The experiments under HAPPI are designed to be as similar as possible in
experimental design as current (or proposed) climate experiments, notably the
International CLIVAR Climate of the 20th Century Plus Detection and
Attribution (C20C+ D&A) project . Synergies
between the experiments allow to minimise the additional computational time
required from modelling centres. The core experiments will be driven with a
spectrum of different leading atmosphere-only Global Circulation Models
(GCMs), the initial participants of which are listed in Table . By
using atmosphere-only models instead of fully coupled models, we are able to
generate larger ensemble sizes (due to decreased computational cost) while
providing more accurate regional climate projections . Boundary
conditions for the models are taken from the CMIP5 experimental design and
from models that participated in that initiative.
There are two tiers of experiments, intended to characterise various climate
scenarios, as well as uncertainties in the specifications of the
temperature-based scenarios.
Tier 1 experiments
Three core experiments are proposed:
Current decade (2006–2015) conditions (50- to 100-member ensembles).
1.5 ∘C warmer than pre-industrial (1861–1880) conditions
(50- to 100-member ensembles) relevant for the 2106–2115 period.
2.0 ∘C warmer than pre-industrial (1861–1880) conditions
(50- to 100-member ensembles) relevant for the 2106–2115 period.
Each simulation within an experiment differs from the others in its initial
weather state. The use of 50–100 10 year time slices provides 500–1000 years
of data per experiment. Simulations are limited to 10 years in length because
the observed ocean temperatures, upon which all HAPPI experiments are based,
have been approximately constant during this period (at least within the
context of the anthropogenic warming scales considered by HAPPI). However,
10 year periods should provide material for some analysis of multi-year
events, such as droughts. The degree to which the output of the simulations
can be used to estimate unbiased return values for a specific return period
will depend on various aspects of the event, such as region and climate
variables. In the extratropical summer, for instance, the 500–1000 years may
be considered an unbiased sample, whereas in the tropics it may be important
to acknowledge the major El Niño and La Niña events during the 2006–2015
period.
Time series of global annual mean surface-air temperature anomalies
(relative to 1861–1880) from CMIP5 RCP2.6 and RCP4.5 experiments. Solid
lines show the multi-model mean and shaded regions show the 5–95 % range
across all 26 models. Only one simulation is used for each model. All models
where the data were available for both scenarios were used, leading to 26
models in total.
![]()
Left: SST warming pattern added to the current decade to produce
the (top) 1.5∘ and (bottom) 2∘ scenarios. Right: the standard deviation
of annual mean delta SSTs across the 23 models. Units are in
∘C.
![]()
Polar stereographic projections of decadal-mean (top) sea ice
concentration from the 1.5∘ experiment and (bottom) the difference in
sea ice concentration between the 1.5∘ experiment and OSTIA. The OSTIA
data cover the decade 2006–2015. Left panels show the Northern Hemisphere, right panels show
the Southern Hemisphere.
![]()
Current decade experiment:
Modelling centres will use observed
forcing conditions as in the DECK AMIP design, including Sea Surface
Temperatures (SSTs) and sea ice . The 2006–2015 decade is
chosen because it is our most recently observed period, but also because it
contains a range of different SST patterns over the decade, allowing for an
assessment of how the ocean conditions vary on inter-annual timescales. From
2017 onward, modelling centres will also have the option of simulating
observed 2016 climate, thereby capturing the large El Niño event in
2015–2016. Note that the C20C project will also perform these experiments.
The 1.5 ∘C experiment:
It is difficult (without many
climate-model-specific iterations) to explicitly design an emissions scenario
that would lead to a world exactly 1.5 ∘C warmer than pre-industrial
conditions. This is because the CMIP community are set up to use particular
emission scenarios or RCP scenarios, rather than a scenario that leads to
some chosen amount of warming. Here, we take 1.5 ∘C to mean
“1.5 ∘C as measured as the near-surface air temperature”, as is the
formal definition of the transient climate response, rather than some mix of
measuring systems (for instance surface ocean) that may have implications for
the energy budget .
By chance, the multi-model average across climate model simulations submitted to CMIP5
under the RCP2.6 forcing scenario results in a global average temperature
response at 1.55 ∘C relative to pre-industrial levels (2091–2100 relative to
1861–1880). Figure shows the average and 5–95 % spread in
global-mean temperature anomaly for all available CMIP5 models for the RCP2.6
scenario (dark blue). Within HAPPI, we assume that this amount of warming is
sufficiently close to inform the call of the UNFCCC on a special report on
the “impacts of global warming of 1.5 ∘C above pre-industrial
levels” , and thus HAPPI adopts the end-of-century
anthropogenic radiative forcing conditions from the RCP2.6 emissions
scenario. Specifically, forcing values for the year 2095 for greenhouse gases, aerosol, land-use, and land-cover changes are repeated for each of the years within the 1.5 ∘C
decade. Natural radiative forcings, however, are set to the same values as in
the current-decade experiment.
Projected SSTs are calculated by adding to the observed 2006–2015 SSTs a
change in SST (ΔSST) between the decadal average of the modelled
2006–2015 period and the decadal average of the modelled 1.5 ∘C
world over 2091–2100. Hence the SST patterns are still time-varying because
they are based on the 2006–2015 observations, but they have an additional
warming added to them. As CMIP5 historical simulations stopped in 2005, the
decadal average of the 2005–2015 SSTs is estimated from RCP8.5 simulations,
as this is the scenario that is closest to observations over this period. The
decadal average of the 2091–2100 SSTs is estimated from CMIP5 RCP2.6
simulations. The spread of these models is shown in Fig. , of
which 23 models have the required data (see Sect. for more details
on the individual patterns). The resulting multi-model average ΔSST,
used in the 1.5 ∘C experiment, is shown in Fig. . The
global-mean SST response is 1.02 ∘C relative to the pre-industrial
period, with larger warming over land providing the global 1.55 ∘C
total. Because of the time period we use as our baseline (2006–2015) some of
the so-called hiatus effect may bias our results cold, and this will be partially
compensated for by the fact that our global-mean temperature is 0.05 ∘C
higher than desired, but we also note that it is the difference between the
0.5 ∘C warming in these relatively low emission scenarios that is
important, rather than the exact magnitude.
Estimated sea ice is more problematic than estimated SSTs, because the
CMIP-projected Arctic and Antarctic sea ice extents vary dramatically between
models . In the Arctic, most climate models show a
decrease at all longitudes in sea ice. In the Antarctic, the overall model
responses show a similar decrease with equally variable projections. The
CMIP5 climate models are also unable to capture the observed increases in
Antarctic sea ice over the satellite era , leading to low
confidence in their ability to predict future changes. As such, we use a
different method to estimate sea ice under 1.5 ∘C and higher
scenarios, which is an adaptation of . In short, we
calculate an anomaly (from 1996 to 2015) for every month from 1996 to 2015 in both
SSTs and sea ice from the operational sea surface temperature and sea ice
analysis (OSTIA) data set and fit a linear
relationship between SSTs and sea ice as a function of month and grid box. We
use as the regressor the meridional average of SST grid boxes, within a
hemisphere, at grid points where there is ice present at some point in time
between 1996 and 2015 (i.e. the climatological monthly mean ice concentration
for the grid box is non-zero). This represents temperature at that longitude
under and near the ice edge, thereby minimising poorly observed values in
ice-covered regions. We use ice cover in an index grid box as the regressand,
and smooth the resultant field with a 500 km smoother. We then apply the sea
ice–SST relationship to the 1.5 ∘C experiment SST anomalies, to give
a projected sea ice concentration anomaly. These anomalies are added on to
the observed OSTIA data spanning the most recent decade. The absolute sea ice
concentration fields and anomalies from observations are given in Fig. .
This methodology has the added benefit that the SSTs and sea ice
are consistent with each other in the HAPPI experiments.
The 2 ∘C experiment:
For the 2 ∘C experiment, no
analogous CMIP5 simulations are available. The RCP scenario resulting in the
second coolest temperatures by the end of the 21st century is RCP4.5, which
reaches ∼ 2.5 ∘C relative to pre-industrial levels by the end of the
21st century (Fig. ). Both RCP2.6 and RCP4.5 have 5–95 % ranges
that overlap a global mean temperature of 2 ∘C, and the mean of both scenarios are a
similar distance from this threshold.
To calculate the future SST and sea ice conditions of a 2 ∘C world
we therefore take a weighted sum of the two RCP scenarios,
W1 × RCP2.6 + W2 × RCP4.5. The weights are calculated such that the global-mean
temperature response is 2.05 ∘C (i.e. exactly half a degree above
the 1.55 ∘C response from the 1.5 ∘C experiment), and
results in W1 =0.41 and W2 =0.59. These weights are used to calculate the
SSTs and sea ice coverage using the same methodology as in the
1.5 ∘C experiment.
The same weightings are applied to the radiative forcing of each well-mixed
greenhouse gas (e.g. CO2, CH4, N2O, CFCs etc). Some concentrations do not
scale linearly with radiative forcing, for instance CO2 concentrations
following a logarithm, and the CH4 and N2O concentrations follow a square
root. All other concentrations are linearly related to the radiative forcing.
A full list of these relationships is given by the IPCC .
Natural forcings remain at the 1.5 ∘C experiment (and current-decade
experiment) values. Land cover and land use are represented in a discretised form in
the climate models, and so cannot be interpolated. Meanwhile, the climate
responses to anthropogenic aerosols and ozone concentrations (or, for some
models, emissions of their precursors) do not follow a simple functional
form, and in the case of aerosols this is further complicated by major
differences in the spatial distributions of concentrations between the two
RCPs. Considering that the parties of the UNFCCC are most concerned about a
CO2-dominated warming, and CO2 is the dominant contributor to changes in the
radiative budget by 2100 e.g. see Fig. 12.3 of , we
chose to set the remaining (i.e. other than CO2, SST, sea ice, and
natural forcings) 2 ∘C experiment forcings to their 1.5 ∘C
experiment values.
In addition to the three core experiments, modelling centres will also run at
least five ensemble members spanning the period 1959–2015, thereby allowing
for a range of biases in the climate models to be assessed (see Sect. ).
Tier 2 experiments
The Tier 2 experiments will replicate the Tier 1 1.5 and 2 ∘C
experiments, but also take into account SST and sea ice uncertainty at the
expense of ensemble size. Individual estimates of SST response patterns from
the 23 different CMIP5 models will be used, the annual means of which are
presented in Appendix A for both scenarios. Each individual model pattern
will be scaled to have the same SST mean response as the multi-model mean
(MMM) response (1.02 ∘C for the 1.5 ∘C experiment); this
will give a measure of the impact of uncertainty in the pattern of
large-scale warming, conditioned on a specific global temperature change,
consistent with research demanded by the UNFCCC call.
Additional Tier 2 experiments will determine the sensitivity of the response
to 1.5 and 2.0 ∘C of warming to the inclusion of
atmosphere–ocean interactions in models, and hence to the choice of an
AMIP-type approach for the Tier 1 HAPPI experiments. This is an important
question, given that air–sea feedbacks have been shown to affect the fidelity
of model representations of key phenomena that control weather and climate
extremes e.g. the Madden–Julian oscillation;. These
Tier 2 experiments use atmospheric GCMs coupled to either one-dimensional
mixed-layer oceans (i.e. with vertical resolution) or zero-dimensional slab
oceans. These models require bias corrections to either the full vertical
profile of temperature and salinity (for mixed-layer oceans) or to SST (for
slab oceans) to represent missing ocean dynamics and to correct for biases in
atmospheric surface fluxes e.g.. A key advantage of
these models is that they can maintain a given global-mean temperature
effectively indefinitely. They do not include modes of coupled
atmosphere–ocean variability that rely on ocean dynamics (e.g. the El
Niño–Southern Oscillation or the Indian Ocean Dipole), which can be an
advantage as it avoids issues of under-sampling natural variability. These
models are also much less computationally expensive than coupled models with
full ocean GCMs.
Here, we describe the experiment design for Tier 2 experiments with the
MetUM-GOML2 model, which comprises the Global Atmosphere 6.0 configuration of
the Met Office Unified Model coupled to the Multi-Column
K Profile Parameterisation mixed-layer ocean (MC-KPP), as described in
. First, we perform a present-day ensemble using forcing
for the 1976–2005 period: greenhouse gases and aerosols are set to the
average values of the period 1976–2005; temperature and salinity corrections
constrain MC-KPP to the ocean climatology from ; and
climatological sea ice extent and concentrations are prescribed.
Climatological SSTs are also prescribed in regions of seasonal sea ice cover
in the high latitudes, where the model is not coupled
see. 1976–2005 differs from the 2006–2015 period
chosen for the Tier 1 experiments, but the objective is to understand the
effect of air–sea coupling on the response to warming, not to compare the
MetUM-GOML2 present-day simulation to any other model.
Secondly, we adjust the CO2 in MetUM-GOML2 to achieve target global-mean
warming levels, relative to the present-day ensemble, consistent with 1.5 and
2.0 ∘C above pre-industrial levels, measured by near-surface air
temperature. The target levels are computed by first finding the observed
global-mean surface temperature difference between 1976–2005 and
pre-industrial values, which is 0.52 ∘C in HadCRUT4. The target levels are
set to 1.5 and 2.0 ∘C minus this difference, or 0.98 and
1.48 ∘C, respectively. This is equivalent to projecting the change
between a 1.5 or 2.0 ∘C warmer world and the 1976–2005 period.
Finding the correct CO2 concentrations involves a trial-and-error
approach, but the effort is mitigated by the fact that warming is a roughly
linear function of CO2 (for small amounts of warming) and the model
reaches steady state in 5–10 years. There are no changes to the temperature
or salinity corrections, which assumes that the mean ocean heat and salt
transports do not change for relatively small warming. However, we impose
changes to sea ice and the prescribed SSTs in uncoupled (seasonally
ice-covered) regions. We compute these using a transient simulation of the
fully coupled MetUM-GC2 with a 1 % yr-1
CO2 increase, by averaging 20 year periods with global-mean warming
closest to our 0.98 and 1.48 ∘C target levels and taking the
difference between these periods and the climatology of the MetUM-GC2
present-day control simulation. We apply these differences to the 1976–2005
observed climatologies.
Thirdly, we perform initial condition perturbation ensembles of MetUM-GOML2
simulations at the target warming levels, using the CO2 concentrations and
sea ice and high-latitude SST boundary conditions determined above. Finally,
we perform AMIP-type experiments with the same atmospheric model, in which we
prescribe the daily SSTs and sea ice from the MetUM-GOML2 ensembles.
MetUM-GOML2 uses a 3 h coupling frequency; converting to daily SSTs
introduces sufficient noise to cause the coupled and atmosphere-only
experiments to diverge.
Comparing the coupled and AMIP-type experiments at the same level of warming
allows one to determine the sensitivity of the response to the presence of
atmosphere–ocean interactions, in a framework in which the mean and
inter-annual variability of SST and sea ice are consistent between the
simulations. Similarly, comparing the relative difference between the 1.5 and
2.0 ∘C simulations in the coupled and AMIP-type experiments allows
one to determine whether the response to an additional half-degree of warming
is sensitive to inclusion of air–sea coupled feedbacks. We expect that
analysis of these experiments will focus mainly on sub-seasonal variability
and extremes (e.g. heatwaves, intense precipitation events), but it is
possible that air–sea coupling will also affect the mean response.