SoylentNews

Last Post writes:

Climate models are numerical simulations of the climate system [usda.gov], which are used to for predicting climate change from emissions scenarios and many other applications. Let's take a closer look at how they work.

Why do we need climate models?

The climate system and its components like the atmosphere and hydrosphere are driven by many processes that interact with each other to produce the climate we observe. We understand some of these processes very well, such as many atmospheric circulations that drive the weather. Others like the role of aerosols and deep ocean circulations are more uncertain. Even when individual processes are well understood, the interaction between these processes makes the system more complex and produces emergent properties like feedbacks and tipping points [nasa.gov]. We can't rely on simple extrapolation to generate accurate predictions, which is why we need models to simulate the dynamics of the climate system. Global climate models simulate the entire planet at a coarse resolution. Regional climate models simulate smaller areas at a higher resolution, relying on global climate models for their initial and lateral boundary conditions (the edges of the domain).

How do climate models work?

A climate model is a combination of several components, each of which typically simulates one aspect of the climate system such as the atmosphere, hydrosphere, cryosphere, lithosphere, or biosphere. These components are coupled together [nature.com], meaning that what happens in one component of the climate system affects all of the other components. The most advanced climate models are large software tools that use parallel computing to run across hundreds or thousands of processors. Climate models are a close cousin of the models we use for weather forecasting [wikipedia.org] and even use a lot of the same source code.

The atmospheric component of the model, for example, has a fluid dynamics simulation at its core. The model numerically integrates a set of primitive equations [wikipedia.org] such as the Navier-Stokes equation, the first law of thermodynamics, the continuity equation, the Clausius-Clapeyron equation, and the equation of state. Global climate models generally assume the atmosphere is in hydrostatic balance at all times, but that is not necessarily the case for regional models. Hydrostatic balance means that the force of gravity completely balances with the upward pressure gradient force, meaning that the air never accelerates upward or downward, which does occur in some instances like inside thunderstorms.

Not all atmospheric processes can be described by these equations, and we also need to predict things like aerosols (particulates suspended in the atmosphere), radiation (incoming solar radiation, and heat radiated upward), microphysics (e.g., cloud dropets. rain drops, ice crystals, etc...), and deep convection like thunderstorms (in models with coarse resolutions) to accurately simulate the atmosphere. Instead, these processes are parameterized [wikipedia.org] to simulate their effects as accurately as possible in the absence of governing equations.

The atmospheric simulations are generally more complex and run at a higher resolution for weather models than in climate models. However, weather models do not simulate the oceans, land surface, or the biosphere with the same level of complexity because it's not necessary to get accurate forecasts. For example, the deep oceans don't change enough on weather time scales to impact the forecast, but they do change in important ways on climate time scales. A weather model also probably isn't going to directly simulate how temperature and precipitation affect the type of vegetation growing in a particular location, or if there's just bare soil. Instead, a weather model might have data sets for land use and land cover during the summer and winter, use the appropriate data depending on the time of year being simulated, and then use that information to estimate things like albedo and evapotranspiration.

The main difference between climate models and weather models is that weather models are solving an initial condition problem whereas climate modeling is a boundary condition problem. Weather is highly sensitive to the initial state of the atmosphere, meaning that small changes in the atmosphere at the current time might result in large differences a week from now. Climate models depend on factors that occur and are predictable on much longer time scales like greenhouse gas concentrations, land use and land cover, and the temperature and salinity of the deep ocean. Climate models are also not concerned with accurately predicting the weather at a specific point in time, only its statistical moments like the mean and standard deviation over a period of time. We intuitively understand that these statistical moments are predictable on far longer time scales, which is why you could confidently insist that I'm wrong if I claimed that there would be heavy snow in Miami, Florida on June 20, 2050.

How and why are climate models coupled?

Information from the various components in the model needs to be communicated to the other components to get an accurate simulation. For example, precipitation affects the land surface by changing the soil moisture, which may also affect the biosphere. The albedo of the land surface affects air temperatures. Soil moisture also affects temperature, with arid areas typically getting warmer during the day and colder at night. If the precipitation is snow, the snow cover prevents heat from being conducted from the ground into the atmosphere, causing colder temperatures. Warm ocean temperatures are conducive for tropical cyclones to form, but the winds in a strong cyclone can churn up cooler water from below, which will weaken a tropical cyclone.

Both weather and climate models are coupled models [princeton.edu], meaning that information is communicated between different components of the system to allow the model to simulate interactions like these and many others. Each component of the climate system (e.g., atmosphere, hydrosphere, lithosphere, etc...) is generally a separate software module that is run simultaneously with the other components and interfaces with them. If the components of weather and climate models weren't coupled together, we couldn't simulate many of the feedbacks and tipping points that arise from these interactions.

What are climate models used for?

Perhaps the most frequently discussed application of climate models is simulating how various emissions scenarios will affect future climates. But climate models are also used for many other applications like sensitivity studies, attribution of extreme events, and paleoclimate studies.

An example of a sensitivity study might be to examine how deforestation of the Amazon affects the climate [nature.com]. A sensitivity study [ametsoc.org] would require two models, one a control simulation with the Amazon rainforest intact, the other with the rainforest replaced by grassland or bare soil. Most of the parameters that define these simulations like greenhouse gas concentrations would be kept identical so that only the presence or absence of the Amazon rainforest would be responsible for the differences in climate. The simulations would be run for a period of time, perhaps years or decades, and then the differences between the simulations are analyzed to determine the sensitivity of the climate to whatever is different between the simulations.

Extreme event attribution attempts to determine to what extent climate change is responsible for a particular extreme event. This is very similar to sensitivity studies in that there's a control simulation and a second simulation where some aspect of the climate system like greenhouse gas concentrations is different. For example, if we want to estimate the effect of climate change on an extreme heat wave in Europe [washingtonpost.com], we might run a control simulation with preindustrial greenhouse gas levels and another simulation with present day levels. In this case, the greenhouse gas concentrations would probably be prescribed at a particular level and not permitted to vary during the simulation. These simulations might be run for hundreds or even thousands of years to see how often the extreme event occurs in the preindustrial and the modern simulation. If the heat wave occurs every hundred years with modern greenhouse gas levels but never occurs with preindustrial conditions, the event might be attributed entirely to climate change. If the event occurs in both simulations, we would compare the frequency it occurs in each simulation to estimate how much it can be attributed to climate change.

For paleoclimate simulations [archive.ipcc.ch], we have much more limited information about the climate. We might know the greenhouse gas concentrations from bubbles of air trapped in ice cores, for example. There may be proxy data [noaa.gov] like fossil evidence of the plants and animals that lived in a particular location, which can be used to infer information about whether a climate was hot or cold, or whether it was wet or dry. On the other hand, we certainly won't have detailed observations of things like extreme events, oceanic circulations, and many other aspects of the climate system. In this case, the climate model can be configured to match the known aspects of the past climate as closely as possible, then using the simulation to fill in the gaps [buildyourownearth.com] where we don't have observations. Paleoclimate simulations can also be used to identify biases and errors in the model when it's unable to accurately reproduce past climates. When these errors are discovered, the model can be improved to better simulate past climates, and that also increases our confidence in its ability to extrapolate future climates.

Can we trust climate models?

All weather models and climate models are wrong. A weather model will never forecast the weather with 100% accuracy, though they do a remarkably good job at forecasting wide range of weather events. The model is still the best tool we have to predict the weather, especially beyond a day or two where extrapolation just isn't going to be reliable. Many components are shared between weather and climate models, and if these components didn't work correctly, they would also prevent us from producing accurate weather forecasts. Weather models often do have some systematic bias, especially for longer range forecasts, but we can correct for these biases with statistical postprocessing. Every time a weather model is run, it's also helping to verify the accuracy of any components that are shared with climate models.

Climate models from a couple of decades ago generated forecasts for the present climate, and once differences in greenhouse gas concentrations are accounted for, they are very accurate at predicting our current climate [nasa.gov]. Climate models are also used to simulate past climates, and their ability to do so accurately means that we can be more confident in their ability to predict the climate under a much wider range of conditions.

Even when there is a known bias in climate models, it does not invalidate all climate model studies. For example, climate models typically underestimate greenhouse gas sinks, resulting in a high bias in greenhouse gas concentrations for a particular emissions scenario. But we may be able to correct for that bias with statistical postprocessing. Also, many applications of climate models like extreme event attribution, many sensitivity studies, and many paleoclimate simulations do not dynamically simulate the carbon cycle. This means that those applications of climate models would be completely unaffected by the issue with underestimating greenhouse gas sinks.

Many of the climate models like the Goddard Institute for Space Studies models [nasa.gov], the Community Earth System Model [github.com], and the Weather Research and Forecasting Model [github.com] (often used in regional climate modeling) are free and open source, meaning that anyone can download the model, examine the source code, and run their own simulations. Data from a large number of climate model simulations is often publicly shared, especially in various intercomparison projects [llnl.gov]. Climate models are not closely guarded secrets, so anyone can examine and test climate models for themselves, and modify the source code to fix bugs or make improvements.

Original Submission