Coastal dune vegetation succession and stabilisation monitoring
Coastal dune systems shift between bare mobile sand, pioneer swards and fixed grassland on timescales from days to centuries. Multispectral time-series from Sentinel-2 and commercial VHR imagery resolve these succession stages and flag where stabilisation is failing.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at mid-latitudes with both satellites. Resolves individual dune ridges in most macrotidal and mesotidal systems. Red-edge bands (705 nm, 740 nm) improve separation of sparse pioneer cover from bare sand. Free archive from 2015.
- Planet SuperDove: 3 m resolution, near-daily revisit globally. Eight bands including two red-edge channels. Useful for narrow foredune zones and small dune systems where 10 m pixels mix too many cover classes. Commercial licence required.
- Maxar WorldView-3: 0.3 m panchromatic, 1.24 m multispectral, 16 shortwave-infrared bands. The only commercial sensor that can map individual marram grass tussocks and detect fine-scale blowout edges. Tasked on demand; archive coverage is patchy for remote coastlines.
- Landsat 8/9 OLI: 30 m resolution, 8-day revisit (combined), free archive to 1984. Too coarse to resolve narrow foredune zones but invaluable for decadal trend analysis of greenness and bare-sand extent across large dune systems. Consistent radiometric calibration across the archive is a genuine asset.
What the spectral gradient across a dune actually tells you
A coastal dune system is, spectrally, a gradient from high-albedo quartz sand to dense chlorophyll-rich vegetation. Bare mobile sand has very high reflectance across visible wavelengths and low near-infrared (NIR) return. Pioneer species such as marram grass (Ammophila arenaria) or sea rocket introduce chlorophyll absorption in red wavelengths and a sharp NIR rise, producing a modest but measurable NDVI signal even at low canopy cover. Fixed dune grassland and scrub push NDVI above 0.5 in summer. That gradient is the classification ladder.
Sentinel-2's red-edge bands are particularly useful here. The 705 nm band sits at the onset of the vegetation red-edge and responds to chlorophyll content at low cover fractions where the standard red/NIR NDVI is still noisy. Studies using Sentinel-2 red-edge indices have separated sparse pioneer communities (10 to 30 percent cover) from bare sand with considerably better accuracy than broad-band NDVI alone, though exact figures vary with dune system and season. The shortwave-infrared bands (1610 nm and 2190 nm) add moisture sensitivity, which helps distinguish wet compacted sand from dry mobile sand.
The wind problem: transient exposure versus persistent mobility
Wind is the confounding variable that makes single-date dune classification unreliable. A storm can strip pioneer vegetation from a foredune crest and expose bare sand within 48 hours. Captured in a single image, that patch looks identical to a chronically mobile blowout. The distinction matters enormously for management: one is a weather event, the other is a failure of stabilisation.
The solution is multi-date compositing. By stacking cloud-free Sentinel-2 acquisitions across a full growing season and computing the median or percentile NDVI, transient exposures wash out. Persistent bare sand, by contrast, stays spectrally cold across all acquisitions. A practical threshold is to flag any pixel that falls below an NDVI of 0.1 in more than 70 percent of cloud-free observations in a season; that distinguishes chronic mobility from episodic disturbance with reasonable confidence. The exact threshold needs local calibration against field survey or VHR imagery, particularly in systems with naturally sparse pioneer cover.
Cloud is a real constraint on Atlantic-facing dune coasts. The British Isles, the Bay of Biscay coast and the Pacific Northwest can have extended cloudy periods in autumn and winter, precisely when storm erosion events occur. Sentinel-2's 5-day revisit helps, but a coastal manager should expect 15 to 30 cloud-free acquisitions per year in persistently overcast climates rather than the theoretical maximum.
Mapping succession stages: classification approaches and their limits
Supervised classification using training polygons from field survey or VHR imagery is the standard approach. A typical scheme distinguishes four to six classes: open water and wet sand, dry mobile sand, embryo dune and pioneer sward, semi-fixed grassland, fixed dune grassland, and scrub or woodland. Random forest classifiers trained on Sentinel-2 spectral bands plus texture metrics derived from the 10 m NIR band typically achieve overall accuracies of 75 to 90 percent in published studies on European dune systems, with the greatest confusion between semi-fixed and fixed grassland.
The 10 m pixel is the binding constraint for foredune mapping. On many Atlantic dune systems the embryo dune and pioneer zone is only 5 to 20 m wide. A single Sentinel-2 pixel can span the entire zone, mixing sand and vegetation signals and causing systematic underestimation of pioneer cover. WorldView-3 or Planet SuperDove imagery resolves this, but at higher cost and with less temporal depth. A practical hybrid is to use Sentinel-2 for system-wide annual change detection and VHR imagery for detailed foredune-zone surveys every three to five years.
Invasive species complicate the classification further. Carpobrotus (iceplant) has a distinctive succulent spectral signature and can be separated from native dune vegetation using SWIR bands, but this is a separate analytical problem covered on the invasive-species page in this library.
Reading long-term stabilisation trends from the Landsat archive
Landsat's 40-year archive is the only freely available satellite record long enough to track full succession cycles or assess the impact of management interventions such as marram planting, fencing or scrub clearance. At 30 m resolution, individual dune ridges blur together on narrow systems, but the aggregate greenness signal across a dune field is still informative.
A standard approach is to compute annual peak-season NDVI composites from Landsat OLI (2013 onwards) and its predecessors, then fit a linear or piecewise trend to each pixel. Increasing NDVI over decades indicates progressive stabilisation or scrub encroachment; declining NDVI indicates renewed mobility or vegetation loss. The Landsat Collection 2 surface-reflectance product provides consistent atmospheric correction across the archive, which matters when comparing 1984 Thematic Mapper data with 2024 OLI-2 data. Absolute NDVI values are not directly comparable across sensors, but anomalies relative to a per-pixel baseline are.
One honest limit: Landsat's 16-day single-satellite revisit (8 days with Landsat 8 and 9 combined) means that cloud-free annual composites on overcast coasts may rest on only two or three usable scenes. Trend uncertainty is higher in those locations and should be reported alongside the trend estimate.
From classification to management outputs
The analytical products that matter to a coastal manager or conservation authority are not spectral indices but decisions. Which dune sections are actively destabilising and need intervention? Where has scrub encroachment reduced open-sand habitat below the threshold needed by specialist invertebrates? Is a recent marram-planting programme showing measurable greenness increase within two growing seasons?
These questions translate into specific deliverables: annual succession-stage maps in GIS format with area statistics per class, change-detection layers flagging net transitions between classes, and pixel-level NDVI trend rasters with confidence intervals. Alert products, triggered when bare-sand extent in a defined management zone exceeds a rolling baseline by more than one standard deviation, are feasible on Sentinel-2 revisit cadence and give managers early warning before a blowout becomes a breach.
Satellize runs this class of multispectral time-series analysis on open constellations and adds commercial tasking where VHR coverage is needed. The Tonga crop-estimation programme uses a similar seasonal compositing and classification pipeline, adapted to a very different vegetation context. Coastal dune work follows the same logic: define the spectral classes, build the composites, track the transitions.
Typical figures
| Spatial resolution (operational) | 10 m (Sentinel-2), 3 m (Planet SuperDove), 1.24 m multispectral / 0.3 m pan (WorldView-3), 30 m (Landsat OLI) |
| Revisit cadence | 5 days (Sentinel-2, mid-latitudes, both satellites); near-daily (Planet SuperDove); on-demand tasking (WorldView-3); 8 days combined (Landsat 8 + 9) |
| Spectral bands relevant to dune vegetation | Visible (490–665 nm), red-edge (705, 740 nm), NIR (842 nm), SWIR (1610, 2190 nm); Sentinel-2 MSI and WorldView-3 SWIR cover all of these |
| Minimum resolvable feature | Individual dune ridges >20 m wide at Sentinel-2 10 m; foredune zones <10 m wide require VHR (WorldView-3 or SuperDove) |
| Cloud-free acquisition frequency (overcast coasts) | 15–30 usable Sentinel-2 scenes per year typical on Atlantic-facing coasts; compositing across full season recommended |
| Archive depth | Sentinel-2: 2015–present; Landsat: 1984–present (Collection 2 surface reflectance); Planet SuperDove: 2021–present (commercial) |
| Succession-stage classification accuracy | 75–90% overall accuracy reported in published studies on European dune systems using Sentinel-2 multispectral + texture features; highest confusion between semi-fixed and fixed grassland classes |
| Delivery formats | GeoTIFF classification rasters, vector polygon shapefiles or GeoPackage, CSV area-statistics tables, GIS-ready change-detection layers |
| Latency (operational monitoring) | Sentinel-2 Level-2A surface reflectance available within 1–3 days of acquisition via Copernicus Data Space |
Analytics Satellize can run
| Annual succession-stage map | Supervised random-forest classification on seasonal median Sentinel-2 composites (10 bands including red-edge and SWIR), trained on field or VHR reference data | GeoTIFF and polygon GIS layer with per-class area statistics; updated annually |
| Bare-sand mobility index | Per-pixel frequency of NDVI below 0.1 across all cloud-free acquisitions in a growing season; distinguishes chronic mobility from storm-driven transient exposure | Raster layer (0–100% frequency) with flagged zones of persistent mobility; seasonal report |
| Long-term greenness trend | Annual peak-season NDVI composites from Landsat Collection 2 surface reflectance (1984–present); per-pixel linear trend with Mann-Kendall significance test | Trend raster with slope and p-value bands; PDF summary of stabilisation or destabilisation patterns by dune zone |
| Foredune-zone detailed survey | Object-based image analysis on WorldView-3 or Planet SuperDove VHR imagery; delineates pioneer tussock patches, blowout edges and embryo dune extent at sub-metre to 3 m scale | High-resolution polygon dataset; integrates with Sentinel-2 system-wide map as a calibration and validation layer |
| Vegetation transition change-detection layer | Post-classification comparison of consecutive annual maps; net area of each class-to-class transition (e.g. fixed grassland to scrub, pioneer to bare sand) computed with uncertainty estimate | Transition matrix CSV and GIS change layer; flags management-relevant transitions exceeding defined thresholds |
| Destabilisation alert | Rolling baseline of bare-sand extent per management zone from Sentinel-2 time series; alert triggered when current 30-day composite exceeds baseline mean by more than one standard deviation | Email or API alert with affected zone ID, magnitude of anomaly and link to current composite image |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.