Climatic Change

, Volume 119, Issue 1, pp 95–109

Modelling the impact of climate change on Pacific skipjack tuna population and fisheries


    • CLS, Space Oceanography Division
  • Inna Senina
    • CLS, Space Oceanography Division
  • Beatriz Calmettes
    • CLS, Space Oceanography Division
  • John Hampton
    • Secretariat of the Pacific Community
  • Simon Nicol
    • Secretariat of the Pacific Community

DOI: 10.1007/s10584-012-0595-1

Cite this article as:
Lehodey, P., Senina, I., Calmettes, B. et al. Climatic Change (2013) 119: 95. doi:10.1007/s10584-012-0595-1


IPCC-type climate models have produced simulations of the oceanic environment that can be used to drive models of upper trophic levels to explore the impact of climate change on marine resources. We use the Spatial Ecosystem And Population Dynamics Model (SEAPODYM) to investigate the potential impact of Climate change under IPCC A2 scenario on Pacific skipjack tuna (Katsuwonus pelamis). IPCC-type models are still coarse in resolution and can produce significant anomalies, e.g., in water temperature. These limitations have direct and strong effects when modeling the dynamics of marine species. Therefore, parameter estimation experiments based on assimilation of historical fishing data are necessary to calibrate the model to these conditions before exploring the future scenarios. A new simulation based on corrected temperature fields of the A2 simulation from one climate model (IPSL-CM4) is presented. The corrected fields led to a new parameterization close to the one achieved with more realistic environment from an ocean reanalysis and satellite-derived primary production. Projected changes in skipjack population under simple fishing effort scenarios are presented. The skipjack catch and biomass is predicted to slightly increase in the Western Central Pacific Ocean until 2050 then the biomass stabilizes and starts to decrease after 2060 while the catch reaches a plateau. Both feeding and spawning habitat become progressively more favourable in the eastern Pacific Ocean and also extend to higher latitudes, while the western equatorial warm pool is predicted to become less favorable for skipjack spawning.

1 Introduction

The oceanic fisheries of the tropical Pacific Ocean are dominated by skipjack Katsuwonus pelamis, yellowfin Thunnus albacares, bigeye tuna T. obesus and albacore T. alalunga, which together represent >90 % of the world catch taken by industrial fleets, and 70 % of the estimated global tuna catch, just above 4 million tonnes in 2010 (Lehodey et al. 2011a). Skipjack dominates the total catch of tuna and almost all the catch of this species is taken by surface fishery (especially purse seine) together with young yellowfin tuna. The subsurface longline fishery targets mature bigeye and yellowfin tuna in equatorial waters. According to the Western and Central Pacific Fisheries Commission (WCPFC) and the Inter-American Tropical Tuna Commission (IATTC), these tuna resources are approaching maximum sustainable yield and in some cases (e.g., bigeye) exceeded this limit.

Even if they have developed elaborated thermoregulation mechanisms (Brill 1994) tuna are strongly influenced by their temperature environment. Other variables like dissolved oxygen, currents and prey concentration are also of key importance. For instance, the El Niño Southern Oscillation (ENSO), known to change drastically the oceanic conditions of the tropical Pacific Ocean, strongly affects migration, abundance and catches of tuna (Lehodey et al. 1997; Lehodey 2001; Senina et al. 2008). The projection of the future climate in the tropical Pacific Ocean, based on a multi-model mean, indicates a “weakshift towardsconditions whichmay bedescribed as‘El Niño-like’,with SSTsin thecentral andeastern equatorialPacific warmingmore thanthose inthe west” (Solomon et al. 2007). Thus, while the fisheries are setting the level of exploitation to reach the maximum sustainable yield, it becomes urgent to assess the likely effects that climate change itself can have on tuna populations.

Projected changes under one of the worst (A2) emission scenario defined for the IPCC fourth assessment report (Nakicenovic et al. 2000) forecast an overall temperature increase in the surface tropical Pacific Ocean of 0.7–0.8 °C by 2035 relative to 1980–1999; and by 2.5–3.0 °C in 2100 (Ganachaud et al. 2011). Sea surface temperatures in the central and east equatorial Pacific are expected to warm more than those in the west. The size of the western equatorial Warm Pool is projected to increase by ~250 % by 2035 and by 800 % under A2 in 2100. ENSO events are projected to continue for the remainder of the twenty-first century at least, although there is little agreement among models about the frequency or amplitude of El Niño and La Niña episodes in the future (Ganachaud et al. 2011). With climate warming, productivity at the primary (phytoplankton) level is projected to decrease in the tropical Pacific Ocean (Steinacher et al. 2010) and to propagate in the food web through zooplankton (prey of tuna larvae and juveniles) and micronekton (prey of adult tuna). Changes in ocean circulation, vertical stratification and mesoscale activity will also likely affect spawning and feeding grounds of tuna (Lehodey et al. 2011a, b).

Due to the strong interactions between environment and tuna physiology and behavior, complex responses should result from these numerous environmental changes. To investigate them, we use the SEAPODYM (Spatial Ecosystem and Population Dynamics Model) modeling framework that describes the spatial dynamics of tuna and tuna-like species under the influence of both fishing and environmental effects (Lehodey et al. 2008). This model includes a definition of habitat indices, spawning, movements in the responses to the habitat quality and basin-scale seasonal migrations, accessibility of forage for tunas within different vertical layers (Lehodey et al. 2010a), predation and senescence mortality and its change due to environmental (food competition) conditions. Thus, taken together these mechanisms describe most of the recognized interactions between tuna and the oceanic environment. However, ocean acidification associated with increasing atmospheric CO2 are not yet included in the model due to the lack of knowledge on its potential impacts on tuna physiology. A data assimilation technique based on adjoint code and maximum likelihood estimation is implemented to assist in parameterization using historical fishing data (Senina et al. 2008).

A preliminary study simulating IPCC climate impact on Pacific bigeye tuna (Lehodey et al. 2010b) demonstrated that the approach was possible but highlighted important sources of uncertainty and gaps in knowledge which limited the confidence in these first results. A major issue was a large anomaly detected in temperature fields which may prevent the model to reach convergence or produce spurious correlations. In this second study devoted to the use of IPCC—type models to explore the impact of future climate change on tuna populations, we corrected the temperature fields before running optimization experiments with an application to the most tropical tuna species, skipjack. The results are compared to those achieved without temperature correction and to an optimization experiment using more realistic environmental forcing based on an ocean reanalysis and satellite derived primary production.

2 Material and method

2.1 Modeling approach

The main features of the SEAPODYM model (Suppl. Inf. S1) include i) environmental forcing such as temperature, currents, euphotic depth, primary production and dissolved oxygen concentration, ii) prediction of the temporal and spatial distributions of functional groups of prey, iii) prediction of the spatial dynamics of age-structured predator (tuna) populations, iv) prediction of the total catch and the size-frequency of catch by fleet, and v) parameter optimization based on fishing data assimilation techniques.

The mid-trophic level model (Lehodey et al. 2010a) describes vertical and horizontal dynamics of prey groups. Dynamics of tuna populations are predicted using habitat indices, movements, growth and mortality. The feeding habitat is based on the accessibility of prey groups to tuna. The spawning habitat combines temperature preference and coincidence of spawning with presence or absence of predators and food for larvae. Successful larval recruitment is linked to spawning stock biomass and mortality during the drift with currents. Older tuna can swim in addition to being advected by currents. Food requirement and intraspecific competition indices are computed to adjust locally the natural mortality of cohorts, based on food demand, accessibility to available forage components and biomass of other tuna cohorts (Lehodey et al. 2008, 2010b; Senina et al. 2008).

The first optimization experiment used physical variables (temperature and currents) extracted from the 1980–2008 SODA reanalysis1 (Carton et al. 2000) and primary production and euphotic depth derived from satellite data (Behrenfeld and Falkowski 1997)2. Dissolved oxygen concentration is a climatology from the World Ocean Atlas (Garcia et al. 2010). The simulation covered the time period 1998–2008 for which satellite ocean color data were available. Then IPCC simulation experiments are conducted using environmental forcing predicted from a biogeochemical model (PISCES) coupled to a climate model (IPSL-CM4) as previously described in Lehodey et al. (2010b). Present experiments use a correction of temperature fields. The scenario used was the SRES A2 IPCC scenario for the 21th century, i.e., atmospheric CO2 concentrations reaching 850 ppm in the year 2100, and historical data between 1860 and 2000. In the following text, we will refer to these three different configurations by their physical forcing SODA, IPSL and IPSLc respectively.

Projecting the dynamics of an exploited species inevitably raises the issue of projecting the future fishing effort. Any change in fish distribution will certainly redistribute also the fishing effort based on available resource by region and various criteria like management measures, access to fishing zones, cost of fuel, etc. Thus, predicting fleet dynamics under environmental changes and fish population redistribution is a challenging modeling task by itself. As a first approximation of projected fishing effort, we simply used an average of previous years. We computed the average effort of a relatively long period (1980–2000) in order to limit the bias due to interannual variability (ENSO) that is known to have strong spatial impact on fisheries for this species (Lehodey et al. 1997). The projected catch and resulting fish biomass remain a crude estimate in absence of more elaborate scenario of future fishing effort and management but is of interest to assess the implications of climate change on the future spatial distribution of the stock and catch.

2.2 Fishing data for parameter optimization

The definition of fisheries was based from the best available fishing data sets that have been provided by the WCPFC (through the Secretariat of the Pacific Community), and the IATTC (public domain data) at a resolution of 1° × month or 5° × month. A total of 13 fisheries (S2; Lehodey et al. 2011b) were defined based on fishing gears (pole-and-line, purse seine, longline) and strategies (sets on free school, floating object, dolphin associated…). These data were used to run optimization experiments at Pacific basin-scale and to take into account the fishing mortality due to all fisheries. The optimization approach included both spatially disaggregated catch by fishery (monthly 1° or 5° square) and associated size frequency of catch (quarterly 5° x 5° or 10° x 20° or other). Fishing data of the domestic Philippines-Indonesia fishery were not used for optimization due to too low accuracy. Nevertheless, total catch in this region was used for computation of fishing mortality. Details on the optimization approach can be found in previous studies (Senina et al. 2008; Lehodey et al. 2010b). Compared to the first optimization study of skipjack (Senina et al. 2008), the definition of fisheries has been improved in both western and eastern Pacific Ocean (13 fisheries instead of 6) and new data added, especially size frequency data for the eastern Pacific fisheries. Each fishery is defined by a constant catchability coefficient allowed to change linearly with time, and a size-selectivity function characterizing the efficiency of the gear to catch the species at different sizes. Thus, the optimization approach estimates 3 to 4 parameters by fishery, with 18 other parameters used to describes the spatial population dynamics of the species, including the definition of habitats, movements, larval recruitment and natural mortality.

Model results are evaluated using the fit between prediction and observation and comparison with independent estimates from the last stock assessment analysis conducted by the WCPFC (Hoyle et al. 2011). The fit to catch, catch per unit of effort (CPUE) and size frequency of catch is checked for all fisheries as well as their residuals (Lehodey et al. 2011b). Overall spatial fit between predicted and observed catch for all the fleets over the whole time series is provided by the standard R-squared goodness of fit (Eq. 1). A value of 1 means a perfect fit.
$$ {R_{{ij}}}^2 = 1 - \frac{{\sum\limits_{{tf}} {{{\left( {C_{{ijtf}}^{{obs}} - C_{{ijtf}}^{{pred}}} \right)}^2}} }}{{\sum\limits_{{tf}} {{{\left( {C_{{ijtf}}^{{obs}} - \overline{C}_{{ijf}}^{{obs}}} \right)}^2}} }} $$
where C is catch, pred and obs mean predicted and observed respectively, i and j are the coordinates of the grid cell, t is the time step and f the fishery.
We also compute the relative error (e) in % of predicted catch (Eq. 2).
$$ e = \frac{{\sqrt {{\sum\limits_{{tf}} {{{\left( {C_{{ijtf}}^{{obs}} - C_{{ijtf}}^{{pred}}} \right)}^2}} }} }}{{\sum\limits_{{tf}} {\left( {C_{{ijtf}}^{{obs}}} \right)} }} $$

2.3 Correction of IPSL-CM4 temperature fields

The IPSL-CM4 A2 climate simulation used in a previous analysis (Lehodey et al. 2010b) demonstrated a good seasonal cycle at high latitudes, internally generated interannual variability close to observed ENSO variability in the tropics, a good coherence between changes in the vertical structure, primary production (Schneider et al. 2008; Steinacher et al. 2010) and dissolved oxygen concentration. However, as other complex climate models it includes some biases and anomalies, and in particular a strong cold temperature anomaly was detected in the mid to high latitudes. Such anomalies are partly linked to too coarse resolution in the atmospheric component of the climate model system (Hourdin et al. 2012) which can create a latitudinal shift in the atmospheric circulation that is propagated to the ocean dynamics.

The temperature anomaly of the IPSL-CM4 A2 simulation has been corrected using a climatology (Locarnini et al. 2010) and a weighting function based on latitude, with the objective of keeping unchanged the original prediction in the tropical region but correcting the temperature in higher latitude according to the climatology from observation (Eq. 3). An illustration of the change due to correction can be seen in the supplementary information (S3).
$$ {T_c} = f\left( {lat} \right) \cdot {T_{{cl \_ obs}}} + \left( {1 - f\left( {lat} \right)} \right) \cdot {T_{{cl \_ mod}}} + {T_{{anom}}} $$

Corrected temperature (IPSL corrected)


Temperature climatology from observation (WOA)


Temperature climatology from biased model (average IPSL 1900–2000)


Temperature anomaly (difference between IPSL temperature and Tcl_mod)

$$ f\left( {lat} \right) = {{1} \left/ {{\left[ {a \cdot {e^{{\left( {\left. { - b} \right|\left. {lat + 5} \right|} \right)}}} + 1} \right]}} \right.} $$
with a = 2e11 and b = 0.95

3 Results

All optimization experiments use an identical population structure. It is defined by age (cohorts) with variable time unit allowing us to save computation time. There is 1- month cohort for larvae life stage, one 2-month cohort for juvenile stage, four 2-month cohorts for young fish (before age at maturity) and 12 cohorts for adult stages (one of 2 months, eight of 3 months, two of 4 months and one of 6 months). The last one is a “+ cohort” accumulating older fish. First age of maturity is set to 10 months. All parameters cannot be optimized simultaneously due to antagonistic and correlated processes, e.g., mortality and larval recruitment. Thus, for each configuration a series of simulations was carried out to achieve the optimal parameterization.

3.1 Optimization based on historical fishing period

3.1.1 Optimization with ocean reanalysis (SODA)

This first optimization experiment using the ocean reanalysis SODA and satellite derived primary production for the environmental forcing was carried with a relatively short time series (1998–2007). Nevertheless, it was possible to achieve convergence thanks to the large fishing dataset and this realistic environment.

Optimal spawning temperature (SST) was not estimated and fixed at 29.5 °C, with standard error being 3.0 °C. The optimal temperature for the oldest cohort was estimated to 20.6 °C, however the standard error parameter was large and reached the upper boundary value 4.5 (Table 1). This parameterization agrees well with current knowledge on the species (S4). Skipjack are known to have spawning activity peaking between 26° and 30 °C (Schaefer 2001), and an overall species distribution identified from all historical occurrences between 17–30 °C (Sund et al. 1981). Like skipjack, yellowfin tuna spawns in warm waters (Schaefer 1998) and it is worth to note that laboratory experiments on this latter species allowed identifying an optimal range of 26–31 °C for rapid growth and moderate to high survival in first feeding larvae (Wexler et al. 2011). Estimation of the optimal temperature for the oldest cohort is a key progress compared to previous optimization and is due in large part to the use of a better fishing dataset with data from regions with contrasted oceanographic conditions (e.g., the eastern Pacific) and to the use of a more realistic environmental forcing.
Table 1

Estimates of habitats and movement parameters of previous simulations and the new one using environmental forcings from SODA and IPSL-CM4 before and after correction of temperature fields

Parameters estimated by the model


Senina et al. (2008)






Optimum of the spawning temperature function







Std. Err. of the spawning temperature function







Larvae food-predator trade-off coefficient






Optimum of the adult temperature function at maximum age







Std. Err. of the adult temperature function at maximum age







Oxygen threshold value at ΨO = 0.5








Maximum sustainable speed

B.L. s−1





aFixed; [val = value close to minimum boundary value; val] = value close to maximum boundary value

For the oxygen function, the threshold value was estimated close to 2.5 mL/L (Table 1), above the lethal levels of 1.87 mL/L (50 cm Fork Length) defined from laboratory experiments and below the value of 3.8 mL/L below which skipjack spend less than 10 % of their time (Brill 1994). Estimates of movement parameters are not yet fully satisfactory since the value for maximum sustainable speed (VM) reached the fixed lower boundary (1 body length s−1) while diffusion coefficient for fish movements reached the upper boundary leading to a theoretical maximum diffusion rate of 1,000 nmi−2 d−1. Nevertheless, the fixed value for VM seems reasonable in the view of direct observations from electronic tagging experiments for other tunas (e.g., Lutcavage et al. 2000). Indeed, it is predicted to decrease with size (S4) due to increasing accessibility to forage biomass of deeper layers, and thus decreasing of the horizontal gradients of the feeding habitat index controlling the movement. Average predicted maximum diffusion rates (S4) also remain in the range of average values estimated from conventional tagging data (Sibert et al. 1999).

Reasonable estimates of catchability and selectivity coefficients were obtained when allowing a linear temporal increase in catchability. Overall fit to predicted catch was generally good (S5) with a mean R2 of 0.7 and relative errors in catch below 10 % in all major fishing grounds, this error being high only in few places where catch is really occasional and represents a very small volume compared to the total.

With this new parameterization, the total skipjack biomass for the Western Central Pacific Ocean (WCPO) is estimated to be around 7 million tonnes (Fig. 1), about one million tonnes above the biomass estimates from the last stock assessment study with the model MULTIFAN-CL used by the WCPFC (Hoyle et al. 2011), but below some of the previous highest estimates (Langley and Hampton 2008). The uncertainty in biomass estimates is inherent to all population dynamics models. As this study is focused on the change of spatial distributions and long term trends, the estimates of absolute biomass is less important. In the case of SEAPODYM, overestimation of biomass estimates may be due to the spatial structure of the model and its numerical treatment, as it was generally observed that coarse spatial resolution and/or fishing data used for optimization can lead to too high diffusion especially in the margin of the core habitat of the species. Another source of uncertainty concerns the Philippine-Indonesia region where only poor quality fishing data are available and cannot be used in the optimization (though accounted for in fishing mortality), despite it being a major spawning ground of this species together with the western equatorial warm pool. Indeed, when the central equatorial region (region 3: 20°N-20°S; 170°E −150°W) is considered alone, both model estimates are much closer (Fig. 1).
Fig. 1

Total skipjack biomass and catch estimates in the WCPO. a estimates with SODA (red line), and corrected IPSL (blue and black lines) forcing with actual fishing effort for the historical period and either no fishing (continuous blueline) or fishing effort scenario based on 1980–2000 period (dotted lines), and comparison for the historical period with the last WCPFC stock assessment estimate (MFCL; dotted purpleline). b comparison by region (shown on figure S5) between biomass estimated with SEAPODYM using ocean reanalysis SODA and the last stock assessment (MFCL 2011) of the WCPFC (Hoyle et al. 2011) region 1 = 40°N‐20°N, 120°E‐150°W; region 2 = 20°N‐20°S, 120°E‐170°E; region 3 = 20°N‐20°S, 170°E‐150°W. c total monthly catch (metric tonnes) for historical and projected periods according to medium (average 1980–2000) and high (1.5 times the 1980–2000 average) fishing effort scenarios;12 months moving averages (thick lines) are superimposed

3.1.2 Optimization with climate model (IPSL and IPSLc)

Despite the use of a longer data time series (1976–2000), the first optimization carried out with the IPSL-CM4 climate model forcing was not satisfactory as several key parameters could not be estimated and had to be fixed and some other estimated at values with weak biological significance (Table 1). In particular, spawning success was predicted to be only controlled by temperature (α ~ 0), while the adult feeding habitat seemed constrained by a spurious anti-correlation between temperature preference and oxygen leading to an adult thermal habitat too restricted to warm temperature (Ta = 25 °C; σa = 1.1 °C) and a sensitivity to dissolved oxygen concentration too high (Ô = 5.46 mL L−1) in disagreement with current knowledge on the species as discussed in previous section.

The correction of temperature bias in IPSL-CM4 climate outputs resulted in a better fit to fishing data (S5) and more coherent parameterization. This latter is much closer to the one achieved with the ocean reanalysis (SODA), though as in all other experiments all parameters cannot be estimated together, likely because there is not enough realism in the forcing and a lack of critical data for some life stage (e.g., larvae) or mechanisms (e.g., movement) to achieve optimization of all parameters. However an encouraging result of using the corrected temperature forcing is a successful and realistic estimate of the adult thermal habitat with parameters similar to those obtained with the ocean reanalysis (Table 1).

The overall fit to catch and CPUE data of all fisheries showed reasonable spatial correlation with this corrected forcing (S5). The skipjack total biomass estimate for the WCPO is around 6 million tonnes, i.e. between estimates of the SODA optimization experiment and the MULTIFAN-CL stock assessment model (Fig. 1).

The modeled average spatial distribution of larvae for the recent decade in both simulations using climate model outputs show a large favorable region in the western equatorial Pacific extending eastward in a narrow north subequatorial band. This overall pattern is similar to the one resulting from the experiment using SODA reanalysis (Fig. 2). In the latter however, the distribution is less homogeneously distributed due to the higher resolution but also because of a much narrower standard error of the temperature spawning function (Table 1). For the same period, there are much more marked differences in spatial dynamics and distributions of young and adult fish, and thus total biomass (Fig. 3), though in all cases the core habitat is in the western and central tropical Pacific. The biomass distribution predicted with the first climate model experiment (IPSL) is less convincing given the very large catch occurring west of 165°E between Philippines, Indonesia, Papua New Guinea and Solomon Islands (S5).
Fig. 2

Average spatial distributions of Pacific skipjack larvae density. Comparison of larvae density (ind. m−2) averaged over 10 years at the beginning and the end of 21st Century for simulations based on ocean reanalysis (SODA-Psat), and climate model (IPSL-CM4-PISCES) before (IPSL) and after (IPSLc) correction of temperature fields
Fig. 3

Average spatial distributions of Pacific skipjack total biomass. Comparison of total biomass (g m−2) averaged over 10 years at the beginning and the end of 21st Century for simulations based on ocean reanalysis (SODA-Psat), and climate model (IPSL-CM4 –PISCES) before (IPSL) and after (IPSLc) correction of temperature fields

3.2 Projections under IPCC A2 scenario

The IPCC A2 projection with both uncorrected (IPSL) and corrected (IPSLc) temperature forcing fields provided strongly divergent trends in the spatial distribution of all life stages. This is clearly illustrated in Figs. 2 and 3 with the average larvae density and total biomass distributions predicted for the last decade of the 21st Century. For larvae, both simulations predicted an extension toward higher latitudes but in the second simulation (IPSLc) the favorable spawning ground is also shifted to the central and eastern Pacific (Fig. 2). The average density of skipjack larvae is projected to decrease in the present-day spawning grounds (western equatorial warm pool and Philippines-Indonesia) with the exception of a narrow equatorial band. This trend is essentially driven by the increasing temperature, e.g., while the optimal spawning temperature is 29.5 °C (std. err. 3 °C), the average SST remains permanently above 32 °C after 2075 in the area defined by latitudes 10°N-10°S and longitudes 110°E-180°E.

The resulting distribution in total biomass is also an extension to higher latitudes, and either a strong contrast between an increasingly favorable central-eastern region and the less and less favorable western region (IPSL), or a more diffuse expansion to the eastern Pacific with a shift of the core habitat from the western to the central equatorial region (IPSLc). Given the analysis of optimization results described above, this second simulation seems more reasonable. The sensitivity to projected change in dissolved oxygen concentration was tested with a simulation replacing predicted oxygen concentrations by the climatology. The results did not change, indicating a lack of sensitivity. Thus, the predicted changes in skipjack dynamics over this climate simulation are driven by changes in temperature, productivity and currents.

When projecting the fishing effort based on the average period 1980–2000, the predicted catch was lower than observed for the recent years. This is logically due to the continuously increasing effort since the 1980s that resulted in an average moderate fishing effort compared to the effort deployed in the last years (Fig. 1). Thus, a second projection was conducted based on the same fishing effort data set and a factor of multiplication set to 1.5. In both projections of fishing effort under this A2 scenario, estimated skipjack tuna catches in the WCPO would increase and the stock biomass remains stable until the 2060s (Fig. 1). After this date, the catch remains stable but the biomass starts to decrease, reaching at the end of the Century about one million below the level in 2000 when starting the projection. As previously noted, the fishing scenario is rather crude and this projection cannot be used as a stock status evaluation. However, since the projected fishing effort is constant, it is clear that the biomass decreases in the western equatorial region (Fig. 3) in relation with environmental changes described above.

4 Discussion

Understanding the impact of climate change on tuna populations is linked to our capacity to explain, model and predict the effect of natural variability for the historical period for which we have data. With realistic environmental forcing, the SEAPODYM model used in this study seems relatively robust with a good fit to observed fishing data under the influence of natural variability, especially the ENSO variability. Thus, the parameter optimization approach provides a good framework to measure the progress and evaluate the outputs. However, even though the modeling of spatial population dynamics is based on a limited number of parameters, the task is challenging, especially when using environmental forcings from Earth Climate models.

Ocean prediction from Earth Climate models are not fully comparable for historical periods to results achieved from ocean reanalysis. These models are still coarse in resolution and can produce significant anomalies, e.g., in water temperature as in the IPSL-CM4 model used here. Consequently, an ecosystem or fish population dynamics model parameterized from ocean observation or ocean reanalyses needs to be adjusted to the climate model environment. However, the parameter optimization approach relies on the distribution of historical fishing data under the influence of actual variability, which does not necessarily coincide with the internal variability of the climate model, despite the fact that this variability is close in frequency to the natural variability (e.g., ENSO, PDO). Though these limitations have no consequence when describing projections of climate, they can have direct and strong effects when modeling the dynamics of marine poïkilotherm species.

A constant bias for a given variable over the whole domain of the model can be compensated by a shift in the new estimated values of the associated parameters. However, when this bias occurs only in part of the domain, there is no other choice than to attempt to correct the bias, as it has been demonstrated in this analysis. On the other hand, bias adjustment of models may remove systematic errors in the mean state but assumes the mean state and climate signal are independent of each other. Since each model has its own bias, another approach could be to use ocean forcing from multi-model averaging. This would likely result in a better simulation of current and future climate than any individual model by cancelling non-systematic errors.

Despite all these uncertainties on the Earth Climate modeling, it is encouraging to see that fairly good results were obtained from optimization experiments after correction of temperature fields. Skipjack is a typical tropical tuna species, and not too surprisingly the species is predicted to thrive of climate change at least in the first half of the Century, through a general expansion of its habitat both in the central and eastern Pacific and towards higher latitudes. Besides, this type of habitat expansion is typically observed during strong El Niño events. The positive trend however is stopped in the western central tropical Pacific in the middle of the second half of the Century and biomass then starts to decrease even under a moderate fishing effort scenario, indicating that the environmental conditions are becoming much less favorable in this region (too warm for spawning and decreasing productivity). Given its thermal preference, this species is mainly associated to the epipelagic layer above the thermocline, a tendency that is reinforced with projected increasing vertical stratification. Therefore, it is not surprising to verify that projected change in dissolved oxygen has little impact on skipjack dynamics, since the decrease in oxygen is predicted to occur mainly in or below the thermocline and is limited in the upper layer (Lehodey et al. 2010b).

In the case of a fishing effort equivalent to the effort of recent years, the decrease would certainly become still more striking with a biomass in the WCPO substantially lower at the end of the Century than today. Of course this scenario is linked to the skills of the model to predict a correct biomass for the population. Compared to the WCPFC stock assessment, the biomass estimate with the climate model experiment is only slightly above but show less amplitude in its variability. It is possible that this estimate still decrease in future experiments, especially with the use of higher resolution and more realistic environment. Compared to previous lower resolution simulations (e.g., Senina et al. 2008), the estimates are roughly 20 % and 30 % lower for total and adult biomasses respectively in the SODA experiment. One mechanism to explain this decrease in biomass estimate is linked to a reduced diffusion of biomass and more patchy spatial density distribution, due to a better match between positions of high and low CPUE with fish density.

Therefore, while efforts need to be continued to increase the realism and resolution of environmental forcing, the use of higher resolution fishing data should be envisaged in the optimization approach to improve the parameterization. The access to these data may become problematic however, due to the usual problem of confidentiality of fishing data. Other key data exist that can be added in the optimization process. In particular, assimilation of existing large datasets from conventional and electronic tagging should provide key information for improving estimation of movement and habitat parameters of the species (by cohort). A rapid improvement of fishing data collection in the Philippines and Indonesian regions is also highly desirable both for current management and future experiments with SEAPODYM. This model itself should continue to be enhanced in the definition of its mechanisms (e.g., spawning migration, food competition), and multi-species optimization experiments need to be conducted to explore possible mechanisms in food competition. New developments are also required to add bioenergetic rules to simulate physiological effects associated with climate change impacts. Finally, the sub-model describing the functional groups of micronektonic prey organisms is a key component in SEAPODYM that requires further validation and development, especially by assimilation of acoustic data.

In conclusion, though the model and its predictions will be continuously upgraded, results from past and present studies as well as those coming from regular updates for evaluating the effects of climate change are an important achievement that can be already used, particularly by the Pacific Island Countries to learn to adapt to the inevitable changes that will occur to their fisheries resources in response to global warming. A fundamental challenge for these Countries will be to develop coherent policies for fisheries management at both regional and national levels. The spatially explicit aspect of SEAPODYM allows the evaluation of the impacts of national level policies on the regional scale and vice versa (Bell et al. 2011). With climate change, the fishing impact will definitely remain a key driver of tuna stocks. Simulations with management scenarios based on multi-decadal projections including climate change are needed to assist in the timely provision of information on the sustainability of current and future harvest strategies and their potential impact on tuna populations and oceanic ecosystems. In addition, it would be beneficial in further studies to define more realistic future fishing scenarios based on economical requirements and accounting for spatial change in the biomass.


The authors wish to thank the Ocean Productivity team for providing the SeaWiFS-derived primary production, Peter Williams (SPC) and Michael Hinton (IATTC) for preparing and supplying catch and size composition data, and Laurent Bopp for assisting in the use of IPSL-CM4 and PISCES outputs.

This work was funded partly by the 10th European Development Fund project SCICOFISH (Scientific Support to coastal and Oceanic Fisheries Management in the western and Central Pacific Ocean), the Australian Department of Climate Change and Energy Efficiency and by the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) project (Enhanced estimates of climate change impacts on WCPO tuna). The views expressed herein are those of the authors and do not necessarily reflect the views of their organizations or funding Agencies.

Supplementary material

10584_2012_595_MOESM1_ESM.doc (1 mb)
ESM 1(DOC 1059 kb)

Copyright information

© Springer Science+Business Media Dordrecht 2012