Sea-surface temperature stress alerts for marine fish farms
Near-daily thermal infrared imagery from VIIRS, MODIS and Sentinel-3 SLSTR can flag dangerous sea-surface temperature excursions at marine fish farms before mortality events occur, provided operators understand what the satellite is actually measuring.
Sensors
- VIIRS (Suomi-NPP / NOAA-20): Day/Night Band plus thermal infrared bands M15 (10.76 µm) and M16 (12.01 µm) at 750 m nadir resolution. Each satellite provides roughly one daytime and one night-time pass per day; the two-satellite tandem cuts the revisit gap to around 12 hours. VIIRS SST retrievals are the primary operational input for NOAA's CoastWatch SST products.
- MODIS-Aqua / MODIS-Terra: Thermal infrared bands 31 and 32 (11 and 12 µm) at 1 km resolution, with two satellites again providing roughly twice-daily coverage. Archive extends to 2000 (Terra) and 2002 (Aqua), giving two decades of baseline climatology against which anomalies can be judged. Retrieval accuracy under clear skies is typically within 0.3–0.5 °C of in-situ buoy measurements.
- Sentinel-3 SLSTR: Sea and Land Surface Temperature Radiometer on Sentinel-3A and -3B, with a dual-view design (nadir and oblique) that reduces atmospheric correction uncertainty. Thermal channels at 1 km resolution; the two-satellite tandem achieves a revisit of under two days at mid-latitudes, better at high latitudes. ESA publishes L2 SST products via the Copernicus Data Space.
- Copernicus Marine Service L4 SST (CMEMS): Not a single sensor but a blended, gap-filled analysis product that merges VIIRS, MODIS, SLSTR and microwave radiometer retrievals through optimal interpolation. Delivered daily at 0.05° (~5 km) resolution. Cloud gaps are filled by the interpolation scheme, which is essential for cloudy coastal regions. The trade-off is that the blended field smooths out fine-scale thermal gradients that may matter for small farm sites.
- Sentinel-3 OLCI (supplementary context): Not a thermal sensor, but the 300 m ocean colour bands on the same platform provide chlorophyll and turbidity context that modulates how quickly surface heat penetrates the water column, useful for interpreting thermal retrievals at farm sites in optically complex coastal water.
What a satellite thermometer actually reads
Satellite thermal infrared sensors measure radiance emitted from the top fraction of a millimetre of the ocean surface, the so-called skin layer. This skin temperature is typically 0.1–0.5 °C cooler than the bulk temperature a metre or two below, because evaporative cooling acts almost exclusively at the surface. At night the difference narrows; on calm, sunny days it can widen further. For a salmon or sea bream farmer worried about the water their fish are actually breathing, this distinction is not academic. A satellite reading of 19.8 °C might correspond to a bulk temperature already above the 20 °C stress threshold for Atlantic salmon, or it might not.
The practical correction is to pair satellite skin retrievals with even a single in-situ temperature logger at cage depth. The logger calibrates the offset for a given site and season. Without that anchor, satellite SST alone gives trend and anomaly information reliably, but absolute threshold crossing must be treated with a margin of at least 0.5 °C in either direction.
Species thresholds and why a day's warning matters
Atlantic salmon (Salmo salar) begin to show physiological stress above roughly 18–20 °C, with appetite suppression, immune compromise and elevated mortality risk increasing sharply above 23 °C. Sea bream (Sparus aurata) tolerate warmer conditions, with growth optimum around 22–25 °C and stress emerging above 30 °C. These thresholds are published in peer-reviewed aquaculture literature and form the basis of national farm management guidelines in Norway, Scotland, Greece and Turkey.
A 24-hour warning is operationally significant. Farmers can increase aeration, reduce feeding to cut metabolic heat load, prepare emergency harvesting logistics or, where farm design permits, lower net pens to cooler water at depth. None of these responses is instantaneous. A thermal alert that arrives after the fish are already stressed is a post-mortem, not a management tool.
Cloud: the honest constraint
Thermal infrared cannot see through cloud. This is not a minor caveat. Many of the world's most productive marine aquaculture regions, including the Norwegian fjords, the Scottish west coast, and the Aegean in summer, experience persistent cloud cover precisely during the warm-season periods when thermal stress risk is highest. A single VIIRS or MODIS pass may be cloud-contaminated for days at a stretch.
Three mitigations exist, and none is perfect. First, the Copernicus Marine Service L4 blended product uses optimal interpolation to fill cloud gaps, drawing on microwave SST retrievals (which penetrate cloud but have coarser resolution, around 25 km) to constrain the interpolated field. Second, a multi-day composite of clear-sky pixels reduces the gap but introduces temporal lag. Third, the climatological baseline approach flags anomalies relative to long-term mean SST for the same calendar week, so even a partially cloudy retrieval can confirm whether conditions are anomalously warm. Operators should expect that during prolonged cloud events, satellite-derived alerts will not be available and in-situ loggers become the primary data source.
Spatial resolution and the small-farm problem
A typical VIIRS pixel at nadir is 750 m across. A standard salmon farm pen cluster occupies perhaps 200–400 m of sea surface. The satellite pixel therefore integrates temperature over an area much larger than the farm itself, mixing farm-adjacent water with open-water conditions. In a sheltered bay with limited tidal flushing, the farm's local thermal environment can diverge significantly from the surrounding pixel average, particularly if warm freshwater discharge or upwelling affects only a portion of the pixel.
Sentinel-3 SLSTR at 1 km is coarser still. The Copernicus Marine Service L4 product at 0.05° (~5 km) is useful for regional trend monitoring and anomaly detection but cannot resolve individual farm conditions. This is the honest resolution floor: satellite SST is a regional early-warning tool, not a replacement for cage-level monitoring. The appropriate use is to flag regional thermal anomalies that prompt operators to check their in-situ sensors, not to replace those sensors.
Building an operational alert pipeline
A practical alert system ingests daily VIIRS and MODIS SST retrievals for a defined region of interest around each farm site, applies a cloud-mask quality filter (VIIRS quality level 4 or better), and computes the anomaly relative to the 20-year MODIS climatology for that pixel and calendar week. When the anomaly exceeds a configurable threshold, say +1.5 °C above the climatological mean for that week, or when the absolute SST retrieval crosses a species-specific warning level, an alert is issued. The L4 blended product fills in on cloud-contaminated days, with the caveat on spatial smoothing noted above.
Satellize runs this kind of pipeline on open constellations for clients who need it delivered as a daily feed rather than built and maintained in-house. The methodology is directly analogous to the thermal anomaly monitoring Satellize applies in its Tonga crop-estimation programme, adapted here from land surface temperature to sea-surface temperature. Alert outputs can be delivered as GeoTIFF anomaly maps, farm-level time-series CSVs, or threshold-breach notifications via API.
The archive depth of MODIS (back to 2000) means that a new farm site can be characterised against two decades of thermal history before the first fish enter the water. That baseline is genuinely useful for site-selection due diligence and for understanding whether a site is trending warmer over time, which several Norwegian and Scottish fjord sites demonstrably are.
What satellite monitoring cannot replace
Cage-level temperature loggers at multiple depths remain essential. Satellite SST provides spatial context and regional early warning; in-situ sensors provide the ground truth at the point of biological consequence. The two are complementary, not competing.
Thermocline depth is invisible to thermal infrared. A satellite pass showing a cool surface may mask a shallow thermocline that prevents fish from descending to cooler water. Acoustic Doppler profilers or CTD casts are needed to characterise the vertical structure. Satellite data can flag when regional conditions make a shallow thermocline likely, but cannot confirm it. Buyers evaluating a satellite-only thermal monitoring product should ask that question directly.
Typical figures
| Spatial resolution (VIIRS SST) | 750 m at nadir, degrading to ~1.6 km at swath edge |
| Spatial resolution (MODIS SST) | 1 km at nadir |
| Spatial resolution (Sentinel-3 SLSTR) | 1 km (thermal channels) |
| Spatial resolution (CMEMS L4 blended) | ~5 km (0.05° grid), cloud gaps interpolated |
| Revisit (VIIRS two-satellite tandem) | ~12 hours (one daytime + one night-time pass per satellite per day) |
| Retrieval accuracy (clear sky) | ±0.3–0.5 °C versus in-situ buoy; skin-to-bulk offset typically 0.1–0.5 °C additional |
| Thermal spectral bands | VIIRS M15/M16: 10.76 µm and 12.01 µm; MODIS bands 31/32: 11 µm and 12 µm; SLSTR: 10.85 µm and 12 µm |
| Cloud penetration | None for thermal infrared; microwave SST (~25 km) used in L4 blending to partially compensate |
| Archive depth | MODIS-Terra from 2000, MODIS-Aqua from 2002, VIIRS from 2012, Sentinel-3A from 2016 |
| Latency (operational products) | VIIRS and MODIS near-real-time SST: 3–6 hours after pass; CMEMS L4: typically next-day delivery |
Analytics Satellize can run
| Daily SST anomaly map for farm region | Pixel-level departure from MODIS 20-year climatological mean for the same calendar week; cloud-masked to quality level 4 | Daily GeoTIFF with anomaly layer and cloud-mask overlay, delivered via API or SFTP |
| Species-threshold breach alert | Absolute SST retrieval (skin-corrected using site-specific offset from operator logger) compared against configurable species threshold; L4 blended product used on cloud-contaminated days | Push notification or API webhook with breach magnitude, confidence flag and cloud-cover percentage for the farm pixel |
| Farm-site SST time series | Pixel extraction and aggregation from VIIRS, MODIS and SLSTR passes over a defined farm bounding box; quality-filtered compositing | CSV time series updated daily, with per-pass quality flags and source-sensor labels |
| Seasonal thermal baseline report | Statistical characterisation of SST distribution at a farm site using full MODIS archive (2000 to present); percentile envelopes, trend detection and warm-season duration statistics | PDF report with charts and GIS layer; suitable for site-licensing submissions or investor due diligence |
| Regional thermal stress probability forecast context | Fusion of satellite SST anomaly with Copernicus Marine Service short-range ocean forecast fields to extend alert horizon from current observation to 3–5 day outlook | Tabular 5-day outlook with probability of exceeding species threshold, updated daily |
| Multi-site comparative dashboard | Aggregated SST anomaly and threshold-proximity scores across a portfolio of farm sites; ranked by current thermal risk | Web dashboard feed or daily summary report for farm network operators managing multiple licensed sites |
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.