Mapping marine habitat suitability and uncertainty of Bayesian networks: a case study using Pacific benthic macrofauna

Resource managers increasingly use habitat suitability map products to inform risk management and policy decisions. Modeling habitat suitability of data-poor species over large areas requires careful attention to assumptions and limitations. Resulting habitat suitability maps can harbor uncertainties from data collection and modeling processes; yet these limitations are not always transparent to resource managers, who increasingly rely on maps for spatial planning and risk assessment purposes. Interpretation of habitat suitability maps can be improved by visually communicating model uncertainty and data foundations. We applied Bayesian networks (BNs) to a small, marine dataset to model the probability of occurrence (PO) of benthic macrofauna. We also used BNs to create maps displaying model parameter uncertainty and data limitations. We developed BN models for three macrofauna species: a marine gastropod, Aystris gausapata, a marine bivalve, Axinopsida serricata, and a marine worm, Sternaspis fossor. We produced three map products from the BN models of each species: (1) a habitat suitability map of the PO projected from regional predictor variables; (2) an uncertainty map, displaying statistical variance of model predictions of occurrence probability; and (3) an experience map, displaying the empirical basis for PO predictions (equivalent sample size). Map results showed occurrence probability to be high and widespread for Ax. serricata, low to moderate and more limited to deeper offshore areas for Ay. gausapata, and low to high in shallow sandy regions and deeper silty regions, respectively, for S. fossor. The uncertainty and experience maps for each species helped identify regions to prioritize for future sampling. Our results are the first to show that BNs can effectively model habitat suitability of benthic macrofauna, and our detailed methods can be applied to a variety of taxa and systems. Visually describing statistical model uncertainty and equivalent sample size in map format improves interpretation of habitat suitability map predictions and supports place-based risk management of marine management.