Recent shifts in the occurrence, cause, and magnitude
of animal mass mortality events
Samuel B. Feya,b,1,2, Adam M. Siepielskic,1, Sébastien Nussléd, Kristina Cervantes-Yoshidad, Jason L. Hwand,
Eric R. Huberd, Maxfield J. Feyb, Alessandro Catenazzie, and Stephanie M. Carlsond
a
Department of Ecology and Evolutionary Biology, Yale University, New Haven, CT 06520; bDepartment of Biological Sciences, Dartmouth College, Hanover,
NH 03755; cDepartment of Biology, University of San Diego, San Diego, CA 92110; dDepartment of Environmental Science, Policy, and Management,
University of California, Berkeley, CA 94720; and eDepartment of Zoology, Southern Illinois University, Carbondale, IL 62901
Mass mortality events (MMEs) are rapidly occurring catastrophic
demographic events that punctuate background mortality levels.
Individual MMEs are staggering in their observed magnitude: removing more than 90% of a population, resulting in the death of
more than a billion individuals, or producing 700 million tons of
dead biomass in a single event. Despite extensive documentation
of individual MMEs, we have no understanding of the major features
characterizing the occurrence and magnitude of MMEs, their causes,
or trends through time. Thus, no framework exists for contextualizing
MMEs in the wake of ongoing global and regional perturbations to
natural systems. Here we present an analysis of 727 published MMEs
from across the globe, affecting 2,407 animal populations. We show
that the magnitude of MMEs has been intensifying for birds, fishes,
and marine invertebrates; invariant for mammals; and decreasing for
reptiles and amphibians. These shifts in magnitude proved robust
when we accounted for an increase in the occurrence of MMEs
since 1940. However, it remains unclear whether the increase in the
occurrence of MMEs represents a true pattern or simply a perceived
increase. Regardless, the increase in MMEs appears to be associated
with a rise in disease emergence, biotoxicity, and events produced
by multiple interacting stressors, yet temporal trends in MME causes
varied among taxa and may be associated with increased detectability. In addition, MMEs with the largest magnitudes were
those that resulted from multiple stressors, starvation, and disease.
These results advance our understanding of rare demographic
processes and their relationship to global and regional perturbations to natural systems.
catastrophes
Allee effects can drive populations to extinction (1). Population
loss through MMEs can alter the structure of food webs by
abruptly generating resource pulses or losses, removing predators or competitors (8), or disturbing mutualist networks (9).
Such events can reshape the ecological and evolutionary trajectories of life on Earth (10, 11). In addition, MMEs can generate
large economic costs as well as disrupt ecosystem services such
as pollination (12). Human populations have also experienced
MMEs, often in response to extreme weather events or geologic
disasters such as the 2004 Indian Ocean Tsunami, which resulted
in an estimated 260,000 deaths (13).
Despite the importance of MMEs as a demographic process
with considerable repercussions, no quantitative analysis has identified the features characterizing contemporary MMEs across
animal taxa. Indeed, individual MMEs are rarely placed into a
broader context (14). This lack of synthesis, commonly associated with the study of infrequent events across academic disciplines (15), presently precludes uncovering mechanisms underlying
MMEs and contextualizing the importance of MMEs relative
to background mortality levels. In addition, because MMEs can
be associated with climatic events (16) and because the severity
of extreme weather-related events such as heat waves, heavy
precipitation, and droughts is expected to increase in the future
as a result of climate change (17, 18), it is especially important
to establish a quantitative framework to understand and gauge
future MMEs.
| defaunation | death | rare demographic events
Significance
Mass mortality events (MMEs), the rapid, catastrophic dieoff of organisms, are an example of a rare event affecting
natural populations. Individual reports of MMEs clearly demonstrate their ecological and evolutionary importance, yet our
understanding of the general features characterizing such
events is limited. Here, we conducted the first, to our knowledge, quantitative analysis of MMEs across the animal kingdom,
and as such, we were able to explore novel patterns, trends, and
features associated with MMEs. Our analysis uncovered the
surprising finding that there have been recent shifts in the
magnitudes of MMEs and their associated causes. Our database
allows the recommendation of improvements for data collection
in ways that will enhance our understanding of how MMEs relate to ongoing perturbations to ecosystems.
D
eath is a ubiquitous demographic process central to understanding ecological and evolutionary dynamics. Resource
limitation, stochastic events, exceeding physiological thresholds,
senescence, and interactions with predators, pathogens, or parasitoids generate death on daily, seasonal, and annual timescales.
Although all organisms eventually die, the timing and magnitude
of deaths within populations varies greatly. In contrast to mortality
that affects one specific age or stage class, mass mortality events
(MMEs) represent demographic catastrophes that can simultaneously affect all life stages (1) and can rapidly remove a substantial
proportion of a population over a short period of time relative to the
generation time of the organism (2, 3). These events are part of the
continuum between background mortality levels, population extirpations (4), and species-level extinctions, such as the five Phanerozoic mass extinctions and the ongoing sixth mass extinction (5).
MMEs of varying spatial and temporal scales affect biological
communities in the present day and are frequently a natural
phenomenon (6, 7). For example, background mortality levels of
sea stars are occasionally punctuated with MMEs driven by outbreaks of wasting syndrome (6), which have led to considerable
and rapid population losses in species such as Pisaster ochraceus
along both coasts of North America (6). MMEs such as these may
trigger local extinction vortices by depressing populations to levels
at which loss of genetic diversity, demographic stochasticity, or
www.pnas.org/cgi/doi/10.1073/pnas.1414894112
Author contributions: A.M.S. conceived the study; S.B.F., A.M.S., and S.M.C. designed
research; S.B.F., A.M.S., K.C.-Y., J.L.H., E.R.H., M.J.F., A.C., and S.M.C. performed research;
S.N. analyzed data; and S.B.F., A.M.S., S.N., and S.M.C. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Freely available online through the PNAS open access option.
1
S.B.F. and A.M.S. contributed equally to this work.
2
To whom correspondence should be addressed. Email: [email protected].
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1414894112/-/DCSupplemental.
PNAS Early Edition | 1 of 6
ECOLOGY
Edited by James A. Estes, University of California, Santa Cruz, CA, and approved December 22, 2014 (received for review August 5, 2014)
As such, we synthesized existing information on MMEs to
examine both the temporal patterns in the occurrence and magnitudes of MMEs among animal taxa and the causes of MMEs. To
accomplish this, we assembled a database of 727 published animal
MMEs affecting 2,407 populations of amphibian, bird, fish,
invertebrate, mammal, and reptile species from throughout the
world (Dataset S1).
Results and Discussion
Reports of MMEs have been increasing through time for all taxa
(Fig. 1). However, contemporaneous heightened scientific awareness of MMEs coupled with an overall increase in total scientific
productivity (Fig. 1 and SI Appendix, Fig. S1) could generate a
perceived increase. This putative publication bias is a well-known
phenomenon when dealing with rare events occurring in time series
data, akin to the challenge of distinguishing ongoing epidemics from
“epidemics of discovery” (19). Notably, however, more than half of
the variation (mean = 54.5%; range = 15–84%) in changes in the
occurrence of MMEs through time was not explained by increases
in publication output alone (SI Appendix, Fig. S2). MMEs were only
sporadically reported during the late nineteenth century and early
part of the 1900s; since the 1940s, however, MMEs have been
documented consistently for birds, fishes, mammals, and marine
invertebrates on all continents and in all major biomes. Overall,
fishes were the largest contributor of reported MMEs, accounting for 56% of all documented MMEs (Fig. 1). Our analysis
also captures a sharp increase in the occurrence of amphibian
and reptile MMEs beginning in the 1970s, coincident with the
growing awareness of global amphibian declines (20), and more
recently, declines in reptiles (21). The recent declines in the
occurrence of MMEs since 2000 for all groups but reptiles (Fig.
1) are likely a result of reporting delays between when events
occur and when they are reported in the literature (SI Appendix,
Fig. S3), and statistically accounting for these delays reduces the
extent of recent declines in MME occurrence (SI Appendix, Fig.
S4). Such delays may increase in length as ecologists continue to
use older datasets.
Our analysis reveals that the magnitudes of MMEs are changing
through time, as measured by the number of animals that died
during each event (Fig. 2). Interestingly, changes in the magnitude
of MMEs are variable among animal taxa: magnitude increases for
birds, marine invertebrates, and fishes; remains invariant for
mammals; and decreases for amphibians and reptiles, despite substantial variation around these patterns (Fig. 2). With the exception
of reptiles and amphibians, in which a slight quadratic trend in the
magnitude of MMEs is present, a nonparametric local regression
of these data reveals comparable trends (SI Appendix, Fig. S5).
These trends in the number of individuals killed per MME likely
do not result from a publication bias, as the observed trends for
all taxa proved robust when the dataset was reanalyzed by resampling to account for increased MME reporting through time (SI
Appendix, Fig. S6) (22). We note that the temporal patterns of
reptile and marine invertebrate MMEs should be interpreted with
caution because of their comparatively low sample size. Overall,
taxon, cause, and year explained the largest amounts of variance in
the magnitude of MMEs (21.0%, 7.4%, and 4.9%, respectively;
multiple linear regression: F21,770 = 18.3; P < 0.001).
The positive trends in MME magnitude for fishes, birds, and
marine invertebrates also runs counter to the tendency for scientists to report the largest, most obvious events, whereas lessobvious events are typically increasingly reported when scientific
awareness and allocated effort escalate (23). As such, the average
increases in MME magnitude for birds, fishes, and marine invertebrates, which on average tended to increase by 0.22, 0.33, and
0.60 orders of magnitude per decade, respectively, are surprising,
given the increased MME reporting through time (Fig. 1).
Fig. 1. Occurrences of animal MMEs and taxon-specific publication trends through time. Colored bars indicate the number of events during a 5-y interval
(e.g., 1940 stands for the 1940–1944 period), and dashed lines show trends in the total number of papers published each year for each taxon. For all taxa, the
increase in the number of MMEs is coincident with an increase in the number of publications (SI Appendix, Fig. S1).
2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1414894112
Fey et al.
1B
100M
10M
1M
100K
10K
1K
100
10
1
Mammals
1950
1960
1970
1980
1990
Amphibians
1940
1B
100M
10M
1M
100K
10K
1K
100
10
1
1950
1960
2000
2010
1940
1970
1980
1990
2000
1960
1970
1980
1990
Reptiles
F1,167 = 37.6
p < 0.001
R2 = 0.18
Fish
Fi
1950
F1,73 = 8.3
p = 0.005
R2 = 0.1
2010
1940
1950
1960
2010
F1,24 = 9.3
p = 0.006
R2 = 0.28
1970
1980
1990
Invertebrates
F1,447 = 71.3
p < 0.001
R2 = 0.14
2000
2000
2010
F1,9 = 38.9
p < 0.001
R2 = 0.81
ECOLOGY
Number of deaths per event
1940
1B
100M
10M
1M
100K
10K
1K
100
10
1
Birds
F1,70 = 0.001
p = 0.97
R2 = 0
1940
1950
1960
1970
1980
1990
2000
2010
1940
1950
1960
1970
1980
1990
2000
2010
Year
Fig. 2. Magnitude of animal MMEs through time. Each point is a single MME (n = 727 total events). Dashed lines represent statistically significant slopes,
shading demarcates slope 95% confidence intervals, and hatched shading indicates extrapolation of regressions before the first reported MME.
Although the proportion of the animal population removed
during an MME remains the most widespread approach for
defining an MME (2, 3, 14), only 9.6% of published MMEs
reported information on how MMEs affect population sizes (SI
Appendix, Fig. S7). This lack of data limits our ability to resolve
temporal patterns in the population-level consequences of MMEs.
However, the available data suggest that reported MMEs frequently remove a substantial proportion of animal populations,
including up to 100% of the population (SI Appendix, Fig. S7).
Because we have no reason to suspect that population sizes for
these taxa have consistently been increasing through time, the
positive trends in MME magnitude documented in numbers of
individuals for fishes, birds, and marine invertebrates likely indicate
an increase in the proportion of the population being removed.
The differences in MME magnitude among animal taxa may
also highlight an important relationship between MMEs and population demographics. Among vertebrate taxa, amphibians, reptiles,
and mammals contain the largest proportion of species that are
threatened, according to the International Union for Conservation
of Nature (IUCN) Red List, and birds and fishes contain proportionally fewer threatened species (24). Thus, one possibility is that
the tendency for MME magnitude among certain taxa to increase
and then decrease over time may reflect an overall recent decrease
in the size of their extant populations (Fig. 2 and SI Appendix, Fig.
S5). Intriguingly, the recent patterns of amphibian and reptile MME
magnitude (SI Appendix, Fig. S5) support this hypothesis; however,
the comparatively small sample sizes for these taxa and lack of
population-specific data suggest caution is warranted in this interpretation. One extreme example of this mechanism is the 1983
sea urchin Diadema antillarum MME, which was likely caused by
a waterborne pathogen that removed an estimated 99% of all Diadema from the Caribbean. Such a large-scale event precludes the
occurrence of similarly large die-offs in the near future (25) and may
be a precursor to local and regional defaunation (26, 27).
Fey et al.
Across all animal taxa, causes of MMEs were most often associated with disease (Fig. 3; 26.3%) and were attributed to viral
(44.5%), bacterial (18.3%), and fungal infections (12.2%). Human perturbation was the second most common cause of MMEs,
accounting for 19.3% of total MMEs, mainly from point source
environmental contamination (42.5%). Biotoxicity was the third
leading cause of MMEs (15.6%), primarily resulting from toxinproducing cyanobacteria and dinoflagellates that dominate
marine and freshwater harmful algal blooms. Processes directly
influenced by climate (weather, thermal stress, oxygen stress,
starvation) also contributed to the occurrence of MMEs and
accounted, collectively, for 24.7% of known cases.
The causes of MMEs also exhibit shifts through time, but importantly, the causes of MMEs do not change uniformly through
time for all taxa. Overall, infectious disease, biotoxicity, and multiple
stressors were the most rapidly intensifying causes of MMEs, increasing from 0–33%, 5–18%, and 0–8% of reported MMEs from
the 1940s to the 2000s, respectively (Fig. 3). One explanation for
these observed shifts is that technological improvements have enabled increased detection of disease and biotoxicity, which rely
heavily on laboratory-based methods for detection (28). As such,
increased efforts in disease and biotoxicity research could also
produce such a pattern. However, if heightened awareness were
responsible for these patterns, we would expect to see a positive
relationship between the number of publications including disease and biotoxicity as keywords and the proportion of MMEs
attributed to these causes. Analysis of this relationship shows
that although scientific attention to these topical areas has increased, these increases are rarely coincident with the proportion
of MMEs attributable to a particular cause for a particular period (SI Appendix, Fig. S8). Thus, although we suspect that
heightened awareness of these issues has increased, both disease
and biotoxicity likely remain important causes of MMEs that
have changed through time.
PNAS Early Edition | 3 of 6
Fig. 3. Causes of MMEs and associated variation in the magnitude of events
for different causes. (Left) Bars quantify the total number of mortality events
for a given cause, and lines within the bars indicate the relative change in the
occurrence of each cause from 1940 to 2009. *Significant temporal trends.
(Right) Variation in the magnitude of mass mortality across causal categories
after taking into account taxon-specific temporal trends. Shown are the residuals (mean ± 1 SE) from an ANCOVA model, including taxa and year; that is,
a model that describes the expected mortality magnitude without taking into
account the causes. Causes with larger magnitudes of death than the average
will have positive residuals (e.g., starvation), whereas causes with smaller
magnitude than the average will have negative residuals (e.g., O2 stress).
However, not all taxa exhibited uniform increases in MMEs attributed to disease and biotoxicity (Fig. 4; ANOVA year × cause ×
taxa; F45,210 = 1.47; P = 0.038). Whereas fishes have had consistent
proportions of MMEs attributed to disease and biotoxicity, birds
have only recently seen an increase in the frequency of MMEs
attributed to these causes. Moreover, fishes and birds exhibited
sustained instances of MMEs caused by direct human perturbations, but mammal MMEs only included such causes in the last
three decades (Fig. 4). Therefore, the observed shifts in MME
causes might not be driven entirely by reporting or by a detectability bias, yet the varied emphasis put forth by the scientific
community on these topics complicates this interpretation (SI
Appendix, Fig. S8). Nevertheless, it remains possible that both
infectious disease and biotoxicity caused by toxic algal blooms,
which are commonly associated with recent land-use alterations
and climate variability (29, 30), may increasingly cause MMEs,
rather than simply being chronic and ongoing perturbations to
natural systems.
In addition, recent shifts in MME causes may result from
underlying patterns in the occurrence of MMEs within certain
taxa. For example, amphibians and reptile MMEs are overwhelmingly associated with bacterial, fungal, and viral infections
(Fig. 4), and reports of amphibian and reptilian MMEs sharply
increased during the past several decades (Fig. 1), thus increasing the proportion of all MMEs attributed to disease.
Similarly, the overall relative occurrence of thermal stress as
a casual factor of MMEs has been declining through time for
birds, fishes, and marine invertebrates (Fig. 4). A closer examination of the patterns associated with thermal stress reveals that
this trend resulted from reductions in the occurrence of cold
thermal stress events (SI Appendix, Fig. S9), whereas events related to hot thermal stress, although infrequent (n = 6 events),
have only appeared since the 1980s. The decrease in MMEs
attributed to cold thermal stress may relate to the concurrent
reductions in the severity of winter temperatures (18), and it is
likely that trends toward increases in summer temperatures may
result in an increased occurrence of hot thermal stress events in
4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1414894112
the future (31). Such trends in thermal stress events raise an
important consideration regarding the underlying mechanisms of
MMEs driven by environmental forces. MMEs arising from
changes in the abiotic environment either may arise from large
changes in environmental conditions (e.g., temperature or toxin
concentration) or, alternatively, may arise from smaller environmental changes that have a disproportionately large negative
fitness effect if biological thresholds are reached (32, 33). Thus,
both incremental and episodic environmental change may contribute to the recent increased occurrence of MMEs (Fig. 1).
Interestingly, the causes of MMEs also varied in their associated magnitudes. After taking into account variation among taxa
in the temporal trends in MME magnitudes (Fig. 2), we found
that multiple stressors, starvation, and disease were associated
with the largest MME magnitudes, whereas oxygen stress, toxicity, and desiccation were associated with the smallest MME
magnitudes (Fig. 3). Identifying the factors generating variation
in MME magnitude is important because it may improve predictions of MME magnitudes, assuming the current trends in
MME causes persist. However, it is possible for MME causes
to shift independent of trends in MME magnitudes. Mammals
exhibited no directional changes in MME magnitude through
time (Fig. 2), yet the occurrence of mammal MMEs caused by
desiccation decreased through time (Fig. 4) and resulted in the
overall decrease in desiccation-related MMEs across taxa (Fig. 3).
Overall, the interpretation of patterns related to MME causes
or taxa should be considered in the context of sources of bias
that may be present. The high proportion of MMEs reported in
North America and Europe (SI Appendix, Fig. S10) reflects
a reporting bias toward areas containing high human densities
and areas where ecologists often conduct fieldwork (4, 34).
Moreover, certain causes of MMEs are likely underrepresented
in the publication records as a result of being difficult to detect
(29). For example, causes of mortality affecting aquatic taxa are
likely underreported because dead organisms can sink out of
sight (7) or occur in the open ocean. In addition, losses of highly
gregarious organisms or organisms that dominate biological
communities in terms of relative abundance may be reported
more often relative to losses of rare organisms, for which losses
may even go undetected. Finally, the geographic location of
MMEs likely determines whether it is encountered, and thus
reported; for example, higher temperatures in tropical ecosystems can accelerate decomposition rates, shortening the window
within which MMEs can be observed.
Fig. 4. Relative frequency of each cause of mortality for each taxon through
time. Bars indicate the relative frequency of an MME cause for particular taxa
for each decade. Decades are defined by their start year (e.g., 1940 represents
the period 1940–1949).
Fey et al.
Materials and Methods
Literature Search. We reviewed the primary literature by searching the ISI Web of
Knowledge database (The Thompson Corporation) for authors self-identifying and
reporting MMEs using a keyword search for one or combinations of the following
terms: mass mortalit*, die off*, die-off*, mass death*, kill*, mass kill*, or unusual
mortality event, combined with a search term to select for the following organisms: birds (bird*, avian), fish (fish*), mammals (mammal*), invertebrate (marine
invert* OR terrestrial invert* OR freshwater invert*), reptiles (lizard* OR snake*
OR turtle* OR reptile* OR crocodile* OR alligator* OR caiman*), or amphibians
(frog* OR toad* OR salamander* OR newt* OR caecilian* OR treefrog* OR
amphibian*). As a second step, we examined the Introduction, Discussion, and
References Cited sections of all relevant papers identified through our Web of
Knowledge search to identify additional studies for inclusion. We did not include
terrestrial invertebrates or freshwater invertebrates as taxa in our database because of the low number of studies that met our inclusion criteria (n = 0 and n = 1
studies, respectively). We finished conducting our search during September 2012,
and as such, our database includes studies published through that time.
We included studies satisfying the following a priori criteria: mortality events
must have occurred in a wild population, the mortality event must not be the
result of an experiment, and each entry must be a unique event. Although we
attempted to be as comprehensive as possible, we recognize that our retention
criteria excluded some MMEs. In particular, coral reef MMEs in the form of
catastrophic bleaching event die offs (25) were not included in our analysis, as
whole-colony die-offs are often reported, instead of individual-level deaths.
In addition, we excluded studies in which the authors indicated that the
observed population declines might be a result of emigration as well as events
based on paleontologic findings. Our final database yielded 727 MMEs.
For each event, we extracted data on the event start and end date, geographic
location, event habitat, number of organisms that died, species identities of the
organisms that died, cause of the event, and percentage of the population
removed during the event, when available. We integrated qualitative information that existed for MME magnitude by making conservative estimates for
such data (e.g., reporting “thousands of individuals died” as 2,000 individuals).
After all information had been entered into the database, two authors
independently examined the specific cause of MME reported by the authors
of the original paper and assigned it to one of 10 previously agreed on
general categories associated with MMEs (biotoxicity, desiccation, direct human perturbation, disease, multiple stressors, oxygen stress, starvation, thermal
stress, toxicity, or weather). The multiple stressors category consisted either
of events described as being caused by two or more stressor categories or
by two independent stressors within the same category (SI Appendix, Fig.
S11). Events caused by multiple stressors on average consisted of events with
2.32 ± 0.60 (mean ± 1 SD) contributing factors, with biotoxicity combined with
oxygen stress and toxicity combined with oxygen stress being the two most
common combinations of contributing stressors (SI Appendix, Fig. S11). It is
possible that this category is underrepresented, as events listed as having
a singular cause may have had additional interacting stressors not reported.
Fey et al.
For example, because oxygen solubility decreases at high temperatures, oxygen stress and thermal stress are often coupled in aquatic ecosystems. When
the cause did not fit into one of these existing categories, the cause was listed
as “other.” For instances in which we disagreed on the appropriate category
of cause (n = 22 of 727), we discussed the paper together to reach a consensus.
Statistical Analyses. To compare temporal trends in the occurrence of MMEs
for a given taxon relative to the number of publications per taxon, we recorded
the number of citations in Web of Knowledge every year that contained
the name of each taxon (singular and plural) as a keyword in the topic field
(e.g., bird*). For the purpose of our study, we considered squamates, turtles,
crocodilians, and tuatara as belonging to the taxon reptiles, and cartilaginous,
ray-finned, lobe-finned, and jawless fish as belonging to the taxon fishes. As
the number of reported MMEs each year is relatively small, we summed our
data over 5-y periods for analyses of both the number of MMEs and the
number of citations. Because temporal trends for both metrics were exponential, we log10-transformed the data, and therefore assessed the linear
trend in order of magnitude and the difference between the number of MMEs
and the number of publications with an analysis of covariance (ANCOVA) (SI
Appendix, Fig. S1). We used the normalized change in order of magnitude (by
subtracting the mean and dividing by the SD in each metric) as the response
variable, year as a linear variable, and type of record (MME or publication
record) as the covariate. To detect nonlinear trends in the magnitude of
mortality over time, we used a local regression (LOESS), which fits simple linear
regressions on each point of the explanatory variable (i.e., time). The difference from classical models is that only a subsample of the data can be used to
calculate the local regression, and the data points are weighted such that the
more distant data are down-weighted. In this particular case, we used all of
the data available, and a standard tricubic weighting proportional to ½1 −
ðdistance=maxdistanceÞ3 3 , with maxdistance being 1.5 times the actual maximal distance (e.g., 70 y with data from 1940 to 2010) (35).
We calculated the order of magnitude of MME size as the log10 of the
conservative estimate of mortality (described earlier). To account for a potential bias linked to uneven publication record, we randomly resampled the
dataset 10,000 times, using the same number of events per decade (22). For
each taxon, we counted the number of events per decade and resampled, with
replacement, within each decade and for each taxon, a number of events
equal to the median number of events per decade and taxon. Within each
resampling, we computed the linear regression of the magnitude of events
(log10-transformed conservative estimate of individual death) over time.
To determine temporal trends in the relative occurrence of MME causes,
we computed the proportion of MMEs in each causal category within each
decade and assessed change over time with linear regression. A small fraction
of records did not determine a cause (n = 61 studies) or were categorized as
“other” (n = 11 studies), and these were not included in this and subsequent
analyses related to the causes of MMEs.
The lag between the occurrence of an MME and its publication was calculated as the difference in years between the year of publication and
the year of the event, and the trend over time was assessed with linear
regression. When an MME spanned more than 1 y, the starting year of the event
was used to assess the lag (n = 4 studies). We used a conservative approach to
estimate the extent of MMEs that have occurred but have not yet been published. We first created an empirical distribution of the lag between the year of
MMEs and the year of their eventual publication. As the actual lag distribution is
skewed on the basis of the fact that many events between 1990 and 2014 have
likely not been published (SI Appendix, Fig. S3), we only used the data between
1940 and 1990. From this distribution of lag time between MME occurrence and
publication, we estimated the proportion of MMEs that have been published
within 2-, 5-, 10-, 15-, 20-, 25-, and 30-y intervals. With these estimated proportions, we could adjust the number of reported MMEs as if all MMEs had been
published. For the expected number of citations in 2005–2009 (the most recent
5-y period of interest), we used a lag of 2 y to generate a conservative measure,
as using a lag of 0 y would show a dramatic increase of predicted events.
To estimate differences in order of magnitude of mortality among causes, we
first performed an ANCOVA of the log10-transformed number of deaths over
time within each taxon. We then computed an analysis of variance with the
residuals of the ANCOVA model as the response variable and the cause of death
as the explanatory variable to determine whether certain causes led to MMEs
of significantly greater or lower mortality than the average expected mortality
based on the ANCOVA model. Finally, we estimated the proportion of variance
in order of magnitude of MME explained by each variable with a multiple regression (using least square estimation) with the log10-transformed mortality
rate as response variable and the cause of death and the interaction between
year and taxa as explanatory variables. All analyses were performed with
R (R Development Core Team), and effects were considered significant at α = 0.05.
PNAS Early Edition | 5 of 6
ECOLOGY
In conclusion, our analysis of published animal MMEs indicates that the magnitude of MMEs has been undergoing taxonspecific shifts and that MMEs associated with multiple stressors
and disease, which are associated with the largest MME magnitudes, are increasing in frequency. However, it is difficult at
this time to determine how much of this increase reflects improved detection capabilities and a greater emphasis on these
research topics. Determining whether or not the upswing in the
occurrence of MMEs is a real phenomenon or simply a result of
increased awareness remains a critical challenge that needs to be
addressed. Such results, combined with lack of studies measuring
MMEs using population-level data, highlights the need for an
improved program for monitoring MMEs. Beyond data standardization, we encourage increased coordination of data collection networks such as citizen scientist contributions, state and
federal agencies, and academic scientists. At this time, the vast
majority of MMEs are presented in newspapers (7) and do not
find their way into the primary literature. Ultimately, enhancing
the study of MMEs will enable an appropriate integration of rare
demographic processes into established ecological and evolutionary
theory. Although MMEs are a natural occurrence, as we continue
to proceed through an era of dramatic changes to Earth’s physical
(17, 18) and biological systems (5, 29), a heightened awareness and
robust detection network (6, 29) may be warranted.
ACKNOWLEDGMENTS. We are grateful to C. Benkman, K. Boersma,
R. Calsbeek, J. Losos, M. McPeek, G. Mittlebach, and three anonymous
reviewers for reading and commenting on earlier versions of the manuscript, and E. Irwin and J. Lawrence for database assistance. S.B.F. was
supported by an Environmental Protection Agency STAR Fellowship and
a James S. McDonnell Postdoctoral Fellowship; A.M.S. was supported by
National Science Foundation (NSF) DEB-1255318; S.N. was supported by
a Swiss National Science Foundation Early Postdoc Mobility Fellowship;
J.L.H. was supported by NSF DGE-0946797; and K.C-Y. was supported by
NSF DGE-1106400.
1. Lande R (1993) Risks of population extinction from demographic and environmental
stochasticity and random catastrophes. Am Nat 142(6):911–927.
2. Reed D, O’Grady J, Ballou J, Frankham R (2003) The frequency and severity of catastrophic die-offs in vertebrates. Anim Conserv 6:109–114.
3. Ameca Y Juárez EI, Mace GM, Cowlishaw G, Pettorelli N (2012) Natural population dieoffs: Causes and consequences for terrestrial mammals. Trends Ecol Evol 27(5):272–277.
4. Ceballos G, Ehrlich PR (2002) Mammal population losses and the extinction crisis.
Science 296(5569):904–907.
5. Barnosky AD, et al. (2011) Has the Earth’s sixth mass extinction already arrived? Nature 471(7336):51–57.
6. Stokstad E (2014) Death of the stars. Science 344(6183):464–467.
7. La VT, Cooke SJ (2011) Advancing the science and practice of fish kill investigations.
Rev Fish Sci 19(1):21–33.
8. Ostfeld RS, Keesing F (2000) Pulsed resources and community dynamics of consumers
in terrestrial ecosystems. Trends Ecol Evol 15(6):232–237.
9. Thébault E, Fontaine C (2010) Stability of ecological communities and the architecture
of mutualistic and trophic networks. Science 329(5993):853–856.
10. Raup DM, Sepkoski JJ, Jr (1982) Mass extinctions in the marine fossil record. Science
215(4539):1501–1503.
11. Jablonski D (1986) Background and mass extinctions: The alternation of macroevolutionary regimes. Science 231(4734):129–133.
12. Potts SG, et al. (2010) Global pollinator declines: Trends, impacts and drivers. Trends
Ecol Evol 25(6):345–353.
13. Morgan OW, et al. (2006) Mass fatality management following the South Asian
tsunami disaster: Case studies in Thailand, Indonesia, and Sri Lanka. PLoS Med 3(6):e195.
14. Young T (1994) Natural die-offs of large mammals - implications for conservation.
Conserv Biol 8(2):410–418.
15. Harding D, Fox C, Mehta J (2002) Studying rare events through qualitative case
studies - lessons from a study of rampage school shootings. Sociol Methods Res 31(2):
174–217.
16. Caughley G (1994) Directions in conservation biology. J Anim Ecol 63(2):215–244.
17. Rahmstorf S, Coumou D (2011) Increase of extreme events in a warming world. Proc
Natl Acad Sci USA 108(44):17905–17909.
18. Donat MG, et al. (2013) Updated analyses of temperature and precipitation extreme
indices since the beginning of the twentieth century: The HadEX2 dataset. J Geophys
Res Atmos 118(5):2098–2118.
19. Grinker RR (2007) Unstrange Minds: Remapping the World of Autism (Basic Books,
Philadelphia, PA).
20. Beebee T, Griffiths R (2005) The amphibian decline crisis: A watershed for conservation biology? Biol Conserv 125(3):271–285.
21. Böhm M, et al. (2013) The conservation status of the world’s reptiles. Biol Conserv 157:
372–385.
22. Hassall C, Thompson DJ (2010) Accounting for recorder effort in the detection of
range shifts from historical data. Methods Ecol Evol 1(4):343–350.
23. Shumway S, Allen S, Boersma P (2003) Marine birds and harmful algal blooms: Sporadic victims or under-reported events? Harmful Algae 2(1):1–17.
24. IUCN (2013) The IUCN red list of threatened species. Version 2013.2. Available at
www.iucnredlist.org. Accessed January 10, 2014.
25. Hughes TP (1994) Catastrophes, phase shifts, and large-scale degradation of a Caribbean coral reef. Science 265(5178):1547–1551.
26. Galetti M, Dirzo R (2013) Ecological and evolutionary consequences of living in a defaunated world. Biol Conserv 163:1–6.
27. Dirzo R, et al. (2014) Defaunation in the Anthropocene. Science 345(6195):401–406.
28. Price SJ, et al. (2014) Collapse of amphibian communities due to an introduced ranavirus. Curr Biol 24(21):2586–2591.
29. Harvell CD, et al. (1999) Emerging marine diseases—climate links and anthropogenic
factors. Science 285(5433):1505–1510.
30. Brookes JD, Carey CC (2011) Resilience to blooms. Science 334(6052):46–47.
31. Pairaud IL, Bensoussan N, Garreau P, Faure V, Garrabou J (2014) Impacts of climate
change on coastal benthic ecosystems: Assessing the current risk of mortality outbreaks associated with thermal stress in NW Mediterranean coastal areas. Ocean Dyn
64(1):103–115.
32. Dillon ME, Wang G, Huey RB (2010) Global metabolic impacts of recent climate
warming. Nature 467(7316):704–706.
33. Sinervo B, et al. (2010) Erosion of lizard diversity by climate change and altered
thermal niches. Science 328(5980):894–899.
34. Siepielski AM, et al. (2013) The spatial patterns of directional phenotypic selection.
Ecol Lett 16(11):1382–1392.
35. Cleveland WS, Grosse E, Shyu WM (1992) Local regression models. Statistical Models
in S, eds Chambers JM, Hastie TJ (CRC Press LLC, Boca Raton, FL), pp 309–376.
6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1414894112
Fey et al.
A
Number of events or of publications (log10)
10
Log. number of publications
slope = 0.025 ± 0.002
F1,12 = 173, r2 = 0.93 [***]
Log. number of reported events
slope = 0.020 ± 0.002
F1,12 = 85.7, r2 = 0.87 [***]
●
8
●
6
●
●
●
●
●
●
●
●
●
●
●
●
●
4
2
0
1940
1950
1960
1970
B
1980
1990
2000
2010
Year
Change in order of magnitude
0.15
0.10
0.05
0.00
F1,17 = 0.8
p = 0.38
F1,24 = 2.27
p = 0.15
F1,24 = 0.32
p = 0.58
F1,19 = 2.52
p = 0.13
F1,22 = 0.23
p = 0.63
F1,14 = 5.44
p = 0.04
F1,24 = 0.01
p = 0.92
Amphibians
Birds
Fish
Inverts
Mammals
Reptiles
Total
−0.05
Figure S1. Examining potential publication bias in reported animal mass mortality events.
(A) Both the number of reported mass mortality events and the number of publications for all
taxa increase through time. (B) ANCOVA results comparing temporal trends between the
number of publications for each taxon and the number of reported mass mortality events for each
taxon. Bars are the slopes of each taxon-specific linear regression ± 1 SE. Darker colors
correspond with reported numbers of mass mortality events and lighter colors correspond with
total citations. The similar temporal trends between reported mass mortality events and
publications related to each taxon suggest that some of the increases in reported mass mortality
events may be attributed to an overall increase in productivity in the scientific community.
2
10
●
6
●
●
●
●
●
1000
2000
20
4000
5000
●
t[, i + 8]
Amphibians
r−squared = 0.41
p−value = 0.12
25
3000
●
●
●
●
●
●
●
●
0
6
50
100
Reptiles
r−squared = 0.85
p−value = 0.079
5
●
●
●
●
●
2
150
200
250
300
●
t[, i + 8]
t[, i + 1]
4
15
●
10
0
70
10
20
30
Fish
r−squared = 0.55
p−value = 0.0025
60
50
40
40
t[, i + 8]
●
●
50
60
70
●
0
●
●
●
10
●
●
●●●
0
●
200
t[, i + 8]
400
500
600
t[, i + 8]
50
60
70
●
●
●
●
●
2
300
40
6
●
0
100
30
8
4
●
20
Invertebrates
r−squared = 0.22
p−value = 0.21
12
20
0
●
10
●
●
0
30
10
●
1
●
0
3
2
●
●●
5
t[, i + 1]
Total number
t[, i + 1]
of events
6
0
0
●
●
8
4
●
0
t[, i + 1]
Birds
r−squared = 0.16
p−value = 0.15
12
●
●
8
14
●
t[, i + 1]
t[, i + 1]
10
4
●
Mammals
r−squared = 0.54
p−value = 0.0068
12
0
● ●
200
Total number of citations [x1000]
400
●
600
800
t[, i + 8]
Figure S2: Variation in mass mortality events occurrence explained by the variation in
taxon-specific publication trends. The number of mass mortality events for a five-year period
for each taxon is plotted against the total number of taxon-related citations for the same five-year
period (as depicted in Fig. 1). The percent of variance in reported mass mortality events not
explained by the publication trends (1-r2 values reported above) was on average 54.5% (Range=
15-84%), suggesting that much of the variation in mass mortality event occurence is not
explained by a publication bias alone.
Lag between event and publication [years]
20
Slope = 0.07 ± 0.01 SE
F1,693 = 53.6
p < 0.0001
15
Amphibians
Birds
Fish
Invertebrates
Mammals
Reptiles
10
5
0
1940
1950
1960
1970
1980
1990
2000
2010
Publication year
Fig. S3. Publication lag between the occurrence and publication of animal mass mortality
events. The line is the least squares linear regression and the shading demarcates slope 95%
confidence intervals. Each point is a single mass mortality event. The increased publication delay
through time may be driven by an increase in the time that scientists take before publishing
findings on a particular event, or by an increased tendency to utilize older, existing data.
Regardless of the mechanisms behind this trend, this lag partly accounts for the recent decreases
observed in the reported number of mass mortality events across all taxa (Fig. S4).
Mammals
20
Birds
5000K 20
4000K
15
350K
280K
15
3000K
210K
10
2000K
5
5
1000K
0
0
1940
Total number of events
140K
1955
1970
1985
0
2000
Amphibians
40
0
1940
75K
60K
30
70K
1955
1970
1985
2000
Reptiles
20
75K
60K
15
45K
20
45K
10
30K
10
30K
5
15K
0
0
1940
1955
1970
1985
0
2000
Fish
120
480K
80
0
1940
600K
100
15K
1955
1970
1985
2000
Invertebrates
20
900K
720K
15
360K
60
Total number of citations
10
540K
10
240K
40
120K
20
0
0
1940
1955
1970
1985
360K
5
180K
0
2000
0
1940
1955
1970
1985
2000
Year
Figure S4: Occurrences of animal mass mortality events and taxon specific publication
trends through time, corrected for predicted publication lags. Colored bars indicate the
number of events over a 5-year interval (e.g., 1940 stands for the 1940-1944 period) and dashed
lines show trends in the total number of papers published each year for each taxon. The faded
portions of bars represent a conservative estimate of the mass mortality events which have
occurred, but have not yet been published based on historical, taxon-specific lag times between
the occurrence and publication of an event (see Materials and Methods).
1B
100M
10M
1M
100K
10K
1K
100
10
1
Mammals
Number of deaths per event
1940
1B
100M
10M
1M
100K
10K
1K
100
10
1
1950
1960
Birds
1970
1980
1990
2000
2010
1940
Amphibians
1940
1B
100M
10M
1M
100K
10K
1K
100
10
1
1950
1960
1950
1960
1970
1980
1990
2000
2010
1970
1980
1990
2000
2010
1970
1980
1990
2000
2010
Reptiles
1970
1980
1990
2000
2010
1940
Fish
1940
1950
1950
1960
Invertebrates
1960
1970
1980
1990
2000
2010
1940
1950
1960
Year
Figure S5: Non-parametric approach to resolving temporal trends in the occurrence of
animal mass mortality events. Each point is a single mass mortality event. Dotted lines
represent the estimated local regression (LOESS), and the fit for each year is computed with a
weighted linear regression using all the points of the dataset. The weight is inversely
proportional to the distance in years between the fitted point and the rest of dataset. Thus, the
more distant the points, the smaller the weight. Here we used a standard tricubic weighting
proportional to 1 − 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒/𝑚𝑎𝑥𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 ! ! , with maxdistance being 1.5 times the actual
maximal distance (e.g., 70 years with data from 1940 to 2010). The shaded intervals demarcate
the 95% confidence intervals.
Mortality change in order of magnitude
0.05
0.00
−0.05
−0.10
Reptiles
Amphibians
Mammals
Birds
Fish
Invertebrates
Fig. S6. Trends in mass mortality event magnitudes through time after accounting for
uneven publication records. The plotted change in mortality rate from 1940 to 2010 was
computed from 10,000 iterations of subsamples of mass mortality events and associated
magnitudes with replacement. Boxes indicate taxon median ± interquartile range; whiskers
indicate the complete range of data. The changes in mortality rate obtained from this iterative
subsampling approach are consistent with the reported temporal trends in mass mortality event
magnitudes (Fig. 2).
35
30
Number of events
25
20
15
10
5
0
10
20
30
40
50
60
70
80
90
100
Percentage of population lost
Fig. S7. Population-level losses from mass mortality events. The distribution of mass
mortality event magnitude as defined by the proportion of the population removed per event. Xaxis labels represent number of events accounted for by the ten percentage points preceding the
listed number (e.g., 100% represents events that reported 91-100% population lost). The large
proportion of high magnitude events may have resulted from reporting bias of researchers
reporting only the most severe mortality events, or shaped by a negative correlation between
small, easy to estimate population sizes, and high relative mortality events. 100000
100%
100%
100
●
●
100%
100%
biotoxicity
100000
disease
100%
100000
100%
●
●
0% ● ●100
●
●
●
●
●
0%
● ●
100
●
100%
100000
0%
●
●
100000
Cit
0%
●
100000
100000
Cit
100
Cit
●
Cit
Cit
●
100%
●
●
●
100000●
●
100000
●
●
100%
●
●
●
●
●
100%
●
●
100
0%
●
●
●
100
●
●
●
100%
●
●
100000
● ●
0%
●
100
100%
●
●
●
●
100%
●
100
●
●
●
year
●
0%
●
●
100
●
100000
100%
●
●
●
●
●
100
●
●
100
100%
year
●
0%
●
●
100000
●
●
●
0%
●
year
100
●
●
●
●
●
●
●
100000
0%
●
0%
●
100
●
●
100
●
●
●
●
●
●
0%
●
100
●
●
●
●
●
●
●
●
0%
●
●
0%
●
●
100
●
●
100
100000
100%
●
●
100
●
●
100000
year
0%
0%
●
●
●
●
100%
●
●
●
●
100000
year
100000
●
●
Cit
Cit
Cit
●
●
●
●
●
●
100000
●
●
0%
100
●
●
100%
●
●
year
100
0%
●
100%
100000
year
●
0%
100
0%
100
●
100%
●
●
100%
100%
year ●
0%
100●
●
●
●
●
0%
●
100000
0%
100
year
●
●
●
●
●
100000
0%
●
100
●
100%
●
●
●
●
●
0%
●
100
●
●
●
●
100%
●
100
●
●
●
0%
●
●
100%
●
●
100000
●
0%
●
100
●
●
●
●
●
100%
●
●
●
●
●
●
●
●
●
●
100%●
●
year
100
0%
●
100000
100
●
0%
100000
100000
● ●
●
●
year
●
0%
100%
●
year
●
●
●
0%
●
●
100
100%
●
●
●
100%
●
100000
●
●
year
year
0%
● ●
●
100000
year
year
●
●
100000
●
●
●
year
●
100%
● ●
●
●
year
year
●
●
●
year
100
100%
●
●
●
0%
100%
●
100
100
●
●
●
100000
●
●
100000
●
100000
●
0%
100%
●
year ●
0%
● ●
●
●
●
●
year
●
100%
●
●
●
100%
●
●
●
●
100000
●
100%
●
100
●
●
year
100
●
year
●
year ●
●
100000
●
0%
●
●
0%
●
100000
●
year
●
●
●
Cit
year
●
●
●
●
●
●
100000
100%
●
100
year
0%
Cit
100000
year
●
●
100000
●
100%
100%
●
year
●
100000
●
100
●
year
0%
0%
●
●
●
100
●
100000
●
100%
0%
100000
●
Cit
100
●
0%
●
100
100%
Cit
0%
●
●
●
year
Cit
100
100%
●
year
Cit
100000
year
0%
●
●
year
●
100000
100
●
●
●
Cit
●
●
year
Cit
●
●
0%
●
●
●
Cit
●
●
●
●
Cit
100%
●
●
● ●
year
Cit
100000
●
●
100000
100
●
100
100%
●
0%
●
●
Cit
●
●
year
Cit
●
year
Cit
0%
●
Cit
Cit
●
Cit
●
year
0%
●
0%
●
0%
●
●
●
100%
Cit
●
●
Cit
●
●
year
100
100%
●
●
100000
●
●
●
100000
●
Cit
●
100000
●
●
●
100
●
oxygen str. + tox. weather + tox.
100%
●
Cit
●
●
100000
0%
Cit
●
Cit
0%
●
●
Cit
●
100000
Cit
●
●
100%
●
year
Cit
Fishes
100
Cit
●
●
year
●
100●
100
●
100%
Cit
●
●
●
0% ●
year
100
100%
100000
●
100%
●
●
●
100000
●
Cit
Reptiles
100
toxicity
100%
Cit
0%
●
100
●
●
year ●
●
year
●
100000
●
Invertebrates
●
●
●
●
●
●
100000
●
0%
Cit
100
Cit
●
●
●
Cit
0%
●
●
●
●
Cit
●
●
●
year
100%
Cit
Amphibians
100
100●
●
●
year
100000
●
●
100000
●
Cit
100
●
●
0%
Cit
●
●
●
●
Cit
●
100%
●
year
Cit
Birds
●
●
100
●
100000
oxygen str.
100%
Cit
0%
100000
●
Cit
●
dessication
100%
Cit
●
100
●
100000
Cit
●
thermal str.
100%
Cit
●
●
100000
Cit
●
human pert.
100%
Cit
Mammals
weather
100000
●
●
●
100%
year
0% ●
●
100
0%
●
Figure S8: Trends in the relative occurrence of animal mass mortality causes compared to
publication trends through time. The x-axis is time binned into 10-year intervals from 19401949 (furthest left) to 2000-2009 (furthest right). Thick lines connected by dots indicate the
number of citations per decade using cause and taxa keywords (note log10 scaling of the left yaxis). Thin smooth lines represent the proportion of each cause (the right y-axis) within each
taxon (data presented in Fig. 4). Columns show the causal categories that represent the nine most
common causes of mass mortality events. Additionally, two of the most common combinations
of causes contributing to events classified as multiple stressors “oxygen stress and toxicity” and
“weather and toxicity” are included to show patterns associated with the “multiple stressors”
category.
25
15
10
0
5
Number of events
20
Cold stress
Hot stress
1940
1950
1960
1970
1980
1990
2000
Decade of event
Figure S9: Temporal trends in the number of reported animal mass mortality events
caused by cold thermal stress (blue bars) and hot thermal stress (red bars). Decade of event
represents the ten-year period following the listed year (e.g., 1940 represents years 1940-1949).
Mass mortality events attributed to cold thermal stress tended to decrease over time, while events
caused by hot thermal stress did not appear in our database until the 1980s.
140
1940
1950
1960
1970
1980
1990
2000
120
Number of events
100
80
60
40
20
Unspecified
Oceania
South America
North America
Europe
Asia
Africa
Antarctica
0
Figure S10: The number of reported mass mortality events through time occurring on each
continent. This graphical analysis reveals a publication bias in where mass mortality events are
reported, as 75% of all reported mass mortality events occur in Europe and North America.
Biotoxicity
0
0
0
0
0
1
0
Weather
0
1
1
Other
11
0
5
0
0
1
0
0
0
5
0
2
1
1
0
24
0
9
3
1
0
Toxicity
Oxygen stress
Figure S11: Commonalities among contributing stressors to animal mass mortality events
categorized as being caused by multiple stressors. Ovals indicate the four most common
contributing factors (biotoxicity, weather, toxicity, and oxygen stress) and the less frequent
contributing factors (combined into ‘other’). The number of animal mass mortality events is
provided in the intersections. The most common combination of two interacting stressors was
oxygen stress and biotoxicity (n = 24 events) and oxygen stress and toxicity (n = 9 events). The
most common combination of three interacting stressors was weather, oxygen stress, and toxicity
(n = 5 events), and weather, other stressors, and toxicity (n = 5 events).