Multi-temporal spectral indices coupled with iterative species distribution models: A powerful tool for land managers in post-disturbance landscapes

Background/Question/Methods

The Squirrel Creek wildfire disturbed 4,450 ha in Medicine Bow National Forest, Wyoming, USA in 2012, during the hottest and driest summer on record for the state. Among the most pressing concerns of land managers in post-wildfire landscapes are the establishment and spread of invasive species, particularly in areas such as the Squirrel Creek burn that encompass crucial winter habitat for mule deer (Odocoileus hemionus) and elk (Cervus Canadensis). We developed four Species Distribution Models for invasive cheatgrass (Bromus tectorum) in the Squirrel Creek burn using eight spectral indices derived from five months of Landsat 8 imagery corresponding to both species’ phenology and time of field data collection. 

Results/Conclusions

The four models were improved using an iterative approach in which a threshold for abundance (i.e. ≥40% foliar cover) was established from an independent dataset, and produced highly accurate maps of current cheatgrass distribution in Squirrel Creek burn with independent AUC values of 0.95 to 0.97. We quantified the area at highest risk for cheatgrass invasion in future seasons given its current distribution, topographic covariates, and seed dispersal limitations, and showed that this area has increased 30% from pre-disturbance observations. These models demonstrate the power of using derived multi-temporal spectral indices as proxies for species occurrence on the landscape, the importance of selecting thresholds for invasive species abundance to evaluate ecological risk from SDMs, and the applicability of Landsat 8 imagery for targeted invasive species management.