1 Introduction

The oceanographic conditions of the northwestern Mediterranean permit the existence of local, high-productivity areas in an oligotrophic sea (Estrada 1996; Siokou-Frangou et al. 2010). In particular, river outflow from the main two rivers in the area, the Rhône in the Gulf of Lions and the Ebre on the eastern seaboard of the Iberian Peninsula (Fig. 1), allows for high localized productivity in the area where small pelagic species spawn (Palomera et al. 2007). In the area, two winter-spawning small pelagics, sardine (Sardina pilchardus) and sprat (Sprattus sprattus), coexist with two summer-spawning species, anchovy (Engraulis encrasicolus) and round sardinella (Sardinella aurita) (van Beveren et al. 2016; Sabatés et al. 2013). The latter two species show peak spawning activity during the summer months, when pelagic productivity is generally at the lowest levels, but their larvae can feed on high-productivity patches on the surface in the vicinity of the major river outflow or at 50–60 m in depth, where the deep chlorophyll maximum (DCM) occurs (Olivar et al. 2001; Sabatés et al. 2013). In recent decades, good knowledge of the spawning habitats of E. encrasicolus and S. aurita and their variation over time has been gathered in the northwestern Mediterranean (Palomera et al. 2007; Maynou et al. 2020). Their spawning success and consequent recruitment are known to be threatened by hydrological changes caused primarily by increased temperature and decreased primary productivity (Sabatés et al. 2013; Maynou et al. 2014.

Fig. 1
figure 1

Study area. The map shows the location of the ichthyoplankton surveys used to estimate the potential spawning habitat and depth contours at 100 m intervals (grey). The dotted line represents the typical path of the Northern Current

Full size image

Changes in the abundance and distribution of marine species in response to climate change in recent decades have been well documented. In European Atlantic waters, for instance, a distributional shift towards higher latitudes in relation to ocean warming has been shown for fish and invertebrate assemblages in the North Sea (Dulvy et al. 2008; Hiddink et al. 2015). Beyond a certain critical threshold of temperature, salinity or pH, less tolerant species will move away or disappear from areas where they were historically abundant to be replaced by species that were rare or altogether new. In the Mediterranean, for instance, a northward shift of warm water species from the southern Mediterranean has been reported (“meridionalization”: CIESM 2008; Sabatés et al. 2009; Durrieu de Madron et al. 2011), as well as the ongoing substitution of native fish fauna by Lessepsian migrants, particularly in the eastern Mediterranean (Azzurro et al. 2011). In the absence of detailed information on the physiological response of marine organisms to environmental changes (Catalán et al. 2019) or integrated ecosystem models (Peck et al. 2018), empirical models relating environmental covariates with species distribution are useful tools for attempting to anticipate these faunal redistributions (Planque et al. 2011; Elith and Leathwick 2009).

Statistical models of organism distributions seek to associate environmental conditions with species presence or abundance (Planque et al. 2007, 2011). These models, which are also referred to as habitat models (Guisan and Zimmermann 2000), habitat suitability models (Hirzel et al. 2006) or species distribution models (SDM; Guisan and Thuiller 2005), have been used to describe the (current) potential spawning habitat of small pelagic fish (Planque et al. 2011; Giannoulaki et al. 2013). However, these models cannot, by their very nature, capture the true realized spatial distribution of a species, which is usually modulated by other factors, such as biological interactions or fisheries exploitation (Planque et al. 2011). SDMs are nevertheless useful for modelling species responses to environmental changes, such as those projected by climate change models over the coming decades. Small pelagic fish are a good case study for understanding the impact of climate change on marine fishes due to their relatively short life span and the dependence of their populations on spawning/recruitment success, which results in fluctuating populations (Checkley Jr et al. 2009). Spatial and temporal fluctuations in abundance can usually be linked to changes in a few key environmental variables, such as water temperature, salinity, biological productivity or oceanographic structures (Nevárez-Martínez et al. 2001; Planque et al. 2007; Brown et al. 2011).

In the northwestern Mediterranean, important changes in abundance and spatial distribution have already been documented for the dominant four small pelagics (Sabatés et al. 2009; Coll et al. 2018). Palomera et al. (2007) and van Beveren et al. (2016) showed that the abundance of the two target species in the fishery (sardine and anchovy) has decreased over the last few decades along the Spanish and French coasts. Conversely, the abundance of the two low-value species, sprat and round sardinella, has increased since the early years of the current century. Although intense fisheries exploitation is partly to blame for these (and other) ecosystem changes (cf. Taboada and Anadón 2016), the roles of increasing water temperature and decreasing river runoff, or atmospheric drivers, such as the North Atlantic Oscillation or the western Mediterranean Oscillation, have been demonstrated (Martín et al. 2012; Brosset et al. 2015; van Beveren et al. 2016). Other indicators of the influence of climate change on oceanographic conditions in the area are the decreasing biological productivity (cf. Colella et al. 2016, who used satellite-derived chlorophyll-a as a proxy) or the expected increase in stratification and deepening of the DCM (Durrieu de Madron et al. 2011; Coma et al. 2009), but these and other potential drivers have not been used in correlative analysis of the spatial distribution of small pelagics in the area.

Our objectives were to map the future potential spawning habitat (“habitat where the environmental conditions are suitable for spawning”, Planque et al. 2007) of anchovy and round sardinella in the northwestern Mediterranean. We applied a regression model of the realized spawning habitat of the early life stages (eggs and larvae) that was parameterized with empirical hydrographic data and coupled with climate model projections for the middle and final decades of the twenty-first century that were obtained from the regionalized POLCOMS-ERSEM model (Kay et al. 2018).

2 Material and methods

2.1 Study area

The study area covered the Catalan Sea continental shelf and the slope from 40 to 42.5°N latitude and from near the coast to 1000 m depth (Fig. 1). This area of the northwestern Mediterranean is characterized by a narrow continental shelf, which becomes wider in the southernmost area (in the vicinity of the Ebre River delta) and in the north (between the main submarine canyons), as well as, notably, in the Gulf of Lions. The general surface circulation in the region was well established with a main shelf-slope current, the Northern Current, which moves from northeast to southwest along the continental slope (Millot 1999). The Northern Current is subjected to high mesoscale variability that causes oscillations, meandering and eddy generation (Sabatés et al. 2013, 2018). The input of continental water plays an important role in this region. The southern shelf receives a significant river outflow from the Ebre River, while the northern area is affected by the outflow of the Rhône River, which is the largest river in the western Mediterranean basin and outflows into the Gulf of Lions. The northern sector, which is more directly influenced by strong northerly winds, is generally colder than the central and southern parts, and a surface thermal front, which exists between 41 and 42°N, is perpendicular to the coast and roughly coincides with the limit of frequent northerly winds (Sabatés et al. 2009).

The water column structure is characterized by a marked seasonal cycle, with strong stratification in summer, which results in limited vertical water movement and practical depletion of surface nutrients. In summer, primary production is mostly restricted to the DCM, a thin layer at the deepest levels of the photic zone (50–60 m depth, Estrada et al. 1993). In that period, the surface productivity is limited to areas under the influence of river runoff, which are identifiable by lower salinity down to ~ 20–30 m, above the thermocline (Salat et al. 2002).

2.2 Species distribution model

We estimated the future potential spatial distribution of the early life stages of the two summer-spawning small pelagic fish (anchovy and round sardinella) based on a regression model parameterized with ichthyoplankton survey data and in situ environmental data (Maynou et al. 2020) to delineate the habitat where the environmental conditions will be suitable for spawning (Fig. 2). The sampling stations are shown in Fig. 1, and field sampling is described in Annex 2 of the Electronic Supplementary Material (ESM) (see Sabatés et al. 2018 for full details). The regression model is a generalized additive model (GAM) that relates the abundance of anchovies and the presence of the early life stages of round sardinella to surface water conditions (temperature, salinity, chlorophyll-a) in the summer months with peak spawning (June and July). These variables are good descriptors of the spawning habitat and larval environment of both species in the study area (Palomera et al. 2007; Maynou et al. 2014). Briefly, Maynou et al. (2020) estimated the spatial distribution of the early life stages of anchovy and round sardinella from the ichthyoplankton survey data obtained in three summer sampling programmes, which occurred in 1983, from 2003 to 2004 and from 2011 to 2012, where ichthyoplankton were sampled simultaneously with hydrographic data on a grid of stations (Fig. 1). In the case of anchovy, the regression GAM model is a predictor of the abundance (n × 10 m−2) of eggs and larvae and uses a quasi-Poisson distribution function and logarithmic transformation of the dependent variable. For round sardinella, due to the high number of absences in the data, only the probability of occurrence was modelled based on a GAM model with a binomial distribution function and logit transformation. The model coefficients were reproduced here in Table S1. Note that the coefficients of the spline smoother that were larger than 1.0 represent non-linear relationships between the environmental predictor and the outcome. The coefficients equal to 1.0 in the round sardinella egg model (chlorophyll-a and salinity) indicate linear relationships at the scale of the predictor. A detailed interpretation of the partial effects of the predictors is given in Maynou et al. (2020), but in summary, the model coefficients show that anchovy eggs and larvae are more abundant at water temperatures between 22 and 24 °C, salinities of 37.25 to 37.75 and chlorophyll-a concentrations of 0.14–0.22 mg m−3. The presence of round sardinella eggs was enhanced by chlorophyll-a higher than 0.20 mg m−3 and salinity lower than 37.5, while the effect of temperature reached a maximum at approximately 25 °C. For round sardinella larvae, chlorophyll-a was not retained in the model, and salinity had an effect comparable to that on the eggs, while the effect of temperature was positive in the range 24–27.2 °C (maximum temperature observed).

Fig. 2
figure 2

Procedure to derive the spatial distribution models (SDM) of early life stages of anchovy (Engraulis encrasicolus) and round sardinella (Sardinella aurita) from ichthyoplankton surveys, local regression models (GAM) and projections based on future climate change scenarios (IPCC scenarios). The procedure to correct the biases of the biogeochemical model is indicated (see methods)

Full size image

2.2.1 Biogeochemical model

The environmental data (sea surface temperature, SST; sea surface salinity, SSS; and sea surface chlorophyll-a; CHL) in the 2006–2015 (referred to herein as the 2010s or current environmental conditions), 2041–2060 (referred to as the 2050s) and 2080–2099 (referred to as the 2090s) periods were obtained from the POLCOMS-ERSEM modelling system (Holt et al. 2012; Kay and Butenschön 2018).

POLCOMS (the Proudman Oceanographic Laboratory Coastal Ocean Modelling System, Holt and James 2001) is a 3D physical model able to capture the oceanographic dynamics of both the deep ocean and the continental shelf. ERSEM (European Regional Sea Ecosystem Model) is a complex marine biogeochemical model that includes bacteria, four phytoplankton and three zooplankton functional groups, variable carbon to chlorophyll-a ratios and independent nutrient pools for carbon, nitrogen, phosphorous and silicate (Holt et al. 2010; Butenschön et al. 2016). The POLCOMS-ERSEM coupled model framework was regionalized and used at a 10-km spatial resolution and for 40 sigma layers in the NE Atlantic and Mediterranean basins. Subsequently, the coupled model was subregionally validated with surface data (Kay and Butenschön 2018; Ciavatta et al. 2019).

Within the frame and efforts of an EU project (CERES; https://ceresproject.eu/), projections were made for the Representative Concentration Pathway RCP4.5 and RCP8.5 greenhouse gas scenarios using data from the global climate model MPI-ESM-LR (http://www.mpimet.mpg.de/en/science/models/mpi-esm.html) and downscaled to a regional climate model. RCP4.5 and RCP8.5 represent moderate and high greenhouse gas concentrations that increase radiative forcing on Earth by 4.5 W m−2 and 8.5 W m−2 at the end of the twenty-first century, respectively (Alexander et al. 2018). The model was provided by the Plymouth Marine Laboratory (PML, UK) (Kay et al. 2018). The projections of the riverine discharge and nitrogen and phosphorus loadings used the E-HYPE v3.2 with the same global climate model and a business-as-usual nutrient scenario. The river models were made available by the Swedish Meteorological Office (SMHI) (https://www.smhi.se/en/research/research-departments/hydrology/hype-1.7994). The coupled model was run using both scenarios for the 2006–2099 period, and weekly outputs were saved for the present study.

For each of the three environmental variables (SST, SSS, CHL), the climatological values corresponding to June–July were extracted over a rectangle covering the study area (bounded by 0.4–4°E and 40–42.5°N) for 2006–2015 (representing the 2010s decade), mid-twenty-first century (2041–2060, coded as the 2050s) and the end of the twenty-first century (2080–2099, coded as the 2090s). In addition, the standard error of the mean of the model predictions was calculated for each grid cell. We chose a 10-year portion of the data for the “present-day” period (i.e. the decade 2010s) starting in 2006 because we had observational data of the environmental variables and early life stages for the summers of 2011 and 2012 from Fishjelly cruises (Sabatés et al. 2018), which bracketed the middle of the 2006–2015 period. This period served both as a baseline for comparison with future changes and to compare the present-day projections of the biogeochemical models with actual observations during the ichthyoplankton surveys.

2.2.2 Bias correction for the projections

POLCOMS-ERSEM model reasonably captured the main oceanographic features of the study area (Sabatés et al. 2013, 2018; Kay et al. 2018), but the SST was generally overestimated in the lower range (below 21.5 °C) and underestimated in the upper range (above 23 °C) (ESM Annex 1). The SSS was generally overestimated, particularly for the percentiles above 50% (approx. 38.75 psu), but the bias decreased for the percentiles above 90% (ESM Annex 1). The CHL represented the worst case of mismatch between field observations from the 2011–2012 cruises and the POLCOMS-ERSEM model. Even if the general picture of high chlorophyll-a in the Ebre delta can be detected, the values produced by the POLCOM-ERSEM model were 2–3× higher than observed, and the bias actually increased with increasing values of chlorophyll-a (ESM Annex 1). The model predicted high values of chlorophyll-a in the northern half of the study area and in the Gulf of Lions that were not observed in summer. Note that the bias problems were practically identical under the RCP4.5 and RCP8.5 scenarios.

The values of the environmental variables were corrected for bias by means of quantile linear transformation (Gudmundsson et al. 2012, qmap library in R v. 3.5.0) of the original values to match the data distribution of the observed in situ values during the ichthyoplankton sampling surveys in the summers of 2011 and 2012 (Sabatés et al. 2018) (ESM Annex 1). For each variable and grid cell, the difference between the original POLCOMS-ERSEM 2006–2015 projection (“current”) and the bias-corrected map was computed (“delta values”, following the quantile linear transformation method in Gudmundsson et al. 2012). These delta values were subtracted from each corresponding map cell representing the future conditions (2050s and 2090s).

2.2.3 Projection of eggs and larvae

For each cell in the study area, the GAM developed by Maynou et al. (2020) was applied to the bias-corrected climatological environmental variables (SST, SSS, CHL) to compute the value of the eggs or larvae of anchovy and round sardinella, which were averaged over each period, for the current (2010s) and future (2050s, 2090s) conditions (Fig. 2). To determine the error in the estimate, the standard error of the mean (S.E.M.) was computed as follows: (i) for each environmental variable and grid cell, 1000 values were randomly produced from the climatological and standard deviation values; (ii) for each grid cell, the GAM predictor was applied over the 1000 values, yielding 1000 estimates of S.E.M.; (iii) the average of the S.E.M. values was used to produce maps of the precision of the estimates. The computations to project the habitats of the early life stages of both species were carried out with functions in the libraries ncdf4, HiClimR, mgcv and qmap in R (v. 3.5.0).

3 Results

3.1 Projection of environmental variables

3.1.1 Current conditions (2006–2015)

ESM Annex 1 shows the spatial distribution of the climatological values of the three environmental variables that were used to define the small pelagic potential habitat, corresponding to current conditions, which are representative of the 2010s, in the two projection scenarios. Overall, the bias-corrected maps for the current conditions reproduced the average environmental gradients observed during the cruises reasonably well (Sabatés et al. 2013, 2018). The bias-corrected maps showed a south to north gradient of temperature, with the colder temperatures in the northern area (approximately 20–21 °C) but also in the waters influenced by the Ebre River outflow (colder waters of continental origin). Temperature was predicted to be slightly lower in the northern sector under the RCP8.5 scenario. The bias-corrected salinity maps showed a clear shallow-deep gradient with the lowest salinities over the southern area and in the northernmost part (values approximately 37.4–37.6) in both areas influenced by freshwater of continental origin (from the Ebre and Rhône, respectively), with little difference between scenarios. With regard to the surface chlorophyll-a, the values of the bias-corrected maps in the 2010s were projected to be very low (typically < 0.1 mg m−3) everywhere, corresponding to a summer situation in Mediterranean waters with very low surface productivity, except in the southern area influenced by outflow of the Ebre River (0.25–0.3 mg m−3) and near the coast in the northernmost part influenced by the Rhône River outflow. In the RCP8.5 scenario, the bias-corrected chlorophyll-a was overall predicted to be lower than in RCP4.5.

3.2 Future conditions (2050s, 2090s)

The environmental conditions derived from the biogeochemical model for RCPs 4.5 and 8.5 in the central and final decades of the twenty-first century are given in Fig. 3. Under RCP4.5 (Fig. 3a, b), temperatures are projected to increase progressively overall in the 2050s and 2090s, but the north-south gradient is expected to remain stable over time, with temperatures lower than 22 °C to the north of 41.5°. The salinity would not vary markedly between the 2010s (ESM Annex 1) and 2050s, but it would show a significant increase in the 2090s (Fig. 3), likely as a consequence of reduced precipitation. The pattern of surface chlorophyll-a is projected to remain stable but to reach higher levels in the northernmost boundary of the study area, which is under the influence of Rhône riverine waters (right panels in Fig. 3). Note that the biogeochemical model projects high chlorophyll-a values overall in the northern part of the study area, as mentioned above, which contrasts with the observed current conditions of low surface productivity in the summer months. The quantile regression correction proved to be insufficient to correct for such bias.

Fig. 3
figure 3

Bias-corrected projections of environmental conditions (SST: sea surface temperature, SSS: Sea surface salinity, CHL: chlorophyll-a) under scenarios RCP4.5 (a, b) and RCP8.5 (c, d) for the decades 2041–2060 (2050s) and 2080–2099 (2090s)

Full size image

In RCP8.5 (Fig. 3c, d), the surface temperature was projected to follow a similar spatial pattern but with higher absolute values than under RCP4.5 (cf. Fig. 3a, c). For the end of the century, surface temperatures below 22 °C are expected to be found only northwards of 42°N. Salinity in 2050 would be overall higher than present, except in the northernmost continental shelf where values closer to the current climatological average are expected. In 2090, salinity increases significantly in the northern half of the study area but is projected to be relatively low offshore of the continental shelf of the Ebre delta. Under RCP8.5, chlorophyll-a is expected to show lower values than the current climatological averages in the southern half of the study area, but as in RCP4.5, the chlorophyll-a values in the northern area appear unrealistically high, and the suspected bias could not be successfully corrected with the quantile regression model.

The boxplots in Fig. 4 summarize the range of projected values of the environmental variables. Temperature projections under the two scenarios differ markedly, although an overall increase is predicted, higher for RCP8.5 than for RCP4.5, as can be expected. Salinity is expected to reach overall higher values in RCP4.5 than in RCP8.5 in the 2090s. Chlorophyll-a is projected to be higher in any future scenario than current bias-corrected values of approximately 0.1 mg m−3, although scenario 4.5 systematically projects higher chlorophyll-a than RCP8.5. In the RCP4.5 scenario, chlorophyll-a would reach its highest value in the 2050s and decrease in the 2090s, while the average chlorophyll-a under RCP8.5 would be closer to the present values. Note, however, the large variability of the chlorophyll-a projections for both future scenarios.

Fig. 4
figure 4

Boxplots with 0.25, 0.50 (median) and 0.75 quantiles of the distribution of values of environmental variables forecast in the study area. Whiskers represent the minimum and maximum values forecast. All data are averages for the June and July months. Key: 2010s: model projection 2006–2015 data; 2050s and 2090s: forecast values for the middle (2041–2060) and final (2080–2099) decades of the twenty-first c

Full size image

3.3 Projections of early life stages

3.3.1 Current conditions (2006–2015)

ESM Annex 2 shows the observed spatial distribution of anchovy and round sardinella early life stages during the 2011–2012 summers from field ichthyoplankton sampling. Anchovy eggs were mainly located over the Ebre shelf and in the northern part of the area, while larvae showed a wider distribution being more abundant in the northern half of the study area. Note differences in the spatial distribution between 2011 (June) and 2012 (July): In the latter, the early life stages were located more offshore of the Ebre River continental shelf, and the two northernmost transects were not sampled. Round sardinella showed a patchier distribution, eggs were preferentially located very close to the coast, and larvae showed a wider distribution extended offshore.

Figure 5 shows the maps of the distribution of early life stages for anchovy and round sardinella in the baseline period (2010s) under the two RCPs. In general, the projections of the spatial distribution of early life stages of both species are not markedly different between scenarios in the baseline period. Eggs and larvae of anchovy are predicted to be more abundant at present in waters offshore of the Ebre River delta, the central coast and offshore of the northern sector. Comparing the projections in Fig. 5 with the observations of anchovy eggs and larvae during the summer 2011–2012 cruises (ESM Annex 2), we observe that the general distribution pattern is well captured by the models, except in the grid cells near the coast in the Ebre delta and adjacent to the Gulf of Lions, where abundance is projected to be very low. Eggs and larvae of round sardinella in the field samplings showed a high degree of patchiness but were mostly localized in the southern half of the study area (ESM Annex 2). In the model, round sardinella eggs are projected to be more abundant immediately adjacent to the Ebre River delta, while the habitat for larvae extends further northwards to approximately 41.5°N latitude (Fig. 5).

Fig. 5
figure 5

Suitability maps for anchovy and round sardinella early life stages in the 2010s, computed by combining projected bias-corrected environmental variables (Electronic Supplemental Material, Annex 1) with the statistical spatial distribution models based on GAMs (Table 1)

Full size image
Table 1 Estimated non-parametric coefficients of GAM models relating environmental variables to the abundance of anchovy or presence of round sardinella early life stages (from Maynou et al. 2020)
Full size table

3.3.2 Future distribution (2050s, 2090s)

Under RCP4.5, the abundance of anchovy eggs is projected to increase from approximately 400 × 10 m−2 estimated for the 2010s to be over 500 × 10 m−2 (Fig. 6a). In RCP8.5, the results show that egg abundance would increase in the 2050s but decrease to values slightly lower than the current values for the 2090s (Fig. 6a). However, under both future scenarios, the abundance of larvae would be lower than at present, suggesting that even if conditions for spawning can remain constant or even improve in the future, larval habitat is likely to shrink (Fig. 6b). In the case of round sardinella, the probability of occurrence of both eggs and larvae is expected to have a large increase in both scenarios.

Fig. 6
figure 6

Boxplots with 0.25, 0.50 (median) and 0.75 quantiles of the distribution of values of environmental variables projections in the study area. Whiskers represent the minimum and maximum values projections. All data are averages for the June and July months. 2010s, 2050s and 2090s: forecast values for the current (2006–2015), middle (2041–2060) and final (2080–2099) decades of the twenty-first c

Full size image

The maps of the suitability differences between the 2050s and 2010s and the 2090s and 2010s under RCP4.5 and RCP8.5 are shown in Figs. 7 and 8, respectively (ESM Annex 3 for projected absolute values). Under RCP4.5, the potential spawning habitat of anchovy, as measured by the distribution of eggs, was not projected to vary significantly spatially, although higher occupancy is projected for the continental shelf immediately adjacent to the Ebre delta and relatively lower densities in the northern part of the area (Fig. 7). However, a reduction in the suitability of the habitat in the southern area, which is offshore of the Ebre delta, is projected for larvae. This was particularly accentuated under RCP8.5 for the 2090s (Fig. 8). Under RCP8.5, the projections for larvae suggest higher abundance on the northern half (north of 41.5°N) of the study area than those projected by RCP4.5. The difference maps for the anchovy in Figs. 7 and 8 show how habitat suitability would increase in areas adjacent to the coast (in shallower water, especially near the Ebre River delta). This effect is even clearer for larvae, which would actually increase only in the coastal waters immediately adjacent to the Ebre delta and in the northernmost part.

Fig. 7
figure 7

Suitability difference map for anchovy (n × 10 m−2) and round sardinella (probability of presence) early life stages under RCP4.5 between the mid or final decades of the twenty-first century and the 2010s

Full size image
Fig. 8
figure 8

Suitability maps (corrected for environmental bias) for the projected difference for anchovy (n × 10 m−2) and round sardinella (probability of presence) early life stages under RCP8.5 between the mid or final decades of the twenty-first century and the 2010s

Full size image

In the case of round sardinella, the figures in ESM Annex 3 show that the probability of finding eggs and larvae is very low north of approximately 41.75°N, but there are important differences between the projections for eggs or larvae. Under both scenarios, eggs are projected to distribute northwards well beyond the current range around the Ebre delta continental shelf (cf. ESM Annex 3 and Fig. 5). In the 2090s under RCP8.5, round sardinella eggs are projected to be found even in the Gulf of Lions area (but note the very low probability to the south and west of the Ebre continental shelf). The probability of finding larvae of round sardinella will not vary spatially between either scenario and the current conditions (i.e., from approximately 40° to 41.75°N), but the probability values will be significantly higher than 0.5 in many areas, whereas at present, the probabilities higher than 0.5 were only estimated for the areas immediately south and west of the Ebre River.

The projected spatial error of the mean estimate (ESM Annex 4) suggests higher uncertainty in the offshore stations and to the south of the study area, particularly for the anchovy larvae and in the 2090 scenarios. For round sardinella, the standard error of mean occurrence was generally low (below 0.15), suggesting that the results had high confidence.

4 Discussion

4.1 Biogeochemical projection model

The POLCOMS-ERSEM bias-corrected model projections showed increased SST under both scenarios; these changes would be larger under RCP8.5 than under RCP4.5 and reaching higher values by the end of the twenty-first century (Fig. 4a). The projected increases were naturally in accordance with the global ensemble models because this biogeochemical model is based on a regionalization of the CMIP5 projections, which suggest an average SST increase of approximately 0.4 °C per decade and identify the Mediterranean Sea as the European Sea with the most extreme projected warming (Alexander et al. 2018). However, this temperature increase is projected to show important spatial differences, with lower increases in the northern half of the study area, close to the Gulf of Lions, which agrees with the historically different thermal regimes at that spatial scale during the summer months (Pastor et al. 2018). Salinity (bias-corrected) is expected to increase compared to the present conditions under both scenarios and more acutely towards the end of the century for RCP4.5, particularly offshore (Fig. 4b). The most recent reanalyses of the climatic trends in salinity show an increase in the area at different depths (Iona et al. 2018), but the sources of uncertainty associated with these projections (including river discharges) suggest that they should be interpreted with caution. The projections of surface chlorophyll-a presented even higher uncertainties than salinity for our study area. Although the (bias-corrected) POLCOMS-ERSEM projections predicted reasonable values (i.e. consistent with historical observations) for the continental shelf around the Ebre delta and in the central part of the area, the projections of high chlorophyll-a concentrations for the northern part (north of approximately 41.50°N) seem unrealistically high. This has been identified as being due to the excessive riverine nutrient load modelled for the Rhône River (Ciavatta et al. 2019). The results of other modelling efforts (e.g. Macias et al. 2015) suggest that the western Mediterranean should become more oligotrophic in future decades due to increased stratification (strength and duration), and Kay et al. (2018) already recognize a positive bias of POLCOMS-ERSEM in chlorophyll-a projections. Our attempt to correct for bias in the northernmost part of our region with the help of observational data for the summers of 2011 and 2012 solved the problem for the 2010s projections but not for the 2050s or 2090s.

4.2 Potential spawning habitat

The potential spawning habitat of the anchovy, as indicated by the distribution of anchovy eggs, is projected to decrease importantly in the traditional spawning grounds of the Ebre River continental shelf. It is known that this area receives inflow of low saline and productive surface runoff waters, which offer favourable conditions for spawning and larval survival (Palomera et al. 2007; Sabatés et al. 2013). Therefore, it is to be expected that under the climate change scenarios examined here (RCP4.5 and 8.5), higher salinity and lower productivity around the Ebre delta resulting from a decrease in runoff water inputs will likely decrease the spawning habitat suitability of anchovy. Under both future scenarios, the abundance of larvae would be lower than at the present time, suggesting that even if conditions for spawning could be maintained in the future, larval habitat is likely to shrink. In this regard, during the successive heat waves affecting southwestern Europe in summer 2003 (Schär and Jendritzky 2004), which resulted in exceptionally high sea water temperatures, the abundance of anchovy eggs and larvae was much lower than during the years with non-extreme conditions (Maynou et al. 2014). In the northern part of the area, the low abundance of anchovy eggs and larvae projected by the model under both scenarios is initially surprising and points to the limitation of extrapolating SDM models ignoring dynamic processes. In our case, a projection model taking into account changes in the potential spawning habitat of anchovy in the Gulf of Lions and transport of larvae towards the south would provide a more realistic picture of the future distribution of anchovy larvae in the northernmost part of our study area. It must be considered that the high concentrations of anchovy larvae that have been regularly found in the northern area are due to, in addition to local production, their transport by the Northern Current, which flows southwards along the continental slope (Sabatés et al. 2007). These larvae are advected by the current from the northern spawning ground in the Gulf of Lions and aggregated by mesoscale anticyclonic eddies generated by instabilities in the current (Sabatés et al. 2007, 2013; Ospina-Alvarez et al. 2015). Hence, the above considerations suggest that the projected low abundance of anchovy larvae in the northernmost part of the area was not due to temperature, salinity or the (unbiased) chlorophyll-a values projected by the model but rather due to the absence in the model of larval transport due to currents and phenomena of mesoscale dynamics. Combining biogeochemical models with suitably calibrated larval transport models would help to clarify the importance of larval transport and connectivity between spawning and recruitment locations (Lacroix et al. 2018). The number of larvae advected to a given area results from the complex interaction between temperature, current direction and speed and the trophic resources in the dynamic environment surrounding larvae. In simulation models of climatic change, the net result appears highly variable and species-specific, with some studies predicting increased larval duration and reduced recruitment success (Lacroix et al. 2018), while other research results highlight the role of larval retention for successful recruitment and population persistence (Patti et al. 2020).

In the case of round sardinella, the probability of encountering eggs or larvae is projected to increase under both scenarios, and more so under RCP8.5 conditions. The effect was more noticeable for larvae than for eggs. This trend was related to the temperature increase favouring this thermophilic species (Sabatés et al. 2006). Note, however, that under RCP4.5, the highest probability of encountering larvae was in the mid-century projection, while in RCP8.5, the probabilities were not significantly different between the mid and the late twenty-first century. It is likely that an increase in salinity combined with decreasing primary productivity will decrease the northward expansion of this species, which has strictly coastal spawning grounds. In any case, our modelling results were consistent with the increasing presence and abundance of round sardinella (a tropical to subtropical species that is distributed globally) in the area, linked with progressive sea water warming (Palomera et al. 2007; Sabatés et al. 2009; Maynou et al. 2014; van Beveren et al. 2016).

4.3 Study limitations and future prospects

The modelling results presented here show a high degree of uncertainty because they attempt to capture only the statistical relationship between the surface water conditions and the abundance or presence of early life stages of fish. The surface variables considered have been proven to be good descriptors of the spawning habitat and larval environment of both species in the area (Palomera et al. 2007; Sabatés et al. 2013; Maynou et al. 2014). Coupling our spatial distribution model derived from present-day conditions into the future environmental conditions projected by the POLCOMS-ERSEM biogeochemical model can only provide an estimate of the average values of the potential fish habitat, without adequately considering the transient climatic phenomena or mid- to low-frequency forcing by climatic oscillations, such as the NAO or the AMO, which are known to significantly affect the climate of European waters (Alexander et al. 2018). Other important factors determining the spatial distribution (both horizontal and vertical) of fish eggs and larvae, such as mesoscale phenomena or the vertical structure of the water mass (Bakun 2006; Olivar et al. 2010; Ospina-Alvarez et al. 2015), could not be incorporated into the analysis due to the lack of synoptic data, but as mentioned before, larval transport and mesoscale dynamics have been shown to be important factors in explaining the spatial distribution of early anchovy life stages. Future developments in this field would require the incorporation of dynamic models to predict the spatial distribution of fishes, with the caveats and difficulties already identified by Checkley Jr et al. (2017).

In our modelling approach, the values of the environmental variables were taken as accurate, with no consideration of the possible uncertainty associated with the estimation of these variables in the biogeochemical model(s). Naturally, incorporating biogeochemical model uncertainty and structural uncertainty into the spatial distribution model would amplify the variability in the estimations (Payne et al. 2016). To this end, there is a need to create (i) large calibration datasets for biogeochemical models in the Mediterranean by gathering in situ observations that help validate the performance and skills of coupled biogeochemical models in three dimensions and through time, and (ii) ensemble models for the Mediterranean; that is, making available to the research community a set of comparable models aiming to capture the same oceanographic processes.

More accurate and precise projections of the spatial distribution of marine fishes will be obtained by considering mechanistic/physiological, or density-dependent effects, among other factors (Checkley Jr et al. 2017), but we should not forget that past or current conditions may not repeat in the future (“non-ergodicity” and other limits to predictability; Planque 2016). Other aspects, such as changes in the phenology of spawning linked to both temperature and plankton production, are expected and are typically estimated on the order of a shift of 1–2 weeks under RCP8.5 by the end of the century (e.g., Asch et al. 2019). Given the protracted spawning season of these species, changes in phenology would probably not affect our main projections of changes in the suitability of the spawning area during the June–July months used in the model. However, we cannot overlook the potential effects of prolonged heat waves or other extreme weather anomalies (Schär and Jendritzky 2004), which could have a more pronounced impact on the interannual variability of the reproductive success of these species.