Tue, Aug 16, 2022: 9:15 AM-9:30 AM
516A
Background/Question/MethodsThe compounded effects of land-use/land cover change (LULC) and climate change on biodiversity can cause distributional vegetation shifts and alter ecosystem function, subsequently impacting human well-being (e.g., food security, health). Accurate predictions of future and current vegetation distributions are therefore crucial to guide effective biodiversity conservation and land-use policies. Ecological niche models (ENMs) have proved useful to predict vegetation distributions. However, since ENMs do not typically account for interactions that preclude more than one vegetation type from predominating in the same area, they yield overly broad predictions. We aimed to improve vegetation ENMs by postprocessing to account for interactions among three vegetation types in eastern Mexico: cloud forest, pine-oak forest, and submontane scrubland. We built Maxent models using the Wallace v.1.0.6 software, employing occurrence points drawn from land-use and land cover data along with bioclimatic and soil variables as predictors. To indicate which vegetation type was most likely to occupy an area where predictions overlapped, we post-processed ENMs using three different classification methods: winner-by-cell, spatial support vector machines, and spatial-environmental support vector machines. Lastly, we compared omission, commission, and overall misclassification rates for unprocessed and post-processed vegetation ENMs to evaluate how accounting for interactions affects model accuracy.
Results/ConclusionsWe found that post-processing vegetation ENMs with classification methods removed areas that were likely overpredicted by individual models, resulting in more ecologically realistic predictions. Applying a classification mask reduced predicted suitable area between 34.9% - 54.4% for cloud forests, 31.7% – 45.0% for pine-oak forest, and 51.6% - 59.5% for submontane scrublands. Additionally, post-processing greatly reduced model commission errors (CF = -0.39, PO = -0.68, SS = -0.31) while increasing omission error rates only slightly to moderately (CF = +0.2, PO = +0.07, SS = +0.09; both error rates range from 0-1). By providing a generalizable method capable of reducing commission errors without greatly increasing omission rates, this novel approach holds promise for holistic predictions of vegetation distributions that provide more accurate and ecologically realistic estimates. Potential extensions of this approach include using it to predict changes in vegetation distributions and landscape configuration under future climate and/or land-use change scenarios. Outputs from such research could provide insights on the potential socio-ecological impacts of LULC change in conjunction with climate change (e.g., disease emergence, invasive species, food security).
Results/ConclusionsWe found that post-processing vegetation ENMs with classification methods removed areas that were likely overpredicted by individual models, resulting in more ecologically realistic predictions. Applying a classification mask reduced predicted suitable area between 34.9% - 54.4% for cloud forests, 31.7% – 45.0% for pine-oak forest, and 51.6% - 59.5% for submontane scrublands. Additionally, post-processing greatly reduced model commission errors (CF = -0.39, PO = -0.68, SS = -0.31) while increasing omission error rates only slightly to moderately (CF = +0.2, PO = +0.07, SS = +0.09; both error rates range from 0-1). By providing a generalizable method capable of reducing commission errors without greatly increasing omission rates, this novel approach holds promise for holistic predictions of vegetation distributions that provide more accurate and ecologically realistic estimates. Potential extensions of this approach include using it to predict changes in vegetation distributions and landscape configuration under future climate and/or land-use change scenarios. Outputs from such research could provide insights on the potential socio-ecological impacts of LULC change in conjunction with climate change (e.g., disease emergence, invasive species, food security).