Aquaculture farm mapping and licensing compliance monitoring
SAR and multispectral satellites can delineate marine cage arrays, longline buoy patterns and coastal pond boundaries at operational scale, making it possible to verify licensed footprints and flag unlicensed expansion without sending an inspector offshore.
Sensors
- Sentinel-1 SAR (C-band, ESA): 10 m ground range detected resolution in Interferometric Wide Swath mode, 6-day repeat at mid-latitudes with both satellites, cloud-independent. Floating cage arrays and longline buoy fields appear as bright point or linear scatterers against a darker sea background. Minimum detectable structure size is roughly 20-30 m in practice, though individual buoys smaller than one resolution cell can still produce detectable returns if clustered.
- Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands, 5-day revisit with both satellites, free archive from 2015. Useful for pond aquaculture delineation in intertidal zones where water colour and berm geometry distinguish ponds from natural mudflat. Cloud cover is a hard constraint; in persistently overcast tropical coasts, usable scenes may arrive only every few weeks.
- Planet SuperDove multispectral: 3 m resolution, 8 spectral bands (coastal blue through near-infrared), near-daily revisit globally. The finer pixel size resolves individual cage collar geometry and separates tightly packed pond berms that merge at Sentinel-2 resolution. Archive depth is shallower than Sentinel-2 and tasking is commercial, but the daily cadence is well suited to change-detection workflows.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, 8 VNIR bands plus SWIR. At this resolution individual cage nets, mooring lines and access walkways are visible, enabling structural audits rather than just footprint checks. Revisit is irregular and tasking is expensive; practical use is for targeted verification of sites already flagged by coarser sensors.
What a cage array looks like to a radar
Sentinel-1 operates at C-band (5.4 GHz, roughly 5.6 cm wavelength). Metal cage collars, mooring buoys and net support frames are efficient radar reflectors. In calm-to-moderate sea states they produce backscatter returns well above the surrounding water surface, which appears dark in SAR imagery because specular reflection deflects the radar pulse away from the sensor. The contrast is reliable enough that automated detection is feasible.
The practical floor matters. A single circular cage collar of 20 m diameter sits comfortably within one or two 10 m pixels and produces a detectable bright return. Cages smaller than roughly 15 m across become ambiguous at Sentinel-1 resolution because the return competes with wind-roughened sea clutter and occasional ship wakes. Clustered cage arrays are easier: the spatial pattern itself is a discriminator. A 3x4 grid of cages separated by 30 m gaps has a recognisable signature even when individual cages are marginal. Isolated small structures near navigation channels are harder, and that ambiguity is worth stating plainly.
Separating aquaculture from everything else that floats
The classification problem is real. Fishing gear, navigation buoys, fish aggregating devices and debris fields all produce SAR returns that can resemble aquaculture infrastructure. Three features help separate them: geometric regularity, persistence through time, and spatial context.
Licensed cage farms are arranged in grids or arcs dictated by mooring geometry. They do not drift. Comparing SAR scenes across weeks or months reveals whether a bright cluster stays fixed relative to the seabed (aquaculture) or shifts with current and wind (drifting gear or vessels). Optical imagery, when cloud permits, adds colour and shape confirmation. A circular cage collar in a Sentinel-2 or SuperDove image is unambiguous in a way that a SAR blob alone is not. The workflow that works in practice pairs SAR for all-weather persistence detection with optical cross-validation on clear days, rather than relying on either alone.
Pond aquaculture in intertidal zones: a different problem
Shrimp and fish ponds cut into mangrove and mudflat are mapped primarily with multispectral imagery. The spectral signature of a managed pond differs from open water: suspended sediment, algal turbidity and the geometric linearity of berm edges all contribute. At Sentinel-2's 10 m resolution, ponds larger than roughly 0.5 hectares are reliably delineated. Smaller ponds, common in smallholder aquaculture in Southeast Asia, require Planet SuperDove or WorldView-3 to resolve individual berm widths.
The licensing compliance question for pond aquaculture is usually about lateral expansion into protected intertidal habitat rather than structural change within an existing footprint. That makes change detection the primary analytic: a baseline polygon from the licence boundary is compared against current imagery, and any pond area outside the licensed perimeter is flagged. This is straightforward in principle. The complication is that tidal stage affects which pond surfaces are water-covered versus exposed at the time of any given image, so multi-date compositing or tidal-corrected analysis is needed to avoid false positives from tidal variation.
Building a compliance baseline that holds up
A licensing authority needs more than a snapshot. It needs a dated, georeferenced record of farm extent that can be compared against the licence polygon and that carries enough provenance to be defensible in a regulatory proceeding. That means the analytic output must include: the source imagery date and sensor, the detection method, the measured area and its uncertainty, and a clear visual overlay against the licence boundary.
Archive depth matters here. Sentinel-1 data runs back to 2014, Sentinel-2 to 2015. That is enough to reconstruct expansion history for most currently operating farms, which is useful when a regulator wants to establish when an unlicensed structure first appeared. Planet's archive is shallower and less consistent in early years, but its 3 m resolution makes it the right tool for measuring the current state precisely. WorldView-3 is reserved for targeted audit rather than routine monitoring, given the cost of tasking.
Satellize runs exactly this kind of multi-sensor compliance stack for government clients. The Tonga crop-estimation programme is a different domain, but the underlying approach, combining open Sentinel data with commercial tasking for targeted verification, transfers directly to coastal aquaculture licensing.
Honest limits of the method
Cloud cover is the most obvious constraint on optical sensors, and it is severe in tropical regions where much of the world's aquaculture sits. The SAR fallback is real but incomplete: SAR confirms presence and rough footprint, not pond water quality or net integrity. Very small cage operations, below roughly 15 m in their longest dimension, are likely to be missed or misclassified at Sentinel-1 resolution unless they cluster.
Longline aquaculture, particularly for shellfish and seaweed, is the hardest target. The structures are often below SAR detection thresholds individually, and their optical signature in multispectral bands is subtle. Published research has demonstrated detection of longline buoy fields using high-resolution SAR (COSMO-SkyMed or ICEYE at 1-3 m) rather than Sentinel-1, but that requires commercial tasking and is not yet a routine operational product. Anyone claiming reliable automated longline detection from free Sentinel-1 data alone is overstating the current state of the art.
Typical figures
| SAR spatial resolution (Sentinel-1 IW mode) | 10 m ground range detected; 20 m azimuth |
| Optical resolution (Sentinel-2 / SuperDove / WorldView-3) | 10 m / 3 m / 0.31 m panchromatic |
| SAR revisit (Sentinel-1, both satellites) | 6 days at mid-latitudes; up to 12 days near equator |
| Optical revisit (Planet SuperDove) | Near-daily globally |
| Minimum detectable cage diameter (Sentinel-1) | Approximately 15-20 m reliably; smaller structures ambiguous |
| Minimum pond area (Sentinel-2) | Approximately 0.5 ha; smaller ponds require SuperDove or WorldView-3 |
| Archive depth | Sentinel-1/2 from 2014-2015; Planet from 2016 (variable early coverage) |
| Cloud constraint | SAR: none. Optical: hard limit; tropical coasts may yield only a few clear scenes per month |
| Typical analytic latency | 24-72 hours from image acquisition for automated change detection; longer for manual audit layers |
| Delivery formats | GeoJSON / GeoPackage polygon layers, GeoTIFF change maps, PDF compliance report with imagery overlays |
Analytics Satellize can run
| Licensed footprint compliance map | Polygon overlay of detected farm extent against licence boundary GIS layer; area discrepancy calculation | GeoJSON layer with per-farm compliance status and measured overage in hectares |
| Unlicensed expansion alert | Bi-temporal SAR or optical change detection; new bright-object or pond-edge pixels outside existing licence polygons trigger alert | Email or API alert with scene date, coordinates and thumbnail image |
| Historical expansion timeline | Time-series analysis of Sentinel-1 and Sentinel-2 archive; farm footprint measured at quarterly intervals from 2015 | PDF report with annotated imagery and area-over-time chart |
| Cage array geometry audit | Object-based image analysis on WorldView-3 or SuperDove imagery; individual cage collar detection and count | GIS layer of individual cage polygons with count and area statistics |
| Pond boundary delineation (intertidal) | Supervised classification of multispectral imagery using spectral and geometric features; tidal-stage correction via co-registered tide model | GeoPackage of pond polygons with area, perimeter and licence-overlap attributes |
| Ambiguity classification report | Multi-date SAR persistence analysis combined with optical cross-check to separate fixed aquaculture structures from drifting gear or vessels | Flagged feature list with confidence score and supporting imagery evidence |
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.