Mapping Percent-Cover of the Invasive Species Bromus tectorum (Cheatgrass) over a Large Portion of Nevada from Satellite Imagery

Peterson, Eric

Ground data were collected for developing the mapping model (training data) and for assessing the accuracy of the model (assessment data). Satellite images were obtained over the area for early- and late-season dates. The training data were attributed with corresponding data available across the entire geographic area (satellite imagery spectrum bands, elevation, and derived variables). The mapping model was developed from the training data with Tobit Regression, a statistical procedure similar to Multiple Linear Regression, and with estimation limits based on visual interpretation of data graphs. The model was applied to estimate B. tectorum cover for each pixel in the satellite images, resulting in the map of estimated B. tectorum cover. Post-modeling edits included cultivated fields, urban areas, cloudsand their shadows, and a few shallow water bodies where the estimated cover was greater than zero. Error for the final map was calculated from the independent assessment data... Summary: Bromus tectorum (cheatgrass) is an annual grass from Eurasia that has invaded western rangelands. It frequently invades landscapes after wildfire, forming dense stands with fine fuels that shorten typical fire intervals. Thus B. tectorum and wildfire form a positive feedback mechanism that threatens native ecosystems and often impact the rare and endangered species those ecosystems host. It is not an understatement to say that B. tectorum is among the greatest threats to conservation in the intermountain west. Elimination of B. tectorum from the landscape is unrealistic with current technology. Instead, we must learn to manage rangelands so as to minimize the dispersal and establishment of B. tectorum, and to promote the reestablishment of native vegetation after disturbances such as wildfire. Key to such management is a strong understanding of where B. tectorum has invaded the landscape and how dense it has become. With this need in mind, the Nevada Natural Heritage Program set out to map B. tectorum over a large portion of the state of Nevada. A portion of this area was used to test how well the mapping method could be extended to areas beyond the training area. The mapping method involved developing a statistical model for the estimation of B. tectorum cover at training plots with variables derived from satellite imagery and matching topographic data. The model was then applied to each 28.5 X 28.5 meter pixel in the satellite images. Map accuracy was then determined as Root-Mean-Square Error (RMSE) at additional accuracy assessment plots. Sample plots involved an ocular estimation of percent ground-cover of B. tectorum over a circular area of 0.1 hectares. Training data included 262 sampled plots scattered throughout an original (northern 2/3) mapping area. These plots were located arbitrarily to include a broad variety of landscape conditions. For accuracy assessment, an additional 75 plots were sampled. These assessment plots were located throughout the entire mapping area with a constrained-random distribution: random within public lands and within 1 kilometer of a significant road. Landsat 7 ETM+ satellite data were obtained for two time periods in order to use the unusual early phenology (green-up and senescence) of B. tectorum as the primary component of the mapping method. The Normalized Differential Vegetation Index (NDVI) was calculated for each time period and the difference in NDVI between the time periods was then used as the measure of phenology across the landscape. This measure of phenology corresponded well to the cover of B. tectorum at the training plots and formed the most important factor in the mapping model...

Mapping Percent-Cover of the Invasive Species Bromus tectorum (Cheatgrass) over a Large Portion of Nevada from Satellite Imagery

Collections