Boosting builds better species models
Adding the weight of more data makes better species models.
When conservationists work to understand how some species of fish are distributed in habitats, Michigan State University scientists have found putting on some weight is beneficial.
In this week’s Ecological Modelling, researchers in the Aquatic Landscape Ecology Lab for the first time add the weight of species abundance to species distribution to better predict where targeted species are living, and how many likely are there.
“Improving species distribution model predictive accuracy using species abundance: Application with boosted regression trees” by fisheries and wildlife research associates Hao Yu and Arthur Cooper and professor Dana Infante outlines their development of weighted boosted regression tree (BRT) models for 55 fish species native to the northeastern United States that incorporated information on their abundances.
Commonly, BRTs are developed using just presence or absence of species at a subset of sites throughout a region of interest to model site suitability, even though species abundances may be a more robust indicator of site suitability than just presence or absence.
They say the results clearly demonstrated that these “weighted” BRT models outperformed BRTs developed just from presence-absence data for species that are rarer or have smaller distributions.
The work suggests that weighted BRTs may be more suitable for rarer species and provide richer insights for their conservation and management than traditionally applied BRTs. They note this innovative approach holds promise for providing new insights into the conservation of both aquatic and terrestrial organisms.
This study was funded by the USGS Aquatic Gap Analysis Program. Infante is a member of MSU’s Center for Systems Integration and Sustainability.