A landscape scale assessment of cheatgrass density effects on fire occurrence and severity in central Nevada

Background/Question/Methods

The presence of an invasive species can alter many ecosystem processes including disturbance.  Globally, invasive annual grasses have altered fire regimes by increasing frequency (i.e. grass/fire cycle hypothesis).  Cheatgrass (Bromus tectorum) is a dominant invasive annual grass across the Great Basin.  Previous studies conducted at watershed scales or smaller have indicated that fire may become more frequent in cheatgrass dominated areas.  In addition to frequency, changes in vegetation structure (e.g. density) may affect other aspects of fire such as severity.  In this study, we took a landscape approach to answer the questions 1) how does cheatgrass density affect fire occurrence probability in relationship to other environmental factors, and 2) how does cheatgrass density affect changes in severity within a fire.   We used Landsat derived map of cheatgrass densities across north central Nevada from 2001, PRISM weather data, and SW ReGap data to map environmental variables.  Fire perimeters and severity classes were collected from the Monitoring Trends in Burn Severity database for the summer of 2001.  The Bayesian Model Averaging (BMA) package in R was used to reduce the candidate variable list, and MaxEnt was run to model the probability of fire occurrence and severity. 

Results/Conclusions

For fire occurrence analysis, the top model from BMA included: cheatgrass density, average maximum summer temperature, spring precipitation, summer precipitation, secondary road density, slope, and vegetation type.  When run in MaxEnt, the top model had an area under the curve (AUC) of 0.747 based on the receiver operating characteristic curve.  Cheatgrass density had the lowest percent contribution and permutation importance, 0.011 and 0.021, respectfully.  Climate and vegetation type had the highest percent contribution and permutation importance.  When four severity classes were modeled in BMA, variables that described topography were supported in the top model.  The AUC increased with higher severity from 0.786 in the unburned to low class to 0.942 in the high severity.  Cheatgrass density had the highest percent contribution (0.361) and permutation importance (0.448) in the high severity model.  Our results indicate that at landscape scales cheatgrass does not greatly affect probability of a fire; its effects are masked by other environmental variable.  Cheatgrass densities did have a strong effect on which severity class a pixel burned, especially in the highest severity.   Also, the complex response curves of individual variables indicate that probabilities of occurrence and severity are not based on the same environmental factors at landscape scales.