promoted as an alternative, or at least complementary, source of information. In particular, hyperspectral imagery can detect functional and compositional characteristics of vegetation across large spatial extents by measuring subtle differences in reflected light.
The spectral variance hypothesis (SVH) expects spectral diversity, or variability in light reflected across a plot, to predict vegetation diversity. However, there is mixed evidence for the SVH in herbaceous ecosystems, potentially due to a scale mismatch between the size of small-statured grassland plants and the coarse pixels used in most analyses. Alternatively, inconsistencies may be the result of context dependency related to the biological characteristics of observed ecosystems; spatial heterogeneity, invasion and trait plasticity can all impact reflectance. Here, we test the SVH using an approach that overcomes scale mismatch and allows us to explore biological context dependencies. We use very high resolution hyperspectral imagery (pixel size ~1 mm) to measure reflectance in the visible and near-infrared across grasslands of variable composition, heterogeneity and disturbance history. We compare our estimates of cover and diversity with in-situ measures collected by the city of Boulder’s Open Space & Mountain Parks.
We find that NDVI is a consistent indicator of vegetation cover across sites dominated by both tall and shortgrass species (R2 = 0.86, p < < 0.001). In contrast, we find that spectral diversity predicts taxonomic diversity only in some grasslands (here, those mesic sites dominated by C4 grasses; R2 = 0.60, p < < 0.001), and only in grasslands that were not recently disturbed (here, unburned; R2 = 0.36, p < < 0.001). The context dependency of this relationship, even with high resolution data, reveals that additional research is necessary to tease apart the biological conditions underpinning the SVH and to reconcile the contradictory relationships found in the remote sensing of biodiversity literature.