Science - USA (2020-09-25)

OCEAN TEMPERATURE

High-impact marine heatwaves attributable to

human-induced global warming

Charlotte Laufkötter1,2*, Jakob Zscheischler1,2, Thomas L. Frölicher1,2

Marine heatwaves (MHWs)—periods of extremely high ocean temperatures in specific regions—have
occurred in all of Earth’s ocean basins over the past two decades, with severe negative impacts
on marine organisms and ecosystems. However, for most individual MHWs, it is unclear to what extent
they have been altered by human-induced climate change. We show that the occurrence probabilities
of the duration, intensity, and cumulative intensity of most documented, large, and impactful MHWs
have increased more than 20-fold as a result of anthropogenic climate change. MHWs that occurred only
once every hundreds to thousands of years in the preindustrial climate are projected to become
decadal to centennial events under 1.5°C warming conditions and annual to decadal events under
3°C warming conditions. Thus, ambitious climate targets are indispensable to reduce the risks
of substantial MHW impacts.

A

notable number of record-breaking ma-
rine heatwaves (MHWs) ( 1 – 3 )haveoc-
curred in the past decades worldwide,
including the Northeast Pacific 2013 to
2015 MHW ( 4 ); the Northwest Atlantic
2012 MHW ( 5 ); and extreme temperatures in
the Tasman Sea ( 6 ), the Indonesian-Australian
Basin ( 7 ), and the Southern Ocean ( 8 )in2015
and 2016. Recently, a large heatwave emerged
in the Northeast Pacific, which raised con-
cerns that an event similar to the Northeast
Pacific 2013 to 2015 MHW will reappear ( 9 ).
The evolving literature on the impacts of
MHWs ( 10 ) indicates serious consequences for
marine life. Recent MHWs have caused entire
ecosystems to restructure, have led to major
socioeconomic impacts, and have disrupted
ecosystem services ( 1 , 10 ). As an example, the
Northeast Pacific 2013 to 2015 MHW (often
referred to as the blob) has caused increased
mortality of sea birds, salmon, and marine
mammals ( 11 ); very low ocean primary prod-
uctivity ( 12 ); harmful algal blooms ( 13 ); and
large alterations to open-ocean and coastal
ecosystems ( 11 , 14 ). Recent heatwaves in trop-
ical and subtropical waters have caused sub-
stantial coral bleaching ( 15 ), defoliation of sea
grass ( 16 ), and decreases in kelp biomass ( 1 ).
Further impacts include alterations in biodi-
versity patterns of sessile invertebrates and
demersal fish, shifts in community structure,
andpolewardshiftsintropicalfishcommun-
ities. Beyond their impacts on marine eco-
systems, MHWs have led to the closure of
fisheries ( 17 ) and have caused locally-large
carbon outgassing events ( 16 ). Furthermore,
substantial decreases in sea ice have been
linked to individual MHWs ( 18 ).

Understanding whether and by how much

anthropogenic climate change has contrib-

uted to the intensity and likelihood of extreme

events is of substantial scientific and public

interest—for example, to motivate efforts to

limit global warming. Atmospheric heatwaves,

droughts, and floods are routinely attributed

to climate change ( 18 , 19 ); however, there have

been only a few attempts to attribute MHWs

to climate change ( 6 , 20 – 25 ). These publica-

tions have differed strongly in their method-

ologies, in their framings of the attribution

question, and in their treatments of the un-

certainties of the attribution process, which

makes their results difficult to compare and

interpret ( 18 , 26 ). The link between anthro-

pogenic climate change and the occurrence

probability of individual MHWs is therefore

currently not well understood.

Here, we first present an overview of all

large MHWs that have occurred during the

satellite era (September 1981 to December 2017).

We then focus on all large, recent MHWs that

have had a substantial, documented impact on

marine ecosystems or ecosystem services. For

each of those large MHWs, we quantify the

contribution of anthropogenic climate change

to their respective occurrence probability

using a consistent attribution framework. To

this end, we first compute the duration, inten-

sity, and cumulative intensity (the area-weighted

sum of all daily temperature exceedances) of

each of these MHWs on the basis of satellite

observations of sea surface temperature (SST).

We have chosen these metrics because they

are linked to the impact on marine organisms.

Both the duration and the cumulative inten-

sity of MHWs affect the ability of marine or-

ganisms to cope with heat stress ( 1 , 15 ), and

exceeding a certain temperature threshold—

no matter for how long ( 27 )—can be highly

deleterious for some organisms.

For each MHW, we then separately estimate

the probabilities that an MHW has occurred

that equals or exceeds the duration, inten-

sity, and cumulative intensity of the observed

MHW in preindustrial and present-day model

simulations. These probabilities are denoted

by Ppresentdurationday, Pintensitypresentday, Pcumulativeintensitypresentday ,

Ppreindustrialduration ,Pintensitypreindustrial, and Pcumulativeintensitypreindustrial ,

respectively.

Here, we explicitly take changes in the fre-

quency of heatwaves as well as changes in the

duration, intensity, or cumulative intensity

of heatwaves into account (see materials and

methods). Our approach builds on the work

of Stottet al.( 28 ) and Oliveret al.( 6 ) but

with several modifications. In contrast to most

previous attribution studies, we specifically

calculate the occurrence probabilities of heat-

waves as opposed to the probabilities of ex-

treme seasonal mean temperatures, which

allows us to attribute heatwave duration, heat-

wave intensity, and cumulative intensity. Fur-

thermore, building on recent methodological

advances ( 25 , 29 ), we use a rigorous model

preselection, whereby we only allow models

in the attribution analysis in which we find

no statistical evidence that the simulated heat-

wave distributions are different from those in

theobservationalrecord.Finally,weprovide

confidence intervals and quantify the uncer-

tainty that stems from preindustrial temper-

ature reconstructions to assess the robustness

of our attribution statements.

Using the occurrence probabilities, we then

calculate the fraction of attributable risk (FAR)

( 18 , 28 ) for duration as

FAR¼ 1 

Ppreindustrialduration

Ppresentdurationday

and analogously for intensity and cumulative

intensity. Positive FAR values indicate a high-

er likelihood that the respective extreme event

is the result of human influences. For example,

at FAR > 0.5, the likelihood of an event occur-

ringnowisatleasttwiceasthatofitsoccur-

ring under preindustrial conditions. At FAR >

0.8, it is at least five times likelier.

From September 1981 to December 2017,

>30,000 distinct, spatiotemporally contiguous

MHWs have occurred globally (with temper-

atures above the 99.5th percentile; see mate-

rials and methods for the MHW definition).

The majority (90%) of these MHWs lasted

<14 days and covered <0.2 million km^2. How-

ever, the 300 largest MHWs (with the high-

est cumulative intensities) covered, on average,

1.5 million km^2 , lasted 40 days (Fig. 1A), had a

peak temperature anomaly (Fig. 1C) of 5.0°C

above climatology (see materials and methods),

and had a cumulative intensity of 11,900°C

days km^2 (Fig. 1E). The frequency, dura-

tion, intensity, and cumulative intensity of

large MHWs increased during the observa-

tional period (Fig. 1, A, C, and E). In the first

SCIENCEsciencemag.org 25 SEPTEMBER 2020•VOL 369 ISSUE 6511 1621

(^1) Climate and Environmental Physics, Physics Institute,
University of Bern, 3012 Bern, Switzerland.^2 Oeschger Centre
for Climate Change Research, University of Bern, 3012 Bern,
Switzerland.
*Corresponding author. Email: [email protected]
RESEARCH | REPORTS