Using predictive models to identify kelp refuges in marine protected areas for management prioritisation

Kelp forests serve as the foundation for shallow marine ecosystems in many temperate areas of the world but are under threat from various stressors, including climate change. To better manage these ecosystems now and into the future, understanding the impacts of climate change and identifying potential refuges will help to prioritise management actions. In this study, we use a long-term dataset of observations of kelp percentage cover for two dominant canopy-forming species off the coast of Victoria, Australia: Ecklonia radiata and Phyllospora comosa. These observations were collected across three scuba sampling programs that extend from 1998 to 2019. We then associated those observations with habitat and environmental variables including depth, seafloor structure, wave climate, currents, temperature, and population connectivity in generalised additive mixed effects models and used these models to develop predictive maps of kelp cover across the Victorian MPAs. These models were also used to project kelp coverage into the future by replacing wave climate and temperature with future projections (2090, RCP4.5 and RCP8.5). Once all the past, current, and future predictions were compiled, we calculated percent cover change from 1998-2019, stability over the same period, and future predicted change in percent cover (2019-2090) to understand the dynamics for each species across the MPAs. We also used the current percentage cover, stability, and future percentage cover to develop a ranking system for classifying the maps into high climate vulnerability, climate vulnerability, neutral, potential refugia, and likely refugia. A management framework was then developed to use those refugia ranking values to inform management actions and we applied this framework across three case studies: one at the scale of the MPA network and two at the scale of individual MPAs, one where management decisions were the same for both species and one where the actions were species-specific. This study shows how species distribution models, both contemporary and with future projections, can help to identify potential refugia areas that can be used prioritise management decisions and future-proof restoration actions.

Methods
Parks Victoria Subtidal Reef Monitoring Program (SRMP)

Parks Victoria SRMP targets areas within MPAs and sites directly adjacent to them and uses visual census methods developed by Edgar and Barrett (1997), Edgar and Barrett (1999), and Edgar et al. (1997). There are 80 SRMP sites used in this study that extend from 1998 to 2013. Each site within the sampling program was marked by GPS and visited in subsequent survey years where a 200 m transect was laid along a shallow (< 10 m) contour. To quantify the percentage coverage of kelp, 0.25 m2 quadrats, which are divided into a grid of 7 x 7 perpendicular wires to provide 49 points plus one corner making 50, are placed at 10 m intervals along the transect. The cover of kelp is then estimated by counting the number of times the species intersects with one of the points. These point counts are then divided by 50 and averaged across all quadrats in the transect to come up with a percentage coverage of each kelp for each site. For this project, only the percentage cover of E. radiata and P. comosa were extracted from the dataset.

Reef Life Survey (RLS)

Reef Life Survey (RLS) is a citizen science program where volunteer SCUBA divers conduct visual surveys of underwater reefs. This program began surveying 95 sites along the Victorian coast in 2007 and continues to extend the subtidal biological observation time series, including to 2019, the last year used in this study. See Edgar and Stuart-Smith (2009) for detailed methods. Briefly, a GPS point is taken at each site and it is surveyed using a 50 m line transect along a shallow depth contour with two or more contours targeted at each site. To collect data on kelp coverages, digital photo quadrats taken downward from a 50 cm height above the seabed are acquired at 2.5 m intervals along the transect line. Post-processing of these photo quadrats requires classifying the images into 16 functional or morphological categories of algae using CATAMI (Althaus et al. 2015), which is the standard image classification system for Australia. Once classified, the point counts of E. radiata and P. comosa were used to calculate percentage cover of each species.

Victorian Fisheries Authority (VFA)

To increase the coverage and time series of kelp observations, the VFA data were used. For this dataset, diver surveys are conducted annually at 202 monitoring sites across the coast of Victoria. These surveys began in 2002 and continued until 2016. At each site, the divers estimate the percent cover of E. radiata and P. comosa within six 30 m by 1 m transects at random cardinal directions from central site coordinates taken using a GPS point. The estimates for percent cover are based on visual approximation within 30 sections along each transect and then averaged to the nearest 10% across all transects for each site.

Data from these three sampling programs were collated into a single dataset for each species: Ecklonia radiata and Phyllospora comosa for statistical analyses. The resulting compiled dataset included over 300 locations sampled over 22 years across Victoria, both inside and outside marine protected areas. Each sampling program captured data using SCUBA following slightly different sampling protocols (see Supplementary Information 1.1 for details on differences in sampling). Then, for the sampling methods (either using quadrats, photo quadrats, or multiple transects), an average percentage coverage for each species was calculated per site per year (i.e., all quadrats or transects at each site were averaged to produce one percentage cover value per site), resulting in a time series of percentage cover at each site for both species.