Global Ocean Surface Connectivity Matrix of Minimum Transit Times

Citation Jönsson, B. F. & Watson, J. R. (2016). The timescales of global surface-oceanconnectivity. *Nature Communications*, 7, 11239. https://doi.org/10.1038/ncomms11239 Overview This file contains two global ocean surface connectivity matrices that quantify the minimum transit time — in days — for water parcels to travel between any pair of the 11,116 ocean patches that tile the global surface ocean at roughly 2° × 2° resolution. The matrices are the core dataset of Jönsson & Watson (2016). The key findingof that paper is that the fastest path between two randomly chosen ocean patches is, on average, less than a decade. The dataset has since been widely used in studies of marine biogeography, larval dispersal, pathogen spread, andmicroplastic transport. Methodology Ocean model and particle tracking Particles were advected offline using the TRACMASS Lagrangian particle-tracking code driven by velocity fields from the ECCO2 (Estimating the Circulation and Climate of the Ocean, Phase II) state estimate at 1/4° × 1/4° horizontal resolution. Particles were released at the second model depth layer (≈ 5–20 m) to avoid numerical artefacts at the surface. Seeding strategy Particles were seeded on 9 separate dates, one per year from 2001 to 2009, with 3,456 particles per ocean patch per seeding date (≈ 36 million particles in total). Each particle was advected forward for up to 100 years. Ocean patches (regions) The global ocean was divided into 11,116 non-overlapping patches by aggregating the 1/4° × 1/4° ECCO2 grid into approximately 8 × 8 cell blocks (≈ 2° × 2° each). Patch boundaries follow the ECCO2 grid exactly; patch sizes vary slightly near coastlines and at high latitudes. The centroid latitude and longitude of each patch are stored in the `reglat` and `reglon` coordinate variables, and the lookup map from individual 1/4° grid cells to patch indices is stored in `regmat`. Direct connectivity (mintmat) `mintmat[i, j]` stores the minimum time (days) observed across all 9 seeding years for any particle released in patch j to first arrive in patch i. The matrix is sparse: roughly 17% of patch pairs were directly connected within the 100-year simulation window; all other entries are NaN. The diagonal (same-patch) is NaN because within-patch crossings were excluded to avoid artefacts. Dijkstra shortest-path connectivity (dijkmintmat) To fill the gaps in `mintmat`, Dijkstra's shortest-path algorithm was applied to the direct matrix, finding the fastest multi-hop route through intermediate patches (e.g. A → B → C → D). Connectivity times shorter than one year were temporarily removed before running Dijkstra to prevent within-patch crossings from inflating multi-hop estimates; they were reinstated in the final matrix. `dijkmintmat` is dense (≈ 98% of patch pairs have a finite value). Diagonal elements are 0 (a patch is trivially reachable from itself in 0 days). **Note on directionality.** Both matrices are asymmetric: the time to travel from patch A to patch B generally differs from the time to travel from B to A, reflecting the directionality of ocean currents. Longhurst biogeographic provinces Each patch was assigned to the Longhurst (2007) biogeographic classification using the patch centroid coordinates. Three levels of classification are stored: basin (5 classes), biome (4 classes), and province (up to 57 classes). File contents Dimensions | Dimension | Size | Description ||---|---|---|| `to_reg` | 11,116 | Destination patch index (0-based) || `from_reg` | 11,116 | Source patch index (0-based) || `lat` | 720 | ECCO2 grid latitude (0.25° resolution, 89.875°S–89.875°N) || `lon` | 1,440 | ECCO2 grid longitude (0.25° resolution, 179.875°W–179.875°E) | Variables | Variable | Dimensions | dtype | Description ||---|---|---|---|| `mintmat` | `(to_reg, from_reg)` | float32 | Minimum direct transit time (days); NaN = not connected within 100-year simulation || `dijkmintmat` | `(to_reg, from_reg)` | float32 | Minimum Dijkstra shortest-path transit time (days); NaN = unreachable || `regmat` | `(lat, lon)` | int64 | Patch index for each 1/4° × 1/4° ECCO2 grid cell (0 = land or unassigned) || `regid` | `(to_reg,)` | int64 | Sequential 1-based patch number (1–11,116) || `longhurst_basins` | `(to_reg,)` | float64 | Longhurst ocean basin code (1–5); NaN = unassigned (~6% of patches) || `longhurst_biomes` | `(to_reg,)` | float64 | Longhurst biome code (1–4); NaN = unassigned (~6% of patches) || `longhurst_regions` | `(to_reg,)` | float64 | Longhurst province code (1–57); NaN = unassigned (~6% of patches) | Coordinates | Variable | Dimensions | dtype | Description ||---|---|---|---|| `reglat` | `(to_reg,)` | float64 | Centroid latitude of each patch (degrees north; −78.3° to 84.4°) || `reglon` | `(to_reg,)` | float64 | Centroid longitude of each patch (degrees east; −179.1° to 179.4°) | Key statistics | Variable | Min | Mean | Max | Finite values ||---|---|---|---|---|| `mintmat` | 6 days | 5,323 days (≈ 14.6 yr) | 36,480 days (≈ 99.9 yr) | 17.1% of pairs || `dijkmintmat` | 0 days | 2,041 days (≈ 5.6 yr) | 31,664 days (≈ 86.7 yr) | 97.9% of pairs | Validation The model was validated against the detection of Fukushima-derived radioactiveisotopes on the US Pacific Coast following the March 2011 accident. The modelled minimum transit time from the source region to the US coast was 3.5 years, consistent with the first observed detection at approximately 3.6 years. Notes - For analyses requiring symmetric connectivity, the minimum of `mintmat[i, j]` and `mintmat[j, i]` (or the equivalent Dijkstra values) can be used.- Longhurst NaN values (~6% of patches) occur at patch centroids that fall outside the Longhurst province boundaries (e.g. near coastlines or at high latitudes).