07072010

Climatic Modelling - the problems with getting it right

What is a climatic model?

Simply put, a climatic model is a mathematical model that 'attempts' to model a climatic system over a period of time; the aim being to accurately model future events based on current data and methods.

The models themselves usually break up the climatic system into a series of 'cubes' of atmosphere around the globe, stacked upon each other throughout the atmosphere - similar to the way cardboard boxes can be stacked upon each other to form a wall. Except in this case these 'cubes' are wrapped around a globe - so at the equator their internal area is bigger than those close to the poles. Although some models use different shapes to avoid this difference in internal area, the 'box model' is the most common form.

Now the way the models work is that each box has a set of values associated with it - these are akin the physical characteristics of what ever is in that box and being modeled - be it temperature, wind, humidity, pressure, cloud cover etc. This gives you the 'state' of the box.

During an iteration of the run of the model, these values are processed through a set of equations, with the values from other bounding or affecting boxes to produce the values for the new state. Also in these equations you will have added in external influences that are beyond the model but are known to happen (i.e. the sun, known physical events [such as volcanoes]) - so to ensure the model can take these into account.

Now each iteration represents a certain period of 'real time' - be it a day or month, etc - so multiple iterations of the model allow it to progress forward in time modeling the climate.

So, whats the output?

Well the output is whatever the whole state of the model is at the end of the run. Something important to understand is that the equations are not 'perfect'; we do not have a perfect precision understanding of the climatic system, not a perfect record of measurements to 'seed' the start state of the model. So what often happens is that the model is run with slightly different start states and slightly different constants in the equations to give a 'spectrum' of final end states which can be used to derive a sense of confidence in the results given (i.e. try to factor out the inaccuracies).

But, is it representative?

Ah, depends who you talk to. The trouble is the equations and source data are our 'representation' of how the climatic system works at this moment in time. In other words what the climate does at the end of the day is not limited by our understanding or measurement of it; its very much its own 'agent'. Also the processes that occur in the climate are under constant flux; there is no need for the climate to maintain a steady state in how it works - the significance of certain processes or chemicals in the atmosphere varies over time.

So, in the short term, assuming good availability of data and a good understanding of the short term processes - we should be able to accurately model the climate over the short term. This is indeed the case, as weekly weather forecasting is able to be quite accurate on key metrics.

The difference comes in when modeling over much longer time scales, say years to centuries. Here processes and events which are 'hidden' or 'insignificant' in short term modeling play a stronger role. Also the longer the model run, the more opportunity there is for what were previously unrelated events or relationships to occur and interact. This is sort of akin to the old 'butterfly in the rain forest' quote on chaos theory. The thing is here, due to the events and relationships being unknown, they cannot be accounted for in any long term model.

Now, longer term climatic modelers say they can get round this using 'hind casting' - i.e. model a time period that has already occurred and check for conformance to what happened in reality. The trouble here is two fold:

• Hind casting is usually only done with reference to one recent time period; it is not done across a series of time periods. The danger here is that the model will adopt a 'bias' that made it fit the time period in consideration and carry that through to the future.
• Older data is basically less accurate and more 'sparse' than more recent data - this could lead to in effect a modeling of the underlying error rate.

Also, as previously mentioned, there are very few real 'constants' in a climatic system; the exact mechanisms evolve and change over time in response to what happens within that climate and what is the 'norm' at any given moment in time.

So, what can be done?

Firstly, there needs to be real recognition that long term climatic models are just a modeled indication of what 'might' possibly happen in the future - they do not define the range or scope of future climatic behavior. Our understanding of how the climate system works is constantly improving and being redefined & challenged (see this article on how Co2 could be being over played in current climate models to see what I mean). The media quite often overplays the significance of climate model results and this is leading to a lot of false assumptions.



Secondly, we need to improve the range and accuracy of source data available on our climate - the 'devil' is in the detail as it were, and at the moment a lot of the climatic system is not understood by dint of not being measured to the required accuracy to reveal the detailed workings.

Related Articles

• What is man made global warming?