An integrated platform monitoring shoreline change, coastal habitats, and erosion risk across Cape Cod, Martha's Vineyard, Nantucket, and the Elizabeth Islands.
Built on NOAA, USGS, and NASA data, this platform combines Sentinel-2 satellite imagery, live water temperature and sea level from the Woods Hole tide gauge, vegetation indices, and composite risk scoring to track how our coastline — and the salt marshes, dunes, and barrier beaches it sustains — is changing over time. Two machine-learning layers power the analysis: a deep-learning U-Net for shoreline detection and Gaussian Process regression for ocean forecasts. Here's how to explore it:
Each colored polygon is a coastal zone ranked by its composite risk index. Click to see the score breakdown, zone analysis, erosion rates, habitat status, and conservation contacts.
Switch between Erosion Rates, Protected Lands, and Monitoring Stations to see different views of what's happening along the coast.
From any zone, click "Export Report" to download a printable report with the composite risk score, satellite erosion data, habitats, and conservation contacts.
The top nav has tabs for full zone rankings, live NOAA ocean data with ML-driven "ML Proj." forecasts and 90% confidence bands, methodology details, and data sources.
Live and historical data covering Cape Cod, Martha's Vineyard, Nantucket, and the Elizabeth Islands
Habitat status compiled from published research, agency monitoring, and satellite vegetation analysis
All 19 monitored zones ranked by erosion severity · click any row to open its detail panel
| # | Zone | Region | Survey Rate | Sat Rate (7yr) | Risk |
|---|
Survey rates: USGS CoSMO · Satellite rates: ESA Sentinel-2 via GEE
Federal, state, and satellite data powering this dashboard
How satellite rates, erosion figures, and risk classifications are calculated
The dashboard combines three independent data streams: satellite imagery (for shoreline change), NOAA tide-gauge readings (for ocean conditions), and ground-survey rates (for calibration). Two machine-learning layers sit on top — one for satellite shoreline detection, one for forecasting ocean trends. Everything else is plain arithmetic, transparently scored.
ESA Sentinel-2 imagery (10 m resolution) is pulled through Google Earth Engine for each of the 19 zones, restricted to a September–October window each year so beaches are compared in their late-summer profile (after rebuilding, before nor'easters). This eliminates seasonal noise from the change signal.
Primary method — CoastSat U-Net: a deep-learning model trained on labeled satellite imagery that classifies water/land per pixel and extracts a sub-pixel shoreline polyline for each scene. Robust to wet sand, cliff shadows, seaweed, and breaking waves where simple thresholds fail.
Fallback — NDWI threshold: a spectral index on the Green and Near-Infrared bands. Faster, simpler, less accurate on edge cases. Used where ML output isn't yet available.
Per-year shoreline positions feed a linear regression to produce 2-, 3-, 5-, and 7-year erosion rates. For tidal-flat zones (Barnstable, Dennis & Brewster) where a stable shoreline length can't be defined, rates are reported in ac/yr instead of ft/yr.
Sea surface temperature, sea level, and salinity charts are built from NOAA CO-OPS Station 8447930 (Woods Hole, MA) — the closest continuously-operating tide gauge to the region. SST and SLR are pulled as real annual averages back to 1993 (sea level) and 2005 (temperature). Salinity uses dashboard-maintained historical values because no live API source exists for that station.
The dashed "ML Proj." line on every ocean chart comes from Gaussian Process regression — a real machine-learning model fit to the historical NOAA series. The shaded purple band shows the 90% confidence envelope — the model's range of plausible values around the projected trend, widening into the future as uncertainty grows. Charts forecasted this way (Sea Surface Temperature, Sea Level Rise, Salinity) show an ML badge next to their title.
Outlier years (e.g. partial-year readings or sensor anomalies) are automatically flagged via a residual-based detector and excluded from the fit, so a bad reading doesn't skew the projection.
Each zone receives a composite score (0–10) built from three independent dimensions — the same Hazard × Exposure × Vulnerability framework FEMA, IPCC, and most coastal-risk literature use. The three dimensions are computed separately and then averaged into the composite.
Hazard — how dangerous is the physical process here? Combines the 7-year erosion rate (prefers CoastSat ML rate when available, falls back to USGS/MCZM survey), the trend direction (accelerating / steady / decelerating), and the zone's ocean exposure (open Atlantic vs. sheltered bay).
Exposure — what's in harm's way? Combines building density along the coast, habitat coverage within the zone, and protected-land presence.
Vulnerability — how easily does damage propagate? Combines beach type (cliffs and marshes don't recover; sandy beaches rebuild; barrier islands roll back), shoreline armoring, and current habitat stress.
Thresholds are relative, not absolute. Each zone is ranked against the other 18 in the region — the bottom third are colored Low, the middle third Medium, the top third High. This means "Low" here means lower-risk *relative to other Cape & Islands zones*, not absolutely safe. Zone polygon colors on the map reflect this ranking.
The framework helps surface zones where multiple factors overlap. A zone with high erosion but nothing built behind it scores differently from a zone with moderate erosion and dense development.
Cape and Islands Conservation Dashboard (2026). Coastal erosion and habitat monitoring for Martha's Vineyard, Nantucket, and Cape Cod. Satellite analysis: ESA Sentinel-2 + CoastSat U-Net via Google Earth Engine. Ocean data: NOAA CO-OPS Station 8447930 (Woods Hole, MA). Forecasts: Gaussian Process regression. Survey rates: USGS National Shoreline Change Database. Coastal management data: MA Coastal Zone Management.