Night-light detection of squid and light-fishing fleets
High-intensity fishing lights used to attract squid and other phototactic species are detectable from orbit, giving regulators a vessel census that AIS evasion cannot defeat. This page explains the physics, the sensors, and the honest limits.
Sensors
- VIIRS Day-Night Band (DNB): Suomi NPP and NOAA-20 VIIRS DNB provides roughly 750 m ground sample distance at nadir and a daily global overpass near local midnight. It detects vessel radiance down to approximately 2 × 10⁻¹⁰ W cm⁻² sr⁻¹ in the panchromatic visible range (0.5–0.9 µm), sufficient to identify individual bright-fishing vessels, though the brightest squid fleets saturate the sensor's 14-bit dynamic range.
- Luojia-1 (LJ-1 01): Chinese experimental night-light satellite operating at roughly 130 m ground resolution in the 460–980 nm band. Its finer pixel size partially resolves individual vessels that VIIRS DNB conflates into a single saturated blob, though revisit is irregular and archive depth is limited compared with the VIIRS record.
- Sentinel-1 SAR (C-band): ESA's Sentinel-1A/B pair provides C-band synthetic aperture radar imagery at 10 m resolution in Interferometric Wide Swath mode, with a six-day repeat at the equator (three days with both satellites active). SAR detects vessel hull backscatter regardless of cloud or daylight, enabling a physical vessel count that validates or corrects the light-based census.
- ICEYE X-band SAR: Commercial X-band SAR constellation offering sub-3 m resolution in spotlight mode and same-day tasking. X-band is more sensitive to smaller metallic targets than C-band, improving detection of smaller light-fishing skiffs. Revisit can be contracted on demand, useful for time-sensitive enforcement windows.
Why squid boats are the easiest fleet to find from orbit
Squid, saury and certain other pelagic species aggregate toward light at night. Commercial operators exploit this with arrays of metal halide or, increasingly, LED lamps that can collectively emit several hundred kilowatts per vessel. The resulting radiance is orders of magnitude above the ambient ocean surface. A single large squid jigger in the South Atlantic or East China Sea is visible to VIIRS DNB as a discrete point source against a background that is essentially zero.
This makes light-fishing fleets unusual in the world of maritime monitoring. Most vessel-detection problems require careful signal extraction from background clutter. Here the targets are advertising their position with the brightest artificial lights on the ocean. The challenge is not sensitivity. It is dynamic range, vessel discrimination and the gap between a light detection and a confirmed vessel identity.
What a floating bloom of light gives away, and what it does not
A VIIRS DNB overpass near local midnight produces a snapshot of radiant flux integrated over each 750 m pixel. Where a fleet is dense, multiple vessels fall within one pixel and their signals sum. Published work using VIIRS Nightfire and related algorithms (Colorado School of Mines, EOGDATA) has estimated fleet sizes in the Northwest Pacific squid grounds at hundreds of vessels on a single night, with total radiance correlating broadly with fleet effort. The spatial pattern of lights across weeks or months traces fishing grounds with a fidelity that no AIS-based dataset can match, because a substantial fraction of the global squid fleet either does not carry AIS or disables it.
The limits are real. VIIRS DNB saturates at high radiance values, compressing the brightest vessels into an indistinguishable bright patch. The sensor cannot distinguish a 300 kW jigger from a 150 kW one once both exceed the saturation threshold. Radiance alone cannot confirm vessel type: a gas flare on a drifting platform, a research vessel with deck lighting, or a large trawler with floodlamps can produce comparable signals. Cloud cover blocks the optical path entirely, a persistent problem in the ITCZ and high-latitude winter. And the roughly 750 m pixel means that in a dense fleet, individual vessel positions are smeared into a cluster centroid rather than resolved as discrete hulls.
SAR as the ground truth the DNB cannot provide
Cross-referencing a VIIRS light detection with a near-coincident Sentinel-1 or ICEYE pass resolves most of the ambiguity. SAR detects the physical hull via radar backscatter. It is unaffected by cloud, works day or night, and at 10 m (Sentinel-1) or sub-3 m (ICEYE spotlight) resolution it separates individual vessels that DNB conflates. A light cluster that DNB suggests contains twenty vessels can be confirmed, or corrected, by counting discrete SAR returns within the same bounding box.
The practical complication is temporal offset. Sentinel-1 and VIIRS do not overpass the same location at the same time. A fleet can move tens of kilometres between a midnight DNB pass and a dawn SAR acquisition. Matching detections requires a motion model or, where available, a same-night SAR tasking. ICEYE's on-demand scheduling makes same-night pairing feasible for high-priority areas, at a cost. For routine fleet monitoring across the full global squid grounds, the more common approach is statistical: build a time series of DNB detections and use periodic SAR validation to calibrate the light-to-vessel-count conversion factor for a given fleet type and season.
The archive as evidence: multi-year patterns and treaty compliance
VIIRS DNB data extends back to October 2011 on Suomi NPP, giving over a decade of nightly global coverage accessible through NASA EARTHDATA and the NOAA STAR portal. This archive is long enough to characterise seasonal fishing cycles, detect the entry of new fleets into a zone, and identify whether effort inside an exclusive economic zone has changed after a bilateral access agreement or a moratorium. A single night's detection is a data point. Twelve years of nights is a compliance record.
Regulators in regions such as the Southwest Atlantic, where Argentine and Falkland Islands EEZ boundaries are actively contested by foreign fleets, have used exactly this kind of time-series analysis to document persistent fishing pressure. The light signal does not require the vessel to report. That is its principal value to a fisheries authority that suspects underreporting.
Honest limits a buyer should understand before commissioning this analysis
No current spaceborne night-light sensor can reliably detect a vessel below roughly 50–100 kW of emitted light under typical atmospheric conditions. Small artisanal light-fishing boats using modest lamp arrays may fall below this threshold entirely. The transition of some fleets from high-wattage metal halide to narrower-spectrum LED arrays also changes the spectral signature, and VIIRS DNB's broad panchromatic response means it cannot distinguish lamp type from radiance alone.
Luojia-1's 130 m resolution is a genuine improvement for fleet disaggregation, but its irregular tasking and limited archive make it a supplement rather than a replacement for VIIRS. Hyperspectral night-light sensors capable of lamp-type discrimination exist in research form but are not yet operationally available at global scale. Any fleet-size estimate derived from light data alone carries an uncertainty that should be quantified and reported honestly. Satellize builds that uncertainty range into delivered products rather than presenting a single number as fact, which is the approach we also apply to the Tonga crop-estimation programme and every other analytics engagement.
From detection to a product a fisheries authority can act on
The analytic pipeline runs from raw DNB radiance through cloud masking, background subtraction, and a vessel-detection threshold calibrated against the local dark-ocean baseline. Detected light clusters are assigned a centroid position, an estimated vessel count (with confidence interval), a radiance magnitude, and a flag indicating whether the cluster falls inside or outside a declared EEZ or marine protected area. Where a SAR pass is available within a defined time window, the SAR vessel count is appended as a validation field.
Outputs can be delivered as nightly GeoJSON feeds, weekly aggregated heatmaps in GeoTIFF format, or monthly summary reports with trend analysis. The choice depends on whether the client's priority is real-time patrol tasking or longer-term policy evidence. Both are legitimate uses of the same underlying data.
Typical figures
| VIIRS DNB spatial resolution | ~750 m at nadir (Suomi NPP, NOAA-20) |
| VIIRS DNB revisit | Daily global coverage; one overpass near local midnight per satellite |
| VIIRS DNB spectral range | 0.5–0.9 µm (panchromatic visible/NIR) |
| VIIRS DNB minimum detectable radiance | ~2 × 10⁻¹⁰ W cm⁻² sr⁻¹ under clear-sky conditions |
| Luojia-1 spatial resolution | ~130 m; spectral range 460–980 nm |
| Sentinel-1 SAR resolution (IW mode) | 10 m; C-band (5.405 GHz); 6-day repeat at equator with one satellite |
| ICEYE SAR resolution (spotlight) | Sub-3 m; X-band; on-demand tasking |
| Archive depth (VIIRS DNB) | October 2011 to present (Suomi NPP) |
| Cloud impact | DNB and Luojia-1 blocked by cloud; SAR unaffected |
| Typical delivery formats | GeoJSON (nightly detections), GeoTIFF (heatmaps), PDF/XLSX (monthly reports) |
Analytics Satellize can run
| Nightly vessel-light detection layer | Radiance thresholding and blob detection on VIIRS DNB after cloud masking and dark-ocean background subtraction | GeoJSON point layer with centroid position, radiance magnitude, estimated vessel count with confidence interval, and EEZ membership flag |
| Fleet-size time series by fishing ground | Aggregation of nightly detections into weekly and monthly effort indices, normalised for cloud-free observation days | CSV time series and interactive chart; monthly trend report in PDF |
| SAR-validated vessel count | Co-registration of Sentinel-1 or ICEYE SAR detections with DNB light clusters within a defined spatial-temporal window; hull-count comparison | Validation table appended to the nightly GeoJSON; flagged discrepancies for manual review |
| EEZ incursion alert | Spatial intersection of detected light clusters with EEZ boundary polygons (publicly available from Flanders Marine Institute VLIZ dataset); threshold trigger on cluster persistence | Near-real-time alert (email or API push) with cluster position, estimated vessel count, and entry timestamp |
| Multi-year fishing effort heatmap | Accumulation of cloud-masked DNB detections over user-defined date range; kernel density estimation to produce effort density surface | GeoTIFF raster at 0.05° grid; companion shapefile of high-effort polygons |
| Fleet displacement analysis | Comparison of effort centroids across seasons or regulatory periods to detect spatial shifts consistent with enforcement response or stock movement | Annotated map series and written assessment in monthly summary report |
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.