Post-flood sediment and debris deposition mapping
Spectral turbidity indices and DEM differencing reveal where floods deposited sediment and debris, quantifying volume changes that field surveys rarely capture at scale. Cloud cover and vertical accuracy set hard limits on what satellites can honestly deliver.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with both satellites. Red and green bands support NDTI and turbidity proxy calculation; SWIR bands (20 m) help distinguish wet sediment from dry soil after recession.
- Landsat 8/9 OLI: 30 m multispectral resolution, 8-day revisit per satellite (16-day per sensor alone). Band 4 (red) and Band 3 (green) are the standard inputs for the Normalised Difference Turbidity Index. Archive extends to 1972 across the broader Landsat programme, enabling multi-decadal baseline comparison.
- Planet SuperDove: 3 m resolution, near-daily revisit over most land areas. Eight spectral bands including coastal blue and red-edge. Useful for mapping fine-scale debris boundaries and field-level sediment extent, though the archive is shallower than Landsat and commercial tasking is required.
- TanDEM-X: Bistatic X-band SAR pair producing digital elevation models at 0.4 arcsecond (roughly 12 m) global resolution with a published relative vertical accuracy of better than 2 m and absolute accuracy better than 10 m. DEM differencing against a pre-event reference detects deposition volumes, though thin sheets below roughly 0.5 m remain ambiguous against noise.
- Copernicus DEM (GLO-30 / GLO-90): Derived from TanDEM-X acquisitions and freely available at 30 m and 90 m postings. Serves as the pre-event reference surface for differencing workflows where no contemporaneous TanDEM-X tasking is available. Vertical accuracy varies by terrain and land cover.
What the water leaves behind, and why it matters
A flood is a transport event. The receding water deposits suspended load, bedload and debris across floodplains, agricultural fields and channel beds in patterns that bear almost no resemblance to the inundation extent itself. A field can be drowned for two days and buried under 15 cm of silt for two seasons. That distinction, between where water went and where sediment stayed, is the analytical gap this use case addresses.
The practical consequences are significant for agricultural recovery planning, channel maintenance and infrastructure risk. Deposited fine sediment can seal soil pores, suppressing germination for months. Coarser debris can block irrigation intakes and culverts. Remobilised sediment in the next rainfall event can reactivate landslide scarps or clog reservoirs. Mapping deposition extent and, where vertical accuracy permits, estimating volume gives planners something field surveys rarely provide at catchment scale: a spatially continuous picture within days of recession.
Spectral indices: what turbidity and sediment look like from orbit
Suspended sediment in water increases reflectance in the red and green portions of the visible spectrum while suppressing near-infrared return. The Normalised Difference Turbidity Index (NDTI), defined as (Red minus Green) divided by (Red plus Green), exploits this contrast. High NDTI values indicate turbid, sediment-laden water; values approaching zero or below indicate clearer water or vegetation. On Landsat OLI, Band 4 (red, 0.64-0.67 µm) and Band 3 (green, 0.53-0.59 µm) are the standard inputs. Sentinel-2 Band 4 (red) and Band 3 (green) map directly onto the same formulation.
After waters fully recede, deposited sediment on land surfaces shows as anomalously bright, low-vegetation-index pixels in the red and SWIR bands relative to the pre-event baseline. The Normalised Difference Vegetation Index drops sharply over buried fields not because crops died but because sediment reflectance dominates. Differencing a post-event NDVI image from a pre-event composite isolates these areas with reasonable specificity, though bare-soil agricultural fields in dry seasons can produce false positives. Analyst review against the flood timing and a land-cover reference layer is necessary.
Planet SuperDove's 3 m resolution resolves individual field boundaries and small debris fans that 10 m Sentinel-2 pixels average out. The trade-off is cost and the shallow archive, which limits the quality of pre-event baselines in areas without systematic tasking history.
DEM differencing: estimating what was deposited, not just where
Spectral methods tell you the extent of sediment. DEM differencing tells you the volume. The principle is straightforward: subtract a pre-event elevation model from a post-event one and the residual represents net vertical change, positive for deposition, negative for erosion or scour.
In practice, this is harder than it sounds. The Copernicus DEM GLO-30, derived from TanDEM-X acquisitions taken between 2011 and 2015, has a published relative vertical accuracy of better than 2 m over most terrain. A new TanDEM-X acquisition over the same area after a flood event can be differenced against this baseline, but the noise floor means that sediment sheets thinner than roughly 0.5 to 1 m are statistically indistinguishable from instrument noise and geometric co-registration error. Large avulsion deposits, point bars and thick debris fans, typically exceeding 1 to 2 m, are detectable with reasonable confidence. Thin agricultural silt layers are not.
Co-registration between pre- and post-event DEMs must be done carefully. A horizontal offset of even one pixel at 12 m resolution introduces apparent vertical change on any slope. Standard co-registration workflows using stable off-floodplain reference surfaces are essential before interpreting difference maps.
The cloud problem is real and does not resolve itself quickly
Major flood events in tropical and monsoon-affected regions occur precisely when cloud cover is at its annual maximum. The same atmospheric conditions that produce the flood obscure the optical sensors needed to map its aftermath. Sentinel-2 and Landsat cannot see through cloud. Planet's higher revisit rate improves the odds of catching a clear window, but there is no guarantee.
The practical consequence is that the first usable optical image after a major tropical flood may arrive days to weeks after recession. By then, some surface sediment will have dried, cracked and begun to resemble background soil spectrally. The sediment signal degrades with time. This is an honest constraint, not a solvable one with current open-constellation assets. SAR sensors, covered in the sibling page on flood extent mapping, penetrate cloud but do not directly measure turbidity or sediment reflectance. For deposition mapping specifically, optical data is the primary method, and cloud is its principal enemy.
In practice, analysts should queue automated cloud-fraction monitoring on the area of interest immediately after a flood event is confirmed, ingest the first sub-20% cloud-cover scene, and flag any gap longer than five days as a data-quality caveat in the output.
Combining methods into a deposition product
No single sensor or index produces a complete picture. A credible post-flood sediment deposition product typically layers three inputs: an NDTI-derived turbidity map from the last available image during recession, an NDVI-difference map comparing post-event to a seasonal pre-event composite, and, where available, a DEM difference layer from TanDEM-X. Each layer carries its own uncertainty, and the final product should express confidence classes rather than a single deterministic boundary.
Satellize runs this multi-layer workflow on open Sentinel-2 and Landsat archives, with optional Planet tasking added on client licence. The Tonga crop-estimation programme demonstrated that combining spectral change detection with field-boundary reference layers substantially reduces false positives over bare-soil agricultural areas, a lesson directly applicable to post-flood sediment work. Output is delivered as classified GIS polygons with associated area statistics, plus a DEM-difference raster where TanDEM-X data is available, all accompanied by a methods note that states the cloud-cover fraction of the source imagery and the co-registration residual of any elevation differencing.
What a buyer should ask before commissioning this analysis
Four questions determine whether satellite-derived deposition mapping will be useful for a specific event. First, how many cloud-free scenes exist over the area within two weeks of recession? If the answer is zero, optical methods cannot contribute until conditions clear, and the analysis will reflect a later state of the deposit. Second, what is the minimum deposition thickness of interest? If thin agricultural silt layers below 50 cm matter, DEM differencing alone cannot resolve them and the analysis must rely entirely on spectral change, which has its own ambiguity. Third, is a pre-event DEM or image baseline available at adequate resolution? Without it, change detection has no reference. Fourth, what spatial resolution is needed for the intended use? Field-level agricultural assessment at 3 m is a different commission from catchment-scale channel-change mapping at 30 m, and the cost and latency differ accordingly.
Honest answers to these four questions, before a contract is signed, produce better analysis and fewer disappointed clients.
Typical figures
| Optical spatial resolution | 3 m (Planet SuperDove), 10 m (Sentinel-2 visible/NIR), 20 m (Sentinel-2 SWIR), 30 m (Landsat 8/9 OLI) |
| DEM spatial resolution | 12 m (TanDEM-X 0.4 arcsecond product), 30 m (Copernicus DEM GLO-30) |
| DEM vertical accuracy | Relative: better than 2 m (TanDEM-X); absolute: better than 10 m. Minimum detectable deposition via differencing: approximately 0.5–1 m under good co-registration conditions |
| Optical revisit | ~1 day (Planet SuperDove, commercial tasking); 5 days (Sentinel-2 dual satellite, equator); 8 days (Landsat 8 or 9 individually) |
| Key spectral bands | Green (~0.56 µm), Red (~0.66 µm) for NDTI; NIR (~0.84 µm) for NDVI differencing; SWIR (~1.6 µm, ~2.2 µm) for wet/dry sediment discrimination |
| Cloud penetration | None. Optical methods fail under cloud; SAR-based deposition mapping is not yet operationally mature for this specific product |
| Archive depth | Landsat: 1972–present; Sentinel-2: 2015–present; Planet SuperDove: 2021–present (variable by location); Copernicus DEM: 2011–2015 acquisition epoch |
| Typical analysis latency | 24–72 hours from first cloud-free scene ingestion; scene availability after event is the binding constraint, not processing time |
| Output formats | GeoTIFF rasters, classified GIS polygons (GeoPackage or Shapefile), area and volume statistics (CSV), methods note with data-quality flags |
Analytics Satellize can run
| NDTI turbidity extent map | Normalised Difference Turbidity Index computed on Sentinel-2 or Landsat red/green bands; thresholded against pre-event clear-water baseline | GeoTIFF + classified polygon layer showing turbidity classes during recession phase, with cloud-fraction metadata |
| Post-recession sediment deposition extent | NDVI differencing (post-event minus pre-event seasonal composite) combined with SWIR-based bare-soil discrimination to isolate sediment-covered agricultural and floodplain areas | Classified GIS polygon layer with area statistics per land-cover class; confidence flag (high/medium/low) per polygon |
| DEM-difference deposition volume estimate | Co-registered subtraction of post-event TanDEM-X acquisition from Copernicus DEM GLO-30 baseline; stable off-floodplain reference surfaces used for co-registration correction | DEM-difference raster (GeoTIFF) with positive/negative change classes, volume estimate table, and co-registration residual stated in methods note |
| Field-level sediment impact classification | Intersection of sediment deposition extent layer with cadastral or field-boundary reference data; Planet SuperDove at 3 m used where field boundaries are smaller than Sentinel-2 pixel footprint | Per-field impact table (area affected, estimated sediment class) exportable to agricultural damage assessment workflows |
| Time-series sediment persistence tracking | Sequential NDVI and NDTI composites at each available clear-sky acquisition over a 30–90 day post-event window, tracking spectral recovery toward pre-event baseline | Multi-date GIS stack with recovery timeline chart; flags fields showing no spectral recovery by a user-defined threshold date |
| Channel morphology change summary | DEM differencing and optical change detection combined to identify avulsion deposits, new bar formations and scoured reaches along the main channel and major tributaries | Annotated map report (PDF + GIS layers) summarising channel geometry changes, with honest uncertainty bounds on volume estimates |
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.