Since the 2014 Climate Change in Colorado report, several gridded datasets for
temperature and precipitation have become available that are based on observations
from long-term climate stations, but also apply homogenization methods that account
for changes in observation time, station location, and so forth. Throughout this
report, we use NOAA’s nClimGrid dataset
(Vose et al. 2014),
which includes monthly
temperature and precipitation information on a 4-km latitude/longitude grid across
the contiguous United States, and whose underlying station data and methods are
similar to the NOAA nClimDiv dataset used in the 2014 report. The nClimGrid dataset
is regularly used as an official source for climate monitoring by NOAA and by the
Colorado Climate Center.
All climate datasets have uncertainties and limitations, and to explore these, we
compared the nClimGrid monthly temperature data to a gridded climate dataset
independently developed by the Berkeley Earth project
(Rohde et al. 2013).
Figure A.1 shows that these two datasets provide remark- ably similar estimates of
the temperature change over Colorado during the period 1985-2022. Although differences
exist month-to-month and year-to-year, the two temperature datasets have a
correlation of r = 0.984. This provides confidence that the temperature changes
presented in this report are robust and are not simply an artifact of the choice of dataset.
Annual temperature anomalies (degrees Fahrenheit) for Colorado, with respect to a baseline of 1951-80, for the NOAA nClimGrid and Berkeley Earth datasets. The thick lines show a 5-year running mean.
A.2 Projected temperature, precipitation, and hydrology changes
Climate projections from global climate models (GCMs) - CMIP5 and CMIP6
Projections (i.e., simulations) by global climate models (GCMs) are the foundational
data for assessing the direction and magnitude of physically plausible future climate
changes at global, regional, and local scales. This report uses two sets of climate
model data assembled by the Coupled Model Intercomparison Project (CMIP),
incorporating the efforts of dozens of climate modeling groups around the world. CMIP is
an organized “roundup” of several dozen of the latest generation of climate models
conducted every 7 years or so to support policy-relevant climate assessments as well
as climate research more broadly. (We acknowledge the World Climate Research
Programme (WCRP), which supports and coordinates CMIP, the climate modeling 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 ESGF.)
The CMIP5 (Coupled Model Intercomparison Project, Phase 5) multi-model ensemble was
previously used in the 2014 Climate Change in Colorado report and was used again in
this report. CMIP5 data for Colorado regridded to a common 1-degree grid, but not
downscaled (see section on downscaling below), were obtained through the LLNL GDO-DCP
server (https://gdo-dcp.ucllnl.org/) and used to evaluate statewide temperature and
precipitation change (e.g., Figures 2.5, 2.6, 2.7, 2.12, 2.13). The CMIP5 ensemble used
in this report encompasses 37 projections, one each from 37 models.
In 2020 and 2021, the data from CMIP6 (Coupled Model Intercomparison Project, Phase 6)
were released, representing a new generation of climate models. Because of their
relative newness, CMIP6 climate projections have only recently been added to
public-facing climate portals. Only a handful of datasets of downscaled CMIP6
projections have been produced (as of July 2023), and no watershed-scale CMIP6-based
hydrologic projections for the U.S. have yet been produced.
For this report, the CMIP6 multi-model ensemble was used to supplement and compare
with the CMIP5 projections. CMIP6 data for Colorado regridded to a common 1-degree grid,
but not downscaled, were obtained through the KNMI Climate Explorer
(https://climexp.knmi.nl) and used to evaluate statewide temperature and precipitation
change alongside CMIP5 (e.g., Figures 2.5, 2.6, 2.12). The CMIP6 ensemble used in this
report initially encompasses 37 projections, one each from 37 models, but then was
screened to a final ensemble of 22 projections, one each from 22 models, as detailed below.
GCMs have improved by many measures from one generation to the next, but since CMIP3,
the improvements have diminished, indicating that climate modeling is maturing. While the
CMIP6 models do show general improvements over CMIP5 in reproducing many features of the
climate system and regional climate statistics, the assessed skill of the models by
these benchmarks across the CMIP5 and CMIP6 ensembles show substantial overlap
(e.g., Pierce et al. 2021).
In practical terms, CMIP6 does not make CMIP5 obsolete--in
fact, an issue emerged with CMIP6 models for which the IPCC applied an adjustment that
was not done for previous CMIP ensembles.
When researchers first examined projections across the CMIP6 models, a number of
models showed higher rates of global and regional warming than the upper end of the
CMIP5 models; that is, unexpectedly high climate sensitivity or climate response to a
given increment of greenhouse-gas emissions. Since these models also appear to simulate
excessive warming in recent decades (~1980-present), it is plausible that these “hot”
models’ estimates of future warming are unrealistically high. Accordingly, the IPCC AR6
deemphasized the “hot” CMIP6 model projections, using additional
modeling and analysis to develop an “assessed” range of future global temperatures that
ended up very close to what had come directly from the CMIP5 model ensemble, given a
similar emissions scenario.
Since then, a simpler method has been proposed for screening CMIP6 models to
deemphasize the hot CMIP6 models in projecting future warming at regional scales
(Hausfather et al. 2022).
That method was used for this report to screen the CMIP6 ensemble from 37 models down
to 22 models. If the same method were applied to the CMIP5 ensemble, none of the
models would be screened out.
For Colorado, after the 12 “hot” CMIP6 models are screened out (along with 3 other
models that are too “cold”, according to the screening criteria) CMIP6 still shows
greater warming than CMIP5 for the same emissions increment, though the two ensembles
mostly overlap (Figures 2.6 and 2.7). The reduced range across the CMIP6 temperature
projections relative to CMIP5 primarily results from the screening of CMIP6 and the
resulting smaller ensemble. In any case, there is much more difference among the
models within each CMIP ensemble, than there is between the two CMIPs. The screening
of CMIP6 models has virtually no impact on the projections of precipitation change as
shown in Figure 2.13. Note that there is still considerable discussion within the
climate science community regarding for what applications one should screen out or
otherwise deemphasize the hot models in CMIP6
(Rahimpour Asenjan et al. 2023).
For this report, on balance, we believed it was appropriate to screen out hot models,
consistent with the latest global-scale climate assessment
A major uncertainty in how climate change will unfold in the coming decades stems from
society, not the climate system: How annual global emissions of greenhouse gases, and
thus their atmospheric concentrations, will change in the future. For the CMIPs and
the IPCC reports, the climate modeling community has collectively adopted sets of
assumptions, known as emissions scenarios, whose broad range is intended to capture
this uncertainty (Figure A.2). For the most recent three CMIPs and IPCC report cycles,
three sets of emissions scenarios have been used:
CMIP3 – Special Report on Emissions Scenarios (SRES) scenarios
For CMIP5 (RCP) and CMIP6 (SSP), each of the scenarios is tagged with a number
(e.g., 2.6, 3.4, 4.5, 6.0, 7.0, 8.5) that represents the total radiative forcing in
watts per square meter (W/m2), the extra energy that will be trapped in the climate
system under that scenario, beyond pre-industrial levels.
The 2014 Climate Change in Colorado report focused on outcomes under the
medium-low RCP4.5 emissions scenario, while also reporting selected results under the
high-end RCP8.5 scenario. Since 2010, the year-on-year increase in global fossil-fuel
CO2 emissions—and thus total anthropogenic CO2 emissions—has slowed such that the
trajectory of those emissions through 2022 is on track with the RCP4.5 scenario,
and about 20% below what the RCP8.5 scenario assumes for 2022
(Global Carbon Project 2022).
Fossil-fuel CO2 emissions currently represent about 90% of all CO2 emissions from
human activities, and about 70% of all anthropogenic greenhouse gas emissions.
The current emissions policies enacted by the major emitting countries indicate a
path of global fossil-fuel and total anthropogenic CO2 emissions through 2050 that is
more consistent with the RCP4.5 trajectory, and well below the RCP8.5 trajectory
(Figure A.2; Hausfather and Peters 2020).
While current trends are encouraging,
emitting countries may reverse policies or fail to meet targets. It is also possible
that the total emis- sions of greenhouse gases through mid-century would end up being
closer to RCP8.5 even if fossil-fuel emissions track RCP4.5, if unexpectedly large
carbon-cycle feedbacks occur, e.g., releases of methane from permafrost
(Schwalm et al. 2020).
Annual total anthropogenic CO2 emissions--about 90% of which are from fossil fuel burning--assumed in the emis- sions scenarios used to drive climate model projections in the CMIP5 and CMIP6 ensembles. The black line shows estimated actual annual CO2 emissions through 2021. This report focuses on projections driven by the RCP4.5 scenario (thick dashed orange) and similar SSP2-4.5 scenario (thick solid orange). (Data: IIASA RCP Database v2.0.5; IIASA SSP Database v2.0; Global Carbon Project)
As in the 2014 report, we again focus here on RCP4.5, and also SSP2-4.5, the comparable
scenario used for the CMIP6 projections; both scenarios are approximately in line with
the upper end of combined national pledges under the 2015 Paris Agreement (green box in
Figure A.2; IPCC 2021).
While this focus on 4.5 scenarios excludes an assessment of
high-end warming outcomes seen only under RCP8.5 and its CMIP6 analog SSP5-8.5,
Figures 2.6 and 2.7 showed that under the 4.5 emissions scenarios, there is still a
wide range of projected warming outcomes by 2050, overlapping considerably with the
range of projected warming under 8.5 scenarios. By 2070, the warming ranges under the
4.5 and 8.5 scenarios overlap less. Both of the full CMIP5 and CMIP6 datasets include
projections run under scenarios that are between 4.5 and 8.5 in terms of warming
outcomes (e.g., RCP6.0, SSP4-6.0, SSP3-7.0), and these should also be considered for
use in future climate vulnerability assessments. The limited availability of
downscaled projections under RCP6.0 meant that scenario was not used in this report.
Downscaled climate projections from GCMs
For use at spatial scales smaller than the state of Colorado, GCM output needs to be
downscaled through statistical methods (statistical downscaling), or via higher-resolution
regional climate models (RCMs; dynamical downscaling), in order to better represent
localized changes to weather and climate and to facilitate hydrologic modeling. For
this report, we used the CMIP5-LOCA (LOcalized Constructed Analogs) dataset developed by
Pierce at al. (2014).
Projections from 32 CMIP5 models were statistically downscaling using the LOCA method,
in which multiple daily weather patterns from the historical record are selected,
adjusted, and blended in order to create fine-scale outputs that are consistent with the
coarser-scale weather pattern shown for a given day in the raw GCM output. In this way,
a long-term climate projection is built that is faithful to the way weather and
climate vary (at least historically) across space and time at local scales.
The CMIP5-LOCA dataset was chosen among several options, including the CMIP5-BCSD
(Bias-Correction Spatial Disaggregation) dataset that was used in the 2014 Report.
The BCSD method has since been shown to have a statistical artifact that alters the
GCM-projected precipitation change, causing “wettening” over the Interior West.
Evaluations of downscaling methods have shown that LOCA imposes fewer alterations of the
coarse-scale GCM change signals
(Alder and Hostetler 2019)
while increasing the level
of local detail in a physically meaningful way based on past weather patterns.
The CMIP5-LOCA projections are a 1/16-degree (~6 km) grid, at a daily timestep. The
full CMIP5-LOCA dataset includes 32 projections, one from each of 32 climate models,
for each of two emissions scenarios, RCP4.5 and RCP8.5 (so 64 projections total).
This report only analyzes the projections under RCP4.5, as discussed above.
Watershed-scale hydrology projections
To generate projections of future hydrology for basins in Colorado and elsewhere,
researchers typically take the downscaled future temperature changes and precipitation
changes projected by an ensemble of climate models and then run that set of plausible
trajectories of future climate through a separate watershed-scale hydrologic model,
such as VIC or Noah. That hydrologic model then simulates the changes in snowpack,
streamflow, soil moisture, and other variables associated with each climate model’s
projection of future climate: temperature change and precipitation change.
For this report, we used the set of CMIP5 global climate model (GCM) projections
that were downscaled using the LOCA method and then run through the VIC (Variable
Infiltration Capacity) hydrologic model. These hydrologic projections (CMIP5-LOCA-VIC)
were created by NCAR researchers for a consortium led by the Bureau of Reclamation
(Vano et al. 2020)
and were previously analyzed for some basins in Colorado in
Lukas et al. (2020) and
These projections were obtained through the GDO-DCP server (https://gdo-dcp.ucllnl.org/)
The CMIP5-LOCA-VIC projections are on a 1/16-degree (~6 km) grid, the same resolution
as the underlying CMIP5-LOCA projections, at a daily timestep.
Hirsch, R. M., and K. R. Ryberg, 2012: Has the magnitude of floods across the USA changed with global CO 2 levels? Hydrological Sciences Journal, 57, 1–9, https://doi.org/10.1080/02626667.2011.62 1895.
Hoerling, M., J. Eischeid, J. Perlwitz, X.-W. Quan, K. Wolter, and L. Cheng, 2016: Characterizing Recent Trends in U.S. Heavy Precipitation. Journal of Climate, 29, 2313–2332, https://doi.org/10.1175/ JCLI-D-15-0441.1.
Hoerling, M. P., J. Eischeid, and J. Perlwitz, 2010: Regional precipitation trends: distinguishing natural vari- ability from anthropogenic forcing. Journal of Climate, 23, 2131–2145, https://doi.org/10.1175/ 2009JCLI3420.1.
Hoerling, M. P., J. J. Barsugli, B. Livneh, J. Eischeid, X. Quan, and A. Badger, 2019: Causes for the Century-Long Decline in Colorado River Flow. J. Climate, JCLI-D-19-0207.1, https://doi.org/10.1175/JCLI-D-19-0207.1.
Holden, Z. A., and Coauthors, 2018: Decreasing fire season precipitation increased recent western US for- est wildfire activity. Proc. Natl. Acad. Sci. U.S.A., 115, https://doi.org/10.1073/pnas.1802316115.