A key limitation of GCMs is the fairly coarse horizontal resolution. The climateprediction.net atmospheric resolution is 3.75° by 2.5°. For the practical planning of water resources, flood defences etc., countries require information on a much more local scale than GCMs are able to provide. There are three possible solutions to this problem:

• Run the full GCM at a finer resolution. As the model would then take much longer to complete a simulation, either a very powerful computer (such as the Earth Simulator in Japan) or a much shorter simulation period (e.g. 5 years) is required.
• Use statistical techniques to 'downscale' the coarse, GCM results to local detail (read more about this here). These techniques assume that the relationship between large scale climate variables (e.g. grid box rainfall and pressure) and the actual rainfall measured at one particular raingauge will always be the same. So, if that relationship is known for current climate, the GCM projections of future climate can be used to predict how the rainfall measured at that raingauge will change in the future.
• Embed a Regional Climate Model (RCM) in the GCM. RCMs are a more dynamically consistent way than statistical downscaling to produce a regional forecast.

RCMs work by increasing the resolution of the GCM in a small, limited area of interest. An RCM might cover an area the size of western Europe, or southern Africa - typically 5000km x 5000km. The full GCM determines the very large scale effects of changing greenhouse gas concentrations, volcanic eruptions etc. on global climate. The climate (temperature, wind etc.) calculated by the GCM is used as input at the edges of the RCM. RCMs can resolve the local impacts given small scale information about orography (land height), land use etc., giving weather and climate information at resolutions as fine as 50 or 25km.

In regions where the land surface is flat for thousands of kilometres, and there is no ocean anywhere near, the coarse resolution of a GCM may be enough to accurately simulate weather changes. However, most land areas have mountains, coastlines, changing vegetation characteristics etc. on much smaller scales, and RCMs can represent the effects of these on the weather much better than GCMs.

Winter precipitation over Britain as predicted by a) a GCM with resolution 300km, b) a regional model with 50km resolution and c) a regional model with 25km resolution compared to d) actual observations.

Why can't you just run the RCM on its own? The weather in one part of the world is not independent of the weather elsewhere in the world. For example, the El Nino Southern Oscillation, focused in the South Pacific, has effects which can be detected over most of the planet.

Predicted changes in winter precipitation over central/ Southern Europe between the present day and 2080. The areas of red, where precipitation has fallen by more than 0.5mm/day, indicate large reductions over the Alps and Pyrenees predicted by the RCM (right), but not the large scale GCM (left).

Climateprediction.net has worked together with the PRECIS group to develop a distributed regional climate modelling experiment, using the same numerical models as the PRECIS project. PRECIS is the Met Office Hadley Centre project 'Providing REgional Climates for Impacts Studies'. In the past, regional models have only been run independently of GCMs. A GCM would be run, saving all information generated for the region of interest. This information would then be the input to a subsequent RCM simulation. The climateprediction.net experiment will run the GCM and RCM together, with information being passed between the models as they move forward in time. This means that no where near as much output needs to be saved - making the experiment possible on a home PC. More information about RCMs in general and the PRECIS project in particular can be found here. For further information about the PRECIS project, visit the PRECIS web site.

§ back to top