Dynamic ensemble models to predict distributions and anthropogenic risk exposure for highly mobile species

Abrahms B, Welch H, Brodie S, et al (2019) Dynamic ensemble models to predict distributions and anthropogenic risk exposure for highly mobile species. Diversity and Distributions 25:1182–1193. https://doi.org/10.1111/ddi.12940

Aim Advances in ecological and environmental modelling offer new opportunities for estimating dynamic habitat suitability for highly mobile species and supporting management strategies at relevant spatiotemporal scales. We used an ensemble modelling approach to predict daily, year-round habitat suitability for a migratory species, the blue whale (Balaenoptera musculus), and demonstrate an application for evaluating the spatiotemporal dynamics of their exposure to ship strike risk. Location The California Current Ecosystem (CCE) and the Southern California Bight (SCB), USA.

Methods We integrated a long-term (1994–2008) satellite tracking dataset on 104 blue whales with data-assimilative ocean model output to assess year-round habitat suitability. We evaluated the relative utility of ensembling multiple model types compared to using single models, and selected and validated candidate models using multiple cross-validation metrics and independent observer data. We quantified the spatial and temporal distribution of exposure to ship strike risk within shipping lanes in the SCB. Results Multi-model ensembles outperformed single-model approaches. The final ensemble model had high predictive skill (AUC = 0.95), resulting in daily, year-round predictions of blue whale habitat suitability in the CCE that accurately captured migratory behaviour. Risk exposure in shipping lanes was highly variable within and among years as a function of environmental conditions (e.g., marine heatwave). Main conclusions Daily information on three-dimensional oceanic habitats was used to model the daily distribution of a highly migratory species with high predictive power and indicated that management strategies could benefit by incorporating dynamic environmental information. This approach is readily transferable to other species. Dynamic, high-resolution species distribution models are valuable tools for assessing risk exposure and targeting management needs.