September 27, 2006
Modelling Traps
A serious problem with the climate change - or AGW - research is that it relieas very heavily on mathematical climate models. Even studies aimed at disproving the hypothesis often do it.
Now, mathematical modelling is usaed routinely in a number of different fields; so why is it a problem for climate research? The problem is testing these models.
In the field I'm familiar with - chemical engineering - models are used to design, study and optimize chemical reactors and other units like distillation columns.
These models employ several equations to describe all the phenomena occurring in the specific piece of equipment, and work on user-introduced data to provide the required output - for example, composition of the reactor effluent.
Testing these models is not exactly trivial, but it is amply feasible both as a matter of cost and complexity (for an university or corporation); moreover, there are plenty ofconned and bribed students willing and eager to do the research work. Experimental conditions are set and mantained within narrow limits, and the performance of the unit observed closely.
For example, the conditions to set for a chemical reactor are its dimensions and geometry; amount and type of catalyst; temperature and pressure; feed flowrate and composition (not all of them all the times, tho). What is measured generally is effluent composition, but also temperature profiles in the catalyst bed are of interest. How the reactor responds to a change in experimental conditions is most interesting; the parameters of interest are then varied one at a time to study its effect.
If what is observed in reality differs from what the model predicts (within error limits), it means that something is amiss. Assuming that no mistakes have been made (and assumption that isn't so automatic), a discrepance from reality means that either the model is not applicable in certain conditions, or it is completely wrong.
If a model gives results that are only slightly in disagreement with reality... well, that's a more difficult situation. Usually the model is still employed until something better comes out, and with the warning that predictions may be unreliable.
It should be obvious that climate models cannot be tested in a controlled environment. We cannot take a terracompatible planet, endorse it with a vast array of sensors and measurement equipment, record a suitable baseline and then start fiddling around with its atmosphere (and its star, too) to see what exactly happens to the planetary climate.
The only thing that can be done with climate model is to verify how well they reproduce past climate (and hope they will worke the same in the future; extrapolation is justly regarded as a last ditch technique). Now, we're not even sure of what datasets and what data treatment are the most appropriate for past climate; when the result of models are confronted with, for example, smoothed temperature records there are always are discrepancies. The models are sometimes early, sometimes late; they generally reproduce the main features of the curve, but often miss small ones.
Where does all this leave us, then? It leaves us with a lot of uncertainty.
A certain warming has been observed, but even what part of this warming is true, and what parts are due to variations in land use (the famous urban heat-island effect) and data treatment artifacts, is still largely unknown. And even the true warming can be ascribed to different factors, of which anthropogenic greenhouse gases are only one (personally, I am convinced that there is some contribution from them, anyway).
When it comes to predicting future climate, I think that no models are reliable enough to justify taking action - especially when action is a strongly ideologized treaty such as Kyoto. I think that energy efficiency measures and "reducing carbon footprint" should be taken only if they make economical sense overall (knowing that the tragedy of the commons, the prisoner's dilemma and market failures are still there...) and not just because it seems well and good to do so.
Now, mathematical modelling is usaed routinely in a number of different fields; so why is it a problem for climate research? The problem is testing these models.
In the field I'm familiar with - chemical engineering - models are used to design, study and optimize chemical reactors and other units like distillation columns.
These models employ several equations to describe all the phenomena occurring in the specific piece of equipment, and work on user-introduced data to provide the required output - for example, composition of the reactor effluent.
Testing these models is not exactly trivial, but it is amply feasible both as a matter of cost and complexity (for an university or corporation); moreover, there are plenty of
For example, the conditions to set for a chemical reactor are its dimensions and geometry; amount and type of catalyst; temperature and pressure; feed flowrate and composition (not all of them all the times, tho). What is measured generally is effluent composition, but also temperature profiles in the catalyst bed are of interest. How the reactor responds to a change in experimental conditions is most interesting; the parameters of interest are then varied one at a time to study its effect.
If what is observed in reality differs from what the model predicts (within error limits), it means that something is amiss. Assuming that no mistakes have been made (and assumption that isn't so automatic), a discrepance from reality means that either the model is not applicable in certain conditions, or it is completely wrong.
If a model gives results that are only slightly in disagreement with reality... well, that's a more difficult situation. Usually the model is still employed until something better comes out, and with the warning that predictions may be unreliable.
It should be obvious that climate models cannot be tested in a controlled environment. We cannot take a terracompatible planet, endorse it with a vast array of sensors and measurement equipment, record a suitable baseline and then start fiddling around with its atmosphere (and its star, too) to see what exactly happens to the planetary climate.
The only thing that can be done with climate model is to verify how well they reproduce past climate (and hope they will worke the same in the future; extrapolation is justly regarded as a last ditch technique). Now, we're not even sure of what datasets and what data treatment are the most appropriate for past climate; when the result of models are confronted with, for example, smoothed temperature records there are always are discrepancies. The models are sometimes early, sometimes late; they generally reproduce the main features of the curve, but often miss small ones.
Where does all this leave us, then? It leaves us with a lot of uncertainty.
A certain warming has been observed, but even what part of this warming is true, and what parts are due to variations in land use (the famous urban heat-island effect) and data treatment artifacts, is still largely unknown. And even the true warming can be ascribed to different factors, of which anthropogenic greenhouse gases are only one (personally, I am convinced that there is some contribution from them, anyway).
When it comes to predicting future climate, I think that no models are reliable enough to justify taking action - especially when action is a strongly ideologized treaty such as Kyoto. I think that energy efficiency measures and "reducing carbon footprint" should be taken only if they make economical sense overall (knowing that the tragedy of the commons, the prisoner's dilemma and market failures are still there...) and not just because it seems well and good to do so.
Comments:
Post a Comment
