Physics provides the fundamental language for describing Earth’s climate system. Every process—from the absorption of sunlight by a leaf to the deep overturning circulation of the ocean—obeys physical laws that can be expressed mathematically. Climate models, which are the primary tools for projecting future conditions, are essentially large collections of these physical equations solved on powerful computers. By grounding predictions in well-tested principles, physicists and climate scientists can simulate past climates, attribute observed changes, and explore a range of possible futures. Without physics, climate projections would be little more than guesswork. With it, policymakers, engineers, and communities can plan for rising seas, shifting rainfall patterns, and more frequent extreme events.

The Physical Principles Behind Climate Models

Modern climate models rest on a handful of core physical principles that govern how energy and matter behave on a planetary scale. These principles are not approximations; they are the same laws that describe the motion of a falling apple or the flow of air over an airplane wing. The difference is scale: climate models apply them across the entire globe, over decades to centuries.

Conservation Laws

The most fundamental constraints come from conservation of mass, momentum, and energy. In climate science, conservation of energy ensures that the net energy entering the Earth system (mostly as shortwave solar radiation) must balance the net energy leaving (as longwave thermal radiation plus reflected sunlight). Any imbalance leads to warming or cooling. This framework allows scientists to calculate the planet’s energy budget with high precision. Similarly, conservation of mass governs the cycling of water, carbon, and other key substances through the atmosphere, oceans, and land. When a model incorrectly predicts a mass imbalance, it flags a flaw in the physics.

Fluid Dynamics and the Navier-Stokes Equations

The atmosphere and oceans are turbulent fluids. Their motion is described by the Navier-Stokes equations, which relate changes in velocity to pressure gradients, rotation (the Coriolis effect), viscosity, and external forces such as gravity. Solving these equations on a grid of points around the globe is the core task of a climate model. Because weather and ocean currents are chaotic, small errors grow over time, which is why climate projections focus on statistics (averages, extremes, frequencies) rather than exact forecasts. Fluid dynamics also explains phenomena like jet streams, ocean gyres, and upwelling—all crucial for regional climate.

Thermodynamics and Phase Changes

Thermodynamics governs how heat moves within the climate system. The first law of thermodynamics (energy conservation) is applied to each grid cell in the model to track temperature changes due to radiation, conduction, convection, and latent heat release. Phase changes of water—evaporation, condensation, freezing, and melting—are particularly important. When water evaporates from the ocean, it absorbs heat; when it condenses into clouds or precipitation, that heat is released. This latent heat fuels hurricanes, thunderstorms, and the large-scale circulation of the atmosphere. Models must simulate these processes to get realistic rainfall patterns and storm intensity.

Radiative Transfer

Radiative transfer describes how electromagnetic radiation—both shortwave (solar) and longwave (thermal)—is absorbed, scattered, and emitted by the atmosphere and surface. The key players are greenhouse gases (CO₂, H₂O, methane, ozone), clouds, and aerosols. Each interacts with radiation at specific wavelengths. For example, CO₂ strongly absorbs certain infrared bands, trapping heat and warming the surface. Models use detailed radiative transfer codes to compute the net radiative heating or cooling at every altitude and latitude. This is the physical underpinning of the greenhouse effect and the primary driver of anthropogenic climate change. The NASA Earth Observatory provides a clear overview of these radiative processes.

How Climate Models Are Built and Run

From General Circulation Models to Earth System Models

The earliest climate models simulated only the atmosphere, treating the ocean as a simple swamp. Today’s Earth System Models (ESMs) couple modules for the atmosphere, ocean, land surface, sea ice, and biogeochemical cycles (e.g., carbon, nitrogen). Each module solves its own set of physics equations and exchanges fluxes (heat, water, momentum, carbon) with the others at every time step. The spatial resolution of these models has improved dramatically: typical grid cells are now roughly 100 km across, and some regional models go down to 1 km. Higher resolution captures features like hurricanes and mountain precipitation more faithfully, but requires exponentially more computing power. The IPCC’s Sixth Assessment Report describes the latest generation of models used for policy-relevant projections.

Parameterization: Physics at the Small Scale

No model can resolve every cloud, eddy, or raindrop. Processes that happen below the grid scale must be “parameterized”—represented by simplified mathematical relationships based on physics. For instance, convection in thunderstorms is parameterized by relating the amount of rainfall to the instability and moisture in a grid column. Cloud fraction, microphysics (ice vs. liquid), and turbulence are also parameterized. This is one of the largest sources of uncertainty in climate projections, especially for clouds and their feedback on warming. Ongoing research in cloud physics, often using field campaigns and large-eddy simulations, feeds back into improving these parameterizations.

Ensembles and Uncertainty Quantification

Because models are imperfect and the climate system is chaotic, scientists never rely on a single simulation. Instead, they run ensembles—multiple simulations with small perturbations to initial conditions, slightly different physics parameters, or different emission scenarios. The spread of results quantifies the uncertainty. For example, by running 40 ensemble members of a model, researchers can estimate the probability that global average temperature will exceed 2°C by 2100 under a given scenario. The Carbon Brief explainer on climate models offers a non-technical walkthrough of ensemble methods.

Predicting Future Climate Scenarios

Emissions Pathways: Shared Socioeconomic Pathways (SSPs)

Future climate depends on human choices about fossil fuel use, land use, and technology. To explore this, the climate research community defined a set of Shared Socioeconomic Pathways (SSPs) that describe different possible futures—from rapid decarbonization (SSP1-1.9) to high fossil-fuel growth (SSP5-8.5). Each pathway provides a narrative and a set of emissions and land-use inputs for models. The models then project the resulting changes in temperature, precipitation, sea level, and extreme events. For example, under the high-emissions scenario (SSP5-8.5), global mean temperature could rise by 4–5°C by 2100 relative to pre-industrial, with catastrophic impacts. Under the most aggressive mitigation scenario (SSP1-1.9), warming may be limited to about 1.5°C.

Key Projections and Their Physical Basis

Physics-based models consistently project several robust changes:

  • Global warming proportional to cumulative CO₂ emissions, largely independent of the exact model physics. This linearity arises from the physics of radiative forcing and the carbon cycle.
  • Sea level rise from thermal expansion (ocean warming causes water to expand) and melting of land ice (glaciers and ice sheets). Models include the physics of ice flow and ocean heat uptake, though ice-sheet dynamics remain a frontier.
  • Increased precipitation extremes because warmer air holds more moisture (Clausius-Clapeyron relation, ~7% per °C). This physical law leads to heavier downpours and flooding in many regions.
  • Shifts in storm tracks and jet streams due to altered temperature gradients between the equator and poles. Physics-based fluid dynamics can reproduce these shifts, though uncertainties remain.

Tipping Points and Abrupt Changes

Some of the most concerning projections involve nonlinear thresholds—tipping points where a small change triggers a large, irreversible shift. Examples include the collapse of the Atlantic Meridional Overturning Circulation (AMOC), the dieback of the Amazon rainforest, or the rapid disintegration of the West Antarctic Ice Sheet. Physics is essential to understanding these risks. For AMOC, the driving force is density-driven ocean convection, which depends on temperature and salinity. Models show that if freshwater input from Greenland ice melt is large enough, deep convection can shut down, altering climate across the Northern Hemisphere. The Potsdam Institute for Climate Impact Research maintains a comprehensive overview of known tipping elements.

Challenges in Physics-Based Climate Modeling

Computational Limits and Resolution

Even with exascale supercomputers, global climate models cannot yet fully resolve clouds, ocean eddies, or turbulent mixing. These sub-grid-scale processes must be parameterized, introducing structural uncertainty. Increasing resolution to ~1 km globally would require a thousand-fold increase in computational power, which may become feasible within a decade but is not yet routine. Meanwhile, scientists use smaller-domain, high-resolution simulations to inform parameterizations, but the gap remains a major challenge.

Cloud Feedbacks

The largest source of uncertainty in climate sensitivity (how much warming per doubling of CO₂) comes from clouds. Low clouds reflect sunlight (cooling effect), while high clouds trap heat (warming effect). How cloud cover, height, and microphysics change in a warmer world depends on complex interplay of convection, turbulence, and radiation. Physics-based models handle these interactions differently, leading to a spread in estimated equilibrium climate sensitivity—ranging from about 2°C to 5°C. Observational constraints and process-level studies are narrowing this range but have not eliminated it.

Chaos and Predictability Limits

The climate system is a chaotic dynamical system, meaning its evolution is sensitive to initial conditions. While this limits weather prediction to about two weeks, it does not prevent long-term climate projections because the statistics (mean and variability) are robust. However, chaotic variability—like El Niño–Southern Oscillation—can obscure the forced climate signal on decadal timescales. Physics-based models must simulate internal variability correctly to separate it from the warming trend. The balance between forced and unforced variability is an active area of research.

Carbon Cycle Feedbacks

Physics overlaps with biogeochemistry when modeling the carbon cycle. As the atmosphere warms, soils release more CO₂ (soil respiration increases exponentially with temperature), and the ocean’s ability to absorb CO₂ declines (due to decreased solubility and slowed overturning). These feedbacks amplify climate change. Earth system models include these processes, but they depend on parameterized biological and chemical responses, which add uncertainty. For instance, the future of the Amazon carbon sink may depend on fire physics and drought stress as much as on climate.

Advances in Physics-Driven Climate Modeling

Machine Learning as a Physics Emulator

Rather than replacing physics, machine learning (ML) is increasingly used to accelerate computationally expensive parts of climate models. For example, deep neural networks can emulate radiative transfer codes, cutting calculation times by orders of magnitude. Other applications include downscaling coarse model output to local scales and improving parameterizations of convection and clouds by learning from high-resolution simulations or observations. These hybrid physics-ML models are still emerging but promise higher fidelity at lower cost.

Higher Resolution with Nested Models

Regional climate models (RCMs) use boundary conditions from global models and apply physics at much finer grid spacing (1–10 km). This captures topography, coastlines, and land-sea contrasts that drive local precipitation, heatwaves, and wind patterns. Nested frameworks allow the global model to provide large-scale physics while the RCM handles regional details. Projections for water resources, agriculture, and infrastructure rely heavily on these physics-based regional models.

Data Assimilation and Initialization

For decadal predictions (5–30 years out), accurate initialization of the ocean and atmosphere is crucial. Data assimilation—a technique from numerical weather prediction—combines model physics with observations (satellites, buoys, radiosondes) to create an optimal estimate of the current state. This physically consistent start state improves forecasts of phenomena like Atlantic hurricane activity or the strength of ocean circulation over the next decade. The physics of how information flows from observations to model variables is rooted in estimation theory and dynamical systems.

Conclusion

Physics is not merely a backdrop for climate change discussions; it is the engine that drives all credible predictions. From the conservation laws that govern the energy budget to the turbulent fluid dynamics of the atmosphere and oceans, every line of code in a climate model is a statement about how the world works. While uncertainties remain—particularly around clouds, ice sheets, and carbon cycle feedbacks—the physics-based framework has proven remarkably robust. As computational power grows and our understanding deepens, models will continue to sharpen, providing clearer guidance for the choices that lie ahead. The future of climate prediction is undeniably a future of physics.