Sensors
- Sentinel-2 MSI: 10 m resolution in visible bands, 20 m in red-edge and SWIR. Five-day revisit at the equator under clear skies. Bands B3 (green) and B4 (red) feed NDTI and NDWI calculations; SWIR bands B11/B12 help distinguish turbid water from bare sand stockpiles on bank margins. Severely degraded by cloud during South and South-East Asian monsoon seasons.
- Sentinel-1 SAR C-band: 5.4 GHz, 10 m ground range resolution in IW mode, six-day revisit per satellite (combined A+B gives roughly three days). Cloud-penetrating. Coherence change detection between repeat passes flags surface disturbance on river banks and sandbars; backscatter contrast distinguishes metal pontoon hulls from surrounding water. Useful precisely when Sentinel-2 is blinded by cloud.
- PlanetScope SuperDove: 3 m resolution, daily revisit globally. Eight spectral bands including red-edge. Confirms pontoon positions and stockpile extents identified in coarser data; at 3 m a typical 20–40 m dredging pontoon is clearly resolved. Commercial tasking adds cost; cloud remains a problem at the same latitudes.
- Landsat 8/9 OLI: 30 m multispectral, 16-day revisit per satellite (eight days combined). Lower spatial resolution than Sentinel-2 but a 40-year archive from Landsat 4 onwards allows long-run channel morphology analysis, bankline retreat quantification and comparison of pre-dredging baseline conditions. Bands 3 and 4 support turbidity indices comparable to Sentinel-2 NDTI.
What a turbidity plume gives away
Sand dredging injects suspended sediment directly into the water column. The NDTI, computed from Sentinel-2 green and red reflectance as (B4 minus B3) divided by (B4 plus B3), is sensitive to this suspended load. A clean river reach might return NDTI values around minus 0.1 to minus 0.05; active dredging can push values above zero within the immediate plume, a contrast detectable at 10 m resolution even from a passing overpass that lasts milliseconds. The plume disperses downstream, so the highest NDTI values cluster within a few hundred metres of the extraction point and decay with distance in a pattern that distinguishes a point-source disturbance from diffuse agricultural runoff.
This is not foolproof. Seasonal flood pulses, upstream dam releases and natural bank erosion all elevate turbidity. The analytical discipline is to build a per-pixel seasonal baseline from two or three years of Sentinel-2 archive, then flag anomalies that exceed the expected envelope at a specific location on a specific date. A pontoon working the same reach for several weeks produces a persistent anomaly that seasonal variation cannot explain.
SAR coherence as a record of surface disturbance
Synthetic aperture radar coherence measures how similar the phase of a radar return is between two passes separated by days or weeks. Stable surfaces, dry compacted ground, concrete, undisturbed vegetation, maintain high coherence. A sand stockpile being built up and shifted, or a riverbank being excavated, loses coherence rapidly because the surface geometry changes between passes. Sentinel-1 six-day repeat pairs over an active dredging site will show coherence values dropping toward zero on affected bank areas, while adjacent undisturbed land holds coherence above 0.6.
The C-band wavelength (roughly 5.6 cm) also produces strong specular returns from calm open water and high backscatter from metal structures. A steel-hulled pontoon anchored mid-river shows up as a bright point target against the dark water background. Comparing backscatter images across multiple passes traces the pontoon's movement along a reach, providing something close to a positional log even through solid cloud cover. This matters enormously on the Mekong, Irrawaddy or Ganges during June to September, when Sentinel-2 optical revisit is effectively broken for weeks at a time.
Channel morphology and the longer record
Individual dredging events are detectable in near-real time. The structural damage they cause accumulates over years and is readable in the long Landsat archive. Bankline digitisation from Landsat imagery at 30 m resolution, repeated annually, quantifies lateral channel migration and sandbar loss. INTERPOL's 2021 report on sand crime in South-East Asia noted that some river reaches have lost several metres of bank width per year under sustained illegal extraction, figures consistent with what bankline-change analysis of Landsat time series can resolve.
Morphological change analysis is slower work than plume detection, but it builds the evidentiary record that enforcement agencies and prosecutors actually need. A map showing that a specific sandbar shrank by 40 % between 2019 and 2023, correlated with the presence of dredging pontoons in Sentinel-2 imagery during those years, is a different class of evidence from a single turbidity anomaly.
Where the methods break down
Honesty about limits is not optional here. Sentinel-2's 10 m resolution is adequate for locating a pontoon and its plume but cannot resolve individual workers or equipment type. Cloud cover during monsoon season, which coincides with peak river flow and often peak dredging activity on major South and South-East Asian rivers, can produce data gaps of two to four weeks in the optical record. SAR fills part of this gap but coherence analysis requires two clean passes separated by a known interval; heavy rain can also degrade SAR signal quality over water surfaces.
Small-scale dredging operations using hand pumps or very small motorised suction units may produce turbidity signals below the noise floor of NDTI at 10 m resolution, particularly on already-turbid rivers. The Ganges and lower Mekong carry naturally high suspended loads; distinguishing a small illegal operation from background turbidity on these rivers requires careful baseline construction and is genuinely difficult. Larger commercial-scale illegal operations, the kind documented in UNEP's 2022 Sand and Sustainability report, are far more detectable.
SAR pontoon detection depends on the vessel being metallic and of sufficient size to produce a backscatter return distinguishable from wave clutter. Wooden or fibreglass hulls on narrow tributary rivers may not meet this threshold.
Building an operational monitoring programme
An effective programme combines three temporal layers. Near-real-time optical alerts, Sentinel-2 processed within 24 to 48 hours of acquisition, flag new turbidity anomalies against the seasonal baseline and trigger field or commercial-tasking follow-up. Weekly SAR coherence change maps cover the cloud-blind periods and track pontoon movements. Monthly or quarterly morphological summaries from Landsat or Sentinel-2 composites document cumulative channel change for the enforcement and legal record.
Prioritisation matters. River systems in South and South-East Asia where illegal sand extraction is documented, the Mekong tributaries in Cambodia and Myanmar, stretches of the Ganges and Brahmaputra in India and Bangladesh, the Solo River in Indonesia, span thousands of kilometres. Computational resources and analyst time are finite. A tiered alert system that escalates from automated anomaly detection to human review to commercial tasking is more realistic than attempting full-coverage manual analysis. Satellize applies this kind of layered approach in its analytics work, running open-constellation pipelines as the primary detection layer and adding commercial tasking where confirmation is needed. The workflow is similar in structure to the crop-estimation programme the company runs for the Kingdom of Tonga, adapted for a very different enforcement context.
Coordination with national hydrological agencies is worth pursuing. Many countries in the affected regions operate river gauge networks whose turbidity or discharge data can serve as independent validation, helping to separate dredging signals from flood events in the satellite record.
Typical figures
| Optical spatial resolution (Sentinel-2) | 10 m (visible/NIR), 20 m (red-edge/SWIR) |
| SAR spatial resolution (Sentinel-1 IW) | 10 m ground range |
| Optical revisit (Sentinel-2, equatorial) | 5 days under clear skies; effectively degraded to weeks during monsoon cloud cover |
| SAR revisit (Sentinel-1 A+B combined) | Approximately 3 days; cloud-independent |
| Minimum detectable pontoon size (SAR) | Metal-hulled vessels roughly 20 m or longer on open water; smaller vessels on narrow channels unreliable |
| Turbidity index sensitivity floor | NDTI anomaly detection unreliable on rivers with naturally high suspended load (>200 mg/L background TSS) without careful per-pixel baseline construction |
| Archive depth | Sentinel-2 from 2015; Landsat from 1972 (Landsat 1) with consistent OLI data from 2013 (Landsat 8) |
| Alert latency (Sentinel-2 optical) | 24–48 hours from satellite acquisition to processed anomaly flag |
| Coherence change pair interval (Sentinel-1) | 6 days (single satellite), 12-day pairs also used for stability |
| Spectral bands used | Sentinel-2 B3 (green, 560 nm), B4 (red, 665 nm), B8 (NIR, 842 nm), B11/B12 (SWIR); Sentinel-1 C-band 5.405 GHz VV/VH polarisation |
Analytics Satellize can run
| NDTI turbidity anomaly alerts | Per-pixel seasonal baseline from Sentinel-2 archive; z-score anomaly detection on NDTI time series | Georeferenced alert points with anomaly magnitude and date, delivered as GeoJSON or shapefile within 48 hours of acquisition |
| SAR pontoon position log | Backscatter thresholding and point-target detection on Sentinel-1 IW imagery over designated river reaches | Weekly CSV of detected bright targets with coordinates, acquisition time and polarisation ratio, suitable for import into enforcement case management systems |
| Coherence change map (bank disturbance) | Sentinel-1 six-day interferometric coherence differencing; low-coherence patches on bank margins flagged against stable-land reference | GeoTIFF change layer per 12-day cycle, with classified disturbance polygons and area statistics |
| Bankline retreat time series | Automated water-body extraction (MNDWI) from Landsat 8/9 and Sentinel-2 annual composites; bankline vectorisation and change quantification | Annual bankline shapefiles with lateral change metrics (metres per year) per defined river segment, suitable for legal evidentiary reporting |
| Sandbar volumetric loss estimate | Multi-date sandbar area mapping from Sentinel-2 SWIR composites at low-flow season; area change used as proxy for volume with published grain-size assumptions | Tabular report of sandbar area and estimated volume change per monitoring period, with methodology notes and uncertainty range |
| Cloud-gap-filled river-reach composite | Fusion of Sentinel-2 optical and Sentinel-1 SAR-derived water extent to maintain continuous coverage through monsoon cloud gaps | Monthly gap-filled water-extent raster and anomaly summary report for each designated monitoring reach |
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.