Back to Climate Change Portal

Additional Information

On this page
Real-World Examples Glossary Primer: Climate Models and Projections

Brief real-world examples of specific needs

To illustrate how specific needs lead to different types of climate change information, we illustrate four brief real-world examples of information needed by communities in the Mountain West.

Example 1. Public outreach

Need: Outreach to the public about overall climate adaptation challenges

Basic question: How much warmer will it be, overall?      

Specific information needed: The likely increase in average temperature for 2050 for the city

Example 2. County health authority 

Need: Long-range planning by the county health authority to plan for anticipated increases in heat-related illness

Basic question: How will extreme heat change? 

Specific information needed: The increase in the number of extremely hot (>95°F) days in the county for three future time periods    

Example 3. Water planning

Need: Computer modeling of a water district’s system under future conditions of demand and supply, to support a new integrated water resources plan

Basic question: How will streamflows change? 

Specific information needed: Multiple time-series of annual streamflow change factors for the next 50 years for a local watershed 

Example 4. Stormwater infrastructure

Need: General guidance for a stormwater utility’s planning and operations, including the possibility of up-sizing culverts when they are replaced

Basic question: How will extreme precipitation change? 

Specific information needed: The change in the number of days with over 1″ of precipitation (note that information about precipitation extremes is only available from a few portals, and these data are more uncertain than other climate-change data) 

Obtaining the information listed in these four examples required going to different portals and resources. This guide will show you where to look.

A Brief Glossary

When you navigate climate change information, you enter a dense thicket of technical terms and acronyms. Here is a brief introduction to the most important ones; you can find detailed descriptions and further guidance in the Primer on climate models and projections. For a much more comprehensive glossary of climate-related terms, see the IPCC AR6 WG1 (2021) Glossary). 

Climate model – A complex math- and computer-based simulator of the climate system (atmosphere, oceans, ice sheets, land surface) that uses both fundamental physical laws and observed relationships to model the evolution of climate over time and space. Climate models that also include biogeochemical cycling (e.g., carbon cycle) are also referred to as Earth system models, or ESMs.  

CMIP (Coupled Model Intercomparison Project) – A periodic international “roundup” of the available GCMs and climate projections run under standardized emissions scenarios and conditions. Nearly all climate projection data available on climate portals were initially generated as part of a CMIP. 

CMIP5 – The roundup conducted in 2011-12 of the then-current generation of about 50 GCMs, as featured in the IPCC AR5 reports (2013-14). Even though the CMIP6 climate projections are now broadly available, the CMIP5 climate projections are still appropriate for use in climate assessments and other analyses (see the CMIP6 FAQ for more details on CMIP5 vs. CMIP6).

CMIP6 – The roundup conducted in 2019-20 of the latest generation of about 60 GCMs, as featured in the IPCC AR6 Working Group I report (2021). In the past few years, CMIP6 climate projections have become increasingly available through climate change portals (see the CMIP6 FAQ for more details on CMIP6).

Downscaling – A procedure by which the data from a GCM projection is translated to finer spatial resolution to make it more usable for local and regional analysis and decision-making. LOCA, MACA, BCSD, and BCCA are examples of different types of downscaling procedures and their associated datasets.

Emissions scenario – A potential trajectory of greenhouse gas emissions and concentrations over the next century given particular societal choices; a simulation (projection) of the future by a climate model is driven by a particular emissions scenario (see RCP, SSP)

Ensemble – A group of model simulations of historical or future climate conditions. Most commonly this refers to multi-model ensembles (e.g., from CMIP6) with one or more simulations made by each of several models. The spread of results across a multi-model ensemble can provide an estimate of model uncertainty. Ensembles can also be made with one model using different initial conditions; such single-model ensembles can characterize the uncertainty associated with natural (internal) climate variability. 

ESM (Earth system model) – See climate model

Forcing – An external driver of the climate system; for example, a change in the concentration of CO2 or change in radiation from the sun; can also refer to the net effect of all external drivers, such as in the RCPs and SSPs (e.g., 4.5 W/m2  at 2100). 

GCM (global climate model) – See climate model

Greenhouse gases (GHGs) – Gaseous constituents of the atmosphere, both natural and anthropogenic, that absorb and emit radiation at specific wavelengths within the spectrum of terrestrial radiation emitted by the Earth’s surface, the atmosphere itself, and by clouds. This absorption and emission of energy causes the greenhouse effect. Water vapor (H2O), carbon dioxide (CO2), nitrous oxide (N2O), methane (CH4) and ozone (O3) are the primary GHGs in the Earth’s atmosphere. There are also many entirely human-made GHGs in the atmosphere, such as hydrofluorocarbons (HFCs) and perfluorocarbons (PFCs) and other chlorine- and bromine-containing substances. (IPCC 2021) 

Initial conditions – The starting values of the different variables in a climate model simulation. In a weather model forecast, the model’s initial conditions are derived from current observations. For climate models, the initial conditions for future projections come from time-slices chosen from control runs, or from the end of historical simulations. 

Parameter – A quantitative term in a climate model, derived from observations or other modeling, used to represent a process that cannot be explicitly resolved at the spatial or temporal resolution of the model (i.e., subgrid-scale processes) using the model’s physical equations.

Projection – One simulation of future climate from a single GCM, which assumes a particular future emissions scenario. Because the simulation is conditional on that scenario, technically it’s not a prediction or a forecast. 

RCP (Representative Concentration Pathway) – One of a set of emissions scenarios associated with a specified “climate forcing” (excess energy retained in the Earth system), used to drive the simulations from CMIP5 climate models.

Run – See Projection

SSP (Shared Socioeconomic Pathway) – A broad scenario of future population, policy, economic growth, and technology, that in conjunction with a specific emissions trajectory (see RCP) was used to drive the simulations from CMIP6 climate models.

Primer on climate models and projections

Global climate models (GCMs) 

Global climate models (General circulation models–GCMs or Earth system models–ESMs) are complex, computer-based, mathematical representations of the Earth’s climate system based on fundamental scientific principles. They are designed to capture the dynamics and interactions of the main components of the climate system: atmosphere, oceans, land surface and vegetation, sea ice, land ice. GCMs provide realistic simulations of the key physical phenomena at global down to continental scales:

GCM simulations are not as realistic at finer scales (the size of a Western U.S. state or smaller). This is true especially for precipitation–which is harder to model than temperature, at any spatial scale. 

The latest generation of GCMs includes over 60 models that have been developed by about 30 modeling centers in 13 countries. These GCMs are not fully independent of each other; the modeling centers share model code and parameters for many climate processes, and several modeling centers maintain multiple, related GCMs.

Climate model projections

A simulation of future climate from a GCM is called a projection, rather than a prediction or forecast, because it is conditional on a particular set of assumptions about future greenhouse gases and other climate forcings, i.e., an emissions scenario). For any given climate variable, a future projection will show both (1) simulated climate variability (“internal” or “unforced” variability) and (2) a forced climate change over time, if that variable is affected by changes in global climate “forcing”–most importantly, the higher forcing from increasing greenhouse gases (as depicted in different emissions scenarios, described below).

CMIP5 and CMIP6

What are they?

The Coupled Model Intercomparison Project (CMIP) is an organized “roundup” of the latest GCMs conducted every 7 years or so. Nearly all GCM output used in global, national, and state climate assessments since 2013 has come from CMIP5, whose data was released in 2011-12. Compared to the previous CMIP (CMIP3—there was no CMIP4), CMIP5 included more participating GCMs—models whose spatial resolution was generally higher and captured more climate-system components and processes. 

In 2020, the data from CMIP6 was released, representing a new generation of climate models. CMIP6 was timed to support the IPCC Sixth Assessment Report (AR6), just as CMIP5 supported the IPCC AR5 report. CMIP6 climate projections are now more prevalent in public-facing climate portals than they were a few years ago. That said, there are still fewer downscaled datasets available for CMIP6 projections than for CMIP5 projections shown in most climate portals. At this time (July 2025) many climate portals only show or provide CMIP5 data, though this will continue to change. 

Are the CMIP6 models and projections better than CMIP5? 

Yes and no. As climate modeling continues to develop and mature, the overall improvement represented by a new CMIP has become smaller. On average, CMIP6 models are better than CMIP5 models according to several measures including model resolution and skill in replicating historical climate, but not by so much as to make CMIP5 obsolete. Depending on the specific use case, there may be compelling reasons to use projections from CMIP6 or CMIP5 (see CMIP6 FAQs for more details).

Do the CMIP6 projections show a different picture of climate change than CMIP5?

Yes and no. The ensembles of CMIP6 and CMIP5 climate projections, under comparable emissions scenarios, show very similar average global spatial patterns of temperature and precipitation change. The places that CMIP6 indicates will warm faster or slower, or be wetter, or be drier, are the same places that CMIP5 would change in those ways. 

However, about one-third of the CMIP6 models show greater global and regional warming than the upper end of the CMIP5 models–that is, they have greater climate sensitivity or climate response to the same increment of greenhouse-gas emissions and forcing. Climate scientists now believe that these “hot” models are overestimating future warming, since they are also simulating too much warming in recent decades (~1980-present). Accordingly, the IPCC AR6 report effectively deemphasized the “hot” CMIP6 model projections, using additional modeling and analysis to develop an “assessed” range global temperature projection that was comparable to CMIP5 in the amount of future warming, given a comparable emissions scenario. Since then, simpler techniques have been proposed for deemphasizing the hot CMIP6 models for climate assessment at regional scales.

For the Mountain West, once the hot models are accounted for, CMIP6 still shows slightly greater warming than CMIP5 for the same emissions increment, though the two ensembles mostly overlap. In other words, there are larger differences among the models within each CMIP than between the two CMIPs. It is similar for precipitation: the CMIP6 ensemble average annual precipitation is slightly wetter than for CMIP5, but both CMIPs depict broad ranges of plausible precipitation futures that mostly overlap. The uncertainty (range of projected futures) for both temperature and precipitation is similar between CMIP6 and CMIP5. 

For more detailed overviews of CMIP6 and how it differs from CMIP5, see these two resources:

Emissions Scenarios: RCPs and SSPs

Since anthropogenic greenhouse gases are the primary cause of recent global warming and other climate changes, it is necessary for future climate simulations from GCMs to have inputs that describe how greenhouse gas emissions and concentrations will unfold over the next century. The climate modeling community has adopted a set of trajectories whose range is intended to capture the significant uncertainty in how annual emissions and concentrations of these gases, as well as smaller climate forcings (like aerosol emissions and changes in surface reflectivity), will change in the future.

The trajectories developed for CMIP5 are called Representative Concentration Pathways (RCPs). RCPs provide trajectories of GHG emissions and concentrations that are linked to plausible future trends in demographic, socioeconomic, technological, and political factors. Since those underlying trends cannot be predicted with any confidence, no likelihoods were formally assigned to these RCPs. (That said, it is hard to avoid the question of which scenarios are more plausible than others given current trends, which we address below.)

Each CMIP5 GCM simulation or projection uses one of the four RCPs: RCP2.6, RCP4.5, RCP 6.0, or RCP8.5. The numbers refer to the strength of the global radiative forcing in 2100, in watts per square meter (W/m2)—the extra energy trapped in the climate system by added greenhouse gases and other human-caused changes—above and beyond pre-industrial levels. The divergence in forcing among the four RCPs at mid-century is much smaller than later in the century. The projected increase in global average temperature by 2100 for any given GCM closely corresponds to the radiative forcing of each RCP.

To support CMIP6, further effort went into articulating the socioeconomic assumptions that are associated with emissions pathways; these assumptions were distilled into five “Shared Socioeconomic Pathways” (SSPs), from SSP1 (assumptions generally associated with lower emissions and concentrations of greenhouse gases) to SSP5 (assumptions generally associated with higher emissions and concentrations). Within each SSP, multiple emissions and concentration pathways with specific climate forcings (i.e., RCPs) are possible, so these need to be specified as well. 

Thus, each of the scenarios used to drive CMIP6 modeling is essentially a combination of an SSP and an RCP, but is simply called an “SSP” for convenience. Four of these SSPs– SSP1-2.6, SSP2-4.5, SSP4-6.0, and SSP5-8.5–are updates of their CMIP5 counterparts (RCP2.6, RCP4.5, RCP6.0, and RCP8.5), with the same radiative forcing levels in 2100, though the exact forcing trajectory during the 21st century may differ somewhat. Four new SSPs were introduced for CMIP: SSP1-1.9, SSP4-3.4,  SSP5-3.4OS [OS = “overshoot”], and SSP3-7.0, that fill in some gaps between and below the CMIP5 RCPs. 

For more details about the assumptions behind the RCPs and SSPs, and guidance on which RCPs and SSPs to use, see Q4 and Q5 in the CMIP6 FAQ.

For more about how the SSPs were developed, and how they dovetailed with RCPs to support the CMIP6 modeling, see this Carbon Brief Explainer on SSPs.

Downscaling and other processing of GCM output

The “raw” output from GCMs provides the best estimates of future climate changes at the global down to sub-continental scales. But for informing state, local, and watershed-scale decision-making, GCM output has some real limitations. The spatial resolution of the data, with grid cells typically from 100 km to 200 km on a side, is too coarse to capture complex terrain in the Interior West and elsewhere and most of its effects on climate, or to be used as inputs for regional climate-impact modeling (e.g., watershed hydrology modeling, ecological modeling). Raw GCM output also doesn’t capture certain features of weather and climate very well, such as the statistics of daily precipitation. 

GCM output can be downscaled through statistical methods (statistical downscaling) or via higher-resolution regional climate models (RCMs; dynamical downscaling). To better represent localized changes to weather and climate, and to facilitate impact modeling, all of the climate assessments described in this guide, as well as most of the climate portals, rely on downscaled projections. 

Downscaling is not magic; downscaled projections may be a lot more precise (locally specific), but are not that much more accurate, compared to the underlying raw GCM data. Downscaling also does not reduce the uncertainty that manifests in the somewhat different depictions of future change seen across the various GCM projections. In fact, downscaling introduces a new uncertainty, stemming from the choice of downscaling method.

Statistical downscaling: LOCA, MACA, BCCA, BCSD, etc. 

The vast majority of publicly available downscaled data were produced using statistical downscaling–an approach which is much less computationally intensive, allowing large ensembles of projections under multiple emissions scenarios to be downscaled. In the past several years, two statistical downscaling methods and their associated datasets have seen widespread use in assessments and portals: LOCA (Localized Constructed Analogs) and MACA (Multivariate Adaptive Constructed Analogs). These are both “constructed analog” (CA) methods, 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. BCCA (Bias-Corrected Constructed Analogs) is another such method that is used in some studies and at least one portal. 

The BCSD (Bias-Corrected Spatial Disaggregation) statistical method has seen a lot of use in past CMIP5-based studies and assessments, and BCSD-downscaled CMIP5 data are available through at least one portal in this guide, the GDO-DCHP. BCSD downscaling adjusts the statistical distributions of coarse-scale projected climate values (daily or monthly temperature or precipitation) to match the statistical distributions seen in the fine-scale historical climate data, a process called quantile mapping (QM). The QM step has been shown to alter the future precipitation trends seen in the raw CMIP5 GCM projections, such that the Mountain West is tilted towards higher precipitation in the BCSD projections, compared to the raw projections. This alteration is now believed to be an artifact of the method, and so it is not reliable. Users should either avoid using BCSD data, or compare BCSD projections with other downscaled datasets. (Note that the variant of BCSD used in NASA’s NEX-DCP30 CMIP5 dataset and NEX-DCCP-CMIP6 dataset does not have this issue.)

Dynamical downscaling: CORDEX

Dynamically downscaled climate projections use regional climate models run while “nested” within the future conditions shown by the global models, which imparts more physical-process knowledge to the downscaled data than does statistical downscaling. The only such CMIP5-based dataset readily available is from the CORDEX project, which is available via two portals in this guide, the IPCC WG1 Interactive Atlas and the Copernicus Interactive Climate Atlas, although both have limited options for visualization. The CORDEX CMIP5 projections could be useful for comparing to LOCA-, MACA-, or BCCA-downscaled CMIP5 to see if there are consistent differences in the average climate changes for the location or region of interest. 

Regridded raw GCM data

The other approach to processing the raw GCM projections is not to downscale them at all; instead, to replot the raw data from the different GCMs to a common grid (regridding), which puts them all at the same spatial resolution (typically around 1 degree, or 100 km). Several portals and assessments depict regridded raw CMIP5 data. The caution is that the projected future climate variables are not bias-corrected (shifted so that the GCM-simulated recent values match the observations) as with downscaled data. There are large biases in precipitation for the Mountain West in precipitation; all of the GCMs show much more precipitation in recent decades than actually occurred. However, when the future projections are shown as anomalies (projected future period minus simulated recent period), the subtraction process has the effect of bias-correcting the projected future precipitation or temperature. 

Climate variables 

Virtually all portals provide information about future levels of “the basics” for a given area or location:

Other climate variables, available from many of the portals, include:

These variables can be expressed in one of two ways:

  1. Value – The magnitude of the historical or future climate variable, e.g., 72°F 
  2. Anomaly – The difference between the historical value and the future value, e.g., +4°F (Note that the precipitation anomaly is often expressed as a % difference)

The anomaly succinctly captures the future change in climate in a single number, e.g., 4 additional days above 100°F by 2050. But in this and other cases, also knowing the historical and future values, e.g., 1.5 days/year above 100°F (historical) vs. 5.5 days/year (2050), helps put the change into better context. 

At all of the portals described in this guide, the projected future climate data are available as values, but not all portals provide the anomalies too. 

Dealing with uncertainty in future climate projections

An inescapable characteristic of GCM projections of future climate is that the different models’ projections show different future changes in temperature, precipitation, and other climate variables for a given region or location. Thus, the CMIP5 and CMIP6 ensembles of GCMs, which each include dozens of models run under multiple emissions scenarios, show broad ranges of plausible future climate conditions. In the time-series plots and other visualizations, this range is typically shown by a shaded area around a central line (the mean or median), as in Figure 1.

Figure 1. Example of a time-series plot of projected climate (June mean temperature) displayed by the USGS National Climate Change Viewer, in which the red/blue shading shows the middle 80% of the range of model projections under that emissions scenario (RCP), from the 10th percentile to the 90th percentile. 

Where does this uncertainty come from, and how do we deal with it in interpreting and using the information we get from climate portals?

The first source of uncertainty is the most obvious, already discussed above: Our uncertainty in future greenhouse gas emissions and the resulting anthropogenic forcing, or “push” on the climate system. Emissions scenarios (i.e., RCPs, SSPs) are deliberately constructed to represent a range of plausible anthropogenic influences on climate, and the portals always allow you to distinguish between sets of projections driven by different RCPs or SSPs, both in selecting the data, and in the visualizations like the one above, where RCP4.5 projections are shown in blue, and RCP8.5 projections in red. 

For temperature, the effect of the emissions scenario is predictable and consistent: a higher-emissions pathway (e.g., RCP 8.5 vs. RCP 4.5) leads to yet warmer projected temperatures, both globally and regionally. For projected precipitation change, the influence of the RCP is more variable and may be difficult to discern.

But if we look at a set of projections driven by a single emissions scenario, we still find a large range in projected temperature and precipitation changes, as shown by either the blue shading (the range under RCP4.5)  or the red shading (the range under RCP8.5) in the visualization. So we need to understand the other sources of uncertainty:  

The different climate models collected in the CMIP archives have different approaches to representing, in the model code, several key climate processes and patterns. This diversity reflects that our observations and understanding are not complete enough to have confidence in a single methodology for including these processes in climate models, and also that the coarse resolution of the climate models inhibits more accurate simulation. Thus, different modeling groups use different approaches to represent these processes–which results in somewhat different answers to the same question, like “How much global warming will occur for a given increase in greenhouse gas concentrations in the atmosphere?”

Climate models do not attempt to replicate the actual events and sequences of historical climate; instead, they generate a simulated climate history that captures the key characteristics of historic natural variability. As a model is run into the future, that climate projection likewise includes a sequence of natural climate variability that, because of the randomness inherent in the climate system (and the models), does not match the sequences produced by other models. Because natural variability has annual, decadal, and multidecadal components, analyses of projected change will inevitably include some amount of natural variability–which contributes to the spread of the models. In other words, the breadth of the blue and red shading is due to both structural differences between models and the randomness (natural variability) in climate simulated by each model. 

This last factor comes into play if the climate projection is downscaled. The downscaling procedure, whether statistical or dynamical, can shift the future change in temperature and/or precipitation from that shown in the underlying global climate model output. The direction and amount of this shift differs with the climate variable in question, and the region being downscaled.

The importance of the factors will vary by both spatial scale and the future time frame of the projection. For temperature change for the Mountain West by the mid-21st century, the emissions scenario and the GCM’s representation of key processes tend to be most important. For precipitation change, the emissions scenario is less important, and the GCM’s representation of key processes and simulation of natural variability are more important.

This also brings up another important point about uncertainty in climate projections:

We are extremely confident that the climate will continue to warm considerably, though the magnitude of the warming is less certain. For precipitation, in many places, including most parts of the Mountain West, the direction of future precipitation change is highly uncertain, as well as the magnitude of that change. Simulating temperature change at regional to global scales is a much easier task than simulating the manifold and complex processes that influence precipitation, and thus precipitation change.  

So what do we do with this uncertainty–that is, the blue and red shading on the graph? Can we ignore the shading, and just use the dark lines in the middle (the median or average of the model projections) as our “answer”? That’s not a wise idea, for two reasons:

Embracing a range of future climates rather than one (central) future climate does make things more complicated for both analysis and communication, but embracing that range will also lead to more robust planning that better accounts for future uncertainty.