Adaptive classification of marine ecosystems: Identifying biologically meaningful regions in the marine environment

The move to ecosystem-based management of marine fisheries and endangered species would be greatly facilitated by a quantitative method for identifying marine ecosystems that captures temporal dynamics at meso-scale (10s or 100s of kilometers) resolutions. Understanding the dynamics of ecosystem boundaries, which may differ according to the species of interest or the management objectives, is a fundamental challenge of ecosystem-based management. We present an adaptive ecosystem classification that begins to address these challenges. To demonstrate the approach, we quantitatively bounded distinct, biologically meaningful marine regions in the North Pacific Ocean based on physical oceanography. We identified the regions by applying image classification algorithms to a comprehensive description of the ocean's surface, derived from an oceanographic circulation model. Our resulting maps illustrate 15 distinct marine regions. The size and location of these regions related well to previously described water masses in the North Pacific. We investigated seasonal and long-term changes in the pattern of regions and their boundaries by dividing the oceanographic data into four seasons and two 10-year time periods, one on either side of the 1976–1977 North Pacific Ocean climate regime shift. We compared our results for each season across the regime shift and for sequential seasons within regimes using the Kappa Index of Agreement and the index of Average Mutual Information. Seasonal patterns were more similar between regimes than from one season to the next within a regime, while the magnitude of seasonal transitions appeared to differ before and after the regime shift. We assessed the biological relevance of the identified regions using seasonal maps derived from remotely sensed chlorophyll-a concentrations ([chl-a]). We used Kruskal–Wallis and Wilcoxon rank sum tests to evaluate the correspondence between the [chl-a] maps and our post-regime shift regions. There was a significant difference in [chl-a] among the regions in all seasons. We found that the number of regions with distinct chlorophyll signatures, and the associations between different regions, varied by season. The overall pattern of association between the regions was suggestive of observed, broad-scale patterns in the seasonal development and distribution of primary production in the North Pacific. This demonstrated that regions with different biological properties can be delineated using only physical variables. The flexibility of our approach will enable researchers to visualize the geographic extents of regions with similar physical conditions, providing insight into ocean dynamics and changes in marine ecosystems. It will also provide resource managers with a powerful tool for broad application in ecosystem-based management and conservation of marine resources.