All models are wrong, but some are useful

Posted in: Comment, News and Updates

I attended the I-SEE Sustainable Energy & the Environment Webinar, ‘The Science Behind Climate Models’, yesterday.  It was given by Bill Collins, Professor of Climate Processes, Atmospheric Chemistry and Earth System Modelling at the University of Reading.

Here's his blurb:

We often hear the phrase “Climate models predict …”. Some take this as almost a guarantee that a prediction will occur, whereas others are wary of trusting a black box. The truth is nearer the quotation from the statistician George Box “All models are wrong, but some are useful”. This talk will explain how climate models are built, and why even though they all use the same laws of physics they can give quite different estimates for the future levels of climate change. Climate models can be used for much more than predicting the future. They can help us understand the past and present and help answer questions relevant for making policy decisions. This is why the 2021 Nobel prize for physics was awarded to the original pioneers of climate modelling. Climate change will be one of global society’s greatest challenges over the coming decades and climate models are our main tool to tackle it.  This talk will explain how climate models are built and how they can be used for much more than predicting the future.

This was a history of climate modelling from 1967 onwards and was a tale of increasing sophistication aided and abetted by the growth of computing power.  It's a fascinating subject which led to a brief discussion of the idea of Digital Twin Earths.

I was interested in how such models and modelling are presented to the public, and how the public is helped – or not – to interpret and understand them.  With this in mind, I asked:

"It’s sometimes said that IPCC worst-case projections tend to be chosen by some climate activists in order to boost their case for radical social change.  Is there any substance to this accusation?"

Although the question was selected by the webinar moderator, it was side-stepped by Collins and there was no chance of a follow-up.  This question received a similar fate:

"As a modeller, what did you make of the way that models were used during the pandemic?"

Never mind.  There will be other opportunities as the interface between models and the public is a topic that can only have greater salience as the power of models (and modellers) increases.


Posted in: Comment, News and Updates


  • (we won't publish this)

Write a response

  • Thank you for a simple observation about modelling. Yes, it has improved over the decades, yet the basic assumption of how good are they is simple to see in our weekly weather modelling systems. Data in equals data out, and while our knowledge of variables improves immensely as time goes on, we still tend to find incredible variation in models. These variations stem from which variables we include or anticipate have more value in the modelling equations and variable used. All too often the uncertainty preferred by scientists is bypassed as 'most preferred outcome' by politically motivated people, whether science or non-science based.