Defining landscape-level forest types: application of latent Dirichlet allocation to species distribution models

Context Forest type (FT) classification provides useful information to ecologists and forest managers by representing similar sites based on species dominance. Various methods have been developed using stand-level or plot-level information, however, these classifications are not always effective at representing broader landscape patterns of species diversity. Objectives We classified landscape-level FTs from species habitat models and compared against classifications intended for stand-level information. We used a departure score to assess potential changes to current FT from projected changes in climate and habitat suitability (HS). Methods We applied a text mining algorithm, latent Dirichlet allocation (LDA), to 125 species HS models within the eastern United States to define 11 FTs under current conditions. We compared the LDA model against two summations of relative abundance. We then developed a departure score to characterize potential changes to current FTs under projected climate change. Results The LDA model showed broad spatial agreement with summations of species relative abundance. However, LDA’s landscape-level dominance of species differed from stand-level classifications of species summations. Varying degrees of pressure from climate change and HS indicated that future FTs could face conditions that result in departures. However, the overall departure scores tended to be lower due to reduced pressure from modeled changes in HS for much of the eastern US. Conclusions LDA results are promising for classifying landscape-level FTs. Portraying potential changes in future FTs with departure scores may facilitate better management by aligning the spatial scales of information and not attributing changes to specific species or conditions.