Sensors
- Airbus Pléiades Neo (tri-stereo): 30 cm panchromatic resolution; tri-stereo acquisition in a single pass produces dense point clouds with vertical accuracy in the 20–50 cm range under good contrast conditions, enabling volume differencing of pits larger than a few hundred square metres
- PlanetScope: 3 m multispectral, near-daily global revisit; insufficient for volumetrics but excellent for planform change detection in river channels, tracking bar migration, pit expansion and sediment plume extent between epochs
- Sentinel-2 MSI: 10 m in visible and NIR bands, 5-day revisit at mid-latitudes; free archive from 2015 onward supports long-run time series of channel width, exposed bar area and turbidity proxies; cloud cover is the dominant gap-maker in tropical catchments
- TanDEM-X: X-band SAR interferometric DEM; 12 m global TanDEM-X DEM published at 0.4 m relative vertical accuracy (90th percentile); bistatic repeat acquisitions can be commissioned for change detection where cloud permanently defeats optical stereo
Why volume, not just area
Regulators and prosecutors need quantity, not just presence. A photograph of a digger in a riverbed proves activity; a volume deficit proves theft at scale. The standard method is DEM differencing: subtract a post-extraction elevation model from a pre-extraction baseline, integrate the negative cells, and multiply by bulk density to convert cubic metres into tonnes. That number maps directly onto royalty schedules and criminal thresholds in most jurisdictions.
The honest complication is that no single sensor delivers this cleanly. Optical stereo requires texture and contrast for photogrammetric matching. Bare sand and water surfaces are both problematic: sand is textureless at low sun angles, and any standing water in a pit returns no usable stereo signal. SAR-derived DEMs sidestep cloud but introduce their own noise from layover and foreshortening on steep pit walls. A competent programme combines both, using SAR to bridge cloudy seasons and optical stereo for the high-precision epochs that anchor legal evidence.
What a floating roof gives away
Pléiades Neo tri-stereo is currently the most practical commercial option for pit volumetrics at legally useful precision. Three images acquired in a single pass from fore, nadir and aft angles produce a dense surface model without the temporal baseline that would let the ground move between acquisitions. Published validation of Pléiades stereo DEMs over structured terrain consistently reports RMSE values in the 20–50 cm range vertically, depending on ground control and surface type. Over a 5,000 m² pit excavated to 3 m depth, that translates to a volume uncertainty of roughly 1,000–2,500 m³, or a few per cent of the extracted mass. That is tight enough to support enforcement action.
Tri-stereo tasking is not cheap and cannot be justified across an entire river basin on a routine basis. The practical workflow is to use Sentinel-2 or PlanetScope time series to identify where change is occurring, then direct Pléiades Neo tasking to confirmed hotspots. This two-tier approach keeps commercial tasking costs proportionate to the intelligence value.
Reading the river's planform
Riverbeds tell a different story from borrow pits. Sand extraction from active channels removes the sediment load that maintains bar position, channel width and bed elevation. The geomorphological signal accumulates over months to years and is legible in medium-resolution imagery even when individual extraction events are not. Sentinel-2 at 10 m resolution can track changes in the width of exposed gravel bars, the migration of mid-channel islands and the lateral shift of the thalweg. PlanetScope's near-daily cadence adds the ability to catch low-water windows when bars are exposed and extraction is most active.
Normalised Difference Water Index (NDWI) computed from green and NIR bands is the standard tool for mapping the wet/dry boundary in a channel. Turbidity proxies derived from red-band reflectance can flag active dredging even when the extraction machinery itself is below the resolution floor. Neither method is a substitute for a DEM, but together they provide a continuous surveillance layer that costs nothing in data licence fees for Sentinel-2 and relatively little for PlanetScope archive pulls.
The cloud problem in tropical catchments is real and should not be understated. In humid equatorial basins, Sentinel-2 may return fewer than four cloud-free acquisitions per month during the wet season, precisely when high river stages make extraction easiest and most damaging. SAR backscatter from Sentinel-1 (C-band, 10 m IW mode, 6-day revisit) partially compensates: it penetrates cloud and can detect the geometric signature of excavation pits on exposed bars, though the interpretation requires more analyst skill than optical imagery.
Anchoring the baseline: the archive problem
Volume differencing is only as good as the baseline DEM. The global TanDEM-X DEM, released at 12 m posting with 0.4 m relative vertical accuracy (90th percentile, as published by DLR), is the most consistent free baseline available for most of the world. It was acquired primarily between 2011 and 2015. For sites where extraction predates that window, the baseline is already contaminated. In those cases, analysts must either accept a conservative lower-bound estimate or commission a new bistatic TanDEM-X acquisition as the current reference, then difference against future optical stereo epochs.
SRTM (2000 vintage, 30 m posting, roughly 5–10 m vertical noise in vegetated terrain) is too coarse for pit-scale volumetrics but remains useful for catchment-scale sediment budget analysis and for identifying which river reaches have lost the most bed material over the long term. Combining SRTM, TanDEM-X and Pléiades Neo epochs gives a three-point time series spanning up to 25 years for sites with sufficient archive coverage.
Cadence, cloud and what the method cannot see
Every method described here has a resolution floor that matters in practice. Pléiades Neo at 30 cm can resolve a small excavator but cannot see beneath water. Sentinel-2 at 10 m will miss a pit smaller than roughly 200–300 m² and cannot distinguish a natural scour hole from an extraction pit without ancillary information. PlanetScope at 3 m sits in between, useful for planform mapping but not for volumetrics. None of these sensors sees through the turbid water that fills an active pit during extraction.
The temporal resolution constraint is equally important. Even with PlanetScope's near-daily revisit, a single overnight extraction event that removes 500 m³ of sand may leave no detectable surface signature by the time the next clear-sky image is acquired if the pit refills with water or sediment. Enforcement based on satellite evidence therefore works best as a cumulative case built over multiple epochs, not as a real-time alert system.
Satellize structures this kind of programme as a layered monitoring stack: Sentinel-2 and PlanetScope for continuous screening, Pléiades Neo tasking triggered by confirmed change signals, and a quarterly volume-deficit report delivered as a GIS package with supporting evidence suitable for regulatory use. The Tonga crop-estimation programme demonstrated that this kind of cadenced, evidence-grade delivery is achievable with open and commercial data in combination.
Typical figures
| Optical stereo spatial resolution (Pléiades Neo) | 30 cm panchromatic; 3D point cloud at 50 cm posting typical |
| Vertical accuracy, optical stereo DEM | 20–50 cm RMSE with ground control (published validation range) |
| SAR DEM baseline (TanDEM-X global) | 12 m posting; 0.4 m relative vertical accuracy (90th percentile, DLR published) |
| Planform change detection resolution | 10 m (Sentinel-2), 3 m (PlanetScope) |
| Revisit cadence | Near-daily (PlanetScope); 5-day (Sentinel-2 at mid-latitudes); on-demand tasking (Pléiades Neo, typically 1–3 day response) |
| Minimum detectable pit area (optical) | ~200–300 m² at 10 m resolution; ~30–50 m² at 3 m resolution |
| Cloud cover impact | Severe in humid tropics; wet-season optical gaps of weeks to months common; SAR partially compensates |
| Archive depth | Sentinel-2 from 2015; PlanetScope from ~2016; TanDEM-X baseline 2011–2015; SRTM from 2000 |
| Spectral bands used | Visible, NIR (NDWI, turbidity); panchromatic stereo; X-band SAR (TanDEM-X); C-band SAR (Sentinel-1) |
| Delivery formats | GeoTIFF DEM difference raster, vector pit polygons, volume-deficit CSV, PDF evidence report |
Analytics Satellize can run
| Borrow pit volume deficit map | DEM differencing (optical tri-stereo or SAR) against baseline epoch; negative-cell integration | GeoTIFF raster of elevation change plus per-pit volume table in CSV, updated per tasking epoch |
| River channel planform change layer | NDWI time series on Sentinel-2 and PlanetScope; wet/dry boundary vectorisation per epoch | Vector polygon set showing bar area, channel width and thalweg position per epoch; change magnitude attribute |
| Extraction hotspot alert | Automated change detection on PlanetScope daily mosaic; threshold on exposed-sediment area increase | GeoJSON alert feed with coordinates, date, estimated area change and confidence score |
| Turbidity proxy time series | Red-band and green-band reflectance ratio on Sentinel-2; calibrated against published empirical turbidity relationships | Monthly raster stack and reach-averaged time series chart indicating likely active dredging periods |
| Cumulative sediment deficit estimate | Multi-epoch DEM stack (SRTM, TanDEM-X, Pléiades Neo) integrated over monitored reach length | Quarterly PDF report with volume-by-reach table, uncertainty bounds and annotated imagery panels |
| Enforcement evidence package | Collation of change-detection outputs, tasked imagery and volume calculations into a structured evidentiary record | Dated PDF dossier with georeferenced imagery, DEM cross-sections, volume estimates and metadata chain of custody |
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.