Changes in land cover may cause shifts in biodiversity and alter ecosystem function, potentially impacting many aspects of human well-being. Effective land-use policy design and implementation require accurate representations of current and future land cover. Vegetation classification maps are labor-intensive and typically static, lacking predictive ability. Conversely, species distribution models (SDMs) generate predictions of suitable habitat that can be projected spatially and temporally. However, these models have seldom been employed to predict distributions of biological entities higher than the species level. Additionally, only recently have SDM methodologies incorporated interactions among biological entities that may further constrain their distributions. To help fill such gaps, we tested the utility of SDMs to predict the distribution of two adjacent montane forest types in eastern Mexico: a) cloud forest and b) pine-oak forests. Maxent models were built using the Wallace v.1.0.6 software, employing CHELSA bioclimatic variables and occurrence points drawn randomly from the INEGI-VI land-use and vegetation classification dataset. To delimit boundaries between vegetation classes in areas where predicted suitabilities overlap we implemented two post-processing masks: a) winner-by-cell and b) support vector machines. Lastly, we accounted for deforestation by removing areas with tree cover below 70% using the Global Forest Change dataset.
Results/Conclusions
Both sets of masked predictions present similar spatial patterns, with cloud forest indicated primarily on the windward eastern slopes of these mountain ranges and pine-oak forests on the leeward western ones. However, the two methods of boundary delimitation between vegetation classes led to differences in the predicted extent of both forest types and the spatial continuity of cloud forests. Additionally, after accounting for deforestation we observed a notable decrease in predicted suitable conditions for pine-oak forests. This was particularly evident on the fringes of the prediction in the northern portions of the study region. Surprisingly, the cloud forest prediction was not greatly affected by deforestation. This may be because cloud forest presence is restricted to high-elevation areas that may not be as accessible for human intervention. Future directions will focus on including a lowland vegetation type, likely to constitute the intervening matrix between these montane forests. In summary, simultaneous forecasts of multiple vegetation classes may be useful to assess biodiversity at higher levels of organization, providing a tool to guide policies regarding development and land-use planning.