uncertainty vs. error

As you may know, Climate Scale (CS) is a spin-off of Vortex that will offer a similar kind of high-resolution atmospheric modelling results but, instead of spanning 30 years back in time, it will forecast 50 to 100 years ahead.

At Vortex we are learning a lot from CS’s new and refreshing points of view. The CS team comprises climate modelers with deeper and more sophisticated opinions on apparently obvious concepts such as accuracy, error, measurements and uncertainty.

Lately, in what was quite a “philosophical” internal debate with them, measurements and model results were almost put on a par as “samples of the probability distribution that describes reality”. Ultimately, measurements can also be considered model results, with the model being the transfer function of the sensors. In this sense, their uncertainty or errors should at the end of the day be taken into account in a similar fashion to those of a computer model.

However, even at the risk of oversimplifying our clients’ views, we believe that many of them often consider measurements as the “clean” measure of the “reality”: a target towards which model results should “be aimed” and, hopefully, “hit”.

The latter is intended to be the meaning of the above image. The target is the “reality” (the measurements) and the deviations of the “shots” (the model results) from the bullseye are the “errors”, while the scattering of the different shots is the model “uncertainty” related with the dispersion of different model realizations around a central result.

As we have seen in many cases that uncertainty and error, as defined in the last paragraph, are not related (i.e. all four cases of the image above are equally possible). When one of our customers launch a model run at a new site where “reality” is unknown, how is knowing the dispersion of the “shots” around a central result of use if she is concerned with the deviation of that central result from the (unknown) target?

So I’m afraid that the open question I posed in my previous post remains unanswered. Your opinions are therefore still very welcome.