BISHOP

[PAPER; R PACKAGE] Inferring microclimate from tree occurrences.

The image above is a map of fine-scale climate conditions as experienced by trees, generated using a new modeling framework called BISHOP that infers high-resolution climate patterns from species distribution data. I have a paper out in Ecography this week, in which we propose this new BISHOP method and then apply it in a big empirical analysis using data on 216 tree species. This project was a long time in the making and it’s exciting to have it out!

Here’s the abstract:

Fine-scale spatial climate variation fosters biodiversity and buffers it from climate change, but ecological studies are constrained by the limited accessibility of relevant fine-scale climate data. In this paper we introduce a novel form of species distribution model that uses species occurrences to predict high-resolution climate variation. This new category of “bioclimate” data, representing micro-scale climate as experienced by one or more species of interest, is a useful complement to microclimate data from existing approaches. The modeling method, called BISHOP for “Bioclimate Inference from Species’ High-resolution Occurrence Patterns,” uses data on species occurrences, coarse-scale climate, and fine-scale physiography (e.g. terrain, soil, vegetation) to triangulate fine-scale bioclimate patterns. It works by pairing a climate-downscaling function predicting a latent bioclimate variable, with a niche function predicting species occurrences from bioclimate. BISHOP infers how physiography affects bioclimate, estimates how these effects vary geographically, and produces high-resolution (10 m) maps of bioclimate over large regions. It also predicts species distributions. After introducing this approach, we apply it in an empirical study focused on topography and trees. Using data on 216 North American tree species, we document the biogeographic patterns that enable BISHOP, we estimate how four terrain variables (northness, eastness, windward exposure, and elevational position) each influence three climate variables, and we use these results to produce downscaled maps of tree-specific bioclimate. Model validation demonstrates that inferred bioclimate outperforms macroclimate in predicting distributions of separate species not used during inference, confirming its ecological relevance. Our results show that nearby bioclimates can differ by 5°C in temperature and twofold in moisture, with equator-facing, east-facing, windward-facing, and locally-elevated sites exhibiting hotter, drier bioclimates on average. But these effects vary greatly across climate zones, revealing that topographically similar landscapes can differ strongly in their bioclimate variation. These results have important implications for micrometeorology, biodiversity, and climate resilience.

As a supplement to the paper, we also released a companion R package that lets users make fine-scale tree-focused bioclimate predictions for any landscape in the US, based on the regression models from the paper.