A climate model is a computer program that simulates how energy and matter move through Earth’s atmosphere, oceans, land, and ice. It works by dividing the planet into a three-dimensional grid of cells, then solving physics equations for each cell to project how temperature, precipitation, wind, and other variables change over time. These models are the primary tools scientists use to understand past climate patterns and estimate what future conditions will look like under different scenarios.
How Climate Models Work
At their core, climate models run on the same physics you learned in school: conservation of energy, conservation of mass, and conservation of momentum. The computer calculates how heat, moisture, and motion transfer between grid cells at each time step, typically every few minutes of simulated time. It tracks how sunlight enters the atmosphere, how much gets absorbed or reflected by clouds and surfaces, how ocean currents carry heat from the tropics toward the poles, and how water evaporates, forms clouds, and falls as rain or snow.
The equations governing fluid motion (how air and water flow) are well understood in theory, but they’re continuous, meaning they describe smooth, unbroken changes. Computers can’t solve continuous equations directly, so models approximate them by breaking Earth’s surface into discrete grid cells. Think of it like turning a smooth globe into a mosaic of tiles, then calculating what happens inside each tile and how neighboring tiles interact.
Global climate models divide the planet into grid cells that are typically around 100 to 200 kilometers across. That resolution has improved dramatically over the decades, from roughly 500 km in 1990 to as fine as 70 km in more recent models. Regional climate models zoom in on a smaller area and can push resolution below 10 km, which is useful for projecting conditions in a specific country or watershed. The tradeoff is computing power: doubling a model’s resolution in all three spatial dimensions, plus the necessary time-step adjustments, requires about ten times more processing power.
What Gets Fed Into the Model
Climate models need two types of input. The first is a description of the physical planet: topography, ocean depth, ice sheet extent, vegetation cover, and soil types. The second is a set of “forcings,” the external factors that push the climate toward warming or cooling.
Natural forcings include changes in the sun’s energy output, regular shifts in Earth’s orbital cycle, and large volcanic eruptions that loft reflective particles into the upper atmosphere. Human-caused forcings include emissions of heat-trapping gases like carbon dioxide and methane, along with changes in land use (replacing forests with farmland, for instance, changes how much sunlight the surface reflects). Scientists define standardized future scenarios, known as Representative Concentration Pathways, that specify different levels of greenhouse gas emissions through the year 2100. Running the same model under multiple pathways shows how different emission choices lead to different climate outcomes.
General Circulation Models vs. Earth System Models
The term “climate model” actually covers a range of complexity. General Circulation Models, or GCMs, focus on the physical climate system: atmosphere, oceans, land surface, and sea ice. They simulate how these four components store and move heat and carbon. GCMs are the workhorses of climate science and have been in use since the 1960s.
Earth System Models take things further by adding chemistry, biology, and ecological feedback loops. They simulate how different plant species absorb carbon dioxide at different rates, how ocean plankton populations respond to changing temperatures, how permafrost thawing releases methane, and how farmland and cities alter the energy balance compared to natural vegetation. Because climate shifts affect ecosystems and ecosystems in turn affect climate, these feedback loops matter. A warming Arctic, for example, melts sea ice, exposing dark ocean water that absorbs more heat, which melts more ice. Earth System Models capture these cascading interactions that a purely physical model would miss.
Testing Models Against the Past
Before a climate model is used to project the future, it has to prove it can reproduce the past. Scientists do this through a process called hindcasting: they feed the model known historical forcings (actual volcanic eruptions, measured solar output, recorded greenhouse gas concentrations) and let it simulate decades or centuries that have already happened. They then compare the model’s output against observed temperature records, precipitation data, sea ice extent, and other measurements.
If a model can reproduce the broad patterns of the 20th century, including the warming trend, the temporary cooling after major volcanic eruptions, and regional precipitation shifts, that builds confidence in its ability to project forward. No model is expected to reproduce every weather event, because weather is chaotic at short timescales. What matters is whether the model captures the right statistical patterns: average temperatures, seasonal cycles, and the frequency of extremes.
Why Models Disagree (And Why That’s Useful)
Different research groups around the world build their own climate models, each making slightly different choices about how to approximate processes that are too small or complex to simulate directly. Cloud formation is a classic example: individual clouds are far smaller than a typical grid cell, so modelers use simplified formulas to estimate how clouds form and how much sunlight they block. These approximations vary from one model to another, which is one reason model projections don’t all give the same answer.
Scientists handle this by running large ensembles of models together. The latest coordinated effort, called CMIP6 (Coupled Model Intercomparison Project, phase 6), includes dozens of models from institutions worldwide. Researchers evaluate and rank these models using statistical measures of how well they reproduce observed climate. Top performers in recent assessments include models from French, American, and Australian research centers. The spread across the ensemble gives scientists a range of outcomes rather than a single number, which is a more honest representation of what we know and don’t know.
One key measure of uncertainty is equilibrium climate sensitivity: how much global average temperature would eventually rise if atmospheric carbon dioxide doubled from pre-industrial levels and the climate system fully adjusted. The IPCC’s current best estimate puts this at about 3°C, with a likely range of 2.5 to 4°C. That range reflects genuine uncertainty in how strongly feedback loops like cloud changes and ice loss amplify warming.
What Climate Models Are Actually Used For
Climate models serve several practical purposes beyond academic research. City planners use downscaled model projections to design stormwater systems for future rainfall patterns. Agricultural agencies use them to anticipate shifts in growing seasons and drought risk. Insurance companies use them to price flood and wildfire risk decades out. International climate negotiations rely on model projections to estimate the consequences of different emission reduction targets.
The scenarios matter as much as the models themselves. A model doesn’t predict the future the way a weather forecast does. It answers “if-then” questions: if global emissions follow this pathway, then temperatures, sea levels, and precipitation patterns will likely change by this much. The range of scenarios lets decision-makers see the difference between aggressive emission cuts and business as usual, giving them concrete numbers to weigh against the costs of action or inaction.
Resolution and Computing Power
Running a global climate model is computationally intense. A single simulation might cover hundreds of years of climate across millions of grid cells, solving physics equations at every cell for every time step. This work has traditionally required room-sized supercomputers at facilities like NASA’s Center for Climate Simulation. Recent advances in hardware and software have brought some modeling capability to smaller machines, but the highest-resolution global simulations still push the limits of the world’s fastest computers.
Higher resolution lets models capture smaller-scale features like mountain ranges, coastlines, and individual storm systems more accurately. A 200 km grid cell can’t represent the difference between a rain shadow valley and a wet mountain slope just 50 km apart. Pushing resolution down to 10 or 25 km, as regional models do, reveals local details that matter for real-world planning. The challenge is always the same: finer grids demand exponentially more computing power, so scientists constantly balance the detail they want against the resources available.