A global climate model ensemble for downscaled monthly climate normals over North America

Abstract

Use of downscaled global climate model projections is expanding rapidly as climate change vulnerability assessments and adaptation planning become mainstream in many sectors. Many climate change impact analyses use climate model projections downscaled at very high spatial resolution (~1 km) but very low temporal resolution (20‐ to 30‐year normals). These applications have model selection priorities that are distinct from analyses at high temporal resolution. Here, we select a 13‐model ensemble and an 8‐model subset designed for robust change‐factor downscaling of monthly climate normals, and describe their attributes in North America. All models are selected from the Coupled Model Intercomparison Project Phase 6 (CMIP6) archives. The 13‐model ensemble is representative of the distribution of equilibrium climate sensitivity, grid resolution, and transient regional climate changes in the CMIP6 generation. The 8‐model subset is consistent with the IPCC's recent assessment of the very likely range of Earth's equilibrium climate sensitivity. Our results emphasize several principles for selection and use of downscaled climate ensembles: (a) the ensemble must be observationally constrained to be meaningful; (b) analysis of multiple models is essential as the ensemble mean alone can be misleading; (c) small (<8‐member) ensembles should be region‐specific and used with caution; (d) higher grid resolution is not necessarily better; and (e) multiple simulations of each model/scenario combination are necessary to represent precipitation uncertainty. Although we have focused our documentation on North America, our model selection uses primarily global criteria and is applicable to downscaling climate normals in other continents. Downscaled projections for the selected models are available in ClimateNA (http://climatena.ca/). An accompanying web application (https://bcgov-env.shinyapps.io/cmip6-NA/) provides tools for further model selection and visualization of the ensemble.

Full text

RESEARCH ARTICLE
A global climate model ensemble for downscaled monthly
climate normals over North America
Colin R. Mahony
1
| Tongli Wang
2
| Andreas Hamann
3
| Alex J. Cannon
4
1
British Columbia Ministry of Forests,
Lands, Natural Resource Operations and
Rural Development, Victoria, British
Columbia, Canada
2
Centre for Forest Conservation Genetics,
Department of Forest and Conservation
Sciences, Faculty of Forestry, University
of British Columbia, Vancouver, British
Columbia, Canada
3
Department of Renewable Resources,
Faculty of Agricultural, Life, and
Environmental Sciences, University of
Alberta, Edmonton, Alberta, Canada
4
Climate Research Division, Environment
and Climate Change Canada, Victoria,
British Columbia, Canada
Correspondence
Colin R. Mahony, British Columbia
Ministry of Forests, Lands, Natural
Resource Operations and Rural
Development, Victoria, BC, Canada.
Email: colin.mahony@gov.bc.ca
Abstract
Use of downscaled global climate model projections is expanding rapidly as
climate change vulnerability assessments and adaptation planning become
mainstream in many sectors. Many climate change impact analyses use cli-
mate model projections downscaled at very high spatial resolution (~1 km)
but very low temporal resolution (20- to 30-year normals). These applica-
tions have model selection priorities that are distinct from analyses at high
temporal resolution. Here, we select a 13-model ensemble and an 8-model
subset designed for robust change-factor downscaling of monthly climate
normals, and describe their attributes in North America. All models are
selected from the Coupled Model Intercomparison Project Phase 6 (CMIP6)
archives. The 13-model ensemble is representative of the distribution of
equilibrium climate sensitivity, grid resolution, and transient regional cli-
mate changes in the CMIP6 generation. The 8-model subset is consistent
with the IPCC's recent assessment of the very likely range of Earth's equilib-
rium climate sensitivity. Our results emphasize several principles for selec-
tion and use of downscaled climate ensembles: (a) the ensemble must be
observationally constrained to be meaningful; (b) analysis of multiple
models is essential as the ensemble mean alone can be misleading; (c) small
(<8-member) ensembles should be region-specific and used with caution;
(d) higher grid resolution is not necessarily better; and (e) multiple simula-
tions of each model/scenario combination are necessary to represent pre-
cipitation uncertainty. Although we have focused our documentation on
North America, our model selection uses primarily global criteria and is
applicable to downscaling climate normals in other continents. Downscaled
projections for the selected models are available in ClimateNA (http://
climatena.ca/). An accompanying web application (https://bcgov-env.
shinyapps.io/cmip6-NA/) provides tools for further model selection and
visualization of the ensemble.
KEYWORDS
climate change, CMIP6, downscaling, ensembles, North America
Received: 7 September 2021 Revised: 6 January 2022 Accepted: 2 February 2022 Published on: 6 March 2022
DOI: 10.1002/joc.7566
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any
medium, provided the original work is properly cited and is not used for commercial purposes.
© 2022 Her Majesty the Queen in Right of Canada. International Journal of Climatology published by John Wiley & Sons Ltd on behalf of Royal Meteorological Society.
Reproduced with the permission of the Minister of Forests, Lands, Natural Resource Operations and Rural Development.
Int J Climatol. 2022;42:5871–5891. wileyonlinelibrary.com/journal/joc 5871

1|INTRODUCTION
The Sixth iteration of the Coupled Model Inter-
comparison Project (CMIP6; Eyring et al., 2016)isa
once-in-a-decade update to projections of climate change.
CMIP6 provides a larger number of simulations from a
new generation of global climate models, at higher spa-
tial resolution, and using an improved set of emissions
scenarios relative to its predecessor, CMIP5 (Taylor
et al., 2012). These new climate simulations contribute to
and are put into broader context by the Sixth Assessment
Report from Working Group I of the Intergovernmental
Panel on Climate Change (Lee et al., 2021). CMIP6 simu-
lations are rapidly being incorporated into downscaled
climate data products for use in regional climate change
impacts and adaptation initiatives. These initiatives can
benefit from careful selection of climate model projec-
tions that are suited to broad classes of end uses
(e.g., Karmalkar et al., 2019), and their wide application
requires transparency on the attributes of these
ensembles.
Many climate change impact analyses, particularly in
ecology, use projections of climate change that are down-
scaled to very high resolution (~1 km) but very low tem-
poral resolution (20- to 30-year climate normals). The
prevalence of this type of analysis is evident from the
widespread use of WorldClim (Hijmans et al., 2005; Fick
and Hijmans, 2017; 23,340 citations) and ClimateNA
(Wang et al., 2012;2016; Hamann et al., 2013; 1,678 cita-
tions). The low temporal resolution of these applications
simplifies downscaling; both WorldClim and ClimateNA
use change-factor downscaling, also called the climate
imprint method (Hunter and Meentemeyer, 2005) and
simple mean bias correction (Maraun, 2016). This
method adds low-spatial-resolution anomalies from the
climate model to a high-resolution gridded climate map
(Tabor and Williams, 2010). The best practices for
change-factor downscaling to high-spatial and low-
temporal resolution are different than those for the more
sophisticated statistical downscaling techniques neces-
sary for high temporal resolution downscaling (Wilby
et al., 2004), leading to distinct model selection priorities.
One consideration in model selection for change-
factor downscaling is the number of simulation runs for
each candidate model. The change-factor method is sen-
sitive to the influence of natural variability in the histori-
cal reference period against which anomalies are
calculated and bias correction is applied. Performing
change-factor downscaling with multiple simulation runs
of each model reduces the confounding influence of natu-
ral variability in bias correction and improves the signal-
to-noise ratio (Milinski et al., 2019). Further, providing
multiple simulations for each model and scenario can
improve the representation of climate change uncertainty
in downstream analysis by accounting for natural vari-
ability (Deser et al., 2012). Consequently, models with
multiple simulations for the historical period and each
future scenario are preferable in this context.
Another consideration is model bias. All climate
models exhibit biases—systematic differences between
observations and simulations—at the regional scale.
Removal of these biases is a basic step in downscaling
(Maraun, 2016). Change-factor downscaling performs
univariate bias correction and therefore may not conserve
the physical (e.g., thermodynamic) interdependence
between variables such as temperature and precipitation
(Cannon, 2018). The associated potential for univariate
downscaling to produce physically implausible climatic
conditions presumably increases with the size of the
biases in the simulation. For this reason, models with
small biases are preferable to models with large biases,
all else being equal.
Finally, the spatial resolution of climate models is of
interest to high spatial resolution downscaling. Some
models contributing to the CMIP6 ScenarioMIP (O'Neill
et al., 2016) experiment (the candidate pool for ensemble
selection in this study) have horizontal grid resolutions of
70–100 km. These higher-resolution models are able to
resolve macrotopography, for example, to differentiate
the major mountain ranges within the Western Cordil-
lera. The opportunity to better represent the influences of
water bodies and topography on climate change trends,
such as elevation-dependent warming (Salathé
et al., 2008; Palazzi et al., 2019), is appealing for climate
change impact analyses. Conversely, models with very
low spatial resolution (>300 km) can conflate the climate
change signals of distinct regions, particularly at land/
ocean transitions (Lanzante et al., 2018). Very low resolu-
tion therefore is a consideration for exclusion from
ensembles designed for high-resolution change-factor
downscaling.
Collectively, the three considerations described above
suggest an ensemble that prioritizes number of simula-
tions per model rather than number of models, low-to-
moderate bias, and moderate-to-high spatial resolution.
Once a general-purpose ensemble is selected, it is use-
ful to structure the ensemble for further user-specific
model selection. Many applications of projected climate
normals are computationally intensive analyses at
regional scales. In these cases, it can be desirable to use a
small number (3–8) of models that represent the approxi-
mate range of a more comprehensive ensemble.
Cannon (2015) describes a method for structuring an
ensemble into an order of subset selection that optimally
represents the ensemble spread. Alternatively, analysts
may wish to select a custom subset of the ensemble.
5872 MAHONY ET AL.

Documentation of the attributes of the ensemble mem-
bers can help analysts to identify subsets that are best
suited to specific applications.
The purpose of this study is to select and describe an
ensemble of CMIP6 model projections of 21st century cli-
mate change over North America. The focus of model
selection is on facilitating robust downscaling of projec-
ted climate normals at very high spatial resolution. We
characterize the attributes, biases, and climate change
trends of the ensemble and highlight features of interest
in individual climate models. We further screen this
selected ensemble to an 8-model subset consistent with
IPCC assessed constraints on equilibrium climate
sensitivity (Arias et al., 2021). Finally, we provide a selec-
tion order for the 8-model subset for regional analyses.
Downscaled projections for the selected 13 CMIP6 models
are available in ClimateNA (http://climatena.ca/), which
provides downscaling at user-specified spatial resolution
and various temporal intervals (annual, 20-year, and
30-year periods). An accompanying web application
(https://bcgov-env.shinyapps.io/cmip6-NA/) provides tools
for further model selection and visualization of the
ensemble.
2|METHODS
2.1 |Criteria for model selection
We assessed all models in the Earth System Grid Federa-
tion (ESGF) archive for the CMIP6 ScenarioMIP as of
December 15, 2020. We selected models using six objec-
tive criteria, listed below with rationale:
•Criterion 1: T
min
and T
max
available. Mean daily mini-
mum temperature (T
min
) and mean daily maximum
temperature (T
max
) are the directly measured elements
of the long-term temperature record, and are the fun-
damental temperature elements in many climate
change impact analyses.
•Criterion 2: Minimum of three historical runs avail-
able. This criterion ensures robust downscaling by
reducing the confounding influence of natural variabil-
ity in bias correction.
•Criterion 3: Complete scenarios. Models need to have
at least one simulation for three of the four major
shared socioeconomic pathways (SSP) marker scenar-
ios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5).
•Criterion 4: One model per institution. This criterion
is a widely applied best practice in ensemble selec-
tion (Leduc et al., 2016) as one measure to increase
independence among ensemble members. For the
purposes of this criterion, different physics or forcing
schemes of the same model were considered differ-
ent models.
•Criterion 5: No closely related models. Models that
share components were excluded, following Brunner
et al.(
2020, fig. 5).
•Criterion 6: No large biases. Bias is the degree to which
a model simulation differs from the observed climate
over a historical reference period). Models with large
biases relative to the rest of the ensemble in one or
more variables were excluded.
2.2 |Ensemble subset criteria
Users of the ensemble may wish or need to use a lesser
number of models in their analyses. To support the selec-
tion of subsets, we structure the ensemble by defining an
order of exclusion of models. Models are excluded in two
phases: first based on screening criteria and second using
the method of Cannon (2015) to represent the range of
climate changes in the remaining models.
2.2.1 | Screening criteria
Priority for exclusion from model subsets was established
using four screening criteria. The screening criteria are
more subjective than the six selection criteria defined
above. They generally are not sufficient in isolation but
combinations of the criteria provide some justification for
model exclusions.
•Criterion 7: Constraints on equilibrium climate sensi-
tivity (ECS). Multiple lines of evidence indicate that
the Earth's ECS is likely (probability >66%) between
2.5 and 4C and very likely (p>90%) between 2 and
5C (Sherwood et al., 2020; Arias et al., 2021). The evi-
dence is robust for the lower bound, and weaker for
the upper bound. From one perspective, inclusion of
models with ECS outside this very likely range biases
the multimodel ensemble mean and unnecessarily
increases the modelling uncertainty in downstream
analyses (Ribes et al., 2021). An alternate perspective is
that high-sensitivity models are useful as a representa-
tion of high-impact, low-likelihood scenarios (Sutton
and Hawkins, 2020). To accommodate both perspec-
tives, we provide structured subsets with and without
high-sensitivity models.
•Criterion 8: Model resolution. Some ScenarioMIP
models have sufficiently high spatial resolution to
resolve macrotopography, for example, to differentiate
the major mountain ranges of the Western Cordillera.
These models are weighted towards inclusion from the
MAHONY ET AL.5873

ordered subsets. Models with very low spatial resolu-
tion are weighted towards exclusion from the subsets.
•Criterion 9: Number of simulation runs. Models with
only one run per emissions scenario are weighted for
exclusion.
•Criterion 10: Grid cell artefacts. Models exhibiting spa-
tially anomalous climate changes in individual grid
cells are problematic for many of the intended uses of
this ensemble, and are weighted for exclusion from the
ordered subsets.
2.2.2 | Ordered subsets
After exclusion of models using the screening criteria
above, an order of exclusion for the remaining models
is defined using the Katsavounidis–Kuo–Zhang (KKZ)
algorithm, using the application to climate model
ensemble selection described by Cannon (2015). KKZ
deterministically selects models that best represent the
spread of multivariate climate changes projected by the
ensemble. KKZ subset selection is ordered, starting
with the model closest to the ensemble centroid, and
incrementally adding models to a region of the ensem-
ble variation that is poorly represented by each succes-
sive subset.
Since the spatial patterns of climate change differ
among models, we provide separate KKZ subsets for each
of the seven IPCC climate reference regions (Iturbide
et al., 2020) within North America. We also provide an
ordered subset for North America as a whole, but caution
that ensembles of less than eight models are likely
insufficient to represent spatial variation in modelling
uncertainty at continental scales (Pierce et al., 2009;
McSweeney et al., 2014; Cannon, 2015). The implemen-
tation of KKZ in this study used the mean of the
z-standardized seasonal changes in T
min
,T
max
, and pre-
cipitation in three consecutive 20-year time periods
starting with 2041–2060 and two emissions scenarios
(SSP2-4.5 and SSP3-7.0).
2.3 |Representation of the full CMIP6
ensemble
We use the ECS of the CMIP6 models (Meehl et al., 2020)
as a basic assessment of whether the selected ensemble is
globally representative of the full CMIP6 ensemble. How-
ever, matching the CMIP6 ensemble ECS does not guar-
antee that the selected ensemble is representative of
transient temperature and precipitation changes at
regional scales (Karmalkar, 2018). We compare transient
climate change in the selected ensemble to 33 CMIP6
models for which we were able to obtain mean monthly
temperature (tas) and precipitation (Table 1). Transient
climate changes are calculated as the mean 2061–2100
SSP2-4.5 climate for each model simulation relative to
the grand mean 1961–1990 climate of multiple historical
simulations for each model. After visually identifying
and removing outliers, we measured representiveness as
the simple ratio of the univariate ranges in seasonal tem-
perature and precipitation changes spanned by the
selected ensemble relative and the larger CMIP6 ensem-
ble (following the approach of Karmalkar, 2018).
2.4 |Analysis of model bias
We assessed model biases against the ClimateNA
composite of PRISM and WorldClim observed gridded
climate normals for the 1961–1990 period (Wang
et al., 2016). We measured model bias as the mean
absolute bias over North America in each monthly
climate variable (T
min
,T
max
, and precipitation). For
each grid cell, i, the mean simulated 1961–1990 cli-
mate normal of the Khistorical model runs, fik is cal-
culated as
fi=1
KX
K
k=1
fik:ð1Þ
The absolute value of the difference between the sim-
ulated 1961–1990 normal, fi, and the observed 1961–1990
normal, o
i
, aggregated onto the native model grid is cal-
culated for each grid cell:
jeij=fi−oi


:ð2Þ
The mean absolute bias, jej, over all Nprojected grid
cells in North America is calculated as
e
jj
=1
NX
N
i=1
jeij:ð3Þ
To equalize the area of grid cells, we projected abso-
lute bias in the native model grid onto a Lambert Confor-
mal Conic grid with 0.5resolution prior to calculating
this mean.
For precipitation variables, Equations (1) and (2) were
performed on log-transformed normals. Subsequent to
Equation (3), this log-transformation was reversed by tak-
ing the exponent of absolute bias. Doing so expresses
absolute bias of precipitation as a factor of magnitude.
For example, simulated precipitation normals of 50% and
200% relative to observed precipitation both have an
absolute bias of 2.
5874 MAHONY ET AL.

TABLE 1 Candidate models, model exclusion criteria, and number of simulation runs
Model Criterion for exclusion ECS
ESGF holdings Analysed
Historical ssp126 ssp245 ssp370 ssp585 Historical ssp126 ssp245 ssp370 ssp585
ACCESS-CM2 2 <3 historical runs 4.7 2 1 1 1 1 5 5
ACCESS-ESM1-5 3.9 30 10 30 10 10 10 10 10 10 10
AWI-CM-1-1-MR 6 Very high bias 3.2 5 1 1 5 1 5 1
BCC-CSM2-MR 3.3 3 1 1113 1111
CAMS-CSM1-0 1 No Tmax/Tmin 2.3 3 2 2 2 2 2 2
CESM2 1 No Tmax/Tmin in historical 5.2 11 3 3 3 3 3 3
CESM2-WACCM 1 No Tmax/Tmin in historical 4.8 3 1 5 3 5 3 3
CIESM 3 Incomplete scenarios 3 1 1
CMCC-CM2-SR5 1 No Tmax/Tmin 1 1 1 1 1
CNRM-CM6-1 4 Same institution 4.9 30 6 10 6 6 10 6
CNRM-CM6-1-HR 2 <3 historical runs 4.3 1 1 1 1 1 1 1
CNRM-ESM2-1 4.8 11 5 10 5 5 11 5 5 5 5
CanESM5 5.6 65 50 50 50 50 10 10 10 10 10
CanESM5-CanOE 4 Same institution 3 3 3 3 3 3 3
E3SM-1-1 3 Incomplete scenarios 5.3 1 1
EC-Earth3 4.3 73 7 30 7 58 5 5 555
EC-Earth3-AerChem 2 <3 historical runs 2 1
EC-Earth3-Veg 4 Same institution 4.3 8 7 8 6 6
FGOALS-f3-L 1 No Tmax/Tmin 3 3 3 3 3 3 3 1
FGOALS-g3 1 No Tmax/Tmin 2.9 6 4 4 5 4
FIO-ESM-2-0 3 Incomplete scenarios 3 3 3 3 3 3
GFDL-CM4 3 Incomplete scenarios 3.9 1 1 1 1 1
GFDL-ESM4 2.7 3 1 3 1 1 3 1 3 1 1
GISS-E2-1-G r*i1p3f1 variants 2.7 47 7 30 19 7 4 4 4 4 4
HadGEM3-GC31-LL 5 Shared components 5.6 5 1 4 4 4 1
HadGEM3-GC31-MM 3 Incomplete scenarios 5.4 4 1 4
IITM-ESM 1 No Tmax/Tmin 1 1 1 1 1
INM-CM4-8 2 <3 historical runs 1.8 1 1 1 1 1 1 1
INM-CM5-0 1.9 9 1 1 5 1 9 1 1 5 1
IPSL-CM6A-LR 4.6 9 5 6 9 5 9 5 6 9 5
(Continues)
MAHONY ET AL.5875

TABLE 1 (Continued)
Model Criterion for exclusion ECS
ESGF holdings Analysed
Historical ssp126 ssp245 ssp370 ssp585 Historical ssp126 ssp245 ssp370 ssp585
KACE-1-0-G 1 No Tmax/Tmin 3 3 3 3 3 3 3
KIOST-ESM 2 <3 historical runs 1 1 1 1
MCM-UA-1-0 1 No Tmax/Tmin 2 1 1 1 1 1 1
MIROC-ES2L 4 Same institution 2.7 3 3 3 3 3 3 3
MIROC6 2.6 50 50 50 3 50 10 10 10 3 10
MPI-ESM-1-2-HAM 3 Incomplete scenarios 3 2
MPI-ESM1-2-HR 3 10 2 2 10 2 8 2 2 10 1
MPI-ESM1-2-LR 4 Same institution 3 10 10 10 10 10 10 10
MRI-ESM2-0 3.1 7 1 5 5 2 5 1 5 1 1
NESM3 5 Shared components 4.8 5 2 2 2 5 2
NorESM2-LM 1 No Tmax/Tmin 2.6 3 1 3 3 1 3 3
NorESM2-MM 1 No Tmax/Tmin 2.5 1 1 2 1 1 1 1
TaiESM1 1 No Tmax/Tmin 4.4 2 1 1 1 1
UKESM1-0-LL 5.4 19 16 17 16 5 10 5 5 5 5
Note: Grey shading indicates models evaluated for bias. Bold font indicates the 13 models that passed Criteria 1-6. Model list and number of simulations per scenario are ESGF holdings as of December 15, 2020. ECS is
equilibrium climate sensitivity (long-term temperature change in response to an instant doubling of CO
2
); ECS values are quoted from Meehl et al.(2020), supplemented by Schlund et al.(2020). See Table 2for
citations and institutions of selected models.
5876 MAHONY ET AL.

2.5 |Cluster analysis
For visualization of similarity among models, we perform
a cluster analysis on six climate variables—T
min
,T
max
,
and precipitation for winter (DJF) and summer (JJA)—at
325 locations by resampling all models to a common
300 km resolution. To reduce dimensions for clustering,
we used three principal components instead of the origi-
nal six variables, resulting in 975 variables for the con-
struction of the dendrogram (325 locations ×3 principal
climate components). We used Ward's hierarchical clus-
tering algorithm with a Euclidean distance of standard-
ized principal components (i.e., a Mahalanobis distance
metric), implemented with the hclust package for the R
programming environment.
3|RESULTS
3.1 |Ensemble selection
There were 44 models in the CMIP6 ScenarioMIP hold-
ings as of December 15, 2020 (Table 1). Twelve of these
candidates were excluded because they did not provide
monthly means of T
min
and T
max
(Criterion 1). Notably,
CESM2 does provide T
min
and T
max
in its future projec-
tions, but due to an archiving error these variables are
not available for historical runs. An additional 11 models
were excluded because they had less than three historical
runs (Criterion 2) or an incomplete scenario set
(Criterion 3). Of the 21 models that passed these first
three objective criteria, we excluded two more models on
the basis of having a clear choice between models from
the same institution (Criterion 4): CanESM5-CanOE in
favour of CanESM5 and EC-Earth3-Veg in favour of EC-
Earth3. In addition, of the several variants of the GISS-
E2-1-G model, we selected the r*i1p3f1 variant because it
had the most complete set of scenario simulations. We
downloaded historical simulations from the remaining
19 models for further evaluation. For practical purposes,
we limited downloads to 5 historical simulations for EC-
Earth3 due to its relatively high resolution, and 10 simula-
tions for other models.
To assist with choosing among models from the same
institution (Criterion 4) or with shared components
(Criterion 5), we conducted an analysis of bias in T
min
,
T
max
, and precipitation (Figure 1). We excluded AWI-
CM-1-1-MR on the sole basis of its very high temperature
0
2
4
6
8
Absolute bias (°C)
(a) Tmin
0
2
4
6
8
Absolute bias (°C)
(b) Tmax
Ensemble mean
ACCESS−ESM1−5
BCC−CSM2−MR
CanESM5
CNRM−ESM2−1
EC−Earth3
GFDL−ESM4
GISS−E2−1−G
INM−CM5−0
IPSL−CM6A−LR
MIROC6
MPI−ESM1−2−HR
MRI−ESM2−0
UKESM1−0−LL
AWI−CM−1−1−MR
CNRM−CM6−1
HadGEM3−GC31−LL
MIROC−ES2L
MPI−ESM1−2−LR
NESM3
1.0
1.2
1.4
1.6
1.8
2.0
Absolute bias (factor)
(c) PPT
Excluded from ensemble (Criteria 4−6)
FIGURE 1 Model biases in
monthly means of (a) daily minimum
temperature, (b) daily maximum
temperature, and (c) precipitation.
Each box represents 12 values of
mean absolute bias over North
America, one for each month.
Absolute bias for precipitation is
expressed as a factor of magnitude.
For example, relative biases of 50 and
200% both have an absolute bias of 2
MAHONY ET AL.5877

bias (Criterion 6). NESM3 also has high bias relative to
the other models and was excluded due to shared compo-
nents with MPI-ESM1 (Criterion 5). None of the other
related models were sufficiently biased to warrant
exclusion.
Final choices from among related models were:
UKESM1.0-LL selected over HadGEM3-GC31-LL due to
higher resolution and more simulations; MIROC6 over
MIROC-ES2L due to higher number of runs and region-
ally high biases in the Pacific Northwest; MPI-
ESM1.2-HR over MPI-ESM1-2-LR to improve representa-
tion of high-resolution models in the ensemble; and
CNRM-ESM2-1 arbitrarily selected over CNRM-CM6-1 in
favour of the earth system model (ESM) configuration. In
summary, the six criteria reduced the 44 candidate
models to a 13-model ensemble (Table 2).
3.2 |Attributes of the 13-model
ensemble
3.2.1 | Representation of the full CMIP6
ensemble
The 13-model ensemble has a mean ECS of 3.7C and a
range of 1.9–5.6C, which matches ECS of the full CMIP6
ensemble (3.7C; 1.8–5.6C) (Meehl et al., 2020). The
13-model ensemble is moderately representative of
regional transient temperature and precipitation changes
(2061–2100, SSP2-4.5) found in the larger 33-model
CMIP6 ensemble (Figure 2), with some key exceptions.
Two models in the larger ensemble are distinct outliers:
KACE-1-0-G exhibits cooling over the contiguous US
(WNA, CNA, and ENA regions) and Mexico (NCA
region); and AWI-CM-1-1-MR exhibits an outlying 80%
reduction in precipitation over CNA and NCA. We
removed these two outlier models from further analysis
of representation. Notwithstanding the outliers, the
13-model ensemble is conspicuously unrepresentative of
the larger CMIP6 ensemble in summer precipitation
change over western North America (WNA). The models
with highest precipitation increases in this region, in
decreasing order, are NorESM2-LM, NorESM2-MM,
CESM2, and CESM2-WACCM. These four models were
ineligible for inclusion in the selected ensemble because
they did not archive T
min
and T
max
. The model exhibiting
extreme drying in the WNA region is MPI-ESM1-2-LR,
which was eligible but excluded in favour of MPI-
ESM1-2-HR on the basis of higher spatial resolution in
the latter. The content of Figure 2can be explored inter-
actively in the supplemental web application (https://
bcgov-env.shinyapps.io/cmip6-NA/).
Within each IPCC region, there is considerable spatial
variation in the selected ensemble's representation of the
larger CMIP6 ensemble (Figure 3a–d). Representation of
model-mean change is lower for precipitation (Figure 3a,
TABLE 2 Institution and citation for each model in the 13-model ensemble
Model Institutions Citation
ACCESS-ESM1.5 Commonwealth Scientific and Industrial Research Organisation (Australia) Ziehn et al.(
2020)
BCC-CSM2 Beijing Climate Center (China) Wu et al.(
2019)
CanESM5 Canadian Centre for Climate Modelling and Analysis (Canada) Swart et al.(
2019)
CNRM-ESM2-1 CNRM (Centre National de Recherches Meteorologiques) and CERFACS
(Centre Europeen de Recherche et de Formation Avancee en Calcul
Scientifique) (France)
Séférian et al.(
2019)
EC-Earth3 EC-Earth Consortium (European Community) Döscher et al.(
2021)
GFDL-ESM4 National Oceanic and Atmospheric Administration, Geophysical Fluid
Dynamics Laboratory (USA)
Dunne et al.(
2020)
GISS-E2.1 Goddard Institute for Space Studies (USA) Kelley et al.(
2020)
INM-CM5.0 Institute for Numerical Mathematics (Russia) Volodin et al.(
2017)
IPSL-CM6A-LR Institut Pierre Simon Laplace (France) Boucher et al.(
2020)
MIROC6 JAMSTEC (Japan Agency for Marine-Earth Science and Technology), AORI
(Atmosphere and Ocean Research Institute), NIES (National Institute for
Environmental Studies), and R-CCS (RIKEN Center for Computational
Science) (Japan)
Tatebe et al.(
2018)
MPI-ESM1.2-HR Max Planck Institute for Meteorology (Germany) Müller et al.(
2018)
MRI-ESM2.0 Meteorological Research Institute (Japan) Yukimoto et al.(
2019)
UKESM1 Met Office Hadley Centre and Natural Environment Research Council (UK) Sellar et al.(
2019)
5878 MAHONY ET AL.

b) than for temperature (Figure 3c,d), and areas of poor
representation of precipitation change are concentrated
in subregions. Some of these areas may be due to natural
variability in models with only one run. However, the
prominent representation gap for summer precipitation
change in the Pacific Northwest USA is induced by
NWN NEN WNA CNA ENA NCA
Summer (JJA)
246
0
10
20
30
40 (a)
1357
−10
0
10
20
30 (b)
0246
−80
−60
−40
−20
0
20
40 (c)
0246
−80
−60
−40
−20
0(d)
0246
−10
0
10
20 (e)
01234
−60
−40
−20
0
20 (f)
Winter (DJF)
246810
20
30
40
50 (g)
46810
20
30
40
50
60
70 (h)
0246
0
10
20
30 (i)
0246
0
10
20
30
40 (j)
1357
5
10
15
20
25
30
35 (k)
−1 1 2 3 4
−50
−40
−30
−20
−10
0
10 (l)
Precipitation change (%)
Temperature change (°C) Large ensemble (33 models)
Selected ensemble (13 models)
Model runs
Model mean
KACE−1−0−G
AWI−CM−1−1−MR
FIGURE 2 Representation of regional mean temperature and precipitation change in the selected 13-model ensemble relative to a
larger 33-model CMIP6 ensemble. Individual model simulations (runs) are shown as open circles, and the single-model means of these
simulations are shown as larger filled circles. Outlying models are given different symbols for ease of identification. Codes for IPCC regions
of North America (Figure 7h) are northwestern (NWN), northeastern (NEN), western (WNA), central (CNA), eastern (ENA), and northern
Central America (NCA)
13−model means vs. 31−model means
Precipitation
Summer (JJA)
(a)
83%
Winter (DJF)
(b)
82%
Temperature
(c)
85%
(d)
89%
13−model means vs. 13−model runs
Precipitation
Summer (JJA)
(e)
76%
Winter (DJF)
(f)
72%
Temperature
(g)
92%
(h)
87%
40%
50%
60%
70%
80%
90%
100%
Range ratio
FIGURE 3 Range of mean temperature and precipitation change in the selected 13-model ensemble relative to a larger 31-model
CMIP6 ensemble. Change for each model simulation is the 2061–2100 mean SSP2-4.5 climate relative to the 1961–1990 mean of all historical
simulations by the model. (a–d) Ratio of the range of single-model ensemble means in the 13-model and 31-model ensembles. (e–h) Ratio of
the range of single-model ensemble means to the range of all individual simulations in the 13-model ensemble. Region boundaries are IPCC
regions (Figure 7h) [Colour figure can be viewed at wileyonlinelibrary.com]
MAHONY ET AL.5879

exclusion of the NorESM, CESM2, and MPI-ESM1-2-LR
models, all of which have multiple simulations and are
therefore robust to natural variability. Reduced representa-
tion of the temperature change in central North America
(Figure 3c,d) is associated with low warming in CAMS-
CSM1-0 and high warming in HadGEM3-GC31-LL.
The range ratio analysis also provides a useful assess-
ment of the degree to which the ensemble spread in indi-
vidual model simulations (runs) is represented by the
single-model means of these simulations (Figure 3e–h).
Model means of temperature change are generally repre-
sentative of the individual model runs, with exceptions of
summer in northern Central America (Figure 3g) and
winter in NW North America (Figure 3h). In contrast,
model means of precipitation change are poorly represen-
tative of the variation in individual model runs, particu-
larly in winter (Figure 3e,f).
3.2.2 | Projected climate change
A visual comparison of projected seasonal changes in
T
min
,T
max
, and precipitation (Figure 4) indicates some
basic attributes of the ensemble simulations. All models
exhibit Arctic amplification of winter temperatures,
though it is relatively subtle in EC-Earth3. Most models
project the strongest summer warming at mid-latitudes.
All models, with the exception of UKESM1, have a simi-
lar pattern of warming in T
min
and T
max
, though the mag-
nitude of warming is greater for T
min
in most models.
Continental-scale patterns of winter (December–
February) precipitation change are somewhat consistent
among models, with declines in Mexico and increases in
the Arctic and Boreal regions. Deviations from this pat-
tern are strongest in models with few (1–3) historical
runs for SSP2-4.5 (BCC-CSM2-MR, GFDL-ESM4, and
INM-CM5.0), likely due to internal variability. This result
emphasizes the benefit of multiple runs in smoothing out
natural variability to reveal the anthropogenic climate
change signal in noisy climate variables like precipitation
and winter temperature.
Most models project a reduction in summer precipita-
tion in the coastal areas of the Pacific Northwest
(California, Oregon, Washington, and southern BC).
There is substantial disagreement among models in sum-
mer precipitation change over the rest of the continent.
The muted summer precipitation change in the ensemble
mean hides this ensemble disagreement, and underscores
the importance of assessing climate change impacts with
an ensemble of model projections rather than solely
using the ensemble mean.
The two high-ECS models CanESM5 and UKESM1
have similar patterns and magnitudes of change in
winter temperature and precipitation. However, they
differ substantially in the summer, with UKESM1 show-
ing much higher increases in daytime temperatures
(T
max
) in temperate and boreal regions and stronger
declines in precipitation in central North America.
Although CanESM5 has a higher ECS and stronger
trend in 1970–2014 global heating (Liang et al., 2020),
UKESM1 projects stronger mid-century heating over
North America.
3.2.3 | Spatial resolution and model
orography
The selected 13-model ensemble has a mean latitudinal
grid resolution of 1.4(range of 0.7–2.8)(Figure5). Four
models (EC-Earth3, GFDL-ESM4, MPI-ESM1.2-HR, and
MRI-ESM2.0) resolve the macrotopography of the Western
Cordillera, namely the Sierra Nevada, Cascade Range,
Rocky Mountains, and British Columbia (BC) Coast
Ranges. BCC-CSM2-MR does not resolve these ranges,
despite having sufficient grid resolution to do
so. CanESM5 has a distinctly low resolution of 2.8×2.8.
3.2.4 | Elevation-dependent warming
There are large differences among models in the repre-
sentation of elevation-dependent warming (EDW).
These differences are illustrated using spring (MAM)
T
max
in a subset of the ensemble over the Coast Range
and Rocky Mountains of southwestern Canada
(Figure 6). EC-Earth3 and MRI-ESM2.0 both resolve
these mountain ranges in their model orography
(Figure 6a,d). EC-Earth3 has a strong signal of
elevation-dependent warming over the Coast Range and
Rocky Mountains, with more than double the warming
in the Rocky Mountains than the adjacent plateaus
(Figure 6b,c). MRI-ESM2.0 exhibits a relatively weak
relationship between elevation and warming (Figure 6e,
f). ACCESS-ESM1.5 and MIROC6 represent models with
lower spatial resolution that represent the Coast Range
and Rocky Mountains as a single feature in their model
orography. ACCESS-ESM1.5 does not exhibit EDW.
MIROC6 has a strong EDW signal despite being a
moderate-resolution model with no differentiation of
the two mountain ranges. The example of spring T
max
is
purely illustrative and does not represent each model's
EDW in other elements and seasons. For example, the
pattern of EDW among models is reversed for summer
T
min
(not shown), where MRI-ESM2.0 and ACCESS-
ESM1.5 exhibit EDW, while EC-Earth3 and MIROC6
do not.
5880 MAHONY ET AL.

FIGURE 4 Spatial variation in climate change responses among the 13-model ensemble. Mapped climate changes are for the mean
projected climate of the 2041–2060 period (SSP2-4.5). Precipitation is log-scaled to provide proportional magnitude of positive and negative
changes. Models are structured by a cluster dendrogram of spatial similarity in seasonal climate changes in all three climate elements
[Colour figure can be viewed at wileyonlinelibrary.com]
MAHONY ET AL.5881

3.2.5 | Diurnal temperature range
The models consistently underestimate the diurnal tem-
perature range (DTR), measured as the difference
between T
min
and T
max
(Figure 7). However, the
13-model ensemble and the 8-model subset (described in
section 3.3.2) reproduce the observed seasonal cycle in all
regions. Models that deviate most from the ensemble
mean seasonal cycle generally are those excluded from
the 8-model subset, namely IPSL-CM6A-LR (high ampli-
tude in Arctic regions and underestimated elsewhere),
BCC-CSM2-MR (high amplitude at mid-latitudes), and
UKESM1-0-LL (high amplitude in Arctic regions and
WNA). Among the 8-model subset, MIROC6 is distinct in
overestimating the amplitude of the seasonal cycle in
most regions.
3.3 |Ensemble subset selection
3.3.1 | Screening exclusions
The following four models are prioritized for exclusion
from subsets of the ensemble based on combinations of
the four screening criteria: (a) CanESM5,duetoits
very high climate sensitivity (Criterion 6) and its very
low horizontal resolution (Criterion 8); (b) INM-
CM5.0, because it has very low climate sensitivity (ECS
1.9C)andisanoutlieramongCMIP6modelsfor
under-representing the observed 1975–2014 global tem-
perature trend (Liang et al., 2020) (Criterion 7). In
addition, this model has only one simulation for most
scenarios, producing a less robust climate signal
(Criterion 9); (c) BCC-CSM2-MR, due to having a single
simulation for each scenario (Criterion 9) and low
topographic resolution (Criterion 8); and (d) IPSL-
CM6A-LR, due to isolated grid cells with very high
summer warming in the BC Coast Ranges and South-
east Alaska (Figure 8; Criterion 10). The warming in
these cells may be physically plausible in the model's
simplified topography, but is problematic for downscal-
ing to higher spatial resolutions.
UKESM1 also has very high climate sensitivity (Crite-
rion7),similartoCanESM5,thatisassessedbytheIPCC
as very unlikely based on observational evidence
(Sherwood et al., 2020;Ariaset al., 2021). Some
researchers may wish to constrain their ensemble subset
to observations by excluding this model. Others may wish
FIGURE 5 Effective topographic resolution of the 13 selected models. (a–m) Model orography (elevation of land surface) in the native grid of
each model. The extent of the map is central-western North America (106 –142 W, 37 –62 N). The common grid (panel n) is the 0.5grid used for
extraction of observations from ClimateNA for the purpose of the bias assessment
5882 MAHONY ET AL.

to include a high-sensitivity model in their subset as a
representation of the long tail of uncertainty in the upper
limit of climate sensitivity (Sutton, 2018). To accommo-
date both perspectives, we provide ordered subsets with
and without UKESM1 in the ordered ensemble subsets.
We preferred UKESM1 over CanESM5 as a representa-
tive of high-sensitivity models due to its higher grid reso-
lution and closer alignment with the observed post-1970
global heating trend (Liang et al., 2020).
The resulting 8-model subset is ACCESS-ESM1.5,
CNRM-ESM2-1, EC-Earth3, GFDL-ESM4, GISS-E2-1-G,
MIROC6, MPI-ESM1.2-HR, and MRI-ESM2.0. This
8-model ensemble has a mean global ECS of 3.4C
(2.6–4.8C), using ECS values provided by Meehl
et al.(
2020). The 9-model subset that includes UKESM1
has a mean global ECS of 3.6C (2.6–5.4C).
3.3.2 | Ordered subsets
Table 3specifies ordered subsets of the models that pas-
sed screening criteria 7–10. For a desired region and sub-
set size, the ensemble subset for each region includes all
models listed at and above the desired subset size. For
example, a 4-model ensemble for the NEN region would
include CNRM-ESM2-1, UKESM1.0-LL, EC-Earth3, and
FIGURE 6 Relationships between elevation and warming (spring (MAM) T
max
) over southwestern Canada in four CMIP6 models.
Projected warming is for the 2061–2080 period under SSP2-4.5 for all models. Coastal cells (elevation <500 m) are excluded to reduce the
maritime influence on the analysis [Colour figure can be viewed at wileyonlinelibrary.com]
MAHONY ET AL.5883

MPI-ESM1.2-HR. The considerable variation among
regions in the order of the subsets underscores the spatial
variation in climate change responses across North
America. The exception to this variation in model order
is that UKESM1 is the second model in all regions in the
9-model subset. Since the first position in the order is the
model closest to the ensemble centroid and the second
position is the model furthest from the centroid, this
result indicates that UKESM1 projects the most extreme
climate changes in all IPCC reference regions of North
America.
4|DISCUSSION
We selected 13 CMIP6 models from a candidate pool of
44 models contributing to the CMIP6 experiment. This
13-model ensemble is representative of the distribution of
equilibrium climate sensitivity in the full CMIP6 ensem-
ble and adequately represents the CMIP6 range of tran-
sient regional changes in precipitation and temperature.
The 13-model ensemble facilitates robust downscaling by
using multiple historical simulations per scenario for
each model and excluding models with high bias. We
provided rationale for an 8-member subset of the ensem-
ble based on screening criteria and order these 8 models
for selection of smaller ensembles for regional analysis in
North America. We also highlighted some tradeoffs
among the models in terms of grid resolution, number of
simulation runs, climate sensitivity, regional biases, and
local artefacts. These results, and the accompanying web
application (https://bcgov-env.shinyapps.io/cmip6-NA/),
help readers to make model selections appropriate to
their specific research objectives.
4.1 |Model bias
The bias assessment was a useful way to identify models
with extreme divergence from the observed climate. Excep-
tionally high temperature biases were a sufficient reason for
the exclusion of AWI-CM1-1-1-MR. High temperature
biases are an attribute of concern in two of the models
selected for the ensemble, ACCESS-ESM1.5 and MIROC6,
but without a process evaluation are not sufficient basis for
exclusion. The moderate biases in the rest of the ensemble,
however, do not necessarily indicate a problem with the
models. Bias is the difference between model simulations
Observed
13−model ensemble
8−model subset
13−model ens. mean
Jan Apr Jul Oct Jan
5
10
15
Diurnal temperature range (°C)
(a) NWN
ACC
BCC Can
CNR
EC
GFD
GIS
INM
IPS
MIR
MPI
MRI
UK
Jan Apr Jul Oct Jan
5
10
15
(b) NEN
ACC
BCC
Can
CNR
EC GFD
GIS
INM
IPS
MIR
MPI
MRI
UK
Jan Apr Jul Oct Jan
5
10
15
Diurnal temperature range (°C)
(c) WNA
ACC
BCC
Can CNR
EC
GFD GIS
INM
IPS
MIR
MPI
MRI
UK
Jan Apr Jul Oct Jan
5
10
15
(d) CNA
ACC
BCC
Can CNR
EC
GFD
GIS
INM
IPS
MIR
MPI
MRI
UK
Jan Apr Jul Oct Jan
5
10
15
(e) ENA
ACC
BCC
Can
CNR
EC
GFD
GIS
INM
IPS
MIR
MPI
MRI
UK
Jan Apr Jul Oct Jan
5
10
15
Diurnal temperature range (°C)
(f) NCA
ACC
BCC
Can
CNR EC
GFD
GIS
INM
IPS
MIR
MPI
MRI
UK
Jan Apr Jul Oct Jan
5
10
15
(g) SCA
ACC
BCC
Can
CNR
EC
GFD
GIS
INM
IPS
MIR
MPI MRI
UK
CNA ENA
NCA
NEN
NWN
SCA
WNA
(h)
FIGURE 7 Seasonal cycle of the mean diurnal temperature range in observations and the 13-model ensemble, averaged over each IPCC
reference region (h) (Iturbide et al., 2020). Mean diurnal temperature range is calculated as the difference between monthly 1961–1990
normals of T
min
and T
max
. Observations are the ClimateNA composite of PRISM and WorldClim gridded climate normals (Wang
et al., 2016). Model abbreviations are the first two to three letters of the model name [Colour figure can be viewed at wileyonlinelibrary.com]
5884 MAHONY ET AL.

and the observed climate. We controlled the confounding
influence of natural variability in each model by calculating
bias using the mean of several simulation runs. This mea-
sure is not possible for observations since there is only one
realization of the observed climate. Natural variability in
the observed climate, therefore, could produce the appear-
ance of bias even in a hypothetical “perfect”model
(Lanzante et al., 2018). The ensemble mean absolute bias of
2C in temperature and by a factor of 1.5 in precipitation
cannot be definitively attributed to the models or the
ensemble; it is to some extent an artefact of natural variabil-
ity in the observed climate. A process-based evaluation of
models (e.g., Karmalkar et al., 2019) can be helpful in fur-
ther assessing the reliability of individual models for spe-
cific purposes and regions.
4.2 |Representation of climate change
uncertainty
Climate change uncertainty can be decomposed into three
components: scenario uncertainty, modelling (process)
uncertainty, and internal (natural) variability (Hawkins
and Sutton, 2009). Ideally, a small multimodel ensemble
like the one selected here should conserve these uncer-
tainties as expressed in the larger ensemble of candidate
simulations. We explicitly conserved scenario uncertainty
by selecting models that provide simulations for the four
core SSP marker scenarios. The 13-model ensemble ade-
quately conserves modelling uncertainty—approximated
globally by ECS and regionally by the range of model mean
seasonal temperature and precipitation changes in a
screened 31-model CMIP6 ensemble (Figures 2and
3a–d)—with some regional exceptions. These exceptions
should be noted as caveats for interpretation of regional cli-
mate change uncertainties in downstream analyses. Finally,
we explicitly prioritized models that allow assessment and
control of internal variability by selecting models with a
minimum of three historical simulations. However, five of
the selected models have only one or two simulations for
each emissions scenario. For the purpose of representing
modelling uncertainty, model means from these models
should be assumed to be somewhat confounded by internal
variability.
The internal variability component of uncertainty is
only conserved if the individual simulations of each model
0
3.5
7
Change in Summer mean daily
maximum temperature (K)
ACCESS−ESM1−5 (10)
(a)
BCC−CSM2−MR (1)
(b)
CanESM5 (10)
(c)
CNRM−ESM2−1 (5)
(d)
EC−Earth3 (5)
(e)
GFDL−ESM4 (3)
(f)
GISS−E2−1−G (4)
(g)
INM−CM5−0 (1)
(h)
IPSL−CM6A−LR (6)
(i)
MIROC6 (3)
(j)
MPI−ESM1−2−HR (2)
(k)
MRI−ESM2−0 (5)
(l)
UKESM1−0−LL (5)
(m)
Ensemble mean
(n)
FIGURE 8 Summer (JJA) daytime warming in the 13-model ensemble over central-western North America (106 –142 W, 37 –62 N).
Values are the change in summer T
max
for the 2041–2060 period (SSP2-4.5), relative to 1961–1990,inthenativemodelgrid.Changeis
calculated from the mean of multiple simulation runs per model, specified next to the model name [Colour figure can be viewed at
wileyonlinelibrary.com]
MAHONY ET AL.5885

are provided in a downscaled ensemble. Downscaled cli-
mate normal products (e.g., ClimateNA and WorldClim2)
typically provide model means calculated from multiple
simulations, but not the individual simulations themselves.
We demonstrated that the ensemble of model means
obscures substantial internal variability in precipitation
throughout the continent, and in temperature in specific
regions and seasons (Figure 3e–h). In other words, an
ensemble of model means artificially reduces climate
change uncertainty (Deser et al., 2012) and precludes assess-
ment of the full range of potential climate impacts. Within
the limits of practicality, it is preferable to provide multiple
simulations of each model/scenario combination in prod-
ucts of downscaled climate normals.
4.3 |Grid resolution and elevation-
dependent warming
Four of the models in the ensemble have horizontal grid
resolution sufficient to resolve the major mountain ranges
of western North America. One model (EC-Earth3) has
relatively high resolution (0.7×0.7) approaching the
previous generation of regional climate models used for
dynamical downscaling. The trend towards higher resolu-
tion is encouraging, but the benefits of moderate resolu-
tion models for km-scale downscaling are ambiguous. On
one hand, resolving mountain ranges allows for stronger
differentiation of maritime/continental transitions
(Lanzante et al., 2018), windward and leeward dynamics
(Kanehama et al., 2019), and elevation-dependent climate
changes (Palazzi et al., 2019). On the other hand, these
resolved mountain ranges are still highly simplified fea-
tures in even the highest resolution models in the ensem-
ble. The higher-resolution models do not necessarily
reduce the challenges of km-scale downscaling, but
instead can shift these challenges to finer spatial scales.
Our case study of elevation-dependent warming
(EDW) is an illustration of the trade-offs of increased grid
resolution. EDW is a poorly-understood phenomenon
with several hypothesized causes including the snow-
albedo feedback, longwave water vapour feedbacks,
TABLE 3 Ordered subsets of the 13-model ensemble
Subset size
IPCC reference region
NEN NWN WNA CNA ENA NCA SCA NAM
Including UKESM1-0-LL
1 CNRM CNRM MRI ACC EC MRI GISS CNRM
2 UKUKUKUKUKUKUKUK
3 EC EC MPI CNRM GFDL GFDL ACC GFDL
4 MPI MPI GISS GFDL MRI EC MIR EC
5 MRI ACC EC MIR MIR MIR EC MRI
6 ACC GISS CNRM EC GISS CNRM GFDL GISS
7 GISS MRI MIR GISS MPI MPI MPI MIR
8 GFDL MIR GFDL MPI ACC ACC CNRM ACC
9 MIR GFDL ACC MRI CNRM GISS MRI MPI
Excluding UKESM1-0-LL
1 CNRM CNRM MRI MRI GISS MRI GISS CNRM
2 EC EC MPI GFDL ACC GFDL ACC GFDL
3 GFDL ACC GISS MIR MRI EC MIR EC
4 MRI MPI MIR CNRM GFDL MIR EC GISS
5 ACC GISS EC GISS CNRM CNRM GFDL MIR
6 GISS MIR CNRM EC EC MPI MPI ACC
7 MPI MRI GFDL MPI MIR ACC MRI MRI
8 MIR GFDL ACC ACC MPI GISS CNRM MPI
Note: Subsets are provided for North America (NAM) and the seven IPCC reference regions (Figure 7h). The order of each subset indicates the models
recommended for each user-determined ensemble size. For example, a four-member ensemble for the NEN region would comprise CNRM, UK, EC, and MPI.
Exclusion of UKESM1 provides an ensemble that is consistent with the IPCC AR6 assessed constraints on equilibrium climate sensitivity (Arias et al., 2021).
Model abbreviations are ACC (ACCESS-ESM1.5), CNRM (CNRM-ESM2-1), EC (EC-Earth3), GFDL (GFDL-ESM4), GISS (GISS-E2-1-G), MIR (MIROC6), MPI
(MPI-ESM1.2-HR), MRI (MRI-ESM2.0), and UK (UKESM1.0-LL).
5886 MAHONY ET AL.

aerosols, and changes in cloud cover (Pepin et al., 2015;
Minder et al., 2018; Palazzi et al., 2019). The large differ-
ences in EDW evident among models in Figure 6are
unsurprising given the complexity of the phenomenon.
In addition to the process uncertainties, however, grid
resolution itself is a source of model variation in EDW.
Since the drivers of EDW are local in scale, the simula-
tion of EDW over the highly generalized topographies of
global climate models is a source of error for downscal-
ing: EDW will be applied uniformly to all locations—
valley floors to mountain peaks—within a GCM grid cell
during change-factor downscaling. The comparison of
MIROC6 and EC-Earth3 demonstrates that increasing
model resolution can reduce this source of error at
regional scales (i.e., by resolving gaps between mountain
ranges) while increasing it at local scales (i.e., by applying
more fully resolved EDW to unresolved valleys). In the
absence of additional downscaling measures to explicitly
account for EDW and other localized climate change
drivers, we do not view the higher-resolution models in
the ensemble as intrinsically more valuable or valid. The
tradeoffs of model resolution are a component of model-
ling uncertainty, and it is beneficial to include a range of
grid resolutions in a downscaling ensemble.
4.4 |Diurnal temperature range
Underestimation of DTR is a persistent feature of climate
models (Wang and Clow, 2020). Intermodel differences in
DTR can be attributed to differences in parameterizations
for clouds, aerosols and soil moisture, among others
(Lindvall and Svensson, 2015). However, the consistent
underestimation of DTR relative to observations has not
been definitively explained. Part of the underestimation of
DTR may be due to differences in the timescale of DTR
measurement in observations and models; since T
min
and
T
max
are measured instantaneously in observations but sim-
ulated overlonger timesteps in models, models are expected
to have lower DTR (Wilson et al., 2008;Ruppet al., 2013).
To the extent that underestimation of DTR is an artefact of
the different timescales of measurement in observations
and models, rather than of systematic biases in the driving
processes, some overestimation of T
min
and underestima-
tion of T
max
would be expected even from a perfect model.
4.5 |Reconciling the equilibrium
climate sensitivity of the ensemble with
observational constraints
The 13-model ensemble selected here, like the full CMIP6
ensemble, has a mean (3.7C) and upper limit (5.6C) of
equilibrium climate sensitivity that substantially exceeds
the IPCC AR6 assessed best estimate ECS of 3C and very
likely upper limit of 5C (Arias et al., 2021). In other
words, the 13-model ensemble contains models that sim-
ulate stronger global warming than is supported by mul-
tiple lines of observational evidence. Five (38%) of the
13 models are above the IPCC AR6 assessed likely upper
limit on ECS of 4C, and two (15%) of the models are
above the very likely upper limit of 5C. If the ensemble
were to strictly conform to the IPCC assessed range, there
would be only two models exceeding 4C ECS and no
models exceeding 5C, following the IPCC's probabilistic
definitions of likely (one-sided p>83%) and very likely
(one-sided p>95%).
The need to reconcile the CMIP ensemble ECS range
with observational constraints is a new dilemma for cli-
mate change impacts and adaptation researchers. It is
long been agreed that model democracy (one model, one
vote) is not a strictly valid method of assessing climate
change uncertainty (Knutti, 2010; Leduc et al., 2016).
However, prior to CMIP6 this objection was somewhat
academic since the distribution of ECS in CMIP ensem-
bles approximately matched the (wider) range of ECS
supported by other lines of evidence (Schmidt, 2021). For
practical purposes it was reasonable for analysts to use
the multimodel ensemble spread in previous CMIP gen-
erations as a proxy for scientific uncertainty on climate
change. This approach is no longer valid given the incon-
gruence between the CMIP6 ensemble range of ECS and
the IPCC assessed range (Schmidt, 2021). Careful model
selection is now required to avoid biasing regional cli-
mate change analyses.
There are several viable approaches to constrain
CMIP6 ensembles in downscaled regional analyses.
Weighting the models based on observational constraints
is possible for regional analyses (Ribes et al., 2021). How-
ever, in practice many analyses will require simply
selecting a subset of the CMIP6 ensemble that is closer to
the IPCC assessed range, as we have done with the
8-model subset. The disadvantage of this approach is that
it discards valuable information from the excluded
models. The CanESM5 and UKESM1 models are
advanced models from respected modelling centres, with
demonstrated skill in modelling many Earth system pro-
cesses (Eyring et al., 2021). Expressing variables of inter-
est relative to the amount of regional or global warming
is a widely practiced technique that facilitates inclusion
of high-ECS models by removing the timing of the
warming as a factor in the ensemble spread (Arias
et al., 2021). It is conceivable that both techniques could
be used in a single study; to use the 8-model ensemble
for time-relevant analyses and a larger ensemble for ana-
lyses where the warming level is more relevant. These
MAHONY ET AL.5887

considerations highlight that the full CMIP6 ensemble is
a somewhat arbitrary collection of non-independent
models, and careful ensemble selection is necessary to
achieve a meaningful representation of modelling
uncertainty.
5|CONCLUSION
Use of downscaled global climate model projections is
expanding rapidly as climate change vulnerability assess-
ments and adaptation planning become mainstream in
many sectors. This increasingly diverse user base can
benefit from a basic understanding of the attributes and
limitations of the climate model output they are working
with. The results of this study reinforce several best prac-
tices for the selection and use of global climate models
for high spatial resolution downscaling of climate
normals:
•The ensemble must be observationally constrained to
be meaningful. The CMIP6 ensemble and the
13-model ensemble have a biased distribution of equi-
librium climate sensitivities. This bias can be addressed
by model exclusions (as in our 8-model subset), model
weighting, or analysis relative to global warming levels
rather than time.
•An ensemble of multiple models is essential for impact
analysis; the ensemble mean projection alone can be
misleading. For example, small summer precipitation
changes indicated by the ensemble mean throughout
North America are an artefact of averaging across
much larger opposing trends among the models.
•Downscaled climate data products should include mul-
tiple simulations of each model/scenario combination.
This allows downstream analyses to account for the
contribution of natural variability to climate change
uncertainty.
•Higher grid resolution is not necessarily better.
Increased grid resolution can intensify downscaling
errors in some locations while reducing them in
others. A range of grid resolutions is desirable in a
multimodel ensemble.
•Small ensembles should be used with caution. Even
the 13-model ensemble leaves region-specific gaps in
the distribution of climate changes projected by the full
CMIP6 ensemble. Users of downscaled data can bene-
fit from tools to identify these gaps for their variables
and regions of interest.
ACKNOWLEDGEMENTS
We acknowledge the World Climate Research Programme,
which, through its Working Group on Coupled Modelling,
coordinated and promoted CMIP6. We thank the climate
modelling groups for producing and making available
their model output, the Earth System Grid Federation
(ESGF) for archiving the data and providing access, and
the multiple funding agencies who support CMIP6 and
ESGF. We are grateful to Caren Dymond and Ambarish
Karmalkar for their helpful comments on the manuscript.
CONFLICT OF INTEREST
The authors declare no potential conflict of interest.
AUTHOR CONTRIBUTIONS
Colin R. Mahony: Conceptualization; formal analysis;
methodology; software; visualization; writing –original
draft. Tongli Wang: Conceptualization; data curation.
Andreas Hamann: Conceptualization; formal analysis;
writing –review and editing. Alex J. Cannon: Methodol-
ogy; software; writing –review and editing.
DATA AVAILABILITY STATEMENT
A supplementary web application (https://bcgov-env.
shinyapps.io/cmip6-NA/) provides tools for model selec-
tion and visualization of the ensemble. Downscaled projec-
tions for the selected 13 CMIP6 models are available in
ClimateNA (http://climatena.ca/), which provides down-
scaling at user-specified spatial resolution and various
temporal intervals (annual, 20-year and 30-year periods).
In addition to the basic monthly values of Tmin, Tmax,
and precipitation, ClimateNA uses these elements to esti-
mate derived bioclimatic and engineering variables such
as heat sums and frost periods. Gridded (1-km) climate
normals for the 8-model subset are provided at https://
adaptwest.databasin.org/pages/adaptwest-climatena/.
ORCID
Colin R. Mahony https://orcid.org/0000-0002-6111-
5675
Tongli Wang https://orcid.org/0000-0002-9967-6769
Andreas Hamann https://orcid.org/0000-0003-2046-
4550
Alex J. Cannon https://orcid.org/0000-0002-8025-3790
REFERENCES
Arias, P.A., Bellouin, N., Coppola, E., Jones, R.G., Krinner, G.,
Marotzke, J., Naik, V., Palmer, M.D., Plattner, J.G.-K.,
Rogelj, M.R., Sillmann, J., Storelvmo, T., Thorne, P.W.,
Trewin, B., Rao, K.A., Adhikary, B., Allan, R.P., Armour, K.,
Bala, G., Barimalala, R., Berger, S., Canadell, J.G., Cassou, C.,
Cherchi, A., Collins, W., Collins, W.D., Connors, S.L., Corti, S.,
Cruz, F., Dentener, F.J., Dereczynski, C., Di Luca, A., Niang, A.
D., Doblas-Reyes, F.J., Dosio, A., Douville, H., Engelbrecht, F.,
Eyring, V., Fischer, E., Forster, P., Fox-Kemper, B.,
Fuglestvedt, J.S., Fyfe, J.C., Gillett, N.P., Goldfarb, L.,
Gorodetskaya, I., Gutierrez, J.M., Hamdi, R., Hawkins, E.,
5888 MAHONY ET AL.

Hewitt, H.T., Hope, P., Islam, A.S., Jones, C., Kaufman, D.S.,
Kopp, R.E., Kosaka, Y., Kossin, J., Krakovska, S., Lee, J.-Y.,
Li, J., Mauritsen, T., Maycock, T.K., Meinshausen, M.,
Min, S.-K., Monteiro, P.M.S., Ngo-Duc, T., Otto, F., Pinto, I.,
Pirani, A., Raghavan, K., Ranasinghe, R., Ruane, A.C., Ruiz, L.,
Sallée, J.-B., Samset, B.H., Sathyendranath, S., Seneviratne, S.I.,
Sörensson, A.A., Szopa, S., Takayabu, I., Treguier, A.-M., van
den Hurk, B., Vautard, R., von Schuckmann, S.Z.K., Zhang, X.
and Zickfeld, K. (2021) Technical summary. In: Masson-
Delmotte, P.Z., Pirani, A., Connors, S.L., Péan, C., Berger, S.,
Caud, N., Chen, Y., Goldfarb, L., Gomis, M.I., Huang, M.,
Leitzell, K., Lonnoy, E., Matthews, J.B.R., Maycock, T.K.,
Waterfield, T., Yelekçi, O., Yu, R. and Zhou, B. (Eds.) Climate
Change 2021: The Physical Science Basis. Contribution of Work-
ing Group I to the Sixth Assessment Report of the Intergovern-
mental Panel on Climate Change. Cambridge and New York,
NY: Cambridge University Press, pp. 1–150.
Boucher, O., Servonnat, J., Albright, A.L., Aumont, O., Balkanski, Y.,
Bastrikov, V., Bekki, S., Bonnet, R., Bony, S., Bopp, L.,
Braconnot, P., Brockmann, P., Cadule, P., Caubel, A., Cheruy, F.,
Codron, F., Cozic, A., Cugnet, D., D'Andrea, F., Davini, P., de
Lavergne, C., Denvil, S., Deshayes, J., Devilliers, M.,
Ducharne, A., Dufresne, J.L., Dupont, E.,

Ethé, C., Fairhead, L.,
Falletti, L., Flavoni, S., Foujols, M.A., Gardoll, S., Gastineau, G.,
Ghattas, J., Grandpeix, J.Y., Guenet, B., Guez, L.E., Guilyardi, E.,
Guimberteau, M., Hauglustaine, D., Hourdin, F., Idelkadi, A.,
Joussaume, S., Kageyama, M., Khodri, M., Krinner, G., Lebas, N.,
Levavasseur, G., Lévy, C., Li, L., Lott, F., Lurton, T., Luyssaert, S.,
Madec, G., Madeleine, J.B., Maignan, F., Marchand, M., Marti, O.,
Mellul, L., Meurdesoif, Y., Mignot, J., Musat, I., Ottlé, C.,
Peylin, P., Planton, Y., Polcher, J., Rio, C., Rochetin, N.,
Rousset, C., Sepulchre, P., Sima, A., Swingedouw, D.,
Thiéblemont, R., Traore, A.K., Vancoppenolle, M., Vial, J.,
Vialard, J., Viovy, N. and Vuichard, N. (2020) Presentation and
evaluation of the IPSL-CM6A-LR climate model. Journal of
Advances in Modeling Earth Systems,12,1–52.
Brunner, L., Pendergrass, A.G., Lehner, F., Merrifield, A.L.,
Lorenz, R. and Knutti, R. (2020) Reduced global warming from
CMIP6 projections when weighting models by performance
and independence. Earth System Dynamics, 11, 995–1012.
Cannon, A.J. (2015) Selecting GCM scenarios that span the range of
changes in a multimodel ensemble: application to CMIP5 cli-
mate extremes indices. Journal of Climate, 28, 1260–1267.
Cannon, A.J. (2018) Multivariate quantile mapping bias correction:
an N-dimensional probability density function transform for
climate model simulations of multiple variables. Climate
Dynamics, 50, 31–49.
Deser, C., Knutti, R., Solomon, S. and Phillips, A.S. (2012) Commu-
nication of the role of natural variability in future North Ameri-
can climate. Nature Climate Change, 2, 775–779.
Döscher, R., Acosta, M., Alessandri, A., Anthoni, P., Arneth, A.,
Arsouze, T., Bergmann, T., Bernadello, R., Bousetta, S.,
Caron, L.-P., Carver, G., Castrillo, M., Catalano, F.,
Cvijanovic, I., Davini, P., Dekker, E., Doblas-Reyes, F.,
Docquier, D., Echevarria, P., Fladrich, U., Fuentes-Franco, R.,
Gröger, M., Hardenberg, J.V., Hieronymus, J., Karami, M.P.,
Keskinen, J.-P., Koenigk, T., Makkonen, R., Massonnet, F.,
Ménégoz, M., Miller, P., Moreno-Chamarro, E., Nieradzik, L.,
van Noije, T., Nolan, P., O'Donnell, D., Ollinaho, P., van den
Oord, G., Ortega, P., Prims, O.T., Ramos, A., Reerink, T.,
Rousset, C., Ruprich-Robert, Y., Le Sager, P., Schmith, T.,
Schrödner, R., Serva, F., Sicardi, V., Madsen, M.S., Smith, B.,
Tian, T., Tourigny, E., Uotila, P., Vancoppenolle, M., Wang, S.,
Wårlind, D., Willén, U., Wyser, K., Yang, S., Yepes-Arb
os, X.
and Zhang, Q. (2021) The EC-Earth3 earth system model for
the Climate Model Intercomparison Project 6. Geoscientific
Model Development Discussions,1–90. https://doi.org/10.5194/
gmd-2020-446
Dunne, J.P., Horowitz, L.W., Adcroft, A.J., Ginoux, P., Held, I.M.,
John, J.G., Krasting, J.P., Malyshev, S., Naik, V., Paulot, F.,
Shevliakova, E., Stock, C.A., Zadeh, N., Balaji, V., Blanton, C.,
Dunne, K.A., Dupuis, C., Durachta, J., Dussin, R., Gauthier, P.P.
G., Griffies, S.M., Guo, H., Hallberg, R.W., Harrison, M., He, J.,
Hurlin, W., McHugh, C., Menzel, R., Milly, P.C.D., Nikonov, S.,
Paynter, D.J., Ploshay, J., Radhakrishnan, A., Rand, K., Reichl, B.
G., Robinson, T., Schwarzkopf, D.M., Sentman, L.T.,
Underwood, S., Vahlenkamp, H., Winton, M., Wittenberg, A.T.,
Wyman, B., Zeng, Y. and Zhao, M. (2020) The GFDL Earth Sys-
tem Model version 4.1 (GFDL-ESM 4.1): overall coupled model
description and simulation characteristics. Journal of Advances in
Modeling Earth Systems, 12, e2019MS002015.
Eyring, V., Bony, S., Meehl,G.A., Senior, C.A., Stevens, B., Stouffer, R.
J. and Taylor, K.E. (2016) Overview of the Coupled Model
Intercomparison Project Phase 6 (CMIP6) experimental design
and organization. Geoscientific Model Development,9,1937–1958.
Eyring, V., Gillett, N.P., Rao, K.M.A., Barimalala, R., Parrillo, M.B.,
Bellouin, N., Cassou, C., Durack, P.J., Kosaka, Y., McGregor, S.,
Min, S., Morgenstern, O. and Sun, Y. (2021) Human influence
on the climate system. In: Masson-Delmotte, V., Zhai, P.,
Pirani, A., Connors, S.L., Péan, C., Berger, S., Caud, N.,
Chen, Y., Goldfarb, L., Gomis, M.I., Huang, M., Leitzell, K.,
Lonnoy, E., Matthews, J.B.R., Maycock, T.K., Waterfield, T.,
Yelekçi, O., Yu, R. and Zhou, B. (Eds.) Climate Change 2021:
The Physical Science Basis. Contribution of Working Group I to
the Sixth Assessment Report of the Intergovernmental Panel on
Climate Change. Cambridge and New York, NY: Cambridge
University Press, pp. 1–202.
Fick, S.E. and Hijmans, R.J. (2017) WorldClim 2: new 1-km spatial
resolution climate surfaces for global land areas. International
Journal of Climatology, 37, 4302–4315.
Hamann, A., Wang, T., Spittlehouse, D.L. and Murdock, T.Q.
(2013) A comprehensive, high-resolution database of historical
and projected climate surfaces for western North America. Bul-
letin of the American Meteorological Society, 94, 1307–1309.
Hawkins, E. and Sutton, R. (2009) The potential to narrow uncer-
tainty in regional climate predictions. Bulletin of the American
Meteorological Society, 90, 1095–1107.
Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. and Jarvis, A.
(2005) Very high resolution in terpolated climate surfaces for global
land areas. International Journal ofClimatology, 25, 1965–1978.
Hunter, R.D. and Meentemeyer, R.K. (2005) Climatologically aided
mapping of daily precipitation and temperature. Journal of
Applied Meteorology, 44, 1501–1510.
Iturbide, M., Gutiérrez, J.M., Alves, L.M., Bedia, J., Cerezo-
Mota, R., Cimadevilla, E., Cofiño, A.S., Di Luca, A., Faria, S.H.,
Gorodetskaya, I.V., Hauser, M., Herrera, S., Hennessy, K.,
Hewitt, H.T., Jones, R.G., Krakovska, S., Manzanas, R.,
Martínez-Castro, D., Narisma, G.T., Nurhati, I.S., Pinto, I.,
MAHONY ET AL.5889

Seneviratne, S.I., van den Hurk, B. and Vera, C.S. (2020) An
update of IPCC climate reference regions for subcontinental
analysis of climate model data: definition and aggregated
datasets. Earth System Science Data, 12, 2959–2970.
Kanehama, T., Sandu, I., Beljaars, A., van Niekerk, A. and Lott, F.
(2019) Which orographic scales matter most for medium-range
forecast skill in the Northern Hemisphere winter? Journal of
Advances in Modeling Earth Systems, 11, 3893–3910.
Karmalkar, A.V. (2018) Interpreting results from the NARCCAP
and NA-CORDEX ensembles in the context of uncertainty in
regional climate change projections. Bulletin of the American
Meteorological Society, 99, 2093–2106.
Karmalkar, A.V., Thibeault, J.M., Bryan, A.M. and Seth, A. (2019)
Identifying credible and diverse GCMs for regional climate
change studies—case study: northeastern United States. Cli-
matic Change, 154, 367–386.
Kelley, M., Schmidt, G.A., Nazarenko, L.S., Bauer, S.E., Ruedy, R.,
Russell, G.L., Ackerman, A.S., Aleinov, I., Bauer, M., Bleck, R.,
Canuto, V., Cesana, G., Cheng, Y., Clune, T.L., Cook, B.I.,
Cruz, C.A., Del Genio, A.D., Elsaesser, G.S., Faluvegi, G.,
Kiang, N.Y., Kim, D., Lacis, A.A., Leboissetier, A., LeGrande, A.
N., Lo, K.K., Marshall, J., Matthews, E.E., McDermid, S.,
Mezuman, K., Miller, R.L., Murray, L.T., Oinas, V., Orbe, C.,
García-Pando, C.P., Perlwitz, J.P., Puma, M.J., Rind, D.,
Romanou, A., Shindell, D.T., Sun, S., Tausnev, N., Tsigaridis, K.,
Tselioudis, G., Weng, E., Wu, J. and Yao, M.S. (2020) GISS-E2.1:
configurations and climatology. Journal of Advances in Modeling
Earth Systems, 12, e2019MS002025.
Knutti, R. (2010) The end of model democracy? Climatic Change,
102, 395–404.
Lanzante, J.R., Dixon, K.W., Nath, M.J., Whitlock, C.E. and Adams-
Smith, D. (2018) Some pitfalls in statistical downscaling of future
climate. Bulletin of the American Meteorological Society,99,
791–803.
Leduc, M., Laprise, R., de Elia, R. and Separovic, L. (2016) Is insti-
tutional democracy a good proxy for model independence?
Journal of Climate, 29, 8301–8316.
Lee, J.Y., Marotzke, J., Bala, G., Cao, L., Corti, S., Dunne, J.P.,
Engelbrecht, F., Fischer, E., Fyfe, J.C., Jones, C., Maycock, A.,
Mutemi, J., Ndiaye, O., Panickal, S. and Zhou, T. (2021) Future
global climate: scenario-based projections and near-term infor-
mation. In: Masson-Delmotte, V., Zhai, P., Pirani, A.,
Connors, S.L., Péan, C., Berger, S., Caud, N., Chen, Y.,
Goldfarb, L., Gomis, M.I., Huang, M., Leitzell, K., Lonnoy, E.,
Matthews, J.B.R., Maycock, T.K., Waterfield, T., Yelekçi, O.,
Yu, R. and Zhou, B. (Eds.) Climate Change 2021: The Physical
Science Basis. Contribution of Working Group I to the Sixth
Assessment Report of the Intergovernmental Panel on Climate
Change. Cambridge and New York, NY: Cambridge University
Press, pp. 1–195.
Liang, Y., Gillett, N.P. and Monahan, A.H. (2020) Climate model pro-
jections of 21st century global warming constrained using the
observed warming trend. Geophysical Research Letters,47,1–10.
Lindvall, J. and Svensson, G. (2015) The diurnal temperature range
in the CMIP5 models. Climate Dynamics, 44, 405–421.
Maraun, D. (2016) Bias correcting climate change simulations—a
critical review. Current Climate Change Reports,2,211–220.
McSweeney, C.F., Jones, R.G., Lee, R.W. and Rowell, D.P. (2014)
Selecting CMIP5 GCMs for downscaling over multiple regions.
Climate Dynamics, 44, 3237–3260.
Meehl, G.A., Senior, C.A., Eyring, V., Flato, G., Lamarque, J.F.,
Stouffer, R.J., Taylor, K.E. and Schlund, M. (2020) Context for
interpreting equilibrium climate sensitivity and transient cli-
mate response from the CMIP6 Earth system models. Science
Advances,6,1–11.
Milinski, S., Maher, N. and Olonscheck, D. (2019) How large does a
large ensemble need to be? Earth System Dynamics, Discussions,
11, 885–901.
Minder, J.R., Letcher, T.W. and Liu, C. (2018) The character and
causes of elevation-dependent warming in high-resolution sim-
ulations of Rocky Mountain climate change. Journal of Climate,
31, 2093–2113.
Müller, W.A., Jungclaus, J.H., Mauritsen, T., Baehr, J., Bittner, M.,
Budich, R., Bunzel, F., Esch, M., Ghosh, R., Haak, H.,
Ilyina, T., Kleine, T., Kornblueh, L., Li, H., Modali, K.,
Notz, D., Pohlmann, H., Roeckner, E., Stemmler, I., Tian, F.
and Marotzke, J. (2018) A higher-resolution version of the Max
Planck Institute earth system model (MPI-ESM1.2-HR). Journal
of Advances in Modeling Earth Systems, 10, 1383–1413.
O'Neill, B.C., Tebaldi, C., Van Vuuren, D.P., Eyring, V.,
Friedlingstein, P., Hurtt, G., Knutti, R., Kriegler, E., Lamarque, J.F.,
Lowe,J.,Meehl,G.A.,Moss,R.,Riahi,K.andSanderson,B.M.
(2016) The scenario model intercomparison project (ScenarioMIP)
for CMIP6. Geoscientific Model Development,9,3461–3482.
Palazzi, E., Mortarini, L., Terzago, S. and von Hardenberg, J. (2019)
Elevation-dependent warming in global climate model simula-
tions at high spatial resolution. Climate Dynamics, 52, 2685–2702.
Pepin, N., Bradley, R.S., Diaz, H.F., Baraer, M., Caceres, E.B.,
Forsythe, N., Fowler, H., Greenwood, G., Hashmi, M.Z., Liu, X.
D., Miller, J.R., Ning, L., Ohmura, A., Palazzi, E., Rangwala, I.,
Schöner, W., Severskiy, I., Shahgedanova, M., Wang, M.B.,
Williamson, S.N. and Yang, D.Q. (2015) Elevation-dependent
warming in mountain regions of the world. Nature Climate
Change,5,424–430.
Pierce, D.W., Barnett, T.P., Santer, B.D. and Gleckler, P.J. (2009)
Selecting global climate models for regional climate change
studies. Proceedings of the National Academy of Sciences of the
United States of America, 106, 8441–8446.
Ribes, A., Qasmi, S. and Gillett, N.P. (2021) Making climate pro-
jections conditional on historical observations. Science
Advances,7,1–10.
Rupp, D.E., Abatzoglou, J.T., Hegewisch, K.C. and Mote, P.W.
(2013) Evaluation of CMIP5 20th century climate simulations
for the Pacific Northwest USA. Journal of Geophysical Research:
Atmospheres, 118, 10884–10906.
Salathé, E.P., Steed, R., Mass, C.F. and Zahn, P.H. (2008) A high-
resolution climate model for the U.S. Pacific northwest: meso-
scale feedbacks and local responses to climate change. Journal
of Climate, 21, 5708–5726.
Schlund, M., Lauer, A., Gentine, P., Sherwood, S.C. and Eyring, V.
(2020) Emergent constraints on equilibrium climate sensitivity
in CMIP5: Do they hold for CMIP6? Earth System Dynamics,
11, 1233–1258.
Schmidt, G.A. (2021) #NotAllModels. RealClimate. Available at:
https://www.realclimate.org/index.php/archives/2021/08/
notallmodels/ [Accessed 4th January 2022].
Séférian, R., Nabat, P., Michou, M., Saint-Martin, D., Voldoire, A.,
Colin, J., Decharme, B., Delire, C., Berthet, S., Chevallier, M.,
Sénési, S., Franchisteguy, L., Vial, J., Mallet, M., Joetzjer, E.,
Geoffroy, O., Guérémy, J.F., Moine, M.P., Msadek, R.,
5890 MAHONY ET AL.

Ribes, A., Rocher, M., Roehrig, R., Salas-y-Mélia, D.,
Sanchez, E., Terray, L., Valcke, S., Waldman, R., Aumont, O.,
Bopp, L., Deshayes, J.,

Ethé, C. and Madec, G. (2019) Evalua-
tion of CNRM earth system model, CNRM-ESM2-1: role of
earth system processes in present-day and future climate. Jour-
nal of Advances in Modeling Earth Systems, 11, 4182–4227.
Sellar, A.A., Jones, C.G., Mulcahy, J.P., Tang, Y., Yool, A.,
Wiltshire, A., O'Connor, F.M., Stringer, M., Hill, R.,
Palmieri, J., Woodward, S., de Mora, L., Kuhlbrodt, T.,
Rumbold, S.T., Kelley, D.I., Ellis, R., Johnson, C.E., Walton, J.,
Abraham, N.L., Andrews, M.B., Andrews, T., Archibald, A.T.,
Berthou, S., Burke, E., Blockley, E., Carslaw, K., Dalvi, M.,
Edwards, J., Folberth, G.A., Gedney, N., Griffiths, P.T.,
Harper, A.B., Hendry, M.A., Hewitt, A.J., Johnson, B.,
Jones, A., Jones, C.D., Keeble, J., Liddicoat, S.,
Morgenstern, O., Parker, R.J., Predoi, V., Robertson, E.,
Siahaan, A., Smith, R.S., Swaminathan, R., Woodhouse, M.T.,
Zeng, G. and Zerroukat, M. (2019) UKESM1: description and
evaluation of the U.K. earth system model. Journal of Advances
in Modeling Earth Systems, 11, 4513–4558.
Sherwood, S., Webb, M.J., Annan, J.D., Armour, K.C., Forster, P.
M., Hargreaves, J.C., Hegerl, G., Klein, S.A., Marvel, K.D.,
Rohling, E.J., Watanabe, M., Andrews, T., Braconnot, P.,
Bretherton, C.S., Foster, G.L., Hausfather, Z., von der Heydt, A.
S., Knutti, R., Mauritsen, T., Norris, J.R., Proistosescu, C.,
Rugenstein, M., Schmidt, G.A., Tokarska, K.B. and Zelinka, M.
D. (2020) An assessment of Earth's climate sensitivity using
multiple lines of evidence. Reviews of Geophysics, 58,
e2019RG000678.
Sutton, R.T. (2018) ESD ideas: a simple proposal to improve the
contribution of IPCC WGI to the assessment and communica-
tion of climate change risks. Earth System Dynamics, 9, 1155–
1158.
Sutton, R.T. and Hawkins, E. (2020) ESD ideas: global climate
response scenarios for IPCC assessments. Earth System Dynam-
ics, 11, 751–754.
Swart, N.C., Cole, J.N.S., Kharin, V.V., Lazare, M., Scinocca, J.F.,
Gillett, N.P., Anstey, J., Arora, V., Christian, J.R., Hanna, S.,
Jiao, Y., Lee, W.G., Majaess, F., Saenko, O.A., Seiler, C.,
Seinen, C., Shao, A., Sigmond, M., Solheim, L., Von Salzen, K.,
Yang, D. and Winter, B. (2019) The Canadian Earth System
Model version 5 (CanESM5.0.3). Geoscientific Model Develop-
ment, 12, 4823–4873.
Tabor, K. and Williams, J.W. (2010) Globally downscaled climate
projections for assessing the conservation impacts of climate
change. Ecological Applications, 20, 554–565.
Tatebe, H., Ogura, T., Nitta, T., Komuro, Y., Ogochi, K.,
Takemura, T., Sudo, K., Sekiguchi, M., Abe, M., Saito, F.,
Chikira, M., Watanabe, S., Mori, M., Hirota, N., Kawatani, Y.,
Mochizuki, T., Yoshimura, K., Takata, K., O'ishi, R.,
Yamazaki, D., Suzuki, T., Kurogi, M., Kataoka, T.,
Watanabe, M. and Kimoto, M. (2018) Description and basic
evaluation of simulated mean state, internal variability, and cli-
mate sensitivity in MIROC6. Geoscientific Model Development,
12, 2727–2765.
Taylor, K.E., Stouffer, R.J. and Meehl, G.A. (2012) An overview of
CMIP5 and the experiment design. Bulletin of the American
Meteorological Society, 93, 485–498.
Volodin, E.M., Mortikov, E.V., Kostrykin, S.V., Galin, V.Y.,
Lykossov, V.N., Gritsun, A.S., Diansky, N.A., Gusev, A.V. and
Iakovlev, N.G. (2017) Simulation of the present-day climate
with the climate model INMCM5. Climate Dynamics, 49, 3715–
3734.
Wang, K. and Clow, G.D. (2020) The diurnal temperature range in
CMIP6 models: climatology, variability, and evolution. Journal
of Climate, 33, 8261–8279.
Wang, T., Hamann, A., Spittlehouse, D. and Carroll, C. (2016)
Locally downscaled and spatially customizable climate data for
historical and future periods for North America. PLoS One, 11,
e0156720.
Wang, T., Hamann, A., Spittlehouse, D.L. and Murdock, T.Q.
(2012) ClimateWNA: high-resolution spatial climate data for
western North America. Journal of Applied Meteorology and
Climatology, 51, 16–29.
Wilby, R.L., Charles, S.P., Zorita, E., Timbal, B., Whetton, P. and
Mearns, L.O. (2004) Guidelines for Use of Climate Scenarios
Developed from Statistical Downscaling Methods. IPCC Task
Group on Data and Scenario Support for Impact and Climate
Analysis (TGICA). Geneva: IPCC.
Wilson, L.J., Vallee, M. and Montpetit, J. (2008) Comments on
“Hydrometeorological accuracy enhancement via post-
processing of numerical weather forecasts in complex terrain”.
Weather and Forecasting, 24, 892–894.
Wu, T., Lu, Y., Fang, Y., Xin, X., Li, L., Li, W., Jie, W., Zhang, J.,
Liu, Y., Zhang, L., Zhang, F., Zhang, Y., Wu, F., Li, J., Chu, M.,
Wang, Z., Shi, X., Liu, X., Wei, M., Huang, A., Zhang, Y. and
Liu, X. (2019) The Beijing Climate Center Climate System
Model (BCC-CSM): the main progress from CMIP5 to CMIP6.
Geoscientific Model Development, 12, 1573–1600.
Yukimoto, S., Kawai, H., Koshiro, T., Oshima, N., Yoshida, K.,
Urakawa, S., Tsujino, H., Deushi, M., Tanaka, T., Hosaka, M.,
Yabu, S., Yoshimura, H., Shindo, E., Mizuta, R., Obata, A.,
Adachi, Y. and Ishii, M. (2019) The meteorological research
institute Earth system model version 2.0, MRI-ESM2.0: descrip-
tion and basic evaluation of the physical component. Journal of
the Meteorological Society of Japan, 97, 931–965.
Ziehn, T., Chamberlain, M.A., Law, R.M., Lenton, A., Bodman, R.
W., Dix, M., Stevens, L., Wang, Y.P. and Srbinovsky, J. (2020)
The Australian Earth System Model: ACCESS-ESM1.5. Journal
of Southern Hemisphere Earth Systems Science, 70, 193–214.
How to cite this article: Mahony, C. R., Wang,
T., Hamann, A., & Cannon, A. J. (2022). A global
climate model ensemble for downscaled monthly
climate normals over North America. International
Journal of Climatology,42(11), 5871–5891. https://
doi.org/10.1002/joc.7566
MAHONY ET AL.5891