Data and code from: Thresholding species distribution models: Simple approaches for land-use planning in multifunctional landscapes

Species distribution models (SDMs) are often used to understand changes to
species’ distributions and their habitats under different land-use
scenarios, enabling decision-makers to prioritize areas for management
efforts and balance environmental conservation with socio-economic demands
on the landscape. However, application of SDMs in land-use planning and
Environmental Impact Assessments (EIAs) remains limited due to challenges
in interpreting and communicating continuous predictions resulting from
these SDMs. Although different binarization methods have been used to
overcome such challenges, the choice of threshold can profoundly alter the
resulting binary habitat map, and most methods lack simplicity and require
access to underlying species occurrence and environmental data used to
develop the SDMs. Hence, there is a demand for testing simple alternative
binarization methods to enable in-house application of SDMs by
practitioners and to facilitate interpretation and communication. Using
SDMs of 103 boreal bird species in Alberta, Canada, we transform species
relative abundance predictions of SDMs into direct estimates of habitat
area, a proxy for habitat suitability, using four simple and three complex
thresholding methods. We compare the performance of the binarized models
for each bird species and between forest specialists
vs. generalists under land-use change scenarios. We found that
thresholded models reflect losses in suitable habitat under industrial
disturbance scenarios more realistically compared to continuous relative
abundance models. Notably, simple thresholding methods, particularly the
mean predicted relative abundance, performed similarly to complex
thresholding methods in predicting suitable habitat areas, and as
indicated by model evaluations using the area under the curve. These
findings suggest that using the mean as a binarization threshold can
effectively bridge the gap between complex SDMs and their application in
policy and planning, without sacrificing predictive accuracy. We conclude
that simple threshold binarization methods, such as the mean, can leverage
the strong predictive power of SDMs to provide insights into future
changes in species’ habitat during land-use planning scenarios, account
for their uncertainties, and expand their utility to facilitate
interpretation for science-informed decision-making in multifunctional
landscapes.