Mixed-layer depth variability as a prey concentration index for fisheries
When the mixed layer shallows, zooplankton and small pelagics are compressed toward the surface and become available to predators. Satellite altimetry and blended SST products can estimate that shoaling, with important caveats about accuracy.
Sensors
- Jason-3 altimeter (CNES/EUMETSAT/NOAA/NASA): Measures sea-surface height along repeat ground tracks with ~1 cm RMS precision at 1 Hz (roughly 7 km along-track). A 10-day exact repeat cycle. Sea-level anomaly (SLA) derived from Jason-3 is the primary input for estimating the depth of the 20°C isotherm as a thermocline proxy in the tropics, using the linear SLA-to-D20 relationship documented in the literature.
- Sentinel-6 Michael Freilich (ESA/EUMETSAT/NASA/NOAA): Successor to Jason-3, flying the same reference orbit with an interleaved 10-day repeat. Its high-resolution mode (HR-LRM) improves coastal and inland water retrievals. Combining Sentinel-6 with Jason-3 SLA fields reduces the temporal gap between track crossings, improving the spatial picture of dynamic height anomalies.
- GHRSST blended SST products (multiple contributing sensors): Group for High Resolution Sea Surface Temperature L4 blended products combine infrared (MODIS, VIIRS, AVHRR) and microwave (AMSR2) observations into daily gap-filled grids at 0.01° to 0.25° resolution. SST gradients from these products indicate frontal boundaries where mixed-layer depth changes sharply. Cloud cover is the principal limit: infrared channels are blind under overcast skies, and microwave channels resolve only to roughly 25 km.
- CMEMS ocean reanalysis and near-real-time models (Copernicus Marine Service): The Global Ocean Physical Reanalysis (GLORYS12) assimilates altimetry, SST and Argo profiles at 1/12° resolution (~8 km). It provides modelled mixed-layer depth (MLD) fields that can be used alongside satellite-derived proxies. Reanalysis MLD has known biases in regions with weak stratification and sparse Argo coverage; near-real-time products carry larger uncertainty than hindcast fields.
- Argo float profiles (international programme, not a satellite): Included here as the ground-truth reference. Argo floats profile temperature and salinity from 2000 m to the surface roughly every 10 days. They are the primary validation dataset for satellite-derived MLD estimates. Float density is uneven: the tropical Pacific is well sampled; the Southern Ocean and marginal seas are not.
Why depth matters more than colour
Fisheries remote sensing tends to focus on sea-surface temperature and chlorophyll as proxies for productivity. Both are useful. But they describe conditions at or very near the surface; they do not directly tell you where the food is vertically distributed, or whether a predatory fish can reach it without expending more energy than the meal is worth.
Mixed-layer depth is the missing dimension. When the thermocline sits at 80 m, zooplankton and the small pelagics that feed on them are spread across a large volume. When it shoals to 20 m, the same biomass is compressed into a thin layer that surface-feeding predators, including tuna, billfish, seabirds and commercial fishers, can exploit with far less effort. This compression effect is well documented in the Pacific tuna literature and underpins the design of several operational fishing advisory systems.
What a satellite altimeter actually measures, and what you infer from it
An altimeter measures the height of the sea surface above a reference ellipsoid. After removing tidal, atmospheric and mean dynamic topography signals, what remains is the sea-level anomaly (SLA), which reflects the integrated density structure of the water column beneath. In the tropical oceans, where a warm, light surface layer sits above a cold, dense thermocline, SLA and the depth of the 20°C isotherm (D20) are strongly correlated. A positive SLA typically means a deeper thermocline; a negative anomaly indicates shoaling.
The relationship is not universal. It holds best in the equatorial Pacific and Indian Ocean, where the thermocline is sharp and the warm layer dominates the density signal. In subpolar regions, salinity stratification can decouple SLA from thermal structure entirely. The correlation also degrades near strong boundary currents such as the Kuroshio or the Gulf Stream, where mesoscale eddies mix the signal. Users working outside the tropical Pacific should treat D20 estimates as indicative rather than precise without regional Argo validation.
Jason-3 and Sentinel-6 together provide SLA along interleaved ground tracks with a combined revisit at any given track of roughly five days. Between tracks, spatial gaps are filled by interpolated gridded products such as CMEMS DUACS, which blend all available altimeters into a daily 0.25° grid. That interpolation smooths out mesoscale features smaller than roughly 100 to 150 km. Sub-mesoscale shoaling events, which can be ecologically significant, are largely invisible to the current altimeter constellation.
SST gradients as a corroborating signal
A shoaling thermocline often brings cold water close enough to the surface that SST drops measurably, particularly during upwelling events. GHRSST blended products at 0.01° resolution can resolve frontal gradients of less than 0.5°C over tens of kilometres, which is sufficient to identify the surface expression of thermocline shoaling. Combining a negative SLA anomaly with a concurrent SST cold patch strengthens the inference considerably.
The limit is cloud. Infrared SST retrievals fail under overcast conditions, which are persistent in many productive fishing regions, including the Humboldt Current and the West African upwelling zone. Microwave SST fills some of the gap but at coarser resolution (roughly 25 km for AMSR2). In practice, a blended product that weights recent clear-sky infrared observations alongside microwave data gives the most useful daily picture, at the cost of some spatial sharpness.
Translating a depth estimate into a fishing advisory
The operational chain is straightforward in concept. Compute daily or weekly SLA anomaly maps from CMEMS DUACS. Apply a regional SLA-to-D20 regression calibrated against Argo profiles from the same basin. Identify areas where D20 is shoaling below a threshold, say 40 m in a system where 60 m is the seasonal mean. Overlay GHRSST fronts to find the surface expression. The resulting spatial layer identifies candidate areas where prey concentration is elevated.
Thresholds are region-specific. In the western Pacific warm pool, a D20 of 40 m would be unusually shallow; in the eastern equatorial Pacific during a La Niña, it is common. Any advisory system needs a climatological baseline for the target region before anomaly detection is meaningful. CMEMS reanalysis provides a 30-year hindcast that can serve as that baseline, though the reanalysis itself has biases of 10 to 20 m in MLD in some regions compared with Argo.
Satellize can build and maintain this pipeline, integrating open CMEMS and GHRSST streams with regional Argo validation, and delivering gridded prey-concentration index layers on a schedule matched to a fleet's operational planning cycle. The Tonga crop-estimation programme demonstrated the same principle in a different domain: a physically grounded index, derived from open satellite data, validated against in-situ observation, and delivered as an operational product rather than a research output.
Honest limits of the method
Satellite-derived MLD is not a substitute for Argo profiles. The RMS error of SLA-based D20 estimates against in-situ profiles is typically 10 to 30 m in well-validated tropical regions, and can exceed 50 m in poorly sampled or dynamically complex areas. That uncertainty matters: the difference between a thermocline at 25 m and one at 55 m is ecologically large.
Revisit is a second constraint. The 10-day Jason-3 repeat and the 0.25° DUACS grid mean that a rapid shoaling event driven by a passing eddy or a brief wind burst may be captured only partially, or with a lag of several days. For species that respond to thermocline position on timescales of hours, this is a real limitation. Near-real-time CMEMS products reduce the lag but do not eliminate it.
Finally, the method says nothing about actual biomass. A compressed layer is a necessary but not sufficient condition for a productive fishing ground. Chlorophyll concentration, prey species composition and local fishing history all matter. This index is most useful as a first filter that directs survey effort or vessel routing, not as a standalone catch predictor.
Validation is not optional
Any operational deployment of this method should include a validation protocol before the index is used for fleet routing or stock assessment. The minimum requirement is a comparison of satellite-derived D20 estimates against co-located Argo profiles from the target region, spanning at least one full seasonal cycle. Where Argo coverage is sparse, expendable bathythermograph (XBT) deployments or moored buoy data can supplement.
Validation should be repeated after major oceanographic events. An El Niño shifts the SLA-to-D20 relationship across the entire Pacific basin; a calibration built on neutral-year data will be systematically biased during the event. Treating the satellite product as a static, pre-validated tool is the most common failure mode in operational oceanographic advisory services.
Typical figures
| Altimeter SLA spatial resolution (gridded DUACS product) | 0.25° (~25 km); sub-mesoscale features below ~100 km are smoothed out |
| Altimeter revisit (Jason-3 or Sentinel-6 single satellite) | 10-day exact repeat; combined interleaved tracks reduce effective gap to ~5 days on reference tracks |
| GHRSST blended SST resolution | 0.01° (~1 km) for L4 infrared-dominated products; ~0.25° for microwave-only fields |
| SST product latency (near-real-time) | Typically 6–24 hours for NRT L4 blended products via CMEMS |
| D20 estimate accuracy (tropical Pacific, validated regions) | RMS error 10–30 m versus Argo; degrades to >50 m in subpolar or boundary-current regions |
| CMEMS reanalysis MLD archive depth | GLORYS12 hindcast from 1993 to present at 1/12° (~8 km) resolution |
| Cloud impact on SST retrieval | Infrared channels fully obscured under cloud; microwave fills gap at ~25 km resolution (AMSR2) |
| Minimum detectable thermocline anomaly | Approximately 10 m shoaling in well-calibrated tropical regions; not reliably detectable in weakly stratified or salinity-dominated water columns |
Analytics Satellize can run
| Regional D20 anomaly map | SLA-to-D20 linear regression calibrated against Argo profiles, applied to CMEMS DUACS gridded SLA fields | Daily or weekly GIS layer (GeoTIFF or NetCDF) showing thermocline depth anomaly relative to climatological baseline |
| Prey concentration index layer | Composite of negative D20 anomaly and cold SST front co-location, thresholded against regional seasonal climatology | Gridded index map with confidence band, delivered as GIS layer or API feed for fleet routing integration |
| Thermocline shoaling event alerts | Anomaly detection against GLORYS12 30-year baseline; alert triggered when D20 crosses a user-defined threshold in a defined area of interest | Email or API alert with spatial polygon and estimated duration, issued within 24 hours of DUACS product availability |
| Seasonal MLD climatology and percentile bands | Statistical summary of CMEMS reanalysis MLD fields (1993 to present) for a defined region, with 10th, 50th and 90th percentile envelopes by week of year | Reference dataset (CSV and GIS) for calibrating operational thresholds and anomaly detection |
| Validation report against Argo profiles | Spatial and temporal co-location of satellite-derived D20 estimates with quality-controlled Argo float profiles from the target region | PDF report with bias, RMS error and correlation statistics; recommendation on regional applicability and threshold settings |
| Multi-year prey-concentration index time series | Retrospective application of the composite index to the CMEMS reanalysis archive, aligned with available catch or survey records | Annual summary report and machine-readable time series for use as environmental covariate in stock assessment models |
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.