November 19, 2026:


Suitable habitat for European sardine (Sardina pilchardus) along Iberia’s Atlantic coast could expand by roughly two-thirds by mid-century, with gains concentrated in Galicia, northern Spain and the Gulf of Cádiz, smaller gains near Peniche and localized losses between Lisbon and the Algarve, according to a study published Nov. 18 in Frontiers in Marine Science. The figure describes where conditions would suit the fish, not how many fish there will be, and the authors conclude that water-column structure and upwelling may matter more than temperature.
The paper comes from Tatiana Reis of AQUALOGUS, a Lisbon consultancy, and Lígia Pinto and Marcos Mateus of the MARETEC research group at Instituto Superior Técnico, University of Lisbon. For fisheries managers and seafood buyers in Portugal and Spain, the practical finding is the geography: the model’s gains sit mostly in Galicia and the Gulf of Cádiz, while its losses fall off central and southern Portugal.
The team compared 2040–2069 and 2070–2099 with a 2003–2022 baseline under two climate scenarios. Projected habitat grows 66.2% and 69.3% under SSP2-4.5, and 62.6% and 76.5% under SSP5-8.5, so the mid-century gain is slightly smaller under the high-emissions path and the ranking flips late in the century. The paper reports no area totals, so the percentages cannot be converted into a footprint on a map.
What the model leaned on is the surprise. Summer mixed layer depth, the thickness of the turbulence-stirred surface layer where temperature and salinity are nearly uniform, carried a relative influence of 75.8% and a permutation importance of 94.6%. Bathymetry followed at 10.5% and summer upwelling at 3.4%, while the two sea surface temperature variables contributed 0.3% and 0.1%. Suitability approaches its maximum when the summer mixed layer is shallower than about 11 meters (36 feet).
That does not prove temperature is irrelevant to sardines. The authors note the weak temperature signal may reflect the narrow thermal range in their training data. Correlative models of this kind are poor at extrapolating beyond observed conditions and reveal little about cause, so “not temperature” is partly a statement about what the model saw.
The projected gains are concentrated in Galicia and the Gulf of Cádiz, with both scenarios showing a marked increase across northern Spain and smaller gains near Peniche on Portugal’s west coast. Localized losses run from Lisbon to the Algarve and appear less extensive later in the century. According to the authors’ discussion, projected summer mixed layers in the north and in the Gulf of Cádiz fall below the 11.25-meter (37-foot) threshold, which is consistent with the projected expansion there.
The Gulf of Cádiz is the weakest part of the map. The model had only three occurrence records there, and projected summer temperatures exceed the range associated with sardine hotspots, so the authors caution that the model extrapolates flatly beyond what it was trained on.
For the stock Portugal and Spain share, the authors say a shift from Portuguese to Galician waters could redistribute fishing opportunities, and losses off central and southern Portugal could push fishers farther offshore, raising fuel costs and requiring specialized gear.
Only partly, according to this model. The authors report a partial poleward shift, with strong northern expansion but also gains in the south, so suitable conditions are not confined to higher latitudes.
Earlier work by Lima and colleagues used different approaches. A 2021 range-wide model projected habitat suitability rising on the Iberian Peninsula by the end of the century, and a 9.6% range expansion under the highest of three emissions pathways, extending along Norway’s coast into the White Sea. A 2022 spawning-habitat study found sea surface temperature and distance to the coast were the most influential variables and projected Iberian winter spawning area dropping about 18% under RCP 8.5. The studies differ in life stage, scenario generation and method, so they are not a head-to-head contradiction. They do show that which variable “drives” sardine habitat depends on how the model is built.
The team used boosted regression trees, a machine-learning method that builds a prediction from many simple decision trees, each fitted to the errors left by the trees before it. They trained it on presence records from OBIS and GBIF, which left 168 occurrences after cleaning, and cut 21 candidate variables to 11. Historical conditions came from Copernicus Marine, NASA Ocean Color, NOAA and GEBCO data, and future conditions from CMIP6 projections accessed through NOAA and Zenodo.
The two scenarios bracket a wide range of futures. SSP2-4.5 corresponds to about 2.7°C (4.9°F) of warming by 2081–2100 and SSP5-8.5 to about 4.4°C (7.9°F), though a 2025 assessment found current-policy emissions track closest to SSP2-4.5 or slightly below it.
Because the sardine records are presence-only, the authors generated 168 pseudo-absences at least 50 km (31 miles) from any presence. Pseudo-absences stand in for the available environment, not for places where the fish was proven missing. Wider exclusion distances of 70 km (43 miles) and 100 km (62 miles) produced perfect metrics that the authors judged a sign of overly distant pseudo-absences. Five-fold cross-validation gave the selected model a mean AUC of 0.9931, and the authors themselves caution it may be inflated by the pseudo-absence design. Ecologists have long noted there is no consensus on how to sample pseudo-absences.
Other limits compound it. Future environmental layers are coarser than historical ones, so part of the projected mixed layer shallowing, and therefore part of the habitat gain, may be a resolution effect. The upwelling index was derived differently for the two periods, summer and spring mixed layer depth are highly correlated (r = 0.91), and the authors say the 11 predictors may omit variables such as dissolved oxygen or subsurface nutrients.
The authors stress that habitat expansion should not be read as population growth or sustainable fisheries. Stronger upwelling could bring oxygen-poor water to the surface, and projected declines in primary production could limit food supply. Separate modeling work suggests the Iberian upwelling region may become more vulnerable to hypoxia.
The stock itself is managed on a different clock. ICES advises catches in 2026 of no more than 38,978 tonnes (about 42,966 US tons) under the maximum sustainable yield approach, 25% below the sum of Portugal’s and Spain’s 2025 national quotas. ICES considers the stock to have been in a low-productivity state since 2006, according to a 2019 evaluation requested by Portugal and Spain.
The Marine Stewardship Council says its certificate now covers 317 vessels along the coast from the Bay of Biscay to the Strait of Gibraltar, after the Portuguese certificate was suspended in 2014. The MSC credits the recovery partly to several years of strong juvenile additions, alongside a management plan with catch limits and closed seasons. On March 4, 2026, Spain’s and Portugal’s fisheries officials agreed to set up a working group to renew the Iberian sardine management plan from 2027 onward.
The work was part of the CE2COAST project, funded through the 2019 Joint Transnational Call on Next Generation Climate Science in Europe for Oceans, with support from national agencies including Portugal’s FCT and France’s ANR. The paper discloses that Reis is employed by AQUALOGUS and that ChatGPT helped with graphical design of Figure 1.
The study cannot say. It projects habitat suitability, not fish numbers, and it finds localized losses between Lisbon and the Algarve that shrink in the later period, alongside gains elsewhere on the Iberian coast. Stock size depends on recruitment and catch limits, which ICES and the MSC track separately.
Not in this model, where the temperature variables each carry under 1% of the influence. The authors say that may reflect the limited thermal range in the training data, so the result leaves the role of warming unresolved rather than ruling it out.
Treat it with caution. The authors flag that it may be inflated by how pseudo-absences were generated, and wider exclusion distances produced perfect scores they considered unrealistic.
The model had only three occurrence records there, and projected summer temperatures exceed the range linked to sardine hotspots, so the model extrapolates flatly beyond what it was trained on.