A century of fish biomass decline in the ocean

Abstract

We performed a global assessment of how fish biomass has changed over the last 100 yr, applying a previously developed methodology using ecological modeling. Our assessment built on more than 200 food web models representing marine ecosystems throughout the world covering the period from 1880 to 2007. All models were constructed based on the same approach, and have been previously documented. We spatially and temporally distributed fish biomasses delivered by these models based on fish habitat preferences, ecology, and feeding conditions. From these distributions, we extracted over 68000 estimates of biomass (for predatory and prey fishes separately, including trophic level of 3.5 or higher, and trophic level between 2.0 and 3.0, respectively), and predicted spatial-temporal trends in fish biomass using multiple regression. Our results predicted that the biomass of predatory fish in the world oceans has declined by two-thirds over the last 100 yr. This decline is accelerating, with 54% occurring in the last 40 yr. Results also showed that the biomass of prey fish has increased over the last 100 yr, likely as a consequence of predation release. These findings allowed us to predict that there will be fish in the future ocean, but the composition of fish assemblages will be very different from current ones, with small prey fish dominating. Our results show that the trophic structure of marine ecosystems has changed at a global scale, in a manner consistent with fishing down marine food webs.

Full text

MARINE ECOLOGY PROGRESS SERIES
Mar Ecol Prog Ser
Vol. 512: 155–166, 2014
doi: 10.3354/meps10946 Published October 9
INTRODUCTION
Opinions regarding the state and future of marine
resources differ. Many fear that we are losing ground
in this last frontier on the globe, and that our impact
is so devastating that all fish stocks will be collapsed
by 2048 (Worm et al. 2006), while alternative inter-
pretations of data conclude that conditions are
improving and we are seeing improvements in fish
populations (Worm et al. 2009). Such conflicting find-
ings have, while creating headlines, sparked a hea -
ted discussion within the scientific community and
have created confusion for fisheries managers and
the general public.
In this study, we evaluate how the abundance of
fish has changed in the world ocean over the last
100 yr. Building on a methodology that combines food
web modeling, statistical analysis and geographic in-
formation systems, as well as physical and ecological
global spatial datasets, our study is the first to eva -
luate trends in global fish biomass based on stratifica-
tion of the world’s oceans using food web models.
We used 200 detailed descriptions of ecosystems in
the form of ecosystem food web models to provide
© Inter-Research 2014 · www.int-res.com*Corresponding author: v.christensen@fisheries.ubc.ca
A century of fish biomass decline in the ocean
Villy Christensen1,*, Marta Coll2,3, Chiara Piroddi4, Jeroen Steenbeek3,
Joe Buszowski3, Daniel Pauly1
1Fisheries Centre, University of British Columbia, 2202 Main Mall, Vancouver, BC V6T 1Z4, Canada
2Institut de Recherche pour le Développement, UMR EME 212,
Centre de Recherche Halieutique Méditerranéenne et Tropicale, Avenue Jean Monnet, BP 171, 34203 Sète Cedex, France,
and Institute of Marine Science, ICM-CSIC, Passeig Marótim de la Barceloneta, 37-49, Barcelona 08003, Spain
3Ecopath International Initiative Research Association, Barcelona, Spain
4European Commission - DG JRC, Institute for Environment and Sustainability, Water Resources Unit, Via E. Fermi,
2749 - TP 272, 21027 Ispra, VA, Italy
ABSTRACT: We performed a global assessment of how fish biomass has changed over the last
100 yr, applying a previously developed methodology using ecological modeling. Our assessment
built on more than 200 food web models representing marine ecosystems throughout the world
covering the period from 1880 to 2007. All models were constructed based on the same approach,
and have been previously documented. We spatially and temporally distributed fish biomasses
delivered by these models based on fish habitat preferences, ecology, and feeding conditions.
From these distributions, we extracted over 68000 estimates of biomass (for predatory and prey
fishes separately, including trophic level of 3.5 or higher, and trophic level between 2.0 and 3.0,
respectively), and predicted spatial−temporal trends in fish biomass using multiple regression.
Our results predicted that the biomass of predatory fish in the world oceans has declined by two-
thirds over the last 100 yr. This decline is accelerating, with 54% occurring in the last 40 yr. Results
also showed that the biomass of prey fish has increased over the last 100 yr, likely as a conse-
quence of predation release. These findings allowed us to predict that there will be fish in the
future ocean, but the composition of fish assemblages will be very different from current ones,
with small prey fish dominating. Our results show that the trophic structure of marine ecosystems
has changed at a global scale, in a manner consistent with fishing down marine food webs.
KEY WORDS: Fish biomass · Global distribution · Trends · Ecosystem models · Fishing down
marine food webs
Resale or republication not permitted without written consent of the publisher
Contribution to the Theme Section ‘Trophodynamics in marine ecology’
F
REE
REE
A
CCESS
CCESS

Mar Ecol Prog Ser 512: 155–166, 2014
snapshots of how much life was in the ocean at given
points in time and space. We then evaluated how the
conditions at each point related to environmental
parameters, based on which we developed a regres-
sion model to predict biomass trend over time.
Finally, we used global environmental databases to
predict the spatial distribution of fish biomass. For
the study, we used an established methodology pre-
viously applied to the North Atlantic, Southeast Asia,
and West Africa (Christensen et al. 2003a,b, 2004).
This study allowed us to predict the biomass trends
for higher trophic level fish, i.e. the larger predatory
fish or ‘table fish’, as well as for the lower trophic
level prey fish, such as small pelagic fish (e.g. sar-
dines, anchovies, capelins), which are used mainly
for fishmeal and oil (Pikitch et al. 2013). Local and
regional studies have reported important declines
of large predators in the ocean (Christensen et al.
2003b), including large finfish and sharks (Baum et
al. 2003, Myers & Worm 2003). In parallel, some stud-
ies have indicated that the abundance of forage
fishes and invertebrates may have increased due to
cascading effects caused by decreasing predator
abundance as a result of human exploitation (Car -
scadden et al. 2001, Worm & Myers 2003, Myers et al.
2007, Coll et al. 2013).
Given the recent controversy over whether ‘fishing
down the food web’ is a phenomenon actually occur-
ring in nature (Pauly et al. 1998) or a sampling arti-
fact (Branch et al. 2010) with no or little relation to the
underlying ecosystem structure, our study adds to
the discussion by specifically evaluating how the bio-
mass of high trophic level species has changed rela-
tive to the biomass of low trophic level species.
MATERIALS AND METHODS
Ecosystem models
We built a database with 230 ecosystem models, all
of which were based on the Ecopath with Ecosim
(EwE) approach and software (Christensen & Pauly
1992, Christensen & Walters 2004), currently the
most widely used ecosystem modeling approach in
the marine environment (Coll et al. 2008a, Colléter et
al. 2013, Heymans et al. 2014). EwE integrates a
large body of information from the ecosystem in a
coherent description of aquatic food webs, placing
human activities within an ecosystem context, and
accounts for environmental changes (Christensen &
Walters 2004). For detailed information about the
EwE approach, its main advantages and limitations,
see the literature (e.g. Christensen & Walters 2004,
Plagányi 2007), and www.ecopath.org.
Each of these EwE models was constructed by eco-
system modelers (i.e. fisheries scientists, ecologists or
marine biologists) using local and regional datasets,
usually with the aim of providing a snapshot of an
ecosystem in a given year, and which had been pre-
viously published and well documented in the scien-
tific literature (see the Supplement at www.int-res.
com/articles/suppl/m512p155_supp.pdf). These mo -
dels jointly cover 44% of the world’s ocean surface
(Fig. 1), representing the years between 1880 and
2007 (the number of models in 3 time periods within
those years are indicated in Table 1).
For the analysis, we first used the snapshot Ecopath
models. We evaluated and quality-controlled the
available Ecopath models, eliminating 30 of the can-
didate models from our analysis. As criteria for inclu-
sion, models must have been well documented in the
literature, included biomass information for different
fish groups as inputs, and the modeled food webs
must have accounted for all living components in
traceable detail. For example, we eliminated models
that only specified fish groups as ‘fish’. For each of the
200 remaining ecosystem models, we evaluated
which of its functional groups belonged to the wider
category of ‘exploitable fish species’. We excluded
mesopelagic species, which mainly occur in the deep,
open ocean and which cannot be exploited economi-
cally with existing technology (Valinassab et al. 2007).
We also excluded juvenile fish, which cannot form the
basis of sustainable fisheries (Libralato et al. 2008).
We performed the analysis separately for fish with a
trophic level of 3.5 or higher, and for fish with a
trophic level between 2.0 and 3.0. The higher-trophic
level group includes predatory fish and represents
the larger ‘table fish’ that tend to be used directly for
human consumption. The lower trophic level groups
tend to be ‘forage fish’, a major prey for the table fish
and other marine predators (Cury et al. 2011, Pikitch
et al. 2012). Humans tend to use forage fish mainly for
fishmeal and oil (Alder et al. 2008, Pikitch et al. 2012).
We also excluded invertebrate species from this ana -
lysis since there is less information about them, and
ecosystem models tend to lack information on these
organisms (but see Pauly et al. 2009, Coll et al. 2013).
There were approximately 3000 types (unique
names) of functional groups in the ecosystem models.
Each functional group represents individual species,
life stages of species, or a collection of similar species
(for instance, ‘small pelagic fishes’). We assigned
each of the functional groups to depth categories,
based on information from FishBase (Froese & Pauly
156

Christensen et al.: Fish biomass decline in the ocean
2013), SeaLifeBase (Palomares & Pauly 2013), Aqua -
maps (Kaschner et al. 2013), other online searches,
and general knowledge of the species involved. The
depth stratifications we used were <10, 10−49,
50−99, 100−199, 200−499, 500−999, 1000−1999, and
>2000 m, and did not include habitat type. Each
functional group could be assigned to any number of
depth strata, with most pelagic species (for example)
being assigned to all strata.
Secondly, we used the spatial-dynamic Ecospace
module of EwE (Walters et al. 1999) to spatially dis-
tribute the functional groups within each Ecopath
ecosystem model based on standardized habitat dis-
tributions as described above, using the depth strati-
fication categories. We used a time−spatial dynamic
simulation of 30 yr per Ecopath model. Default values
of the vulnerability parameter were used. In addition
to depth stratification, the models used relative pri-
mary production as input, derived from SeaWiFS
satellite data. The ecosystem models were then dis-
tributed spatially to a global half-degree grid, based
on information in the model descriptions. The distri-
bution by the available ecosystem models (Fig. 1)
highlights the wide coverage of the global ocean. It is
also important to note that for many locations, several
models were available from different periods, which
was important for obtaining estimates of how fish
abundance has changed over time (Fig. 1 and the
Supplement).
From the spatial models, we extracted 68039 pre-
dictions of fish biomass by cell and year, which sub-
sequently were used as a basis for multiple regres-
sion analyses.
Multiple regression analyses
Regression analyses to model the biomass of pre -
datory and forage fish were done using the open
source R-package. We considered the following
predictor variables: (1) year, (2) latitude, (3) bottom
depth (from www.ngdc.noaa.gov/mgg/fliers/ 01mgg
04.html), (4) distance from coast (modified from
www.ngdc.noaa.gov/mgg/fliers/01mgg04.html), (5)
density of seamounts (Christensen et al. 2011, www.
seaaroundus.org), (6) absolute primary production
(http://oceancolor.gsfc.nasa.gov/SeaWiFS/), (7) aver-
age of surface and bottom temperature (from www.
noaa.gov/), (8) zooplankton biomass (Bogorov et al.
157
Period Number
1880−1969 35
1970−1989 97
1990−2010 108
Table 1. Number of ecosystem models by time period that
were included in the analyses
Number of models
No coverage
1
2
3
4
5
6
7
8
9
Fig. 1. Spatial distribution of the ecosystem models used in this study, illustrating the wide global coverage. Color density is
indicative of the number of models at each location

Mar Ecol Prog Ser 512: 155–166, 2014
1968, FAO 1981, www.seaaroundus.org, Christensen
et al. 2011), (9) macrobenthos biomass (Christensen
et al. 2011, www.seaaroundus.org), (10) mesopelagic
fish biomass (Gjøsaeter & Kawaguchi 1980, Lam &
Pauly 2005, www.seaaroundus.org, Christensen et
al. 2011), (11) upwelling index (from www.pfeg.noaa.
gov/ products/las/docs/global_upwell.html), and (12)
FAO statistical areas (from www.seaaroundus.org).
Our methodology builds directly on 3 previous
studies of trends in fish biomass in the North Atlantic,
Southeast Asia, and off West Africa (Christensen et
al. 2003a,b, 2004). A detailed flow diagram of the
procedure is provided by Christensen et al. (2003b).
As in previous applications of this methodology, the
environmental parameters were treated as static in
the analysis, while in reality they show considerable
inter-annual variability. However, we do not have ac -
cess to a time series of all the environmental data
used in this analysis covering the time period of inter-
est (as such data do not exist).
We used additive and variance stabilizing transfor-
mations as implemented in the AVAS (additivity and
variance stabilization for regression) module of the
acepack R-library to check for linearity between the
predictive variables and the dependent variable (i.e.
the biomass of potentially exploitable fish with tro -
phic level of 3.5 or higher or trophic level between
2.0 and 3.0).
After model exploratory selection using the lm
module of R (R-library gamair) to assess the contribu-
tion of each independent variable to the model, we
excluded the following independent variables: lati-
tude, because of its covariance with temperature and
FAO area; depth, because of its covariance with
distance from coast; density of seamounts, because
our sample of the ecosystem model collection did not
have a good coverage of seamount models; zoo-
plankton biomass, because of its covariance with pri-
mary production; macrobenthos biomass, because of
its covariance with depth; and mesopelagic fish bio-
mass, because it did not significantly correlate with
the independent variable.
We included 19 marine FAO statistical areas as
potential factorial variables, but ignored 4 of these
areas for which we had less than 5 models. We thus
did not use FAO areas 51, 58, 81, and 88 as factorial
variables in the regressions; however, models from
these areas were used for the predictions, i.e. they
were treated as if the FAO areas in which they occur
were not specified.
The AVAS transformations (Fig. 2) indicated that
the dependent variable (biomass) should be log-
transformed, along with 2 of the predictor variables
(primary production and distance from the coast).
The temperature transformation had a peculiar
shape indicating divergence from linearity, which is
likely due to the very limited observations available
for waters where the average of bottom and surface
temperature exceeded 20°C.
After transformation of the variables, we used the
biomass results from the 200 ecosystem models to
evaluate the time trend in fish biomass over the last
100 yr. We also estimated how fish biomass had
changed during 3 time periods: 1910 to 1970, 1970
to 1990, and 1990 to 2010 (see Table 3). We chose
these splits since the North Atlantic fish catches
peaked around 1970, and this was the period when
fisheries expansion gained momentum throughout
the world (Coll et al. 2008b, Swartz et al. 2010). By
1990, this expansion had reached a new level and
fisheries resource depletion had become a global
phenomenon.
The multiple linear regressions had the following
form:
loge(biomass) = a+ b1× year + b2× loge(distance) +
b3× loge(primary production) + b4× temperature +
b5× upwelling index + bi× factor(FAO)
where ais the regression intercept, b1to b5are the
regression coefficients, and biis a coefficient for each
of the categorical FAO variables. In the regressions,
we weighted each of the 68039 estimates of fish bio-
mass by time and space with 1 / loge(number of spa-
tial units), i.e. with the inverse of the log of the num-
ber of half-degree cells covered by each of the
ecosystem models. This was done to limit the influ-
ence of models covering very large spatial areas
(Christensen et al. 2003a,b, 2004). Since the spatial
resolution of the ecosystem models was a half-degree
latitude by half-degree longitude, and because each
cell was sampled only once for each model, spatial
autocorrelation and pseudoreplication were avoided
(Legendre 1993, Guisan & Zimmermann 2000).
Since results from regressions depend on what
ecosystem models are included as data inputs, we
evaluated the robustness of the regression by jack-
knifing (leaving out one ecosystem model at a time),
and found that this had no noteworthy effects on the
results, as previously found in regional studies
(Christensen et al. 2003a,b, 2004). We further evalu-
ated uncertainty by resampling: we randomly selec -
ted 30% of the 68039 estimates of fish biomass 1000
times, and evaluated predicted biomass trends from
the subsampling.
The regression coefficients and test statistics were
analyzed from the regression analysis.
158

Christensen et al.: Fish biomass decline in the ocean
Finally, we used the regression analyses
(based on the total data set and on resam-
pled sets), jointly with a global database
with the predictor parameters (with half-
degree resolution) to estimate global, spatial
biomass distributions of fish from 1950 and
from 2010.
RESULTS
Biomass of predatory fishes
Our evaluations indicated that the bio-
mass of predatory fish has declined signifi-
cantly over the last 100 yr. For the 200 mod-
els covering the entire time period from
1880 to 2010, the multiple regression coeffi-
cient, r2= 0.70, indicating that the regres-
sion explained 70% of the variation in the
data set, which is highly significant
(Table 2). The predictor variables were all
highly significant, apart from the factorial
variable for FAO areas 18 and 31 (represent-
ing the Amerasian Arctic and the Carib-
bean, respectively).
159
Estimate t-value Pr(>|t|) Significance
Intercept 24.2500 54.8 2.00E-16 ***
Year −0.0151 −69.7 2.00E-16 ***
loge(distance) −0.1008 −28.0 2.00E-16 ***
loge(prim. prod.) 1.1040 142.8 2.00E-16 ***
Temperature −0.0608 −69.6 2.00E-16 ***
Upwelling index 0.0002 42.4 2.00E-16 ***
FAO 18 0.0978 2.0 0.0407 *
FAO 21 0.6361 19.9 2.00E-16 ***
FAO 27 0.7966 28.4 2.00E-16 ***
FAO 31 0.0605 1.7 0.0907 ns
FAO 34 −0.1952 −6.0 2.33E-09 ***
FAO 37 −0.4279 −8.4 2.00E-16 ***
FAO 41 1.0460 31.0 2.00E-16 ***
FAO 47 0.6778 18.2 2.00E-16 ***
FAO 48 1.1660 32.8 2.00E-16 ***
FAO 57 1.1920 26.1 2.00E-16 ***
FAO 61 1.1250 35.6 2.00E-16 ***
FAO 67 1.5880 51.4 2.00E-16 ***
FAO 71 1.2270 36.1 2.00E-16 ***
FAO 77 0.4832 14.9 2.00E-16 ***
FAO 87 0.3341 9.7 2.00E-16 ***
Table 2. Parameter coefficients and associated test statistics for multiple
linear regressions to predict the global marine biomass of predatory
fishes (multiple r2= 0.70). The t-value is the ratio between an estimate
and its standard error, and Pr(>|t|) indicates the probability of obtain-
ing a larger t-value. The smaller this probability is, the higher the sig-
nificance of this parameter. ns: not significant; *p < 0.05, ***p < 0.001
Fig. 2. Additivity and
variance stabilization for
regression (AVAS) trans-
formation of independ-
ent variables included in
the regression analysis.
The ana lysis indicates
that loge-transformation
should be used for bio-
mass, primary produc-
tion, and distance from
coast

Mar Ecol Prog Ser 512: 155–166, 2014
Our results showed that the signs of the predictor
variable coefficients were negative for biomass, dis-
tance, and temperature, and positive for primary pro-
duction and the upwelling index (Table 2). Notably,
the regression indicated that we have lost 1.5% of
the biomass of higher trophic level fish annually
since 1880, and that biomass declines 5 to 6% for
every degree of higher water temperature.
If we examine the relationship between observed
and predicted values based on the regression in
Table 2, we see that the regression overestimated
abundance at low biomass and underestimates at
high biomass (Fig. 3). This indicates that the regres-
sion was conservative, i.e. it did not overestimate the
trend in biomass. It also means that the residuals
(predicted less observed values) were negative at
low biomasses and positive at high biomasses. Such a
structure in the residuals suggests that there are ‘hid-
den’ predictors, i.e. variables that have not been
included in the multiple regressions. This should not
come as a surprise given that we seek to predict the
biomass based on a limited number of variables:
year, distance from coast, primary production, tem-
perature, upwelling, and FAO area. While these
parameters do lead to a skewed residuals overall,
only the upwelling index shows indi-
cation of a divergence from linearity
(Fig. 4). These results suggest, over-
all, that our predictor variables are
suitable for use in the regressions.
Using a resampling methodology,
we randomly drew 30% of the esti-
mates of biomass over space and time
and performed a multiple regression
with each subsample for the 3 time
periods analyzed. Based on this ana -
lysis, we obtained a distribution for
each predictor variable (Fig. 5). We
then used each of the resampled re -
gressions and the database of envi-
ronmental parameters to predict
global biomasses. From this, we esti-
mated that the biomass of predatory
fishes has declined by around 75%
during the 100 yr from 1910 to 2010.
Dividing the models into 3 time
periods to obtain higher temporal res-
olution (Table 1), and with the splits
made in 1970 and 1990, we obtained
multiple linear regressions similar to
that reported above for the entire
time period (Table 3). Again, the pre-
dictor variables were highly signifi-
cant (p < 10−16) and the regressions explain 66 to 91%
of the variability in the biomass data.
Evaluating the time trends based on resampling
the 3 regressions 1000 times based on randomly
selecting 30% of the biomass estimates, our results
showed that the biomass of predatory fishes has
declined by two-thirds (66.4% with 95% confidence
intervals ranging from 60.2−71.2%) over the last
100 yr (Fig. 6). This decline is estimated to have been
slow during the first period (10.8% or 0.2% yr−1 up to
1970), then severe during the second period (41.6%
or 4.0% yr−1 during 1970 to 1990), and slower during
the third period (14.0% or 2.9% yr−1 since 1990).
Biomass of lower trophic level fish
We repeated the same procedure as described
above to predict the biomass of lower trophic level
fish (with trophic level between 2.0 and 3.0). Results
from the multiple linear regressions for the entire
time period (Table 4) showed a positive and signifi-
cant regression coefficient for year (0.0085). This in -
dicated that the biomass of prey fish had been in -
creasing over time by 0.85% yr−1. Over a 100 yr time
160
–2 0 2 4
–2
0
2
4
Observed biomass (loge)
Predicted biomass (loge)
Fig. 3. Predicted versus observed biomass (loge-scales, t km−2) for higher
trophic level fish in the world ocean. Solid line indicates the 1:1 line; dotted
line is the observed average trend. The predicted variables underestimate the
observed variability

Christensen et al.: Fish biomass decline in the ocean 161
Year
Frequency
–0.0150 –0.0145 –0.0140 –0.0135 –0.0130
0
50
100
150
Distance
–0.13 –0.12 –0.11 –0.10 –0.09
0
40
80
Prim. prod.
1.02 1.04 1.06 1.08 1.10
0
40
80
120
Temperature
–0.060 –0.058 –0.056 –0.054 –0.052 –0.050
0
40
80
120
Upwelling
0.00018 0.00020 0.00022 0.00024
0
50
150
Fig. 5. Frequency distributions for pre-
dictor variables in the regression cover-
ing the entire time period
Fig. 4. Residuals (predicted minus ob -
served values, loge-transformed) for the
predictor variables in the multiple linear
regression analysis covering the entire
time period

Mar Ecol Prog Ser 512: 155–166, 2014
period, this corresponds to a 130% increase, suggest-
ing that there are now more than twice as many prey
fish in the global ocean as there were a century ago.
Spatial biomass distributions
Results of the global spatial biomass distributions
of predatory fish highlighted notable changes with
time (Fig. 7), especially evident in the Northern
Hemisphere and main upwelling systems. Predicted
predatory fish biomass during the 1950s showed
higher values of biomass in the Northern Hemi-
sphere and upwelling systems (Fig. 7a). Results from
2010 showed large changes in these areas, with a
general decrease of biomass spatially notable in the
north and generally in coastal areas and shelf eco -
systems (Fig. 7b).
DISCUSSION
Through this study, we further refined a methodo -
logy we originally introduced to describe how the
biomass of predatory fish has changed in the North
Atlantic, West Africa, and Southeast Asia (Christen -
sen et al. 2003a,b, 2004). Extending this meth odo -
logy, the aim was to provide a first global estimate for
how fish biomass has changed over the last century,
drawing on a vast amount of information made avail-
able through data-rich ecosystem models (Chris-
tensen & Pauly 1992, Christensen & Walters 2004) to
add to the general discussion of the status of marine
resources worldwide.
Our results show major declines in the biomass of
predatory fish (i.e. of the larger fish that humans tend
to eat), amounting to a decline of two-thirds over the
last century, with 55% of the decline occurring in the
last 40 yr. Indications are that the decline was shar -
pest during the period between 1970 and 1990, and
has since leveled off somewhat. This does not mean,
however, that conditions have started to improve
globally; we found no indications of increase in bio-
mass of predatory fish. There may be regional im -
provements (Worm et al. 2009, Lotze et al. 2011),
however, this is not evident yet at a global scale.
These results could indicate that we have been
fishing past the maximum sustainable yield (MSY)
level (Hilborn & Walters 1992). Productivity for high -
er trophic level fish populations (e.g. tuna) may in -
deed be maximized when populations are reduced to
between one-third and half of their original biomass,
but our study shows that this reduction is an overall
average. The implication is that some higher trophic
level species (notably the larger species) may likely
be reduced even more, while smaller species will
have declined less, yet the overall average indicates
a large reduction. This is in accordance with one of
the most thorough studies of top predator abundance
162
1880−1969 1970−1989 1990−2010
Intercept 0.5489 76.5400 51.5800
Year −0.0021 −0.0411 −0.0293
loge(distance) −0.1054 −0.1183 −0.0700
loge(prim. prod.) 1.1000 1.0530 1.1950
Temperature −0.1917 −0.1008 −0.0335
Upwelling index 0.0001 0.0001 0.0004
r20.9117 0.6721 0.6562
Table 3. Regression coefficients for predictor variables for
the 3 time periods considered, used to estimate the loge(bio-
mass) of predatory fishes (trophic level > 3.5). Coefficients
for FAO areas are not listed for clarity
Estimate t
Intercept −14.5200 54.8
Year 0.0085 −69.7
loge(distance) −0.5958 −28.0
loge(prim. prod.) 0.7790 142.8
Temperature −0.1269 −69.6
Upwelling index 0.0001 42.4
r20.5572
Table 4. Regression coefficients for predictor variables for
prey fish (trophic level between 2.0 and 3.0) covering the
entire time period. Coefficients for FAO areas are not listed
for clarity
0
20
40
60
80
100
120
1910 1930 1950 1970 1990 2010
Biomass (%)
Fig. 6. Global biomass trends for predatory fish during 1910
to 2010 as predicted based on 200 ecosystem models and
1000 times random resampling of 30% of data points. The
lines indicate median values and 95% confidence intervals

Christensen et al.: Fish biomass decline in the ocean
(Sibert et al. 2006). For the Pacific Ocean, it was esti-
mated that the largest pelagic fish (>175 cm fork
length) had decreased to 17% of the unexploited bio-
mass, whereas the decline was much lower for the
smaller species. We also note that it indeed is impos-
sible to fish all species at the MSY level; doing so
may have severe impacts on the trophic structure of
the ecosystem (Walters et al. 2005).
Our results are also in line with previous regional
studies using the same methodology. For example, a
decline of high trophic level fishes by two-thirds dur-
ing a 50 yr period and by a factor of 9 between 1880
and 1998 was documented to have occurred in the
North Atlantic (Christensen et al. 2003b). In West
Africa, previous analysis showed fish biomass (ex -
cluding low trophic level and small fishes) has de -
163
Fig. 7. Global spatial biomass distribution of predatory fish for (a) 1950 and (b) 2010

Mar Ecol Prog Ser 512: 155–166, 2014
clined over a 40 yr period from 1960 to 2000 by a fac-
tor of 13 (Christensen et al. 2004). In Southeast Asia,
the abundance of fish with trophic levels of 3.0 or
higher in 2000 was less than half the value in 1960
(Christensen et al. 2003a). It is interesting to note that
this study reports similar, but somewhat lower, esti-
mates of predatory fish biomass declines than previ-
ous studies. It may be possible that by focusing exclu-
sively on high trophic-level fish (and not just on
target species), the present study is less biased
towards detection of large (~90%) declines (Myers &
Worm 2003, Ferretti et al. 2010).
The decline in predatory fish biomass is closely
linked to increased fishing effort. Anticamara et al.
(2011) found that global fishing capacity (measured
in potential kilowatt days) increased 54% from 1950
to 2010 with no indication of a decrease in re cent
years. Note that this latter study did not include any
‘technology creep’ factor indicating an increase in
technology efficiency (which averages 2 to 3% annu-
ally; Pauly & Palomares 2010, Stergiou & Tsikliras
2011). Overall, we conclude that our results are con-
sistent with the study of fishing effort (Anticamara et
al. 2011), which reports the decline in biomass that
can be expected given the described increase in
global fishing capacity. The increase in fishing
capacity has been up dated and confirmed recently
by additional studies (Watson et al. 2013).
It is interesting to highlight that the FAO areas
were an important term in the regression model to
explain the dependent variables. This importance
may be related to differences in historical pressure
and the expansion of fisheries in marine ecosystems
(Coll et al. 2008b, Swartz et al. 2010), and it is in line
with other studies suggesting that the historical evo-
lution of fishing is a key factor to understand the
dynamics seen in ecological indicators (Shannon et
al. 2014, this Theme Section). The predictive vari-
ables we used in this study are not the only important
factors for fish biomass, even if 70% of the variation
can be explained based solely on them. There are
other important variables, for instance ‘rugosity’ (i.e.
depth variability within spatial cells), substrate types,
and fishing ef fort, but we did not have access to
global data layers covering such variables (or they
were not of sufficient quality to be included), and
therefore they were omitted from the analysis for the
time being. In addition, the spatial distributions used
within Ecospace were inferred from depths at which
species are known to occur; other information such as
habitat type was not included. This represents a lim-
itation to our analysis that must be acknowledged.
The implication is not that the present study is likely
to be misleading, but rather that we would have
obtained better predictions if we could have had
added additional suitable predictor variables. In
addition, although the number of ecosystem models
fitted to the time series of data with Ecosim has
grown substantially (Christensen & Walters 2011,
Christensen 2013), these models were not used
because they do not cover the ocean in a representa-
tive way, since such models are only available for
data-rich areas. However, as more fitted-to-time-
series models become available, future worldwide
analysis will become possible.
It is also important to highlight that, even though
spatial autocorrelation of the data was avoided due to
the data sampling procedure, our cells do not provide
true degrees of freedom. In addition, the t-values of
the regressions give an indication of the internal
‘ranking’ of the parameters (i.e. which ones matter
most). However, due to covariation between vari-
ables it must be acknowledged that the interpreta-
tion of these t-values needs to be treated with cau-
tion, even though the results indicated that the
regression was fairly robust.
Our results also highlight that the biomass of
smaller fish at lower trophic level positions has likely
increased. Because our study used a trophodynamic
approach, this increase is likely linked to predation
release, i.e. the mechanism whereby reduction in
predator populations leads to an increase in prey
abundance, as documented in local and regional
studies (Carscadden et al. 2001, Frank et al. 2005,
Ward & Myers 2005, Daskalov et al. 2007, Myers et
al. 2007, Coll et al. 2008c, Baum & Worm 2009), and
which can be captured using trophic models. This
increase in forage fish biomass is also in line with a
growing importance of this resource in global fishing
activities (Pikitch et al. 2012), which warns of the
degradation of marine ecosystems due to the fact that
these organisms are key elements of the food web
(Cury et al. 2011). Further development of this study
should focus on spatially describing the global distri-
bution of smaller fish.
Overall, our findings contribute to the discussion of
whether ‘fishing down the food web’ is a sampling
artifact or something that occurs in reality (Pauly et
al. 1998, Branch et al. 2010). Here, we estimated that
predatory fish have declined by two-thirds, while
prey fish have increased. Combined, the decrease of
high trophic level fish and increase of low trophic
level fish show that the trophic structure of marine
ecosystems has been changed at a global scale in a
manner consistent with fishing down marine food
webs, in addition to that which is occurring on local
164

Christensen et al.: Fish biomass decline in the ocean
and regional scales (Stergiou & Christensen 2011,
and see www.fishingdown.org). To add to the discus-
sion, it should be noted that our methodology is less
dependent on fisheries catch estimates than previous
studies — an issue that has been central to this
debate. This study is based on food web modeling
results, which deliver important information regard-
ing the ‘fishing down the food web’ debate (Shannon
et al. 2014).
Our results have clear implications regarding
changes in marine ecosystem functioning, and thus
management of marine resources, since the structure
of marine ecosystems may have shifted from eco -
systems with larger abundance of high trophic levels
and large body sizes to ecosystems dominated by
lower trophic level and small body size organisms. A
growing importance of forage fish in marine eco -
systems due to a decrease of top-down control is of
concern, since these organisms are short-lived and
are much more vulnerable to environmental fluctua-
tions (Planque et al. 2010).
Acknowledgements. This is a contribution from Sea Around
Us, a scientific cooperation between the University of British
Columbia (UBC) and the Pew Charitable trust. V.C. acknowl-
edges support from the Natural Sciences and Engineering
Research Council of Canada. M.C. was partially funded by
the EC Marie Curie Career Integration Grant Fellowships to
the BIOWEB project and the Spanish National Program
Ramon y Cajal.
LITERATURE CITED
Alder J, Campbell B, Karpouzi V, Kaschner K, Pauly D
(2008) Forage fish: from ecosystems to markets. Annu
Rev Environ Resour 33: 153−166
Anticamara JA, Watson R, Gelchu A, Pauly D (2011) Global
fishing effort (1950–2010): trends, gaps, and implica-
tions. Fish Res 107: 131−136
Baum JK, Worm B (2009) Cascading top-down effects of
changing oceanic predator abundances. J Anim Ecol 78:
699−714
Baum JL, Myers RA, Kehler DG, Worm B, Harley SJ, Do -
herty PA (2003) Collapse and conservation of shark
populations in the Northwest Atlantic. Science 299:
389−392
Bogorov VG, Vinogradov MD, Varonina NM, Kanaeva IP,
Suetova IA (1968) Distribution of zooplankton biomass
within the surficial layer of the world ocean. Dokl Akad
Nauk SSSR 182: 1205−1207 (in Russian)
Branch TA, Watson R, Fulton EA, Jennings S and others
(2010) The trophic fingerprint of marine fisheries. Nature
468: 431−435
Carscadden JE, Frank KT, Leggett WC (2001) Ecosystem
changes and the effects on capelin (Mallotus villosus), a
major forage species. Can J Fish Aquat Sci 23: 191−219
Christensen V (2013) Ecological networks in fisheries: Pre-
dicting the future? Fisheries 38: 76−81
Christensen V, Pauly D (1992) ECOPATH II — a software for
balancing steady-state ecosystem models and calculat-
ing network characteristics. Ecol Model 61: 169−185
Christensen V, Walters C (2004) Ecopath with Ecosim:
methods, capabilities and limitations. Ecol Model 172:
109−139
Christensen V, Walters CJ (2011) Progress in the use of eco-
system models for fisheries management. In: Christensen
V, Maclean J (eds) Ecosystem approaches to fisheries: a
global perspective. Cambridge University Press, Cam-
bridge
Christensen V, Garces LR, Silvestre GT, Pauly D (2003a)
Fisheries impact on the South China Sea large marine
ecosystem: a preliminary analysis using spatially-explicit
methodology. In: Silvestre GT, Garces LR, Stobutzki I,
Ahmed M and others (eds) Assessment, management
and future directions for coastal fisheries in Asian coun-
tries. Proc WorldFish Cent Conf 67: 51−62
Christensen V, Guenette S, Heymans JJ, Walters C, Watson
R, Zeller D, Pauly D (2003b) Hundred-year decline of
North Atlantic predatory fishes. Fish Fish 4: 1−24
Christensen V, Amorim PA, Diallo I, Diouf T and others
(2004) Trends in fish biomass off Northwest Africa,
1960-2000. In: Chavance P, Ba M, Cascuel D, Vakily
M, Pauly D. Marine fisheries, ecosystems, and societies
in West Africa: half a century of change. IRD, Paris,
p 377−386
Christensen V, Lai S, Palomares ML, Zeller D, Pauly D (eds)
(2011) The state of biodiversity and fisheries in regional
seas. Fish Centre Res Rep 19(3): 1−61
Coll M, Bundy A, Shannon LJ (2008a) Ecosystem modelling
using the Ecopath with Ecosim approach. In: Megrey B,
Moksness E (eds) Computers in fisheries research, 2nd
edn. Springer, Berlin, p 225−291
Coll M, Libralato S, Tudela S, Palomera I, Pranovi F (2008b)
Ecosystem overfishing in the ocean. PLoS ONE 3: e3881
Coll M, Palomera I, Tudela S, Dowd M (2008c) Food-web
dynamics in the South Catalan Sea ecosystem (NW
Mediterranean) for 1978–2003. Ecol Model 217: 95−116
Coll M, Navarro J, Olson R, Christensen V (2013) Assessing
the trophic position and ecological role of squids in mar-
ine ecosystems by means of food-web models. Deep-Sea
Res II 95: 21−36
Colléter M, Valls A, Guitton J, Morissette L and others
(2013) EcoBase: a repository solution to gather and com-
municate information from EwE models. Fish Centre Res
Rep 21(1):1−60
Cury PM, Boyd IL, Bonhommeau S, Anker-Nilssen T and
others (2011) Global seabird response to forage fish de -
pletion—one-third for the birds. Science 334: 1703−1706
Daskalov GM, Grishin AN, Rodionov S, Mihneva V (2007)
Trophic cascades triggered by overfishing reveal possi-
ble mechanisms of ecosystem regime shifts. Proc Natl
Acad Sci USA 104: 10518−10523
FAO (1981) Atlas of the living resources of the seas. FAO
Fisheries Series 15, Food and Agriculture Organization
of the United Nations, Rome
Ferretti F, Worm B, Britten G, Heithaus M, Lotze H (2010)
Patterns and ecosystem consequences of shark declines
in the ocean. Ecol Lett 13: 1055−1071
Frank KT, Petrie B, Choi JS, Leggett WC (2005) Trophic cas-
cades in a formerly cod-dominated ecosystem. Science
308: 1621−1623
Froese R, Pauly D (eds) (2013) FishBase. www.fishbase.org
(accessed February 2013)
165
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤

Mar Ecol Prog Ser 512: 155–166, 2014
Gjøsaeter J, Kawaguchi K (1980) A review of the world
resources of mesopelagic fish. FAO Fish Tech Pap No.
193. Food and Agriculture Organization of the United
Nations, Rome
Guisan A, Zimmermann NE (2000) Predictive habitat distri-
bution models in ecology. Ecol Model 135: 186
Heymans JJ, Coll M, Libralato S, Morissette L, Christensen
V (2014) Global patterns in ecological indicators of
marine food webs: a modelling approach. PLoS ONE 9:
e95845
Hilborn R, Walters C (1992) Quantitative fisheries stock
assessment: choice, dynamics and uncertainty. Chapman
& Hall, New York, NY
Kaschner K, Ready JS, Agbayani E, Rius J and others (2013)
AquaMaps: predicted range maps for aquatic species,
version 2013. www.aquamaps.org
Lam V, Pauly D (2005) Mapping the global biomass of
mesopelagic fishes. The Sea Around Us Project News -
letter 30: 4
Legendre P (1993) Spatial autocorrelation: trouble or new
paradigm? Ecology 74: 1659−1673
Libralato S, Coll M, Tudela S, Palomera I, Pranovi F (2008)
Novel index for quantification of ecosystem effects of
fishing as removal of secondary production. Mar Ecol
Prog Ser 355: 107−129
Lotze HK, Coll M, Magera MA, Ward-Paige C, Airoldi L
(2011) Recovery of marine animal populations and eco-
systems. Trends Ecol Evol 26: 595−605
Myers RA, Worm B (2003) Rapid worldwide depletion of
predatory fish communities. Nature 423: 280−283
Myers RA, Baum JK, Shepherd TD, Powers SP, Peterson CH
(2007) Cascading effects of the loss of apex predatory
sharks from a coastal ocean. Science 315: 1846−1850
Palomares MLD, Pauly D (eds) (2013) SeaLifeBase. www.
sealifebase.org
Pauly D, Palomares MLD (2010) An empirical equation to
predict annual increases in fishing efficiency. Fisheries
Centre Working Paper No. 7, University of British Co -
lumbia, Vancouver
Pauly D, Christensen V, Dalsgaard J, Froese R, Torres F
(1998) Fishing down marine food webs. Science 279:
860−863
Pauly D, Graham W, Libralato S, Morissette L, Palomares
MLD (2009) Jellyfish in ecosystems, online databases,
and ecosystem models. Hydrobiologia 616: 67−85
Pikitch E, Boersma PD, Boyd IL, Conover DO and others
(2012) Little fish, big impact: managing a crucial link in
ocean food webs. Lenfest Ocean Program, Washington,
DC
Pikitch EK, Rountos KJ, Essington TE, Santora C and others
(2013) The global contribution of forage fish to marine
fisheries and ecosystems. Fish Fish 15: 43−64
Plagányi É (2007) Models for an ecosystem approach to fish-
eries, Vol 477. Food and Agriculture Organization of the
United Nations, Rome
Planque B, Fromentin JM, Cury P, Drinkwater KF, Jennings
S, Perry RI, Kifani S (2010) How does fishing alter marine
populations and ecosystems sensitivity to climate? J Mar
Syst 79: 403−417
Shannon JL, Coll M, Bundy A, Shin YJ and others (2014)
Trophic level-based indicators to track fishing impacts
across marine ecosystems. Mar Ecol Prog Ser 512:
115–140
Sibert J, Hampton J, Kleiber P, Maunder M (2006) Biomass,
size, and trophic status of top predators in the Pacific
Ocean. Science 314: 1773−1776
Stergiou KI, Christensen V (2011) Fishing down food webs.
In: Christensen V, Maclean J (eds) Ecosystem approa -
ches to fisheries: a global perspective. Cambridge Uni-
versity Press, Cambridge, p 72−88
Stergiou KI, Tsikliras AC (2011) Fishing down, fishing
through and fishing up: fundamental process versus
technical details. Mar Ecol Prog Ser 441: 295−301
Swartz W, Sala E, Tracey S, Watson R, Pauly D (2010) The
spatial expansion and ecological footprint of fisheries
(1950 to present). PLoS ONE 5: e15143
Valinassab T, Pierce G, Johannesson K (2007) Lantern fish
(Benthosema pterotum) resources as a target for commer-
cial exploitation in the Oman Sea. J Appl Ichthyology 23:
573−577
Walters C, Pauly D, Christensen V (1999) Ecospace: predic-
tion of mesoscale spatial patterns in trophic relationships
of exploited ecosystems, with emphasis on the impacts of
marine protected areas. Ecosystems 2: 539−554
Walters CJ, Christensen V, Martell SJ, Kitchell JF (2005)
Possible ecosystem impacts of applying MSY policies
from single-species assessment. ICES J Mar Sci 62:
558−568
Ward P, Myers RA (2005) Shifts in open-ocean fish commu-
nities coinciding with the commencement of commercial
fishing. Ecology 86: 835−847
Watson RA, Cheung WWL, Anticamara JA, Sumaila RU,
Zeller D, Pauly D (2013) Global marine yield halved as
fishing intensity redoubles. Fish Fish 14: 493−503
Worm B, Myers RA (2003) Meta-analysis of cod–shrimp
interactions reveals top-down control in oceanic food
webs. Ecology 84: 162−173
Worm B, Barbier EB, Beaumont N, Duffy JE and others
(2006) Impacts of biodiversity loss on ocean ecosystem
services. Science 314: 787−790
Worm B, Hilborn R, Baum JK, Branch TA and others (2009)
Rebuilding global fisheries. Science 325: 578−585
166
Submitted: March 27, 2014; Accepted: July 13, 2014 Proofs received from author(s): September 3, 2014
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤