UC Berkeley

Plant diversity safeguards wetland ecosystem functions, stability, and resilience, but is threatened by habitat loss and degradation. Remote sensing could support the cost-effective management of biodiversity by providing consistent and frequent data at large scales. While identifying individual species from remote sensing datasets with low spatial and spectral resolution is challenging, studies can focus on factors known to correlate with or promote diversity. We tested the predictive potential of such factors — maximum annual greenness as an indicator of productivity, texture (i.e., spatial arrangement of grey tones) as a proxy for habitat heterogeneity, and spatial autocorrelation — across a dataset of 1115 wetlands in the conterminous United States surveyed by the EPA's National Wetland Condition Assessment. We used multivariate linear regressions to test whether spectral and spatial metrics derived from two open-source datasets — NASA's Landsat 5 TM and 7 ETM+ (30 m, 16-day revisit) and USDA's National Agriculture Inventory Program (1 m, biennial) — can predict wetland plant diversity and richness. Individual texture metrics showed different sensitivity to vegetation evenness, growth form, and spatial distribution and could together predict 35–36% of site variation in richness and diversity. This highlights the impact of habitat heterogeneity on species diversity and spectral variability. While maximum annual greenness and texture metrics had similar predictive capacity, their interactions and combined effects improved the fit of linear models by 11–14%, demonstrating their complementarity. Best results were achieved when including distance-based Moran's Eigenvector Maps (dbMEMs) describing spatial relations among sites at multiple scales and reflecting the role of spatially structured factors (e.g., climate, topography, dispersal) on diversity. Together greenness, texture, and dbMEMs could predict 59% of plant richness and 50% of plant diversity across the entire dataset and up to 71% of the richness of least disturbed sites. These results show the potential of open-source remote sensing datasets to monitor biodiversity resources at a large scale and prioritize the protection and field monitoring of wetlands.