COS 17-8 - Species distribution models for archipelago-scale analysis of Hawaiian marine ecosystems

Monday, August 6, 2012: 4:00 PM
Portland Blrm 254, Oregon Convention Center
Erik C. Franklin, Hawaii Institute of Marine Biology, University of Hawaii, Megan J. Donahue, Hawaii Institute of Marine Biology, Kaneohe, HI and Paul L. Jokiel, Hawaii Institute of Marine Biology, University of Hawaii, Kaneohe, HI
Background/Question/Methods

Macroscale ecosystem analyses can increase our broader understanding of climate change impacts, coupled land-sea interactions, and invasive species dynamics in tropical coastal communities.  Often, the initial barrier to regional analysis in tropical systems, such as coral reefs, is the lack of data in sufficient extents, appropriate formats, or desired resolutions to describe ecosystems in a spatially explicit and realistic manner. Data from remote sensing can not comprehensively address this void due to limitations on sensor depth penetration and taxonomic interpretation of imagery. Thus, an integrated strategy of field observations, sensor data, and modeling is required for a regional synthesis of detailed biological information.

We developed spatially-explicit species distribution models of corals for the shallow waters of the Hawaiian Islands, an island archipelago in the Central Pacific Ocean that extends linearly for 2600 km, with analysis focused on the human-populated main Hawaiian Islands. A data compilation yielded over 12,000 geo-referenced coral field survey samples performed between 2000-2009 for the four most common coral species (Porites compressa, Porites lobata, Montipora capitata, and Pocillopora meandrina), and environmental datasets relating to wave energy, available light, geomorphology, and substrate type in shallow waters (<= 30 m depth).  Data were standardized to high resolution spatial grids (50 m cells) with a weighted average of values assigned to each cell for the study time period. Data were randomly assigned to a training (60%) and validation (40%) set for 10 iterations. Model runs were performed with boosted regression trees, a machine learning method with model performance evaluated using standard metrics.

Results/Conclusions

Probability of coral species occurrence was driven primarily by maximum significant wave heights, a proxy for wave energy experienced by coral communities. Model performance evaluated by receiver operating characteristic area under curve (AUC) and the kappa statistic indicated very good agreement between model construction and validation data sets. The boulder coral Porites lobata was predicted to occur on exposed coastlines with high wave energies (north, west, and south) within shallow to moderate fore reef depths (6-20m). There was good concurrence between the spatial patterns in model output upon visual inspection. Wave variables were stronger predictors for the finger coral Porites compressa than the other three coral species which may reflect the greater vulnerability of this branching coral species to physical disturbance. The Hawaiian coral species distribution maps provide a basis for spatially-explicit, high resolution examinations of coral population dynamics studies, climate change scenario testing, and marine spatial planning exercises at an archipelagic scale.