Riverbank erosion monitoring for riparian property boundary change
Lateral channel migration can erase metres of riparian land in a single flood season, yet property boundaries defined by the watercourse create legal ambiguity the moment the bank shifts. Multitemporal satellite imagery quantifies that shift with sub-metre precision.
Sensors
- Planet SuperDove: 3 m ground sampling distance, 8 spectral bands including red-edge and NIR, sub-weekly revisit at most latitudes. The short revisit is what matters here: it can bracket a flood peak and capture the post-event bank position before vegetation regrowth obscures the scar.
- Sentinel-2 MSI: 10 m in visible and NIR bands, 5-day revisit with both satellites. Archive runs from 2015, giving roughly a decade of consistent, freely available imagery for baseline channel mapping and long-term migration rate estimation.
- Landsat archive (5, 7, 8, 9): 30 m multispectral, 16-day revisit per satellite. The archive extends to 1972 for Landsat 1, making it the only freely available source for multi-decadal channel planform analysis. Useful for establishing historical migration envelopes that inform legal disputes about when erosion began.
- Airbus Pléiades Neo: 30 cm panchromatic, 50 cm multispectral. Tasked on demand for high-stakes parcels where precise bank-edge delineation is needed for legal proceedings or insurance claims. Does not replace the archive sensors; it validates them at the critical moment.
What the water column gives away
Rivers announce their intentions spectrally before they announce them physically. Suspended sediment load raises water reflectance in the red and short-wave infrared bands. A channel widening into a cutbank shows as a turbid plume on the outside of a meander bend; the point bar on the inside brightens as fresh sand accretes. Neither of these signals requires expensive tasking. Sentinel-2's Band 4 (665 nm) and Band 11 (1610 nm) combination has been used extensively in published literature to map water extent and distinguish clear from turbid water at 10 m resolution.
The practical limit is cloud cover. During the flood peaks that drive the most dramatic erosion, cloud is almost guaranteed in temperate and tropical climates. This is not a solvable problem with optical sensors alone. The honest answer is that Planet's sub-weekly cadence increases the probability of at least one usable acquisition around a flood event, but it does not guarantee one. Synthetic aperture radar, which penetrates cloud, is the complement for confirmed flood-peak mapping, though it sits outside the sensor set for this particular use case.
Measuring lateral migration across decades
Channel planform change is measured by comparing the centreline or bank-edge position across two or more dates. The standard method, well-documented in geomorphological literature, digitises the active channel boundary from each image epoch and calculates the perpendicular distance between successive positions at regular transects along the channel. Rates are expressed in metres per year.
Landsat's 30 m resolution sets a detection floor: bank shifts smaller than roughly 15 to 30 m between epochs are below reliable detection, depending on image quality and bank contrast. Sentinel-2 pushes that floor down to approximately 10 to 20 m. Planet SuperDove, at 3 m, can resolve shifts of 5 to 10 m between consecutive acquisitions, which is operationally significant for smaller rivers where a 10 m loss represents a substantial fraction of a narrow riparian parcel.
Archive depth matters enormously for legal contexts. A landowner disputing a boundary needs to know not just the current bank position but the trajectory: how fast has it moved, was it episodic or continuous, and when did the parcel first fall below a legally relevant threshold? Landsat's record back to the 1970s and 1980s is often the only satellite evidence available for that question.
The cadastral problem nobody budgeted for
In many jurisdictions, riparian boundaries are defined by the ordinary high-water mark or the centreline of the watercourse. This is legally tidy until the watercourse moves. Common law doctrines of accretion and avulsion draw a distinction between gradual deposition (which shifts the boundary with the bank) and sudden avulsion (which does not). The satellite record can, in principle, help establish which regime applies: a channel that migrated 2 m per year for fifteen years looks different from one that jumped 30 m in a single flood event.
That said, satellite evidence is not automatically admissible or conclusive. Pixel-level uncertainty, image registration errors and the 30 m resolution floor of the Landsat archive all introduce ambiguity that a competent opposing expert will exploit. The honest position is that satellite analysis narrows the range of defensible interpretations; it rarely eliminates dispute on its own. Pairing the imagery record with gauging station flood data and LiDAR surveys, where available, produces a more defensible package.
The practical buyer for this analysis is typically a property insurer, a riparian landowner in litigation, a water authority managing flood-risk liability, or a government cadastral agency updating its national parcel database after a major flood season.
Rapid erosion events: the sub-weekly window
Most channel migration is slow. But meander cutoffs, bank collapses into undercut clay layers, and ice-jam floods can shift a bank by tens of metres in 48 hours. For these events, the 5-day Sentinel-2 revisit is often too coarse to capture the pre- and post-event bank positions cleanly. Planet's SuperDove constellation, with its sub-weekly global revisit, is the practical instrument of choice for event-scale monitoring.
The workflow is straightforward in concept: establish a baseline bank-edge vector from a cloud-free pre-event acquisition, then compare it to the first cloud-free post-event image. The difference polygon, clipped to the cadastral parcel boundary, gives the lost area in square metres. At 3 m resolution, this is credible for parcels wider than roughly 20 to 30 m. For very narrow riparian strips, Pléiades Neo tasking at 30 to 50 cm is the only option that produces legally defensible precision.
What a monitoring programme actually looks like
A practical riverbank monitoring programme starts with a reach-scale baseline: channel centreline and bank-edge vectors derived from the best available Sentinel-2 or Planet imagery, cross-referenced against the cadastral parcel layer. Transects are placed at regular intervals, typically 50 to 200 m depending on channel sinuosity, and the perpendicular offset is measured at each epoch.
Alert thresholds are set by parcel, not by reach. A parcel that is 500 m deep can tolerate 20 m of bank retreat without immediate legal consequence. One that is 40 m deep cannot. The monitoring system flags parcels where the bank has retreated beyond a user-defined fraction of parcel depth, triggering a Pléiades Neo tasking request for verification imagery.
Satellize structures this kind of programme as a recurring GIS layer delivered at agreed intervals, with event-triggered alerts for rapid erosion. The Tonga crop-estimation programme demonstrated that open-constellation analytics can be operationalised for government clients at national scale; the same pipeline architecture applies here, with the cadastral parcel database substituting for the crop-field polygons. Outputs are delivered as georeferenced shapefiles or GeoPackages compatible with standard cadastral GIS environments.
Honest limits of the method
Cloud is the dominant operational constraint and there is no optical solution to it. Vegetated banks, where riparian trees overhang the water, obscure the true bank edge from nadir-looking sensors; the measured edge is the canopy edge, not the soil edge, and the two can differ by several metres. Bank geometry matters too: a vertical cutbank is easier to delineate than a gently sloping sandy beach where the water margin shifts with stage.
Image co-registration error between acquisition dates introduces a positional uncertainty that is typically 0.5 to 1.5 pixels for well-processed imagery. At Sentinel-2's 10 m resolution, that is 5 to 15 m of potential false migration signal. This is why change detection below roughly 10 to 20 m should be treated with caution for that sensor, and why Planet or Pléiades Neo are necessary for high-stakes single-parcel disputes.
Finally, satellite analysis measures planimetric change: the horizontal position of the bank. It does not directly measure volumetric loss or the depth of undercutting below the waterline. For structural stability assessments or engineering purposes, field survey remains necessary.
Typical figures
| Spatial resolution (routine monitoring) | 3 m (Planet SuperDove), 10 m (Sentinel-2 MSI visible/NIR) |
| Spatial resolution (legal verification) | 30–50 cm (Airbus Pléiades Neo, tasked on demand) |
| Revisit cadence | Sub-weekly (Planet SuperDove); 5 days (Sentinel-2 two-satellite pair); 16 days (Landsat 8/9 per satellite) |
| Archive depth | From 1972 (Landsat 1); from 2015 (Sentinel-2); from 2009 (Planet, partial) |
| Minimum detectable bank shift | ~5–10 m (Planet SuperDove); ~10–20 m (Sentinel-2); ~15–30 m (Landsat 30 m) |
| Spectral bands used | Red (665 nm), NIR (842 nm), SWIR (1610 nm) for water/sediment discrimination; red-edge for vegetation mask |
| Cloud cover constraint | Optical sensors only; cloud-free acquisitions required; sub-weekly revisit improves event capture probability but does not guarantee it |
| Positional co-registration uncertainty | 0.5–1.5 pixels typical for well-processed imagery; ~5–15 m for Sentinel-2 |
| Delivery format | Georeferenced shapefiles or GeoPackages; change-polygon layers clipped to cadastral parcel boundaries |
| Latency (event alert) | 24–72 hours after cloud-free acquisition, depending on tasking and processing pipeline |
Analytics Satellize can run
| Historical channel migration map | Multitemporal bank-edge digitisation and transect-perpendicular offset measurement across Landsat and Sentinel-2 archive epochs | GIS layer showing bank-edge positions by year, migration rate in m/yr per transect, and cumulative displacement polygon |
| Parcel area loss quantification | Intersection of bank-edge change polygons with cadastral parcel boundaries; area calculation in square metres per epoch | Tabular report per parcel: area lost, percentage of original parcel, date range of loss |
| Rapid erosion event alert | Automated pre/post comparison of Planet SuperDove bank-edge vectors triggered by river gauge threshold or flood forecast | Alert notification with estimated bank retreat distance and affected parcel IDs; Pléiades Neo tasking recommendation for parcels above loss threshold |
| Accretion vs. avulsion classification | Migration rate time-series analysis: gradual (<2 m/yr) flagged as potential accretion; step-change events (>10 m in single epoch) flagged as potential avulsion for legal review | Annotated timeline chart per reach segment, suitable for inclusion in legal or insurance documentation |
| Turbidity and sediment plume mapping | Band ratio analysis (Red/NIR or SWIR) on Sentinel-2 and Planet imagery to map suspended sediment concentration and active erosion zones | Raster layer of relative turbidity index per acquisition date; time-series animation of flood-season sediment dynamics |
| High-risk parcel prioritisation | Scoring of riparian parcels by migration rate, parcel depth, and proximity to active meander bends; ranked watchlist updated each monitoring cycle | Ranked parcel watchlist GIS layer with risk score, recommended monitoring frequency, and next-action flag |
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.