21  Summary

The Marine Sensitivity Toolkit (MST) provides a reproducible system for assessing the sensitivity of marine ecosystems to offshore energy development across US waters. This summary describes release v7; figures are read from that release.

21.1 What this release delivers

  • 16,153 valid marine species on a 0.05° grid (~4 km cells) across the US EEZ study area, of which 9,230 fall inside the 20 BOEM Program Areas, integrating 8 source datasets — continuous suitability models, regulatory habitat designations and expert range maps.

  • 7 scored species categories (invertebrate, fish, other, bird, coral, mammal, turtle), assigned from taxonomy rather than from the dataset a model arrived in.

  • Extinction-risk scoring (1–100) folds in ESA listing, the MMPA floor (assigned by taxonomy, so no marine mammal is missed) and the MBTA floor (the FWS CFR 50 §10.13 species list, not all birds), falling back to the IUCN Red List where there is no US listing.

  • Model merging produces one surface per taxon, with the expert range constraining suitability predictions rather than merely competing with them, and the governing extinction risk applied at the range cells. Sea turtles merge multiplicatively, so risk varies spatially within a species.

  • Ecoregional rescaling normalizes each metric to 0–100 within a BOEM Ecoregion, so scores are comparable among Program Areas in one ecoregion without being dominated by naturally species-rich waters.

  • Interactive applications — Scores and Species — render any published release from a ?ver= parameter, so a reader can look at the release a report cited rather than only the newest one.

21.2 Delivered since the first release

Several capabilities listed as future work in earlier editions of this documentation have since shipped:

  • A DAG pipeline. The workflow is a targets graph generated from each notebook’s own metadata, so adding a dataset means adding a notebook, not editing a pipeline definition.
  • Cloud-Optimized GeoTIFFs and a STAC catalog. Distributions and score surfaces are published as COGs with STAC metadata, discoverable through a searchable STAC API, and the score rasters replaced an on-demand SQL-per-tile factory.
  • Partitioned Parquet releases. Each version is published as Parquet on S3 and read directly, including from the browser, rather than shipped as a monolithic database file.
  • One application across all releases. The per-version application forks are retired; version is data, settled by a published manifest, not a code fork.
  • Whole-range models. Distributions are captured globally rather than clipped to the study area, so a species map shows the species instead of the analysis extent.

The H3 hexagonal grid proposed earlier was evaluated and not adopted: v8 stays on a 0.05° raster grid, made global rather than regional. Interpolating suitability onto hexes added a resampling step without improving the analysis, and the raster grid aligns directly with the source environmental layers.

21.3 Future directions

  • OBIS/GBIF validation — refine distributions against occurrence records, with temporal filters.
  • Density models — NOAA SEFSC and NCCOS density surfaces are ingested and catalogued but excluded from scoring, pending an agreed mapping from individuals km-2 onto the 0–100 suitability scale. They are flagged is_scored = FALSE so no count on these pages includes them.
  • Habitat sensitivity — benthic habitats (coral reefs, seamounts, hydrothermal vents) as a component alongside species and productivity.
  • Stressor–receptor matrices — extend from sensitivity alone to full vulnerability by adding exposure for offshore wind, oil and gas, and other activities (see Chapter 11).
  • Application enhancements — arbitrary area selection, taxonomic-tree browsing, automatic figure generation, and scrollytelling explainers.
  • Infrastructure transfer — support moving the data and applications onto BOEM infrastructure for long-term operation.