The climateprediction.net project comprises three separate experiments (details are given below).

All the experiments have three main parts:

1. initial condition ensembles
2. perturbed physics ensembles
3. forcings

Initial condition ensembles involve the same model, with the same forcings, run from variety of different start dates. Because the climate system is chaotic, tiny changes in things such as temperatures, winds, and humidity in one place can lead to very different paths for the system as a whole. We can work around this by setting off several runs started with slightly different starting conditions, and then look at the evolution of the group as a whole. This is similar to what they do in weather forecasting.

Perturbed physics ensembles form the main scientific focus of the whole project. Modern climate models do a good job of simulating many large-scale features of present-day climate. However, these models contain large numbers of adjustable parameters which are known, individually, to have a significant impact on simulated climate. While many of these are well constrained by observations, there are many which are subject to considerable uncertainty. We do not know the extent to which different choices of parameter-settings or schemes may provide equally realistic simulations of 20th century climate but different forecast for the 21st century. The most thorough way to investigate this uncertainty is to run a massive ensemble experiment in which each relevant parameter combination is investigated. Thus the perturbed physics ensemble is the central feature of the climateprediction.net project. You can read more about the rationale for this experiment by following the other links in the menu bar on the left.

Forcings are the things which drive the climate system. The chaotic variability we target in the initial condition ensemble is due to factors internal to the climate system, while things such as solar variability, sulphate (volcanic, etc) forcing and greenhouse gases are treated as external to the climate system. We call them forcings because they force the system from the outside: if these things change, we expect the climate system to respond. [If the sun puts out more energy, we would expect the Earth to heat up, for instance.]

The climateprediction.net strategy to achieve our ultimate objective of predicting potential future climate states is based on the following three steps:

Step	Goal	Methodology
Experiment 0 Pilot experiment to check feasibility of distributed computing [tell me more]	Identify what physics of the HadSM3 model provide the best / an acceptable simulation of present-day climate with prescribed present-day ocean surface temperature.	To run the atmosphere-slab ocean model with different physics to check the model simulation of present day climate. This experiment corresponds to the spin-up run of experiment 1.
Experiment 1 Explore model sensitivity to physical parameters [tell me more]	Identify suitable parameters and ranges by use of the HadSM3 model.	To realise an ensemble of simulation with perturbations made to a number of paramters. Each simulation includes 3 steps: Model spin-up (15yrs) Model control (15yrs) Double CO2 run (15yrs)
Experiment 2 Explore model sensitivity to initial conditions, historical forcings [tell me more]	Assess predictive model skill by making a probabilistic hindcast of the past climate by use of the HadCM3 model.	Make an ensemble of hindcast simulations for the period 1920-2000 by perturbing the initial conditions, and running a range of historical forcings. Compare model outputs with observation to assess how well the model performs.
Experiment 3 The climateprediction.net project to explore potential future climate state [tell me more]	Make a probabilistic forecast of the future climate by use of the HadCM3 model.	Make an ensemble of forecast simulations for the period 2000-2080 period by perturbing the physics of the model, running an initial condition ensemble and running a range of possible future solar, sulphate and greenhouse gas futures.

This experiment uses an atmosphere-slab ocean model (see expt 1) to simulate the present day climate. Parameters will be varied to explore the sensitivity. The main objective is to check the feasibility of distributed computing and get your feedback (learning by doing).

[Note: this has already been completed.]

This experiment will be slightly simpler than the atmosphere-ocean global circulation model (AO-GCM) which will be used in the main climateprediction.net release. It will still have a full atmosphere but the representation of the ocean will be simplified into just a single layer. This is known in the trade as a "slab" model. It is not so useful in forecasting future climate because the "ocean" can only respond in very limited ways. However, it is very useful for investigating the sensitivity of the model to changes in the atmospheric parameters. Because it doesn't have a full ocean model it responds much more quickly to changes in factors such as levels of greenhouse gases. The knowledge we gain from this experiment about model sensitivity will be used to design properly the climateprediction.net experiment. By perturbing parameters which control the models physical processes (such as cloud formation) it is possible to see different realisations of climate change.

As in the main experiment everybody's model will be unique because each will have a different combinations of parameters. We will be asking you to carry out 3 separate phases with each combination of parameters.

Calibration (phase 1)

It involves running the model for 15 years (which should take about 3 weeks on an up-to-date computer) to make sure the atmosphere and the slab ocean are "in balance" (they are in a stable, equilibrium state where large scale effects such as global mean temperature, do not change substantially from year to year). This experiment will provide some forcing fields to the 2 other phases.

Control run (phase 2)

Standard run where the levels of greenhouse gases in the model atmosphere are kept at pre-industrial levels.

Double CO₂ (phase 3)

Double CO₂ run with levels of greenhouse gases about twice what they are in the control phase. The results will give an indication of how sensitive the climate is to the parameters we vary. We will therefore be able to use the results to guide our choice of parameters in the main experiment.

In experiment 1, we have an initial condition ensemble, a pertubed physics ensemble, and a forcing ensemble that includes pre-industrial levels of CO2, and 2xCO2. This is to help us find the climate sensitivity of the models we are studying. [The IPCC defines climate sensitivity as "equilibrium climate sensitivity is defined as the change in global mean temperature, T2x, that results when the climate system, or a climate model, attains a new equilibrium with the forcing change F2x resulting from a doubling of the atmospheric CO2 concentration."]

The second experiment will use the full atmosphere-ocean GCM (the coupled model). This means the ocean is able to respond much more dynamically than in experiment 1, giving us a more complete simulation of the climate. The basic parts of the experiment are:

1. an initial condition ensemble;
2. a perturbed physics ensemble (the interesting and viable models from experiment 1); and
3. historical forcings (from the period 1920-2000).

The initial condition ensemble will be similar to those used in parts 1 and 3- see the descriptions of experiment 1 for details.

The perturbed physics ensemble will comprise those models that are viable, stable climate models, that span the interesting regions of parameter space we have identified in experiment 1. We will be looking to include as many of these as possible, and will really only be throwing out those models which have gone unstable because of the choice of parameter sets. We imagine that there will be regions of parameter space (families of closesly related models with similar parameter settings) that are viable (it's theoretically possible that some will even be better than the standard model!), and regions that are not. We will find those regions in experiment 1 and use the good ones in experiment 2.

For our forcings in this part of the experiment we will use the data from the climate rercord from 1920-2000. We will start a bunch of experiments in 1920 and force them with historical data for fifty years. This process is called a hindcast: it's like a forecast, only you know the outcome. We know what happened 1920-2000, but it's still a challenge for the model to do a good job of simulating it. We can use the models' performance in simulating the past to see how good they are as fully-coupled models simulating recent and present climate. If they're good at that, we'll use them for predicting the future, too.

[Note: To be launched in 2005.]

The prediction experiment. As in the second experiment we will be using the fully coupled model. Using the models that have managed to do a fairly good job of simulating the historical 1920-2000 climate, we run an ensemble prediction of the period 2000-2080. There are three parts to the experiment:

1. an initial condition ensemble;
2. a perturbed physics ensemble (the survivors from experiment 2); and
3. a future forcings ensemble.

The initial condition ensemble will be similar to those used in parts 1 and 2 - see the descriptions of experiment 1 for details.

The perturbed physics ensemble will comprise those models that score well on the climate prediction index. This index (developed in conjunction with the Hadley Centre) gives a measure of how well a model fits observations. Models that score well have performed well in experiment 2, in which they have simulated the second half of the twentieth century (see above). Models that score badly will be discarded (or down-weighted) for experiment 3.

The future forcings ensemble is necessary because we don't know what the sun or the volcanoes are going to do over the next fifty years. We also don't know how levels of greenhouse gases are going to change over that period. So we're going to run a large number of different possible futures, in which we vary solar, sulphate and greenhouse forcing, to span what we hope will be the likely range.

Experiment 3 will thus comprise the best models we have that are consistent with observations, running predictions that span likely behaviour in the three major climate forcings. Each model and each future scenario will be started off from a range of different initial conditions to check that the results we get back are not simply idiosyncratic functions of the choice of start date.

And the result of all this will be, we hope, the world's best guess at a probabilistic climate forecast for 2080.