GMDGeoscientific Model DevelopmentGMDGeosci. Model Dev.1991-9603Copernicus PublicationsGöttingen, Germany10.5194/gmd-9-3161-2016The weather@home regional climate modelling project for Australia and New ZealandBlackMitchell T.mtblack@student.unimelb.edu.auKarolyDavid J.RosierSuzanne M.DeanSam M.KingAndrew D.MasseyNeil R.SparrowSarah N.https://orcid.org/0000-0002-1802-6909BoweryAndyWallomDavidhttps://orcid.org/0000-0001-7527-3407JonesRichard G.OttoFriederike E. L.AllenMyles R.School of Earth Sciences and ARC Centre of Excellence for Climate System Science, The University of Melbourne, Melbourne, AustraliaNational Institute of Water and Atmospheric Research, Wellington, New ZealandEnvironmental Change Institute, Oxford University, Oxford, UKOxford e-Research Centre, Oxford University, Oxford, UKMet Office Hadley Centre, Exeter, UKMitchell T. Black (mtblack@student.unimelb.edu.au)15September2016993161317626April201612May20164August201610August2016This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://gmd.copernicus.org/articles/9/3161/2016/gmd-9-3161-2016.htmlThe full text article is available as a PDF file from https://gmd.copernicus.org/articles/9/3161/2016/gmd-9-3161-2016.pdf
A new climate modelling project has been developed for regional
climate simulation and the attribution of weather and climate extremes over
Australia and New Zealand. The project, known as weather@home Australia–New Zealand, uses public volunteers' home computers to run a
moderate-resolution global atmospheric model with a nested regional model
over the Australasian region. By harnessing the aggregated computing power of
home computers, weather@home is able to generate an unprecedented number of
simulations of possible weather under various climate scenarios. This
combination of large ensemble sizes with high spatial resolution allows
extreme events to be examined with well-constrained estimates of sampling
uncertainty. This paper provides an overview of the weather@home
Australia–New Zealand project, including initial evaluation of the regional
model performance. The model is seen to be capable of resolving many climate
features that are important for the Australian and New Zealand regions,
including the influence of El Niño–Southern Oscillation on driving
natural climate variability. To date, 75 model simulations of the historical
climate have been successfully integrated over the period 1985–2014 in a
time-slice manner. In addition, multi-thousand member ensembles have also
been generated for the years 2013, 2014 and 2015 under climate scenarios with
and without the effect of human influences. All data generated by the project
are freely available to the broader research community.
Introduction
Extreme weather and climate-related events often have a serious impact on our
economy, environment and society. This is particularly true in Australia and
New Zealand where recurring heat waves, floods, droughts and wildfires have
resulted in the loss of life, property and livelihoods
e.g.. In the aftermath of these events the scientific
community is often faced with the task of quantifying the link to different
causal factors, including human-induced climate change
e.g.. The delivery of such information in a timely,
clear and reliable manner is an ongoing challenge. Therefore, developing a
capacity to research extremes and understand their causes continues to be
crucial for predicting and managing their impacts.
There is clear evidence that the climate has changed as a result of human
influence and that some aspects of extremes have
changed across the globe as a result . However, this
does not imply that the occurrence of every recently observed extreme weather
or climate-related event was the result of human influence on the climate
system, as such events may still have occurred (however unlikely) in the
absence of such an influence e.g.. This can be
understood by recognising that our climate is a complex, chaotic system that
is influenced by internal climate variability and external forcings.
Processes that generate internal climate variability include atmosphere–ocean
teleconnections, such as El Niño–Southern Oscillation (ENSO), as well as
chaotic internal variability. Meanwhile, external forcings of climate can be
either natural, such as explosive volcanic eruptions, or anthropogenic, such
as greenhouse gas emissions from the burning of fossil fuels.
Distinguishing between the responses to internal and external climate
forcings becomes increasingly difficult when moving from global to regional
scales, due to a lower signal-to-noise ratio .
That is to say, analysis on smaller spatial scales offers less opportunity to
reduce the magnitude of natural variability through spatial averaging or
other techniques. This is particularly true for Australia, which is
recognised as having one of the most variable climates in the world
e.g.. The variability in Australia's climate is
driven by a number of factors, in particular the year-to-year variations in
sea surface temperatures in both the Pacific and Indian oceans
. ENSO represents the variations in sea-surface
temperatures and atmospheric patterns across the Pacific Ocean, with warm (El
Niño) conditions producing below-average rainfall, above-average
temperatures and often drought over much of northern and eastern Australia
. The reverse is true during cool
(La Niña) conditions. Although New Zealand's climate is not usually
affected as strongly by ENSO as are parts of Australia, there is nevertheless
a significant influence e.g.. In addition to ENSO
there are a number of other drivers of internal climate variability for
Australia and New Zealand, including the Southern Annular Mode
and the Indian Ocean Dipole . Therefore,
any assessment of extreme weather and climate-related events needs to
consider the interplay of both internal climate variability and forced
external changes, such as the warming effect caused by increased greenhouse
gas emissions.
In light of this challenge, an emerging field of climate science (known as
event attribution) is seeking to quantify how the risk of weather and
climate-related extremes has changed as a consequence of particular forcings
acting on the climate system . This is typically achieved by comparing the probability of such
events under the current (historical) climate against that for counterfactual
worlds in which particular forcing factors (such as human-induced climate
change) are absent. We of course are unable to observe a world in which
either anthropogenic or natural forcing is absent; therefore, physically based
climate models are required to estimate how the climate would respond to the
absence of anthropogenic forcings .
Undertaking event attribution studies of extreme weather events is typically
restricted by two important modelling requirements: ensemble size and model
resolution. Extreme weather events are, by definition, rare, and therefore
very large ensembles of climate model simulations are needed in order to
study the event with a high degree of confidence. Meanwhile, as many extreme
events occur at a regional or local scale, the model must have sufficient
resolution to realistically capture the event. Due to these requirements,
such an undertaking would be computationally expensive and typically beyond
the capability of conventional computing resources. However, these demands
may be met through the aggregated power of distributed computing projects.
Proposed by and launched in 2003,
climateprediction.net became the largest climate modelling
experiment to date by running climate models on volunteers' home computers.
While the project was originally focussed on running low-resolution global
coupled atmosphere–ocean and medium-resolution
atmosphere-only models , a recent advancement (known as
weather@home) involves running the global atmosphere-only model with a nested
higher-resolution regional model to generate very large ensembles of model
simulations . This regional model configuration has been
implemented and evaluated over Europe and the western
United States and successfully used in a number of
event attribution studies
e.g.. Given the
success of these existing weather@home regional climate modelling projects it
was decided to implement a regional configuration over Australia and New
Zealand.
The primary purpose of this paper is to provide a description and basic
evaluation of the weather@home Australia–New Zealand modelling setup. This is
achieved by comparing the regional model output with observations from the
recent past over regions of Australia and New Zealand. Because the modelling
setup is intended to be used for event attribution studies, particular focus
is given to an assessment of how well the model represents (1) mean spatial
fields and interannual climate variability, (2) regional teleconnections to
ENSO and (3) the distribution of daily variables at regional and local
scales. For the purpose of this study we have restricted our analysis to only
consider temperature and precipitation as these are the variables most
commonly assessed in event attribution studies.
The remainder of this paper is structured as follows:
Sect. describes the model setup and summarises the
experimental design for the representation of the historical climate, while
Sect. provides details on the evaluation of the system.
As a thorough comparison with observations is beyond the scope of this paper,
we provide some illustrative comparisons of both temporal and spatial
patterns. Section describes how the counterfactual
climate scenarios are constructed for the purpose of undertaking event
attribution studies. The main conclusions are given in
Sect. , including plans for future improvements.
Model description
Domain and elevation of terrain (metres) used in the weather@home
regional model simulations. Land areas have been separated into six regions
for subsequent evaluation: northern Australia (NAUS), central Australia
(CAUS), eastern Australia (EAUS), southwest Australia (SWAUS), southeast
Australia (SEAUS) and New Zealand (NZ). The coastal city of Melbourne and the
inland city of Mildura are identified by asterisks.
Weather@home Australia–New Zealand uses the Hadley Centre Atmospheric
General Circulation Model 3P HadAM3P; with an embedded
regional model HadRM3P; over the Australasian CORDEX
domain (Fig. ). The HadAM3P–HadRM3P model formulation
is based on the atmospheric component of the HadCM3 general circulation model
with a number of improvements with respect to the
calculation of clouds and convection, and a more realistic coupling of
vegetated surfaces with the soil . HadAM3P–HadRM3P is a
grid-point model which solves equations of motion, radiative transfer and
dynamics explicitly on the same scale as the grid. HadAM3P is integrated with
a 15 min timestep, has 19 vertical levels and has a regular
latitude–longitude grid (1.25∘ longitude by 1.875∘ latitude)
with regular poles. HadRM3P has a 5 min timestep, has 19 vertical levels and
uses a rotated grid (0.44∘ longitude by 0.44∘ latitude) with
an artificial North Pole at 60.31∘ N, 141.38∘ E for the
Australia–New Zealand configuration. This allows the region of interest to
lie about the Equator of the rotated grid, thus ensuring that each grid box
in the nested region has approximately the same area. HadAM3P and HadRM3P are run
in an interleaved manner: HadAM3P first runs for a full model day, providing
the lateral boundary conditions to HadRM3P, which then also runs for one full
model day. The coupling is strictly one-way, meaning that there is no
feedback from the regional model to the global model. There is a four-point
buffer zone around the perimeter of the regional model, where the lateral
boundary conditions are relaxed to values temporarily interpolated from
6-hourly output from HadAM3P. The land surface scheme incorporated within the
model is MOSES 1.0 (Met Office Surface Exchange Scheme, ),
with fixed surface types (one vegetation type and one soil type per grid
box). Further details of the HadAM3P–HadRM3P configuration are provided by
but with the European region replacing the Australasian
region.
Weather@home is able to generate very large ensembles of climate model
simulations by harnessing spare CPU time on a network of volunteers' personal
computers. This distributed computing capacity is made possible by the
Berkeley Open Infrastructure for Network Computing
BOINC; open-source infrastructure. Each volunteer
signs up for the weather@home project via the BOINC client software, which
automatically downloads the climate model setup to the volunteer's computer.
Individual work units are then received from the BOINC server and run when the
computer is idle. The work unit contains all necessary configuration inputs
needed by the climate model to run the experiment for one model year
(December–November), under a specific climate scenario. After the completion
of each model month the output is post-processed to retain only a selection
of key meteorological variables. This is required in order to minimise file
size for data transfer and storage. A complete listing of these output
variables is provided as supplementary material. Following this
post-processing stage the final results are returned to a server hosted at
the Tasmanian Partnership for Advanced Computing in Hobart, Australia. On
average, it takes a standard home computer around 4–5 days to integrate over
the model year. At the completion of the model year an additional file (the
restart file) is returned that represents the final state of the atmosphere.
This final state can then be incorporated as the initial conditions for a new
work unit describing the next year of the climate scenario. Therefore, this
allows the system to run for a year at a time, in a time-slice manner, to
generate an extended time series of climate model integrations. As this
resubmission process is not automated, the generation of these continuous
model runs is somewhat restricted by the need for project scientists to
manage restart files and work unit regeneration. Therefore, typical event
attribution studies will only generate multi-thousand ensemble members for a
single model year of interest.
In order to represent the range of internal variability that is possible with
the model, a perturbation is applied to the initial conditions of each
work unit. These perturbations are applied to the global climate model in the
form of slight changes to the three-dimensional potential temperature field.
The initial condition perturbations were generated by calculating the
next-day differences within a 1-year integration of the global model and then
multiplying by five global scaling factors (1.1, 1.2, 1.3, 1.4 and 1.6) (see
, for details). This resulted in the generation of 1740
different initial condition perturbations. In the case of an extended
(multi-year) model experiment, these perturbations are only applied to the
initial condition of the first model year; no perturbation is applied
thereafter so as to allow for the continuous integration of the model under
its specific climate scenario. Further initial condition perturbations are
also applied to the first year of the model integration using a range of
starting conditions with different large-scale circulations and soil moisture
amounts. That is to say, sets of model simulations are initialised using 100
different restart files taken from control simulations for the preceding
year. Although all of the initial condition perturbations are only applied to
the global model, they immediately affect the regional simulations through
the previously described transfer of lateral boundary conditions at the end
of the first global model day.
As HadAM3P and HadRM3P are both atmosphere-only models, they require
specified forcings at the boundary between the atmosphere and ocean. These
lower boundary conditions come in the form of prescribed sea surface
temperature (SST) and sea ice fraction (SIF) fields. For the historical
1985–2014 climate scenario used in this paper for the purpose of model
evaluation, both the SST and SIF fields were sourced from the UK Met Office
Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) dataset
. OSTIA provides global, daily fields with a spatial
resolution of 0.05∘ latitude × 0.05∘ longitude. In
order for these fields to be defined per grid box of the global climate
model, they are regridded to the HadAM3P resolution of 1.875∘
latitude × 1.25∘ longitude using an area-weighted averaging
method. Any discrepancy between the HadAM3P and OSTIA land–sea masks is
resolved by taking the mean of surrounding ocean grid points. In addition to
these lower boundary conditions, the model also requires the atmospheric
composition to be specified. The concentrations of greenhouse gases
(CO2, CH4 and N20), ozone, halocarbons, sulfur species and
solar anomalies are all prescribed to follow the recommendations outlined by
the Coupled Model Intercomparison Project Phase 5
CMIP5;. The halocarbon gases (CFC113, CFC11, CFC12,
HCFC22, HFC124 and HFC134A) are represented as a single value per time point
in the time series, which produces the equivalent radiative forcing as if all
six gases were modelled. Post-2005, the greenhouse gas concentrations and
aerosol emissions follow the RCP 8.5 scenario. An overview of the model
boundary conditions for the counterfactual climate scenarios is presented in
Sect. .
Model evaluation
In order to establish how well weather@home represents the historical climate
over Australia and New Zealand, we generate 75 model simulations for each
year over the period December 1985 to November 2014. While the modelling
setup is capable of generating much larger ensemble sizes, this would be
unnecessary for characterising the climatology and overall distribution of
climate variables over a 29-year period. Therefore, computing resources were
directed towards generating very large ensembles of climate simulations for
the individual years of 2013, 2014 and 2015, for subsequent use in event
attribution studies that are beyond the scope of the current paper.
Seasonal average daily maximum temperatures (∘C) for
1985–2014 for DJF (left), and JJA (right), from the weather@home regional
model HadRM3P (top), and the observational datasets (middle; AWAP over
Australia and VCSN over New Zealand). The bottom panels show the difference
between HadRM3P and the corresponding observational
dataset.
The spatial fields of the regional model output are separated into six
regions (Fig. ) for subsequent examination. Australia
has been broken up into five established regions based on distinct climatic
zones: northern Australia (NAUS), central Australia (CAUS), eastern Australia
(EAUS), southwest Australia (SWAUS) and southeast Australia (SEAUS) (see
, for details). Meanwhile, the North Island and South Island of New
Zealand (NZ) are treated as a single region. While each of the six regions
could have been further broken down for sub-regional detail, it would be
impractical to present such a mass of information here.
Observational datasets
Evaluation of weather@home is undertaken by comparing the regional model
output to two observational datasets: for evaluation over Australia we use
the Australian Water and Availability Project dataset
AWAP; and for New Zealand we use the Virtual Climate
Station Network dataset VCSN;. The AWAP dataset provides
daily and monthly gridded fields of rainfall and temperature extending back
to 1911 on a 0.05∘× 0.05∘ grid and is highly
regarded for studying trends and variability over Australia
e.g.. For the purpose
of this study we have masked the AWAP data over inland regions of Australia
where there is low station density. Analogous to AWAP, VCSN provides high-resolution (0.05∘ latitude × 0.05∘ longitude)
estimates of daily rainfall and temperature over New Zealand extending back
to 1972. While these observational datasets may be subject to uncertainties
in and of themselves, they have nevertheless been found to be appropriate for
use in model evaluation studies e.g..
In order to compare model output and observations, a remapping of the
observational datasets onto the HadRM3P model grid is required. For
temperature this is achieved using bilinear interpolation. For precipitation,
a conservative remapping scheme is used to ensure that the total amount of
precipitation in the remapped data is the same as in the original data.
Climatological mean fields and inter-annual variability
We first examine weather@home's ability to correctly represent seasonal mean
fields of temperature and precipitation. By averaging over the 29-year period
we are attempting to reduce the internal atmospheric model variability about
the mean state. Therefore, any differences between the observations and model
output may be interpreted as model deficiencies. We have separated our
analysis into seasons and for brevity are only showing results for the
Austral summer (December–February) and winter (June–August). While not
shown here, the weather@home model is able to adequately resolve the
onset/cessation seasons for temperature and precipitation (see Figs. S11–S13
in the Supplement).
As in Fig. but showing seasonal average
minimum temperature (∘C).
As in Fig. but showing seasonal average
precipitation (mm day-1). The difference fields are expressed as
percentages relative to the observational datasets.
The spatial fields of seasonal average daily maximum temperature
(Tmax) and minimum temperature (Tmin) are shown in
Figs. and , respectively. In
each of these figures, the top panels show the mean field from HadRM3P,
averaged over the 75 ensemble members for the 29-year period, while the
middle panels show the mean for the observational datasets. The bottom panels
show the difference between the model and observations and can be interpreted
as an indication of model bias. Overall, the model is able to capture the
large-scale spatial patterns of temperature very well, including the regions
of warmest temperature over northern parts of Australia and the persistently
cooler temperatures over southeast Australia and New Zealand, associated with
topography. For Tmax, HadRM3P is capable of representing mean summertime
values to within ±1 ∘C over most parts of Australia and over the
North Island of New Zealand (Fig. e). In winter, the
model underestimates Tmax at almost every land grid point across the model
domain (Fig. f). For Tmin, the model overestimates
summertime values at most locations, particularly over southwest and
southeast Australia and the northern and eastern parts of New Zealand
(Fig. e). In winter, the model overestimates Tmin in
the north and east of Australia and in parts of New Zealand, while it
underestimates temperatures to the west (Fig. f). The
prominent negative bias in temperatures along the western coastline of the
South Island of New Zealand in both seasons may be the result of two
features: an inability of the model to correctly resolve temperature in this
region of complex topography, as well as possible limitations in the VCSN
network due to a lack of stations at high elevations.
Time series of summertime (December–February) average maximum
temperature for the respective study regions (as labelled) for 1986–2014.
Ensemble-mean values from weather@home simulations are shown by the solid
line (5–95th percentile shaded envelope) while the observations (AWAP over
Australia and VCSN over New Zealand) are shown by the dashed line. The time
series are given as anomalies relative to the mean of the entire period. The
bias between the medians for the model and observations is indicated, along
with the Pearson correlation coefficient (r; calculated using a two-sided
test) and p value for testing non-correlation.
As in Fig. but showing average minimum
temperature.
As in Fig. but showing average
precipitation.
The simulated patterns of seasonal average precipitation
(Fig. ) clearly demonstrate weather@home's ability to
capture both seasonal variations and, at least to some extent, the influence
of topography. Over Australia, the regional model is able to capture the
strong summertime monsoon rainfall over the northern parts of the continent,
as well as the rainfall associated with onshore moisture transport along the
eastern seaboard (Fig. a). There is a distinctly
different rainfall distribution over Australia in winter, with the highest
rainfall restricted to the southern parts of the continent, including regions
of topography, where rainfall is often associated with the passage of frontal
systems (Fig. b). Over New Zealand, the model is able to
resolve the strong rainfall gradient along the South Island, reflecting the
region's complex topography. Overall, weather@home tends to underestimate
rainfall in both seasons over Australia and New Zealand, with the exception
of parts of southwest and eastern Australia in summer
(Fig. e), and parts of the South Island of New Zealand in
winter (Fig. f). The prominent differences in wintertime
rainfall along the southern and western coastlines of Australia
(model ≤ 50 % of observations; Fig. f) suggest
that the model may not be able to fully capture the influence of local land–sea breezes and/or the influence of passing frontal systems at those
locations. Meanwhile, the prominent differences over northern Australia in
JJA (Fig. f) are an artefact of expressing the
differences as percentages; the actual rainfall values for both observations
and the model output are both small over this region, meaning that any
resulting small differences equates to a large percentage.
Next, we assess HadRM3P's capacity to represent interannual variability. For
each of the regional clusters identified in Fig. , we
compare time series of annual variations of seasonal average temperature and
precipitation from the model and the observational datasets
(Figs. –). Here, we
express these time series as anomalies relative to the period mean. The solid
lines show the median of the 75 ensemble members, while the shading
represents the 5–95th percentile range. Meanwhile, the dashed line
represents the corresponding observational dataset. For brevity, we only show
time series for summer here (winter time series are included as Supplement).
There is general agreement between the interannual variability captured by
weather@home and the observational datasets
(Figs. –). For each of
Tmax, Tmin and precipitation there are specific years that correspond to
peaks/troughs in both the model estimates and observations, and the overall
shapes of the curves are similar. In addition, the observations lie within
the model ensemble range for each year with only a few exceptions (e.g.
northern Australian rainfall; Fig. a). Because
sea surface temperatures are the only source of interannual variability that
is common to both the weather@home simulations and the observational records,
the agreement between the time series (as represented by the correlation
coefficients between the median values for the model simulations and
observations) in
Figs. – highlights the
importance of sea surface temperatures on driving the climates of Australia
and New Zealand.
Response to ENSO
Given that ENSO is an important driver of internal climate variability for
Australia and New Zealand e.g., we assess
weather@home's ability to correctly simulate ENSO teleconnections. This is
achieved by comparing regional model output against observations for the
different phases of ENSO: La Niña, neutral and El Niño. La Niña
(El Niño) events were defined when the average Nino-3.4 index was at or
below (above) -1 ∘C (+1 ∘C) anomaly for at least 3
months in the September–February period. Neutral events were defined as
periods when the average Nino-3.4 index did not go beyond ±1 ∘C
anomaly in any month of September–February. These criteria allowed an equal
number of events to be selected for each of the three ENSO phases: La Nina
(1988–1989, 1998–1999, 2007–2008, 2010–2011), neutral (1992–1993, 1993–1994,
2003–2004, 2012–2013) and El Niño (1994–1995, 1997–1998, 2002–2003,
2009–2010). Furthermore, these criteria allowed the events to be relatively
evenly spread across the period of available model simulations (1985–2014).
The events were grouped according to their ENSO phase for subsequent
analysis.
September–February average maximum temperature for the respective
study regions (as labelled) during different phases of ENSO: La Niña
(LN), neutral (NU) and El Niño (EN). Observed values are plotted as
coloured circles while values from the weather@home HadRM3P simulations are
shown as box-and-whisker plots. The boxes show the median and interquartile
range while the whiskers extend to the 5th and 95th percentiles. See text for
details.
As in Fig. but showing minimum
temperature.
As in Fig. but showing
precipitation.
Figures – show the distributions of
temperature and rainfall, averaged over September–February, for the
different phases of ENSO. For the purpose of model evaluation we present
results for each of the six study regions. The model-derived distributions
are shown as box-and-whisker plots; each box represents the median and first
and third quartiles, while the whiskers extend to the 5th and 95th
percentiles. Meanwhile, the corresponding values calculated from the
observational dataset are represented as dots. It is worth noting that while
we only have four examples of observed atmospheric response to each of the La
Niña, neutral and El Niño forcings, we have 4 × 75 examples
from weather@home. This large number of model simulations allows us to reduce
the influence of internal chaotic variability by averaging in the modelled
ensemble; thus, differences between the median values of the box plots are
likely to be representative of forced responses to the observed
teleconnections.
Overall, weather@home is able to correctly represent the response of
temperature and rainfall to changes in the phase of ENSO. As ENSO changes
from the La Niña to El Niño phase, there is a warming shift in the
distributions of Tmax over each of the Australian regions
(Fig. a–e). Meanwhile, the model is able to capture the
reverse relationship for Tmax over New Zealand (Fig. f).
The response of Tmin to changes in the phase of ENSO is less pronounced over
Australia (Fig. a–e), while conditions continue to be
warmer over New Zealand during the La Niña phase
(Fig. f). For precipitation, there is a shift towards
higher rainfall totals over each of the study regions during La Niña
conditions (Fig. ). Weather@home seems able to capture
the observed non-linear ENSO–precipitation relationship despite many global
coupled and atmosphere-only models failing to do so . While
the limited observations prevent us from determining whether the magnitude of
these shifts are suitably represented by the model, they do suggest that the
directions of these shifts are correct.
Daily variability
Because the weather@home setup is specifically designed for use in the
attribution of extreme weather events, it is important that the model is able
to correctly represent the distribution of daily values of temperature and
precipitation at regional and local scales. Such an assessment needs to
consider not only the model's ability to correctly resolve the mean state
but also the tails of the distributions where the extreme events lie. By
identifying any limitations of the model, a bias correction approach may be
used to correct for systematic errors.
Here, we compare the distributions of daily Tmax, Tmin and precipitation from
the ensemble of weather@home regional simulations against the distribution of
these variables in the observational dataset. By way of example, results for
the SEAUS region are presented in Fig. for both summer
(DJF) and winter (JJA) in the form of quantile–quantile plots. The
corresponding plots for the other regions are presented as supplementary
material. For brevity, we do not intend to provide a thorough assessment of
the model's representation of daily fields over each of the defined study
domains. Rather, we highlight how the large ensemble provided by weather@home
allows us to systematically identify biases in the modelled distribution.
Figure is constructed by extracting the daily fields of
Tmax, Tmin and precipitation from the regional model simulations and
calculating area averages over the SEAUS region. As the model uses a 360-day
calendar, and there are 2175 model simulations (75 model realisations for
each of the 29 years), this results in a total sample size of 195 750 daily
values for each season. This large sample size provides a thorough sampling
of physically plausible climate states represented by the model and, thus,
allows the tails of the distribution to be resolved with confidence. The
solid blue line in Fig. identifies the percentile values
when considering all of the 2175 model runs together, while the envelope
shows the 5th to 95th percentile range for values at each percentile when
considering each model run separately. Therefore, the range of this envelope
provides an assessment of both sampling uncertainty and internal variability.
Figure shows that the weather@home regional model provides
an adequate representation of the distribution of daily summertime Tmax, Tmin
and precipitation averaged over SEAUS, when compared against the AWAP
observational dataset. That is to say, the solid blue line is almost directly
overlying the 1:1 line of agreement (shown in black). The model is capable
of correctly resolving not only the mean state but also the tails of the
distribution (represented by the 1st and 99th percentiles). The large spread
in the envelope indicates that the model is capable of representing a wide
range of temperatures and precipitation rates; therefore, in order to
fully sample internal variability of the model, a large ensemble is
necessary.
Quantile–quantile plots showing distributions of daily maximum
temperature (a, b), minimum temperature (c, d) and
precipitation (e, f), averaged over southeast Australia.
Distributions are shown for December–February (a, c, e) and
June–August (b, d, f). The solid blue line shows the percentile
values for the entire ensemble of model simulations, while the blue envelope
shows the 5th to 95th percentile range of values for individual ensemble
members.
Estimated sea surface temperature response pattern (∘C) to
anthropogenic forcing, calculated from 10 different CMIP5 models (as
labelled). The temperature responses are calculated for each month
(January–December) but are shown here as annual
averages.
Return periods of daily December–February maximum temperature at
Melbourne, Australia, for historical climate conditions (red) and various
counterfactual climate conditions (grey).
The relatively high resolution of the weather@home regional model, and the
large number of model simulations, allows the performance of the model to
also be assessed at much more local scales. By way of example,
quantile–quantile plots have also been generated for the coastal city of
Melbourne and the inland city of Mildura by extracting and examining the
corresponding nearest model grid point (see Supplement). Overall, the
weather@home setup provides sufficient model resolution and ensemble sizes to
allow the model to be assessed (and where appropriate, bias corrected) for
subsequent use in event attribution at both the regional and local scales.
Creating counterfactual climate scenarios
In order to quantify how human-induced climate change has altered the
likelihood of extreme weather and climate related events, large ensembles of
model simulations are required under two distinct scenarios: under current
(historical) climate and under a counterfactual (natural) climate as might
have been without human influence on atmospheric composition. Up to this
point we have only considered the weather@home model under the observed
climate scenario. Therefore, a brief description of the counterfactual
climate scenarios is provided here.
The key differences between the observed and counterfactual
climate scenarios are the lower boundary conditions used to drive the
weather@home model. As outlined in Sect. , simulations
for the historical climate are driven by historical SSTs and sea ice from the
OSTIA dataset, as well as present day atmospheric composition (well-mixed
greenhouse gases, ozone and aerosols). For the counterfactual climate the
model is driven by different atmospheric composition and different sea ice
and SST specifications; the atmosphere has prescribed pre-industrial (1850)
levels of greenhouse gases, ozone and aerosols, the sea ice extent
corresponds to the year of maximum sea ice extent in each hemisphere of the
OSTIA record, and SSTs are modified to remove estimates of anthropogenic
warming. Meanwhile, forcings common to both scenarios are the natural forcing
factors, such as changes in volcanic aerosols and solar irradiance. There is
no change in the land surface types between the historical and counterfactual
climate scenarios.
As the true climate conditions for the “world without humans” cannot be
observed, weather@home simulations are run under 10 alternative realisations
of the counterfactual climate scenario. These alternative realisations are
derived from different estimates of the underlying SST warming (delta-SST)
due to human influence, which are separately calculated from 10 available
Coupled Model Intercomparison Project Phase 5 (CMIP5) models
(; see Supplement for details). Monthly-average delta-SST
estimates are calculated for each of the CMIP5 models by calculating the
difference between the decadal-average (1996–2005) SSTs from the
“historical” simulations (which include both anthropogenic and natural
forcings) and the corresponding “natural” simulations. The resulting
patterns and magnitudes of warming are seen to differ across the 10
delta-SST estimates (Fig. ). These delta-SST patterns are
then subtracted from the historical OSTIA SSTs to provide the lower boundary
conditions for each of the respective counterfactual realisations.
The use of multiple realisations of the counterfactual scenario allows us to
account for some of the uncertainty in our estimates of a world without
anthropogenic influence. By way of example, Fig. shows
weather@home model estimates of summertime daily Tmax at Melbourne for
2014–2015, under the historical and counterfactual climate scenarios. For
each of these scenarios the model has been run thousands of times and the
daily values of Tmax have been extracted from the nearest model grid point to
Melbourne. The return time curve for the historical climate scenario (shown
in red) is positioned to the left of the curves for the respective
counterfactual scenarios (shown in shades of grey). This suggests that
anthropogenic climate change has shifted the distribution of Melbourne
summertime maximum temperatures towards warmer conditions. However, the
extent of this shift varies when considering each of the separate
counterfactual scenarios. The multiple realisations of the counterfactual
climate scenario allow for uncertainty to be characterised and communicated
in any resulting attribution statement. More detailed examples of event
attribution studies performed using the weather@home Australia–New Zealand
system can be found in the 2015 special issue of the Bulletin of the American Meteorological Society investigating extreme events of 2014
e.g.. The model
evaluation undertaken in each of these studies was tailored to the region of
interest and builds upon the general model evaluation in this paper.
Discussion and conclusions
The weather@home Australia–New Zealand climate modelling setup has been
described and briefly evaluated. By harnessing spare computing power of
volunteers' home computers, weather@home is capable of generating very large
ensembles of regional climate model simulations over Australia and New
Zealand. This provides a unique tool for undertaking attribution studies of
extreme weather and climate events in the region. To date, 75 model
simulations have been successfully integrated over the period 1985–2014 in a
time-slice manner, while multi-thousand member ensembles have also been
generated under both historical and counterfactual climate scenarios for the
years 2013, 2014 and 2015. All of this model output is freely available to
the research community.
The weather@home regional model is seen to be capable of resolving many
climate features that are important for the Australia and New Zealand
regions. This is reflected in the model's ability to provide a good
representation of temperature and precipitation, both spatially and
temporally. Results presented here suggest that the model is capable of
correctly simulating ENSO teleconnections, which is a key requirement given
the importance of ENSO on driving natural climate variability in the region.
The reasonably high resolution of the regional model, and the large ensemble
size achieved through the distributed computing setup, allows certain types
of extreme weather events to be examined with confidence at regional and
local scales. While the model is seen to exhibit varying degrees of bias in
temperature and precipitation for different regions, this bias may be
corrected through a simple scaling and offset approach
e.g., or through more complicated approaches
such as quantile mapping e.g. or ensemble re-sampling
techniques e.g..
Despite the strengths of weather@home it is important to recognise some of
the limitations of the project. Under the current model configuration, land
surface data (e.g. vegetation roughness and type) are fixed and are the same
for the historical and counterfactual climate scenarios. Furthermore,
weather@home only uses a single atmospheric model, meaning that any resulting
attribution statement can only be made within the context of that specific
modelling setup. In order to test the dependence of the model simulation on
physical parameterization, future work will employ a perturbed physics
approach whereby perturbations will be applied to components of atmospheric
and surface physics.
Other areas of current and future work involve generating larger ensembles
for the recent past (1985–2015), as well as sets of future simulations under
varying projections of climate change. In addition, the model will be driven
with idealised SSTs for different phases of ENSO, under both current and
counterfactual climate scenarios, so as to provide a novel framework for
assessing the relative roles of ENSO and anthropogenic climate change on
recent extreme weather events. Overall, the weather@home Australia–New
Zealand modelling setup provides a unique modelling resource and greatly
enhances Australia and New Zealand's capacity for researching extremes and
understanding their causes.
Data and code availability
All data generated by the weather@home Australia–New Zealand project are
hosted on a server at the Tasmanian Partnership for Advanced Computing,
Hobart. These data are freely available to the broader research community and
can be accessed by contacting the authors of this article. The HadAM3P and
HadRM3P models are both available from the UK Met Office as part of the
Providing REgional Climates for Impacts Studies (PRECIS) programme. Access to
standard versions of the software is dependent on attendance at a PRECIS
training workshop, after which all source and other materials will be made
available
(http://www.metoffice.gov.uk/research/applied/applied-climate/precis/obtain).
As a programme for supporting developing countries, this workshop is free for
officially categorised developing countries and incurs a charge for other
country participants. The code to manage and embed these models within the
weather@home project is specific to their utilisation within the BOINC
environment, which we do not consider within the scope of this publication.
The Supplement related to this article is available online at doi:10.5194/gmd-9-3161-2016-supplement.
Mitchell Black's contribution towards this work was performed as part of his PhD project.
The weather@home Australia–New Zealand project was initiated by Myles Allen and David Karoly and was set up with the
assistance of Richard Jones, Neil Massey, Andy Bowery, Mitchell Black, Suzanne Rosier, Sam Dean, Sarah Sparrow and Friederike Otto.
All results were plotted and analysed by Mitchell Black with advice from David Karoly and Andrew King.
The paper was written in its final form by Mitchell Black with input from all contributing authors.
Acknowledgements
The authors would like to thank Dáithí Stone and Andrew Ciavarella
for their assistance in improving the original manuscript. M. T. Black,
D. J. Karoly and A. D. King have been supported by funding from the ARC
Centre of Excellence for Climate System Science (grant no. CE110001028).
Weather@home ANZ is a collaboration among the University of Oxford, the UK
Met Office, the ARC Centre of Excellence for Climate System Science in
Australia, NIWA in New Zealand, the University of Melbourne, the University
of Tasmania and the Tasmanian Partnership for Advanced Computing. We thank
the volunteers who donated their computing time to run
weather@home.Edited by: J. Williams
Reviewed by: D. Stone and A. Ciavarella
ReferencesAllen, M.: Do-it-yourself climate prediction, Nature, 401, 642–642,
10.1038/44266, 1999.Allen, M.: Liability for climate change, Nature, 421, 891–892,
10.1038/421891a, 2003.
Anderson, D.: BOINC: A system for public-resource computing and storage,
in:
5th IEEE/ACM International Workshop on Grid Computing, 8 November,
Pittsburgh, P.A., 2004.
Arblaster, J., Jubb, I., Braganza, K., Alexander, L., Karoly, D., and Colman,
R.: Weather extremes and climate change, Tech. rep., Australian Climate
Change Science Programme, 2015.Bennett, J. C., Grose, M. R., Corney, S. P., White, C. J., Holz, G. K.,
Katzfey, J. J., Post, D. A., and Bindoff, N. L.: Performance of an empirical
bias-correction of a high-resolution climate dataset, Int. J.
Climatol., 34, 2189–2204, 10.1002/joc.3830, 2014.Bergaoui, K., Mitchell, D., Zaaboul, R., Mcdonnell, R., Otto, F., and Allen,
M.: The contribution of human-induced climate change to the drought of 2014
in the southern Levant region, in: Explaining extreme events of 2014 from a
climate perspective, B. Am. Meteorol. Soc., 96,
S66–S69, 10.1175/BAMS-D-15-00129.1, 2015.Black, M. T., Karoly, D. J., and King, A. D.: The contribution of
anthropogenic forcing to the Adelaide and Melbourne, Australia, heat waves of
January 2014, in: Explaining extreme events of 2014 from a climate
perspective, B. Am. Meteorol. Soc., 96,
S145–S148, 10.1175/BAMS-D-15-00097.1, 2015.Cox, M. P., Betts, A. R., Bunton, B. C., Essery, H. R. L., Rowntree, R. P.,
and
Smith, J.: The impact of new land surface physics on the GCM simulation of
climate and climate sensitivity, Clim. Dynam., 15, 183–203,
10.1007/s003820050276, 1999.
CSIRO and Bureau of Meteoroloy: Climate Change in Australia Information
for
Australia's Natural Resource Management Regions: Technical Report, Tech.
rep., CSIRO and Bureau of Meteorology, Australia, 2015.Donlon, C. J., Martin, M., Stark, J., Roberts-Jones, J., Fiedler, E., and
Wimmer, W.: The Operational Sea Surface Temperature and Sea Ice Analysis
(OSTIA) system, Remote Sens. Environ., 116, 140–158,
10.1016/j.rse.2010.10.017, 2012.Frame, D. J., Aina, T., Christensen, C. M., Faull, N. E., Knight, S. H. E.,
Piani, C., Rosier, S. M., Yamazaki, K., Yamazaki, Y., and Allen, M. R.: The
climateprediction.net BBC climate change experiment: design of the coupled
model ensemble, Philos. T. Roy. Soc. A, 367, 855–70,
10.1098/rsta.2008.0240, 2009.Gordon, C., Cooper, C., Senior, C. A., Banks, H., Gregory, J. M., Johns,
T. C., Mitchell, J. F. B., and Wood, R. A.: The simulation of SST, sea ice
extents
and ocean heat transports in a version of the Hadley Centre coupled model
without flux adjustments, Clim. Dynam., 16, 147–168,
10.1007/s003820050010, 2000.
Gordon, N. D.: The Southern Oscillation and New Zealand Weather, Mon.
Weather Rev., 114, 371–387, 1986.Grose, M. R., Black, M. T., Risbey, J. S., and Karoly, D. J.: Attribution of
exceptional mean sea level pressure anomalies south of Australia in August
2014, in: Explaining extreme events of 2014 from a climate perspective,
B. Am. Meteorol. Soc., 96, S158–S162,
10.1175/BAMS-D-15-00116.1, 2015.Hegerl, G. and Zwiers, F.: Use of models in detection and attribution of
climate change, Wiley Interdisciplinary Reviews: Climate Change, 2,
570–591, 10.1002/wcc.121, 2011.Hendon, H. H., Thompson, D. W. J., and Wheeler, M. C.: Australian rainfall
and surface temperature variations associated with the Southern Hemisphere
Annular Mode, J. Climate, 20, 2452–2467, 10.1175/JCLI4134.1,
2007.
Herring, S., Hoerling, M., Peterson, T., and Stott, P.: Explaining extreme
events of 2013 from a climate perspective, B. Am.
Meteorol. Soc., 95, S1–96, 2014.Herring, S. C., Hoerling, M. P., Kossin, J. P., Peterson, T. C., and Stott,
P. A.: Explaining extreme events of 2014 from a climate perspective,
B. Am. Meteorol. Soc., 96, S1–S172,
10.1175/BAMS-D-15-00157, 2015.
Holland, G. J.: Interannual Variability of the Australian Summer Monsoon at
Darwin: 1952–82, Mon. Weather Rev., 114, 594–604, 1986.
Jones, D., Wang, W., and Fawcett, R.: High-quality spatial climate data-sets
for Australia, Australian Meteorological and Oceanographic Journal, 58,
233–248, 2009.
Jones, D. A. and Trewin, B. C.: On the relationships between the El
Nino-Southern Oscillation and Australian land surface temperature,
Int. J. Climatol., 20, 697–719, 2000.
Jones, R., Noguer, M., Hassell, D., Wilson, S., Jenkins, G., and Mitchell,
J.:
Generating high resolution climate change scenarios using PRECIS, Tech.
rep., Met Office Hadley Centre, 2004.Karoly, D. J. and Wu, Q.: Detection of regional surface temperature trends,
J. Climate, 18, 4337–4343, 10.1175/JCLI3565.1, 2005.King, A., Black, M., Karoly, D., and Donat, M.: Increased likelihood of
Brisbane, Australia, G20 heat event due to anthropogenic climate change, in
Explaining extreme events of 2014 from a climate perspective, B.
Am. Meteorol. Soc., 96, S141–S144,
10.1175/BAMS-D-15-00098.1, 2015a.King, A. D., Alexander, L. V., and Donat, M. G.: The efficacy of using
gridded
data to examine extreme rainfall characteristics: a case study for
Australia, Int. J. Climatol., 33, 2376–2387,
10.1002/joc.3588, 2013.King, A. D., Donat, M. G., Alexander, L. V., and Karoly, D. J.: The
ENSO-Australian rainfall teleconnection in reanalysis and CMIP5, Clim.
Dynam., 44, 2623–2635, 10.1007/s00382-014-2159-8,
2015b.Li, S., Mote, P. W., Rupp, D. E., Vickers, D., Mera, R., and Allen, M.:
Evaluation of a regional climate modeling effort for the western United
States using a superensemble from weather@home, J. Climate, 28,
7470–7488, 10.1175/JCLI-D-14-00808.1, 2015.Massey, N., Jones, R., Otto, F. E. L., Aina, T., Wilson, S., Murphy, J. M.,
Hassell, D., Yamazaki, Y. H., and Allen, M. R.: Weather@home – development
and validation of a very large ensemble modelling system for probabilistic
event attribution, Q. J. Roy. Meteor. Soc.,
141, 1528–1545, 10.1002/qj.2455, 2015.
McBride, J. L. and Nicholls, N.: Seasonal relationships between Australian
rainfall and the Southern Oscillation, Mon. Weather Rev., 111,
1998–2004, 1983.Mera, R., Massey, N., Rupp, D. E., Mote, P., Allen, M., and Frumhoff, P. C.:
Climate change, climate justice and the application of probabilistic event
attribution to summer heat extremes in the California Central Valley,
Climatic Change, 133, 427–438, 10.1007/s10584-015-1474-3, 2015.Min, S. K., Cai, W., and Whetton, P.: Influence of climate variability on
seasonal extremes over Australia, J. Geophys. Res.-Atmos., 118, 643–654, 10.1002/jgrd.50164, 2013.
Moise, A., Wilson, L., Grose, M., Whetton, P., Watterson, I., Bhend, J.,
Bathols, J., Hanson, L., Erwin, T., Bedin, T., Heady, C., and Rafter, T.:
Evaluation of CMIP3 and CMIP5 models over the Australian region to inform
confidence in projections, Australian Meteorological and Oceanographic
Journal, 65, 19–53, 2015.Mote, P. W., Allen, M. R., Jones, R. G., Li, S., Mera, R., Rupp, D. E.,
Salahuddin, A., and Vickers, D.: Superensemble regional climate modeling for
the western US, B. Am. Meteorol. Soc., 97,
203–216, 10.1175/BAMS-D-14-00090.1, 2015.NASEM: Attribution of Extreme Weather Events in the Context of Climate
Change, Tech. rep., The National Academies of Sciences, Engineering and
Medicine, Washington, D.C., 10.17226/21852, 2016.Nicholls, N., Drosdowsky, W., and Lavery, B.: Australian rainfall
variability
and change, Weather, 52, 66–72, 10.1002/j.1477-8696.1997.tb06274.x,
1997.Pall, P., Aina, T., Stone, D. A., Stott, P. A., Nozawa, T., Hilberts, A.
G. J.,
Lohmann, D., and Allen, M. R.: Anthropogenic greenhouse gas contribution to
flood risk in England and Wales in autumn 2000., Nature, 470, 382–385,
10.1038/nature09762, 2011.Perkins, S. E. and Alexander, L. V.: On the measurement of heat waves,
J. Climate, 26, 4500–4517, 10.1175/JCLI-D-12-00383.1, 2013.Peterson, T. C., Hoerling, M. P., Stott, P. A., and Herring, S. C.:
Explaining
extreme events of 2012 from a climate perspective, B. Am.
Meteorol. Soc., 94, S1–S74, 10.1175/BAMS-D-13-00085.1, 2013.
Reisinger, A., Kitching, R. L., Chiew, F., Hughes, L., Newton, P. C. D.,
Schuster, S. S., Tait, A., and Whetton, P.: Australasia, in: Climate Change
2014: Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects.
Contribution of Working Group II to the Fifth Assessment Report of the
Intergovernmental Panel of Climate Change, edited by: Barros, V. R., Field,
C. B., Dokken, D. J., Mastrandrea, M. D., Mach, K. J., Bilir, T. E.,
Chatterjee, M., Ebi, K. L., Estrada, Y. O., Genova, R. C., Girma, B., Kissel,
E. S., Levy, A. N., MacCracken, S., Mastrandrea, P. R., and White, L. L.,
Cambridge University Press, Cambridge, United Kingdom and New
York, NY, USA, 1371–1438, 2014.Risbey, J. S., Pook, M. J., McIntosh, P. C., Wheeler, M. C., and Hendon,
H. H.:
On the remote drivers of rainfall variability in Australia, Mon. Weather
Rev., 137, 3233–3253, 10.1175/2009MWR2861.1, 2009.Risbey, J. S., McIntosh, P. C., and Pook, M. J.: Synoptic components of
rainfall variability and trends in southeast Australia, Int.
J. Climatol., 33, 2459–2472, 10.1002/joc.3597, 2013.Rosier, S., Dean, S., Stuart, S., Carey-Smith, T., Black, M., and Massey, N.:
Extreme rainfall in early July 2014 in Northland, New Zealand-was there an
anthropogenic influence?, in: Explaining extreme events of 2014 from a
climate perspective, B. Am. Meteorol. Soc., 96,
S136–S140, 10.1175/BAMS-D-15-00105.1, 2015.Saji, N. H., Goswami, B. N., Vinayachandran, P. N., and Yamagata, T.: A
dipole
mode in the tropical Indian Ocean, Nature, 401, 360–363,
10.1038/43854, 1999.Schaller, N., Kay, A. L., Lamb, R., Massey, N. R., van Oldenborgh, G. J.,
Otto,
F. E. L., Sparrow, S. N., Vautard, R., Yiou, P., Bowery, A., Crooks, S. M.,
Huntingford, C., Ingram, W. J., Jones, R. G., Legg, T., Miller, J., Skeggs,
J., Wallom, D., Weisheimer, A., Wilson, S., and Allen, M. R.: The human
influence on climate in the winter 2013/2014 floods in southern England,
Nature Climate Change, 627–634 10.1038/nclimate2927, 2016.
Seneviratne, S., Nicholls, N., Easterling, D., Goodess, C., Kanae, S.,
Kossin,
J., Luo, Y., Marengo, J., McInnes, K., Rahimi, M., Reichstein, M., Sorteberg,
A., Vera, C., and Zhang, X.: Changes in Climate Extremes and their Impacts
on the Natural Physical Environment, Tech. rep., Intergovernmental Panel on
Climate Change, 2012.Sippel, S. and Otto, F. E. L.: Beyond climatological extremes – assessing
how
the odds of hydrometeorological extreme events in South-East Europe change in
a warming climate, Climatic Change, 125, 381–398,
10.1007/s10584-014-1153-9, 2014.Sippel, S., Otto, F. E. L., Forkel, M., Allen, M. R., Guillod, B. P., Heimann, M., Reichstein, M., Seneviratne, S. I., Thonicke, K.,
and Mahecha, M. D.: A novel bias correction methodology for climate impact simulations, Earth Syst. Dynam., 7, 71–88, 10.5194/esd-7-71-2016, 2016.
Steffen, W., Hughes, L., and Karoly, D. J.: The critical decade: extreme
weather, Tech. Rep. April, The Climate Commission, 2013.Stocker, T., Qin, D., Plattner, G.-K., Tignor, M., Allen, S., Boschung, J.,
Nauels, A., Xia, Y., Bex, V., and Midgley, P. Eds.: Climate Change 2013 –
The Physical Science Basis, Cambridge University Press, Cambridge,
10.1017/CBO9781107415324, 2014.Stott, P. A., Stone, D. A., and Allen, M. R.: Human contribution to the
European heatwave of 2003, Nature, 432, 610–614, 10.1038/nature03089,
2004.Stott, P. A., Christidis, N., and Betts, R. A.: Changing return periods of
weather-related impacts: the attribution challenge, Climatic Change, 109,
263–268, 10.1007/s10584-011-0265-8, 2011.
Stott, P. A., Allen, M., Christidis, N., Dole, R., Hoerling, M., Huntingford,
C., Pall, P., Perlwitz, J., and Stone, D.: Attribution of weather and
climate-related extreme events, in: Climate Science for Serving Society:
Research, Modeling and Prediction Priorities, edited by: Asrar, G. and
Hurrell, J., Springer Science and Business Media, Dordrecht, 307–337,
2013.Tait, A., Henderson, R., Turner, R., and Zheng, X.: Thin plate smoothing
spline interpolation of daily rainfall for New Zealand using a climatological
rainfall surface, Int. J. Climatol., 26, 2097–2115,
10.1002/joc.1350, 2006.
Taylor, K. E., Stouffer, R. J., and Meehl, G. A.: An Overview of CMIP5 and
the
experiment design, B. Am. Meteorol. Soc., 93,
485–498, 10.1175/BAMS-D-11-00094.1, 2012.