Satellite upwelling indices as environmental covariates in stock assessment
Satellite altimetry, sea-surface temperature and ocean colour together quantify upwelling intensity at the spatial and temporal scales that matter for forage-fish recruitment. Feeding these indices into stock-assessment models cuts residual variance and sharpens catch-quota advice.
Sensors
- Sentinel-6 Michael Freilich (radar altimeter): Measures sea-level anomaly (SLA) along repeat ground tracks with ~3 cm range precision; 10-day exact-repeat cycle. SLA fields, combined with Copernicus Marine Service gridded products, yield geostrophic current velocities used to identify upwelling-favourable circulation cells.
- TOPEX/Poseidon and Jason-1/2/3 altimeter series: Provides the multi-decadal SLA archive (1992 to present) essential for climatological baselines. Jason-3 continues the 10-day, 1,336 km repeat orbit at roughly 5 cm SLA accuracy in open ocean; coastal accuracy degrades within ~20 km of shore.
- MODIS-Aqua (ocean colour and SST): Delivers daily global SST at 1 km resolution and chlorophyll-a concentration at 1 km (Ocean Colour bands 8–16, 405–877 nm). Cloud cover is the primary operational limit; weekly or monthly composites are typically required for reliable upwelling index time series in persistently cloudy regions such as the Humboldt system.
- Sentinel-3 OLCI (Ocean and Land Colour Instrument): 21 spectral bands from 400 to 1,020 nm at 300 m resolution, with a ~2-day revisit at the equator. Chlorophyll-a retrieval accuracy is published at roughly ±30% in Case-1 waters; turbid coastal waters introduce larger errors. Complements MODIS for higher spatial detail near shelf breaks.
Why the ocean's plumbing predicts the fish catch
Coastal upwelling systems, the Humboldt, Benguela, California and Canary Currents among them, account for roughly 20% of global wild fish landings despite covering less than 1% of ocean area. The mechanism is well established: equatorward winds drive Ekman transport, pulling cold, nutrient-rich water to the surface. Phytoplankton bloom. Zooplankton follow. Larval anchoveta, sardine and jack mackerel either find food within their first weeks of life or they do not survive. The outcome of that lottery is what stock assessors call recruitment, and it is the single largest source of uncertainty in most pelagic stock models.
The connection to satellites is direct. Upwelling intensity leaves three measurable signatures: a negative sea-surface temperature anomaly relative to climatology, a positive sea-level anomaly depression (the surface dips as cold dense water rises), and a chlorophyll bloom that follows the nutrient pulse by roughly one to three weeks. All three are observable from orbit. None requires a ship.
What altimetry adds that temperature alone cannot
SST has been the default environmental covariate in fisheries models for decades, partly because AVHRR data run back to the early 1980s. The limitation is that temperature is a consequence of upwelling, not its driver. By the time a cold anomaly is visible, the physical forcing has already acted on larvae. Sea-level anomaly from altimetry is more directly tied to the geostrophic current field: a negative SLA depression along the shelf edge indicates active upwelling-favourable circulation, and it can precede the thermal signal by several days.
The Copernicus Marine Service gridded altimetry product (CMEMS, combining Sentinel-6, Jason-3 and other missions) delivers daily, 0.125-degree SLA and derived geostrophic current velocity fields globally. For the Humboldt Current system off Peru and Chile, published studies have used these fields to compute the Bakun upwelling index from satellite-derived curl of wind stress and geostrophic divergence, achieving better predictive skill for anchoveta year-class strength than SST alone. The honest caveat: the 0.125-degree grid (~14 km) smooths mesoscale features smaller than roughly 50 km, and coastal SLA accuracy degrades within two to three grid cells of the shoreline.
Chlorophyll as a lagged recruitment signal
Phytoplankton concentration, retrieved from MODIS-Aqua or Sentinel-3 OLCI, functions as a time-integrated proxy for the food environment experienced by first-feeding larvae. The lag between peak upwelling and peak chlorophyll bloom is typically one to three weeks in the Humboldt system, depending on mixed-layer depth and grazing pressure. This lag is operationally useful: a strong bloom in the spawning quarter can be used as a leading indicator of juvenile survival before acoustic or trawl surveys confirm recruitment.
Sentinel-3 OLCI at 300 m offers meaningful resolution improvement over MODIS's 1 km for shelf-edge blooms that are spatially narrow. The trade-off is that OLCI's atmospheric correction is less mature than MODIS's in some coastal geometries, and both sensors share the same fundamental cloud problem. In the highly productive but persistently stratocumulus-covered Peruvian shelf, cloud-free pixel fractions can fall below 30% in austral winter. Eight-day and monthly composites are standard practice; they smooth the very short-lived bloom events that may matter most for larvae.
Integrating indices into stock-assessment models: the practical steps
The standard workflow has four stages. First, construct anomaly time series: SST anomaly against a 1982-to-present OISST or MODIS climatology, SLA anomaly against the CMEMS multi-mission mean, and chlorophyll anomaly against an MODIS or SeaWiFS climatological baseline. Second, aggregate spatially to the stock's core spawning habitat, typically a defined shelf polygon, using area-weighted means. Third, compute seasonal or quarterly indices aligned to the stock's biological calendar. Fourth, enter these indices as covariates in the stock-assessment model, most commonly a surplus production model or an age-structured model with a stock-recruitment relationship, and test whether they reduce AIC or residual variance in the recruitment equation.
Published work on Peruvian anchoveta (Engraulis ringens) has shown that composite upwelling indices incorporating SLA, SST and chlorophyll explain more recruitment variance than any single variable. The improvement is real but modest: recruitment in small pelagics remains inherently noisy, and no satellite covariate eliminates the need for acoustic biomass surveys. The satellite layer narrows the uncertainty; it does not replace the survey.
Archive depth and the baseline problem
A covariate is only as useful as the historical record behind it. For SST, AVHRR extends usable data back to 1981. For ocean colour, the continuous global record effectively begins with SeaWiFS in 1997 and MODIS-Aqua in 2002. For altimetry, the TOPEX/Poseidon launch in 1992 marks the practical start. These archives are long enough to capture several El Niño and La Niña cycles, which is the dominant interannual forcing for Humboldt and Benguela upwelling systems, but they remain short relative to multi-decadal stock fluctuations. Assessors should treat pre-satellite recruitment estimates as having substantially larger environmental-covariate uncertainty.
Sentinel-6 and the forthcoming Sentinel-3C ensure continuity through at least the 2030s. That matters: a covariate relationship calibrated on 30 years of data and then broken by a sensor gap is worse than no covariate at all. Satellize builds operational upwelling index pipelines on open Copernicus and NASA data streams, with commercial gap-filling on client licence. The Tonga crop-estimation programme demonstrated the same principle in a different domain: environmental indices derived from open constellations, maintained consistently, outperform ad hoc analyses pulled together at assessment time.
Honest limits before you commit the budget
Three limits deserve plain statement. Coastal altimetry accuracy degrades within roughly 20 to 50 km of shore depending on the mission and processing version; for inshore upwelling systems this is a genuine constraint, not a footnote. Chlorophyll retrieval in turbid or optically complex coastal water is unreliable with standard Case-1 algorithms; sites with significant river input or resuspension need dedicated Case-2 processing, which adds cost and uncertainty. And the statistical relationship between an upwelling index and recruitment is empirical: it is calibrated on historical data and can shift under climate change or when stock distribution moves. A covariate that improves model fit over 1993 to 2015 may perform differently over 2025 to 2040.
None of these limits argues against using satellite upwelling indices. They argue for using them as one input among several, updating the empirical relationship regularly, and reporting prediction intervals honestly to fisheries managers rather than presenting satellite-enhanced recruitment forecasts as more certain than they are.
Typical figures
| SLA spatial resolution (gridded product) | 0.125° (~14 km); along-track altimeter footprint ~2–7 km depending on mission |
| SLA temporal resolution | Daily gridded CMEMS product; 10-day exact-repeat ground track for Sentinel-6 / Jason-3 |
| SST spatial resolution (MODIS-Aqua) | 1 km (night-time, 11 µm and 12 µm bands); 4 km daytime product standard for time series |
| Chlorophyll-a resolution (OLCI Sentinel-3) | 300 m; accuracy ±30% in Case-1 open-ocean waters; degraded in turbid coastal zones |
| Chlorophyll-a resolution (MODIS-Aqua) | 1 km; global daily coverage; cloud-free fraction highly variable by region |
| Altimetry archive depth | TOPEX/Poseidon 1992 to present via Jason-1/2/3 and Sentinel-6 continuity |
| Ocean colour archive depth | SeaWiFS 1997–2010; MODIS-Aqua 2002 to present; Sentinel-3 OLCI 2016 to present |
| Latency (operational CMEMS products) | Near-real-time SLA: ~3–5 hours; delayed-mode science-quality: ~60 days |
| Coastal SLA accuracy limit | Degrades within ~20–50 km of shore; use dedicated coastal altimetry processors for nearshore work |
| Delivery formats | NetCDF-4 (CMEMS, NASA Earthdata); GeoTIFF on request; CSV anomaly time series for model ingestion |
Analytics Satellize can run
| Quarterly upwelling intensity index | Area-weighted SLA anomaly plus SST anomaly composite over defined spawning habitat polygon, anomalised against multi-decadal CMEMS and OISST climatologies | CSV time series and PDF summary report, updated quarterly, ready for direct ingestion into surplus production or age-structured stock models |
| Chlorophyll bloom timing and magnitude index | 8-day MODIS-Aqua or Sentinel-3 OLCI chlorophyll composites; bloom onset detected by threshold exceedance above seasonal climatology; magnitude as integrated anomaly over spawning quarter | Annual bloom calendar GeoTIFF stack and tabular bloom metrics per stock assessment zone |
| Geostrophic current anomaly maps | CMEMS multi-mission gridded altimetry; derived zonal and meridional geostrophic velocity anomalies; upwelling-favourable current cells identified by divergence calculation | Monthly GeoTIFF velocity anomaly fields and shelf-edge transport index CSV for model covariate use |
| El Niño / La Niña phase classification and stock-area SLA tracking | Niño-3.4 SST index cross-referenced with local SLA and chlorophyll anomalies in the client's stock polygon; phase-conditional recruitment prior for Bayesian assessment models | Seasonal briefing note with phase classification, local anomaly plots and recommended prior adjustment for the assessment model |
| Multi-variable upwelling composite score | Principal component or weighted linear combination of normalised SLA anomaly, SST anomaly and chlorophyll anomaly; weights calibrated against historical recruitment index where survey data are available | Annual composite score time series (1993 to present) in CSV; model fit statistics (AIC reduction, residual variance) reported alongside |
| Covariate uncertainty and cloud-gap report | Pixel-level cloud-mask audit on MODIS and OLCI composites; flagging of quarters where cloud-free fraction falls below 30% and SLA coastal proximity degrades accuracy; honest uncertainty bounds on each index value | Data quality annex appended to each quarterly index report, flagging unreliable periods for assessors |
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.