The potential of satellite greenness to predict plant diversity among wetland types, ecoregions, and disturbance levels

Abstract

The unprecedented global biodiversity loss has massive implications for the capacity of ecosystems to maintain functions critical to human well‐being, urgently calling for rapid, scalable, and reproducible strategies for biodiversity monitoring, particularly in threatened ecosystems with difficult field access such as wetlands. Remote sensing indicators of spectral variability and greenness may predict the diversity of plant communities based on their optical diversity; however, most evidence is based on narrowband spectral data or terrestrial ecosystems. We investigate how spectral greenness and heterogeneity from publicly available broadband multi‐spectral Landsat satellite imagery explain variation in vegetation diversity across different wetland types, ecoregions, and disturbance levels using 1,138 sites surveyed by U.S. EPA's National Wetland Condition Assessment. We found positive correlations of plant species richness and diversity with indicators of annual maximum spectral greenness and its spatial heterogeneity, explaining up to 43% variation within the global sample, 48% within wetland types or ecoregions, and up to 61% with abiotic covariates. The combined effect of spectral greenness and heterogeneity was stronger than the best‐performing model using climatic, topographic, and edaphic factors alone. When compared among major U.S. watersheds and individual states, the fit of diversity‐greenness models increased when more wetland types were included within the corresponding region's boundaries, up to 61% at the watershed and 77% at the state level, respectively, for diversity models and up to 73% and 80%, respectively, for richness models. Model outliers were characterized by a significantly greater diversity of nonnative species (P < 0.0001), suggesting that changes in model performance and greenness distributions could be used as indicators of shifts in plant community composition, particularly in tidal wetlands making the majority of outliers with significantly lower than predicted diversity. This study represents a first‐time national‐scale effort to use publicly available remote sensing, climatic, and topographic data to predict plant diversity in wetlands, which tend to be understudied compared to terrestrial ecosystems despite being among the most stressed ecosystems on Earth. Our study suggests that multi‐temporal broadband satellite imagery could provide a low‐cost assessment of regional and national wetland biodiversity for prioritization of conservation efforts and early detection of biodiversity loss.

Greenness, texture, and spatial relationships predict floristic diversity across wetlands of the conterminous United States

“…We used the Green Normalized Difference Vegetation Index (GNDVI) -based on the normalized difference between the green band, sensitive to species-specific variation in chlorophyll content, and the near infrared (NIR) band, strongly reflected by mesophyll cellsas an indicator of plant biomass and coverage (Gitelson and Merzlyak, 1998; here after referred to as "greenness"). GNDVI was the best predictor of plant richness and diversity in this dataset among a group of six SVIs (Taddeo et al, 2019b).…”

“…We hypothesized that a multivariate predictive model leveraging both texture and greenness (i.e., maximum SVI value) would enhance predictive potential by accounting for the positive impact of habitat heterogeneity on plant richness and minimizing the confounding effect of background land covers (e.g., soil, water, litter) and introduced species on diversity-productivity relationships. Our previous effort did incorporate standard deviation in maximum greenness measured from Landsat data as a predictor of species richness, with a significant but somewhat low predictive capacity ( i.e., standard deviation in maximum greenness estimated using the Green Normalized Vegetation Index could predict 3% of variation in site richness; Taddeo et al, 2019b). In the present study, we explored this potential more indepth by testing a greater range of texture measures representing complementary aspects of spatial heterogeneity.…”

“…From a remote sensing perspective, this means that high values for a SVI sensitive to plant coverage, biomass, or photosynthetic activity (Huete et al, 1997) should correlate to species richness. This theory has been tested in a variety of ecosystems (Castillo-Riffart et al, 2017;Madonsela et al, 2017) including wetlands (Taddeo et al, 2019b) with sometimes modest yet significant results demonstrating the utility of this approach to help target field monitoring and conservation interventions. 4…”

“…Incorporating texture as a proxy for habitat heterogeneity may also help address an important challenge in the application of the diversity-productivity theory in monodominant wetlands covered by few invasive species associated with high greenness values (e.g., EVI; NDVI) but low species richness (Fig. 1B; Taddeo et al, 2019b). Highly invaded sites, however, might have a low spectral heterogeneity which could be captured by textural metrics.…”

“…Our goal was to compare predictors (i.e., maximum annual greenness, texture, and spatial autocorrelation) of wetland plant richness and diversity derived from open-source databases. This study builds on a previous effort utilizing maximum annual greenness (i.e., maximum annual value for a spectral vegetation index sensitive to plant coverage and abundance) derived from the Landsat archive to predict plant diversity across 1,115 wetlands of the conterminous United States (Taddeo et al, 2019b). We hypothesized that a multivariate predictive model leveraging both texture and greenness (i.e., maximum SVI value) would enhance predictive potential by accounting for the positive impact of habitat heterogeneity on plant richness and minimizing the confounding effect of background land covers (e.g., soil, water, litter) and introduced species on diversity-productivity relationships.…”