Boyte_etal2012 Proceedings Boise, Idaho Pacific Northwest Climate Science 1-2 October 2012 2012 Annual Conference, 3rd BoyteStephenP. WylieBruceK. MajorD.J. Boise, Idaho presentation at Cheatgrass and dieoff mapping in the northern Great Basin [abs.] We modeled and mapped cheatgrass percent cover and cheatgrass dieoff in the northern Great Basin from 2000 to 2010. Cheatgrass (Bromus tectorum L.) is a scourge in the Great Basin, where it transforms a diverse, native ecosystem into one where large areas are dominated by a single species after a disturbance. Cheatgrass dieoff in the Great Basin could be viewed as a windfall; however, dieoff can cause other severe problems such as accelerated soil erosion, loss of spring forage, and unknown recovery pathways. We used a remote sensing product (eMODIS NDVI) integrated into a regression-tree model with variables (e.g., elevation, land cover, and soils) that affect cheatgrass percent cover to develop an ecological model (R2 = 0.85). We trained the model on Petersons cheatgrass dataset for northern Nevada and annual grass dataset for the Owyhee Uplands. The model was input into a mapping application to develop a time series (2000 2010) of cheatgrass percent cover maps for the western and central portions of the northern Great Basin. These maps captured the spatial and temporal variation of cheatgrass percent cover and served as the dependent variable in the development of an expected cheatgrass performance model. Next, we developed the expected cheatgrass performance model (R2 = 0.88) using a regression-tree model with annual weather data and a site potential dataset as independent variables. The weather data captured annual variations in ecosystem performance, and the site potential data captured long-term spatial variation in ecosystem performance. This model was also input into a mapping application that created cheatgrass dieoff maps (2000 2010). The annual cheatgrass percent cover maps, which served as a proxy for actual cheatgrass performance, were statistically compared to the annual cheatgrass dieoff maps at the 80% confidence level. Annual cheatgrass dieoff maps were derived from the statistical comparison of actual cheatgrass performance maps and expected cheatgrass performance maps. Lastly, we modeled the probability of cheatgrass dieoff using the cheatgrass dieoff time series combined with topographic, edaphic, land cover, a latitude proxy, and climate data. These variables were input into a decision-tree model to develop the cheatgrass dieoff probability model. This model was input into a mapping application to create a cheatgrass dieoff probability map. Future projections of cheatgrass dieoff probability based on climate change were estimated by substituting current climate data in the probability model with future climate data.