Ocean mixed-layer depth estimation from altimetry and in-situ fusion
Mixed-layer depth governs how much heat the ocean stores and how much CO₂ it absorbs. Satellite altimetry and SST, fused with Argo float profiles via optimal interpolation, produce gridded MLD fields at near-global scale, though coverage thins sharply in under-floated basins.
Sensors
- Sentinel-6 Michael Freilich (radar altimeter): Ku- and C-band dual-frequency altimeter measuring sea-surface height to roughly 2–3 cm precision along a 10-day repeat ground track at ~7 km along-track sampling; sea-level anomaly (SLA) is a proxy for upper-ocean heat content and, indirectly, mixed-layer depth in dynamically active regions.
- Jason-3 (radar altimeter): Predecessor mission with identical 10-day, 66° inclination orbit to Sentinel-6 MF; running in tandem through the transition period, the two satellites together halve the effective revisit gap and improve SLA mapping in mid-latitudes. Along-track resolution ~6 km.
- MODIS (Terra and Aqua, SST): 11 µm and 12 µm thermal infrared bands deliver sea-surface temperature at 1 km nominal resolution under clear sky, with daily global coverage across both platforms. SST gradients indicate fronts where mixed-layer depth changes sharply; cloud contamination is the dominant data-loss mechanism at high latitudes.
- Argo float network (in-situ reference): Roughly 4,000 active autonomous profiling floats globally, each surfacing every 10 days and transmitting temperature and salinity profiles from 2,000 m to the surface. MLD is computed from each profile using a density- or temperature-threshold criterion (commonly Δσ = 0.03 kg m⁻³ or ΔT = 0.2 °C from 10 m reference depth). Float density varies from one per ~3° × 3° box in the North Atlantic to one per ~6° × 6° or sparser in the Southern Ocean and marginal seas.
What the mixed layer actually is, and why satellite altimeters care
The ocean mixed layer is the near-surface layer kept vertically uniform by wind stirring and convective cooling. Its depth, typically 20–200 m depending on season and location, sets the volume of water exchanging heat and gas with the atmosphere. A shallow summer mixed layer warms quickly and supports stratified phytoplankton blooms; a deep winter mixed layer ventilates carbon-rich water and moderates sea-surface temperature anomalies.
Satellite altimeters do not see the mixed layer directly. What they measure is the sea-surface height anomaly, which integrates the steric (density-driven) expansion of the water column beneath. In regions where upper-ocean heat content dominates the steric signal, SLA correlates meaningfully with mixed-layer depth. The correlation is strongest in the tropics and weakest at high latitudes, where salinity stratification and sea-ice complicate the picture. This indirectness is the central limitation of the altimetric approach, and any honest MLD product must quantify it.
How optimal interpolation stitches floats and satellites together
Optimal interpolation (OI) is a weighted averaging scheme that blends observations according to their estimated error covariances and the spatial correlation structure of the field being mapped. In practice, Argo profiles supply the ground truth for MLD at sparse, irregular locations; satellite SLA and SST fields supply the spatial structure that guides interpolation between float positions.
The workflow runs roughly as follows. Each Argo profile is processed to extract MLD using a threshold criterion. SLA from Sentinel-6 MF and Jason-3 is gridded along-track, then merged into a multi-mission product (such as the CMEMS L4 sea-level product). SST from MODIS provides frontal boundaries that constrain the interpolation anisotropy, preventing the scheme from smearing MLD values across sharp density fronts. The OI then minimises a cost function that penalises departure from the float observations while respecting the satellite-derived spatial covariance. The output is typically a gridded field at 0.25° or 0.5° horizontal resolution, updated weekly or monthly.
Published products following this general approach include the IFREMER mixed-layer depth climatology and the Argo-based MLD climatologies from de Boyer Montégut and colleagues, which remain widely used reference datasets. Operational centres such as Copernicus Marine Service (CMEMS) produce near-real-time upper-ocean heat content and MLD analyses using similar fusion frameworks.
Where the method works well, and where it quietly falls apart
The North Atlantic and North Pacific subtropical gyres are well served. Float density is relatively high, the SLA-to-heat-content relationship is stable, and SST fronts are detectable even under partial cloud. Seasonal MLD cycles of 20–150 m are captured with uncertainties of roughly 10–20 m in well-sampled areas, based on cross-validation studies in the published literature.
The Southern Ocean is a different matter. Float coverage south of 50°S is improving following dedicated deployment campaigns, but remains patchy. The SLA signal in the Southern Ocean is dominated by energetic mesoscale eddies rather than mean upper-ocean heat content, weakening the altimetric proxy. Winter cloud cover suppresses MODIS SST retrieval for weeks at a time. MLD uncertainties in the Southern Ocean routinely exceed 30–50 m, and in some sub-regions the interpolated field is little more than a smoothed climatology.
The Arctic and marginal ice zones present additional problems: sea-ice masks SST retrieval entirely, and altimetric waveform retracking over mixed ocean-ice surfaces introduces height errors. The Mediterranean and other semi-enclosed seas are under-floated relative to their dynamical complexity. Users in these regions should treat gridded MLD products as indicative rather than quantitative, and should examine the associated uncertainty fields, which reputable products always provide.
Uncertainty quantification: the number behind the number
A gridded MLD value without an accompanying uncertainty estimate is operationally incomplete. The main error sources are: the threshold criterion used to define MLD from individual profiles (which can differ by 10–30 m depending on the choice of criterion and the sharpness of the pycnocline); the interpolation error, which grows with distance from the nearest float; and the representativeness error of the SLA proxy, which varies by region and season.
Good OI implementations propagate these errors formally, producing a per-grid-cell posterior variance alongside the MLD estimate. The ratio of that variance to the signal variance is a direct measure of how much the satellite data is contributing versus how much the scheme is simply reverting to climatology. In data-rich regions this ratio can fall below 0.2; in the Southern Ocean it commonly exceeds 0.7. Publishing this diagnostic is not optional for serious operational use: it tells the downstream user whether to trust the field or to treat it as background.
Operational applications: who needs this and at what precision
Climate and ocean modelling centres assimilate MLD analyses to initialise and validate general circulation models. Fisheries managers use MLD as a predictor of habitat suitability for commercially important species that congregate near the thermocline. Naval applications require MLD for acoustic propagation modelling, since the thermocline acts as a sound channel boundary. Carbon-cycle research uses MLD to estimate the depth of the surface reservoir exchanging CO₂ with the atmosphere.
For most of these applications, a 10–20 m MLD accuracy in well-sampled regions is adequate. Naval acoustic work is more demanding and typically supplements satellite-derived products with local expendable bathythermograph (XBT) casts. Climate modelling requires long, consistent time series more than it requires point accuracy, which is where the Sentinel-6 MF and Jason-3 tandem's continuity with the TOPEX/Poseidon lineage (stretching back to 1992) becomes genuinely valuable.
Satellize's analytics stack can ingest CMEMS multi-mission SLA products and Argo data from the Global Data Assembly Centre, apply basin-specific OI configurations, and deliver uncertainty-qualified MLD grids as GIS layers or NetCDF feeds. The same infrastructure that supports the Tonga crop-estimation programme handles the data-fusion pipelines; the domain changes, the workflow logic does not.
What the next decade of sensors will change
SWOT (Surface Water and Ocean Topography), launched in late 2022, maps sea-surface height in two-dimensional swaths at roughly 1 km resolution rather than along narrow ground tracks. This substantially improves the spatial sampling of mesoscale and sub-mesoscale features that drive MLD variability, particularly near fronts and eddies. Early validation work suggests SWOT will improve the spatial fidelity of OI-based MLD products in dynamically complex regions, though the 21-day repeat cycle means temporal sampling remains a constraint.
Biogeochemical Argo floats, now numbering several hundred and growing, add oxygen, nitrate and chlorofluorocarbon sensors to standard T/S profiles. These allow independent MLD estimates from biological tracers, providing a cross-check on the physical threshold method. The combination of denser float networks, wider-swath altimetry and improved OI algorithms will progressively reduce the Southern Ocean uncertainty gap, though it will not close it entirely within this decade.
Typical figures
| Altimeter along-track resolution | ~6–7 km (Jason-3, Sentinel-6 MF); SWOT 2D swath at ~1 km |
| Altimeter repeat cycle | 10 days (Jason-3 and Sentinel-6 MF individually); effective sampling improved by tandem operation |
| SLA measurement precision | ~2–3 cm (Sentinel-6 MF Ku-band, published mission spec) |
| MODIS SST spatial resolution | 1 km (thermal infrared, bands 31–32); cloud-contaminated pixels flagged and excluded |
| Argo float profile spacing | ~10-day surfacing interval; ~3°×3° mean spacing in well-floated basins, 6°×6° or sparser in Southern Ocean |
| Gridded MLD output resolution | Typically 0.25°–0.5° horizontal; weekly or monthly temporal resolution |
| MLD uncertainty (well-sampled basins) | ~10–20 m (published cross-validation estimates) |
| MLD uncertainty (Southern Ocean, marginal seas) | 30–50 m or greater; posterior variance field required for interpretation |
| Altimetry archive depth | TOPEX/Poseidon to present: 1992 onwards (~32 years of merged multi-mission SLA) |
| Delivery formats | NetCDF-4 (CF conventions), GeoTIFF per layer, GIS-ready vector uncertainty contours |
Analytics Satellize can run
| Basin-scale gridded MLD with per-cell uncertainty | Optimal interpolation of Argo threshold-derived MLD, constrained by multi-mission SLA and MODIS SST frontal structure | Monthly NetCDF grids with mean MLD and posterior standard deviation; GIS layer package |
| Anomaly detection: MLD departure from climatology | Differencing of current OI analysis against de Boyer Montégut or IFREMER climatological baseline; z-score normalisation per grid cell | Weekly anomaly map with significance flags; alert feed for anomalies exceeding 1.5 standard deviations |
| Upper-ocean heat content proxy | Integration of SLA-derived steric height over the mixed layer using published SLA-to-heat-content regression coefficients (region-specific) | Gridded heat content anomaly layer; time-series report per user-defined region of interest |
| Mixed-layer depth trend analysis | Linear and seasonal decomposition over the 32-year merged altimetry archive combined with Argo climatology; Mann-Kendall trend test per grid cell | Decadal trend map with confidence intervals; basin-summary PDF report |
| Acoustic environment index (thermocline depth) | Pycnocline depth extracted from OI T/S field as depth of maximum Brunt-Väisälä frequency; MLD used as upper bound | Gridded thermocline-depth product; region-specific depth-profile report |
| Data-gap risk assessment by basin | Posterior variance mapping; float-density audit against JCOMMOPS Argo dashboard; identification of grid cells where OI is reverting to climatology | Coverage-quality report with recommended supplementary deployment zones; uncertainty ratio layer |
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.