Floating plastic and marine debris detection from orbit
Satellite sensors can detect dense floating debris aggregations using spectral indices derived from Sentinel-2 and hyperspectral data, but reliable discrimination from Sargassum, foam, and pumice at the pixel level remains an open research problem.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible bands, 20 m in SWIR bands (B11 at 1,610 nm, B12 at 2,190 nm). The Floating Debris Index (FDI) and Plastic Index (PI) are derived from B8, B11 and B12. Revisit is 5 days at the equator with both satellites. The critical 1,215 nm absorption feature falls between Sentinel-2 bands, so it cannot be directly sampled.
- PRISMA (ASI): Italian Space Agency hyperspectral imager covering 400–2,500 nm at approximately 30 m spatial resolution and ~12 nm spectral sampling. Directly samples the 1,215 nm SWIR absorption feature that distinguishes many common plastics in laboratory conditions. Revisit is roughly 29 days without tasking; on-demand tasking is available.
- Landsat 8/9 OLI: 30 m multispectral resolution with SWIR bands at 1,610 nm and 2,200 nm, broadly equivalent to Sentinel-2 for index-based detection. 16-day revisit per satellite; combined 8-day revisit. Free archive to 2013 enables retrospective accumulation studies.
- Planet SuperDove: 3 m spatial resolution across 8 spectral bands including a red-edge band at ~705 nm. High resolution helps resolve debris patch boundaries and reduces mixed-pixel contamination, but SuperDove lacks SWIR bands, so plastic-specific spectral indices cannot be computed. Useful for confirming extent and texture of anomalies flagged by other sensors.
What a floating roof gives away: the physics of plastic spectra
Plastics are polymers. Their molecular bonds absorb infrared radiation at characteristic wavelengths, producing absorption features that differ measurably from water, algae, and mineral foam. The most discussed feature for marine debris sits near 1,215 nm, where common plastics including polyethylene and polypropylene show a distinct dip in reflectance. Seawater absorbs strongly across most of the near-infrared, so any floating material that reflects in this region stands out against the dark ocean background. That contrast is the physical basis for all orbital detection methods.
The problem is that several other floating materials also reflect in the near-infrared. Sargassum seaweed, sea foam, pumice from volcanic events, and fish-aggregating devices all produce elevated reflectance in overlapping spectral regions. Distinguishing them requires either fine spectral resolution to resolve the shape of the absorption feature, or ancillary information about location, season, and oceanographic context.
The Floating Debris Index and its honest limits
The Floating Debris Index, proposed in published literature by Biermann and colleagues and subsequently refined, is computed from Sentinel-2 bands B8 (NIR, 842 nm), B11 (SWIR1, 1,610 nm), and B12 (SWIR2, 2,190 nm). The index exploits the fact that floating debris, including plastic, reflects more strongly in SWIR than submerged or mixed material. A related Plastic Index uses B4 and B8. Both indices can flag dense accumulations in convergence zones, river plumes, and gyres with reasonable reliability when debris concentrations exceed roughly tens of kilograms per square metre of pixel area.
What neither index can do is confirm plastic composition at the pixel level. Sargassum blooms, which have expanded dramatically in the Atlantic since 2011, produce FDI signals that overlap substantially with debris signals. Foam from breaking waves, oil sheens, and pumice rafts add further ambiguity. Published validation studies consistently report confusion rates that preclude regulatory-grade attribution without in-situ confirmation. This is not a processing failure; it is a fundamental consequence of Sentinel-2's spectral design, which places no band directly on the 1,215 nm feature.
Hyperspectral instruments and the laboratory-to-ocean gap
PRISMA and the ESA-funded Ocean Plastic Sentinel concept address the spectral gap directly. With continuous coverage from 400 nm to 2,500 nm at roughly 12 nm sampling, PRISMA can in principle resolve the shape of the 1,215 nm absorption feature and compare it against spectral libraries of known plastics. Several published studies have demonstrated detection of artificial debris targets placed at sea for validation purposes.
Open-ocean conditions are considerably harder than controlled experiments. Sunglint, whitecaps, sub-pixel mixing with water, and the thin vertical extent of floating debris all reduce signal-to-noise. A plastic film one millimetre thick floating in a 30 m pixel contributes a reflectance increment that can fall below instrument noise in moderate sea states. Published detection limits for PRISMA in realistic ocean conditions are generally quoted for aggregations rather than isolated items: patches of order hundreds of square metres with high areal coverage of debris. Dispersed microplastic, which constitutes the majority of oceanic plastic by particle count, is invisible to any current orbital sensor.
Where detection actually works: convergence zones and river mouths
Satellite-based detection is most credible in two settings. First, oceanographic convergence zones where surface currents concentrate debris into windrows and patches: the subtropical gyres, the North Pacific Subtropical Convergence Zone, and seasonal accumulation areas in the Mediterranean and Caribbean. Second, river plumes immediately after flood events, where high-concentration debris fields extend kilometres offshore and are detectable in Sentinel-2 imagery at 10–20 m resolution.
In both settings, the practical workflow combines the FDI or a similar index with oceanographic masking. Pixels flagged as potential debris are filtered against sea-surface temperature composites, chlorophyll-a maps from ocean colour sensors such as OLCI on Sentinel-3, and wind-field data to exclude foam. What remains is a probability surface, not a confirmed plastic map. Responsible reporting describes it as such.
Planet SuperDove imagery at 3 m resolution can then be tasked over flagged areas to assess patch texture and continuity. Debris fields tend to have irregular, filamentary structure that differs visually from Sargassum mats, though this distinction is qualitative and analyst-dependent.
What buyers should expect from a monitoring programme
A realistic monitoring programme built on current open-constellation data can deliver three things: systematic flagging of anomalous floating material in defined areas of interest, temporal tracking of patch movement and dispersion using Sentinel-2's 5-day revisit, and retrospective analysis of accumulation patterns using the Landsat archive back to 2013. It cannot deliver confirmed plastic mass estimates, species-level discrimination from Sargassum, or detection of dispersed microplastic.
Hyperspectral tasking via PRISMA adds spectral confidence for individual events but at lower revisit and higher coordination cost. The science community is actively working on improved retrieval algorithms and dedicated sensors; the Ocean Plastic Sentinel concept proposed a dedicated 10 m hyperspectral instrument specifically for this application, but no such satellite is yet operational.
Satellize runs index-based debris screening on Sentinel-2 and Landsat archives as part of its open-constellation analytics stack, with the same methodological transparency it applies to its Tonga crop-estimation programme. Clients receive probability-flagged GIS layers and a written assessment of confidence, not a false-precision plastic mass figure.
Typical figures
| Best spatial resolution (multispectral) | 10 m (Sentinel-2 visible); 20 m (Sentinel-2 SWIR, where plastic indices are computed); 30 m (Landsat OLI) |
| Best spatial resolution (hyperspectral) | ~30 m (PRISMA); 3 m texture confirmation available from Planet SuperDove (no SWIR) |
| Revisit at equator | 5 days (Sentinel-2A+B combined); 8 days (Landsat 8+9 combined); ~29 days unscheduled or on-demand (PRISMA) |
| Key spectral bands for detection | NIR ~842 nm, SWIR1 ~1,610 nm, SWIR2 ~2,190 nm (Sentinel-2/Landsat); continuous 400–2,500 nm at ~12 nm sampling (PRISMA) |
| Critical unsampled feature | ~1,215 nm plastic absorption: not covered by Sentinel-2 or Landsat; directly sampled by PRISMA |
| Minimum detectable target | Dense aggregations of order hundreds of square metres with high fractional debris cover; dispersed microplastic is undetectable by any current orbital sensor |
| Cloud and sunglint sensitivity | Optical sensors fully blocked by cloud; sunglint in SWIR bands can mimic or mask debris signals; wind-roughened seas increase foam confusion |
| Archive depth | Sentinel-2: from 2015; Landsat: from 1972 (SWIR from Landsat 4, 1982); PRISMA: from 2019 |
| Latency (open data) | Sentinel-2 Level-2A typically available within 3–5 hours of acquisition via Copernicus Data Space |
| Discrimination confidence | Plastic vs. Sargassum: low to moderate with multispectral; moderate with hyperspectral under good conditions. In-situ validation required for regulatory use. |
Analytics Satellize can run
| Floating debris probability map | Floating Debris Index (FDI) and Plastic Index (PI) computed from Sentinel-2 B8, B11, B12; oceanographic masking using SST and chlorophyll composites | GIS layer (GeoTIFF or GeoJSON) with per-pixel probability score and confidence class, delivered per acquisition |
| Patch trajectory and drift analysis | Multi-date FDI time series correlated with Copernicus Marine Service surface current vectors to track accumulation zone movement | Animated time series and tabular drift report over a defined area of interest and date range |
| River-plume debris event report | Post-flood Sentinel-2 screening within 50 km of defined river mouths; flagged pixels cross-referenced with discharge anomaly data | Event-triggered PDF report with mapped extent, estimated patch area, and uncertainty statement |
| Retrospective accumulation baseline | Annual FDI compositing over Landsat 8/9 archive (2013 onwards) to identify persistent convergence hotspots | Multi-year hotspot map and summary statistics table, suitable for marine spatial planning input |
| Hyperspectral spectral match assessment | PRISMA scene processing with spectral angle mapper against laboratory plastic spectral libraries; sunglint and foam masking applied | Per-patch spectral match report with explicit confidence bounds and list of confounding materials that cannot be excluded |
| High-resolution patch confirmation | Planet SuperDove 3 m imagery tasked over FDI-flagged areas; texture and morphology assessment by trained analyst | Analyst annotation layer and written assessment distinguishing debris-consistent from Sargassum-consistent patch morphology |
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.