In the Great Basin, cheatgrass (Bromus tectorum) and fire rapidly convert Wyoming sagebrush (Artemisia tridentata subsp. wyomingensis) communities to annual grasslands. Meanwhile, plant species diversity may help ecological communities to withstand stressors and recover following fire. This study integrates Landsat time-series data and biophysical covariates including soils, fire, topography, and annual climate data to model plant species diversity in shrublands across a five-state area in the Great Basin. Annual medoid composite images were produced from Landsat data for each year from 1984– 2019; for each annual image, we calculated a suite of vegetation indices—including soil-corrected and tasseled cap indices—and spectral heterogeneity indices—demonstrated to relate to plant diversity. All image processing, calculation of covariates, and extraction of plot-level raster values were conducted in Google Earth Engine (GEE). Field species inventory measurements from 4,729 Bureau of Land Management Assessment, Inventory, and Monitoring (AIM) plots were used as field reference and validation, in addition to 42 plots collected as part of this project. Species richness was selected from a suite of 19 species diversity indices as the most representative response variable of plot-level diversity. Spatial models were constructed and employed to map plant species richness across the study area.
Results/Conclusions
Despite the physiographic diversity of the Great Basin and the difficulty of producing spatially-continuous models of plant species diversity from multispectral imagery, preliminary results have been promising. Random Forest model outputs report a pseudo-R-squared value of .4605 with an RMSE of 8.36, while the actual variance of the species richness dataset is 129.562. Remote sensing indices tasseled cap brightness (TCB), soil-adjusted total vegetation index (SATVI), and normalized difference vegetation index (NDVI) were each ranked highly in terms of variable importance. Additionally, spring temperature, years since fire, and soil moisture capacity covariates were also shown to be important predictors of plant species diversity.
Spatial predictions of plant diversity may contribute to efforts identifying drivers of ecological resistance and resilience within Great Basin shrubland communities and further inform management planning. While these preliminary results are promising, we will continue to evaluate additional predictors that may improve predictions of plant species diversity including seasonal and annual climate indices of aridity and growing degree days. Next, we will leverage the Random Forest models alongside the LandTrendr algorithm in GEE to model ecological recovery directly. These approaches have the potential to shed light on predominant drivers of post-fire recovery in the region.