Our Climate Future: What Models Actually Project

What does climate science say about our future?

Earth will keep warming in the near term, but the amount of warming later this century is not fixed. Climate projections combine physical models with alternative emissions pathways, so they produce conditional ranges rather than one forecast. Some responses, including higher heat extremes and continued sea-level rise, are robust. Their eventual scale, many regional rainfall changes, and the human damage they cause still depend on emissions, natural variability, local exposure, and adaptation.

That answer contains two ideas that are often pulled apart. The future is constrained by physics: adding greenhouse gases changes Earth's energy balance. It is also shaped by choices: people determine a large part of the future emissions fed into the models. A projection is neither a calendar of guaranteed disasters nor a reason to assume everything remains negotiable.

What is the difference between a projection and a prediction?

A prediction says what is expected to happen. A climate projection says what models show if a stated set of conditions occurs. The distinction matters because no model can know future energy systems, land use, technology, policy, population, or consumption in advance.

The US Fifth National Climate Assessment's scenarios and datasets appendix is explicit: climate-model output is neither a forecast nor a prediction. It is a conditional projection whose result depends on human choices and on how the climate system responds.

This does not make a projection empty. Conditional statements can be precise and useful. An engineer may not know how many people will cross a bridge in 2050, but can still calculate its response under specified loads. Climate scientists likewise compare the system under specified greenhouse-gas, aerosol and land-use pathways.

Weather forecasts are a different product. They begin with the atmosphere's current state and try to follow its detailed evolution over days. Climate projections ask how the statistics of weather change over decades as external influences change. Our guide to weather versus climate explains why uncertainty about one Tuesday's rain does not erase information about a warmer long-term distribution.

How are climate projections built?

Global climate models divide the atmosphere, ocean, land and ice into three-dimensional cells. Within and between those cells, equations represent conservation of energy and mass, fluid motion, radiation, clouds, the carbon cycle and other processes. Processes too small or complex to calculate directly at the grid scale are approximated using physically informed relationships called parameterizations.

A projection usually involves five layers:

1. A scenario supplies external conditions. It describes a possible path for greenhouse gases, aerosols, land use and other influences.
2. A model calculates the response. The model moves energy, water and carbon through its representation of the Earth system.
3. Repeated runs sample natural variability. Slightly different starting conditions can produce different sequences of El Nino events, circulation patterns and regional weather while sharing the same long-term forcing.
4. Multiple models sample structural differences. Research groups make different defensible choices about resolution and approximated processes.
5. Assessments combine evidence. Model results are evaluated alongside observations, paleoclimate records, process understanding and other lines of evidence.

Models are not accepted because their maps look sophisticated. Scientists test whether they reproduce important features of past and present climate, examine where they fail, compare independent models, and constrain results with observations. The NOAA climate-model primer describes this cycle of physical equations, model comparison and checks against observed conditions.

A model is necessarily a simplified Earth. The relevant question is not whether it is perfect; no useful model is. The questions are whether it represents the process needed for the decision, whether known biases are addressed, and whether conclusions survive across methods and evidence.

Why do projections use emissions scenarios?

Emissions scenarios are structured alternatives, not prophecies and not ranked betting odds. Modern assessments often use Shared Socioeconomic Pathways, or SSPs, paired with different levels of climate forcing. Each pathway supplies a coherent set of inputs with which models can ask: what follows if emissions fall rapidly, remain intermediate, or rise very high?

The scenario labels do not say which world will occur. The Fifth National Climate Assessment notes that its scenarios do not have relative likelihoods assigned. Calling the highest pathway "the prediction" is therefore wrong; calling the lowest pathway a promise is equally wrong.

The Intergovernmental Panel on Climate Change's Sixth Assessment Synthesis Report gives a useful view of the spread. For global surface temperature averaged over 2081-2100 relative to 1850-1900, its assessed best estimates and very likely ranges are:

These are 20-year global averages, not temperatures for one year, one city or one afternoon. The values also share a historical reference period; silently changing the baseline would change every number. The IPCC Synthesis Report's headline findings summarise the central result: continued greenhouse-gas emissions produce additional warming, while deep and sustained reductions slow that warming.

The table contains two different kinds of spread. The difference between rows is largely scenario uncertainty: society can still take different emissions paths. The range within a row includes uncertainty about how strongly the climate responds and how models and evidence represent that response. Neither should be collapsed into a single number with false precision.

What parts of the climate future are most firmly constrained?

Some statements rest on well-understood mechanisms, observations and agreement across projections.

Continued emissions produce further warming. Carbon dioxide accumulates, and cumulative net carbon-dioxide emissions are closely related to global warming. The greenhouse effect is a measured physical mechanism, not an assumption added only when a model is run.

Hot extremes become more frequent and intense as global warming increases. A warmer temperature distribution shifts the conditions in which heat events occur. The change is not identical everywhere, and individual events still have immediate weather causes.

The water cycle changes, but not uniformly. A warmer atmosphere can hold more moisture, supporting heavier precipitation in many circumstances. Circulation, soils, seasons and geography complicate regional rainfall and drought. "Wet places get wetter" is not a universal local rule.

Oceans and ice respond over long timescales. Ocean heating, glacier and ice-sheet loss, and sea-level rise do not stop on the date printed at the right edge of a 2100 chart. Some long-lived responses continue after atmospheric temperature stabilizes, and their rate and eventual scale depend on the emissions path.

Risks rise with each increment of warming. This does not mean every hazard rises by the same percentage. It means a lower warming path generally carries lower climate-related risks than a higher one, all else equal. NASA's current summary of climate effects notes that effects already observed include ice loss, sea-level rise and more intense heat waves, while the severity of future effects depends on future human activity and total carbon-dioxide emissions.

For specific hazards, the level of confidence differs. The evidence for heat is especially strong. Precipitation, drought, wildfire conditions and tropical-cyclone characteristics require narrower statements about the variable and region. That evidence ladder is set out in our guide to climate change and extreme weather.

What is already constrained, and what remains open?

"Locked in" is useful only if its object and timescale are named. It can otherwise turn a physical constraint into a claim that choices no longer matter.

Even where a physical change is unavoidable, harm is not a fixed output of temperature alone. A flood becomes a disaster through the meeting of a hazard with exposed people, infrastructure and unequal capacity to respond. Adaptation can reduce damage; mitigation limits the amount of change that requires adaptation. The distinction is developed in climate mitigation versus adaptation.

Where do the ranges come from?

Four sources of uncertainty deserve separate names.

Scenario uncertainty

Future emissions and land use depend on decisions that have not been made. Scenario uncertainty grows in importance over longer horizons as low- and high-emissions pathways diverge.

Model uncertainty

Models differ in resolution and in how they approximate clouds, aerosols, vegetation, ice and other processes. An ensemble, meaning a collection of model runs, helps reveal where conclusions are robust and where the spread is wider. Models are not simply counted as equally independent votes; assessments may use observations, model performance and process evidence to constrain results.

Internal variability

The climate system varies naturally through processes such as El Nino and La Nina and through chaotic atmospheric and ocean behavior. This variability can temporarily amplify or mask a long-term trend, especially over smaller regions and shorter periods. Its precise sequence is largely unpredictable.

The Fifth National Climate Assessment chapter on Earth-system processes explains that internal variability becomes more important as spatial scale decreases. At regional and multidecadal scales, it can contribute more uncertainty than either the model or the scenario for some variables.

Observational and downscaling uncertainty

Observations used to evaluate models have measurement coverage and error. Translating a coarse global projection to a county, river basin or street adds further methodological choices. Statistical and dynamical downscaling can provide useful local detail, but a finer-looking map is not automatically a more certain map.

The Met Office's explanation of projection uncertainty identifies internal variability, observational uncertainty, emissions uncertainty and modelling uncertainty, and explains why ensembles are used to explore a range rather than advertise one immaculate line.

What do confidence and likelihood mean in an assessment?

In assessment reports, confidence and likelihood are not synonyms. Confidence describes the strength, quality and consistency of the evidence and the degree of agreement. Likelihood is a probabilistic judgment about an outcome.

The Fifth National Climate Assessment's guide to calibrated language gives the numerical convention behind several familiar terms: "very likely" means a probability from 90% to 100%, "likely" means 66% to 100%, and "as likely as not" means 33% to 66%. These intervals overlap because they are communication categories, not adjacent bins in a spreadsheet.

A high-confidence finding need not include a probability. Conversely, a numerical model range does not automatically establish high confidence in every mechanism behind it. Read the term attached to the exact claim, not as a transferable badge for the whole paragraph.

How should you read a climate projection chart?

Before interpreting the shape or color, check seven items:

1. Variable: Is it temperature, rainfall, sea level, soil moisture or a hazard index?
2. Baseline: Relative to which years is change measured?
3. Time window: Is the point one year, a decade or a 20-year average?
4. Geographic scale: Global means are not local forecasts.
5. Scenario: Which emissions or forcing pathway supplies the input?
6. Band: Does shading show model spread, a likely range, confidence interval or something else?
7. Processing: Has the output been bias-adjusted, weighted or downscaled?

Then ask whether the chart shows an absolute value or a change from a reference. Two charts can both be correct while using different baselines, scenarios and averaging periods. Comparing their headline numbers without reconciling those choices manufactures a disagreement that the underlying studies do not have.

For local decisions, use an assessment or climate service designed for that place and variable. A global multimodel mean is appropriate for understanding the direction of planetary warming; it is not a drainage design value or an emergency forecast.

What is the honest bottom line?

Climate models do not offer one photograph of 2100. They show a set of conditional futures bounded by physical understanding, tested against evidence and widened where knowledge or natural variability demands it.

The robust message is not vague: continued emissions add warming; more warming increases multiple hazards; oceans and ice carry long memory. The open part is consequential: emissions choices strongly affect the late-century range, while adaptation and development affect who is exposed and how much damage follows. Uncertainty belongs in the decision, but it is not evidence that every outcome is equally plausible or that action has no effect.