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 report (IPCC 2021) 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 (IPCC 2021).

Emissions scenarios

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
• CMIP5 – Representative Concentration Pathways (RCP)
• CMIP6 – Shared Socioeconomic Pathways (SSP)

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).

Figure A.2

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)

Download Figure

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 Reclamation (2021). 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.