Cheatgrass die-off in the Great Basin: A comparison of remote sensing detection methods and identification of environments favorable to die-off
AuthorBrehm, Joseph R.
AdvisorWeisberg, Peter J.
Environmental and Natural Resource Sciences
AltmetricsView Usage Statistics
Invasion by cheatgrass (Bromus tectorum L.) has greatly damaged the rangeland ecosystems of the Western United States. While mechanical and chemical control of cheatgrass is possible, control is costly and often ineffective in the long-term. An improved understanding of cheatgrass die-off, or stand emergence failure, could lead to additional means of control for this widespread invasive. The majority of published research has focused on identifying the cause of die-off, on demonstrating its potential for contributing to restoration success, and on mapping its distribution. At this time, we know little about the regional-scale environments favorable to disease development. While remote sensing methods exist for mapping of cheatgrass die-off, more reliable methods are needed to facilitate both further research and restoration efforts. Here we investigate three methods of using remote sensing to identify patches of die-off across two study areas with distinctly different spectral characteristics: (1) a machine learning method classifying die-off from spectra, (2) a spectral change detection method using threshold values of a vegetation index, and (3) a multiple classifier system (MCS), which identifies die-off by combining the results of the prior two models. Though we hypothesize the MCS approach to be the most reliable, we find the spectral signature classifier to have the most accurate and spatially consistent results. We use die-off mapped by this method to identify spatial and temporal patterns in the occurrence of die-off, finding most patches to be small (<1ha) and ephemeral (lasting 1-2 years). While die-off is widespread, occurring across the majority (>80%) of our study areas at least twice within a 35-year time series, in any given year a variable portion of the landscape will experience die-off (range of 5 to 34% of study area). Next we use a boosted regression tree approach to identify patterns in the environments in which cheatgrass die-off occurs, measuring topographic, edaphic, and normal climatic conditions as well as weather prior to and during the die-off. We explore three hypotheses predicting patterns of association: (H1) an accumulation of conditions stressful to cheatgrass at the individual level by reducing the ability to fend off disease, and at the community level by reducing its ability to re-establish each year, (H2) a “Goldilocks” effect where die-off is associated with nonlinear, interactive effects, and a multiseasonal pattern of weather, and (H3) strong spatial limitations within which die-off is triggered by weather conditions. We find that our data best supported Hypothesis 2. The strongest predictors of die-off are precipitation patterns where a wet spring and summer precede a dry winter and spring. We find little evidence of spatial limitation, noting that die-off occurs most frequently within low slope areas with chronic water deficit conditions typical to the Great Basin. These results suggest that die-off research and management involving natural die-off will be most feasible and successful when launched after observing wet conditions early in the year and dry conditions in the winter and subsequent spring.