Oak savanna conservation and restoration planning using habitat models for two focal species: Wild lupine and the Karner Blue butterfly

Background/Question/Methods

Midwestern oak savanna ecosystems have become quite rare in today’s ecological landscape, largely due to a combination of land use change and fire suppression over the past 150 years. There have been active conservation efforts since the 1990s to increase the acreage of high quality habitat.  Despite these efforts, there remain a variety of obstacles and challenges that are difficult to overcome; the most prominent of which are: (A) where (spatially) to allocate limited resources and focus conservation efforts, and (B) the difficulty of conservation planning in the context of climate change.  To address these challenges, we conducted a study with the broad goals of : (1) constructing a series of species distribution models (SDMs) to map potential habitat for two oak savanna focal species, and (2) exploring the utility of these models to inform future oak savanna conservation planning.  The two important oak savanna focal species used in this study were wild lupine (Lupinus perennis), and the federally-endangered Karner Blue butterfly (Lycaeides melissa samuelis).  The SDMs were constructed using spatial data encompassing soils, land cover, physiography, and climate.  Species-location data were field collected from 2006 to 2013.  We used eight different approaches to construct the SDMs, and compared the performance of each; model performance was tested using both the area under the receiver operating characteristic curve (AUC) and the true skill statistic (TSS) on a 70:30 training to testing data split.

Results/Conclusions

Across all models, the most important predictors of suitable habitat for wild lupine included: soil taxonomic subgroup, land cover, elevation, and soil drainage.  For the Karner Blue, the most important predictors included: soil taxonomic subgroup, land cover, elevation, and winter precipitation.  Preliminary analysis of the SDMs using future climate data indicates shifting habitat suitability for the Karner Blue, largely due to changes in winter precipitation.  Preliminary analyses suggest that while the models performed fairly well in general, the Random Forests (RF) and boosted regression tree (BRT) approaches had the best performance with respect to AUC and TSS.  Several restoration projects and trials are currently being planned on the Manistee National Forest based on the results from these two (RF and BRT) modeling approaches.