The climateprediction.net project comprises three separate experiments - one to explore the model we are using, the second to see how well the models replicate past climate and the third to finally produce a forecast for 21st century climate. Each model that we distribute will be used for all three experiments. Each model distributed is unique, and differs from all the others in three ways: the initial conditions it is started from, the attributes which force it to be in one particular climate state and the parameters which make up the actual model.

Every climate model has to make a number of approximations, called parameterisations. To read more about these, click here. Basically this means that there are numbers in the model which are given a certain, fixed value, but this value is not known for sure and a range of values could be equally realistic. The experiments will investigate the effect on the modelled climate of varying the value of 20 of the most poorly understood parameters in the model - such as the relationship between the number of raindrops in a cloud and how much it actually rains (to see what they are, click here). It is possible that some combinations of parameters may replicate the past climate equally well, but produce widely different forecasts for what might happen in the future. Some combinations of parameters will not work at all, produce a completely unrealistic climate ( for example an Earth that boils or freezes, or oscillates between very hot and very cold every couple of years) and probably crash the model. It is not possible for us to tell beforehand what these combinations will be.

Experiment	Goal	Methodology
1 Explore model sensitivity to parameters [tell me more]	Identify suitable ranges of parameters.	Each simulation includes 3 phases: Calibration (15yrs) Pre-industrial CO2 run (15yrs) Double CO2 run (15yrs)
2 Simulation of 1920-2080 [tell me more]	Assess model skill by making a probability based forecast of the past climate. Compare model outputs with observations to assess how well the model performs. Make a probability based forecast of future climate.	Run the model with a range of initial conditions, forcings and parameters .

This experiment is more about learning how the model reacts to changes in initial conditions and parameters than about actually trying to replicate the Earth's climate. For this reason, the model we use has a sophisticated atmosphere, but a simplified ocean (a single layer, 'slab' ocean). This means that some aspects of the climate system (such as oceanic currents, and the El Nino oscillation) are not replicated, but the model runs a lot faster and a lot more calculations can be completed.

The knowledge we gain from this experiment about the way the model reacts to changes to the parameters will be used to design the next phases of the climateprediction.net experiment - combinations of parameters that obviously do not work can be avoided.

The experiment consists of 3 separate phases, each model which is distributed completes all 3 phases from a unique set of initial conditions:

Calibration step(phase 1)

Phase 1 is the calibration phase of the experiment. In this phase, the temperature of the surface of the ocean is artificially held constant. The movement, or flux, of heat, in or out of the ocean that is required to keep the ocean at a constant temperature is calculated. This is an easy solution to having a very simple ocean in the model, which cannot actually store heat in the way that a real, deep, complex ocean can. The dates given to this phase are 1810-1825.

Pre-industrial CO₂ step (phase 2)

This is the control phase. This involves running the model for 15 years with the levels of CO₂ in the model atmosphere kept constant at pre-industrial levels, 282ppm. Unlike phase 1, here the temperature of the ocean surface is allowed to vary, according to how much energy the ocean receives and emits. However, it is safe to assume that the amount of heat flowing into the oceans is the same as in phase 1, so the heat fluxes calculated in phase 1 are applied. Unless the atmosphere starts doing something very different, and the energy balance at the top of the atmosphere is changed, the temperature of the whole atmosphere should therefore stay the same. If this is the case, the globally averaged surface temperature should also be approximately constant and not change substantially from year to year or drift off to a very different temperature, and we say that the model is stable. The dates given to this phase are 1825-1840.

Double CO₂ step (phase 3)

In this phase the levels of greenhouse gases (you can read more about the Greenhouse Effect here) are doubled and the model is run for a further 15 years. In a good model, the atmosphere should adjust to this change in forcing and eventually settle in a new stable, equilibrium state (which may be the same, warmer or cooler). The dates given to this phase are 2050-2065.

The results will give an indication of what combinations of parameters work (in terms of producing an atmosphere that behaves in a similar manner to reality and does not freeze or boil, or oscillate in and out of ice ages on a timescale of a couple of years). We will therefore be able to use the results to guide our choice of parameters in the main experiment.

By comparing the single and doubled CO₂ steps, we can calculate the climate sensitivity of the models - this is the difference between the globally averaged surface temperature in the model with pre-industrial CO₂ and in that with doubled CO₂. This is a useful indicator of how a climate model behaves, although it is slightly artificial, as of course carbon dioxide values in the atmosphere do not remain constant for 15 years, but change continuously.

As extensions to this experiment, we ran a 4 phase thermohaline circulation experiment for a couple of months from June 2004. You can see the first results of this here.

In September 2005, we launched a 5 phase sulphur cycle experiment which will also run for a couple of months.

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

• The combinations of parameters that were identified in experiment 1 as working i.e. that produce a stable, viable climate.

• The range of initial conditions will be the same as that used in experiment 1.

• The experiment will be forced by observations of CO₂, volcanic emissions etc. from 1920-2000 and a range of possible scenarios for what might happen in the next 100 years. Read more about the forcing scenarios here.

By using each model to produce a 'hindcast' for 1920-2000, and then comparing the spread of forecasts with observations of what actually happened, we will get an idea of how good our range of models is - do most of them do a good job of replicating what actually happened? This will also let us 'rank' models according to how well they do. All the models will also be used to produce a forecast for the future - until 2080. When this experiment finishes, we will have a range of forecasts for 21st century climate.

There is a time issue particular to the ocean. The ocean's heat capacity is thousands of times greater than that of the atmosphere; therefore it takes the ocean much longer than the atmosphere to reach an equilibrium. Hence a long "spinup" phase is required before it is possible to perform experiments using ocean models. If this were not done, the ocean part of the model might still be adjusting to the starting conditions of an experiment, rather than changes imposed by the experiment itself, at the end of the experiment.

To read more about the spin-ups that were completed before this experiment could start, follow this link. In preparation for the coupled launch we have tested and compared the results from different (in resolution and topography) models and methods of the spinup, before deciding on the final design. We have also selected physical parameters and their ranges with which to perturb the ocean models. We have created masks covering respective ocean basins and other specific areas of interest, in order to extract the variability in time of ocean diagnostics such as heat and freshwater content. We have also created flux correction fields, which will be used in the "hindcast" phase to maintain a realistic ocean surface climate.

The coupled model runs asynchronously, which means that the atmosphere model runs first for some time then the ocean model runs for some time, taking turns. In this experiment the individual components run for one day at a time.

click here to read about the experimental strategy in more detail.