“But Modulz”

A square around “teh modulz are stoopid.” Most rejoinders reiterate that all models are wrong, but some are useful. Pixel perfection is the enemy of good modelling. See also #ButPredictions, #ButObservations, #ButEvidence.


Maybe people just aren’t buying the decades of failed predictions and wildly inaccurate models anymore.


Interesting. When you use an ML-style method to estimate the effect on climate of CO2 emissions, it comes out fairly small.


Objections and Replies

AGW Proof. The only real proof you got are your modulz—
Not exactly.

Clouds. We don’t know if clouds are an important source of warming and—
☞ While clouds are complex stuff, Dick & Roy’s claim has fallen into disrepute.

COVID. Climate models are just like epidemiological models—
☞ No. Really, not at all. Climate models are another kind of beast altogether. You could look under the hood and see for yourself. Otherwise look at energy models and economical models.

Derivation. I’d like an engineering-quality exposition of sensitivity—
There is no direct calculation to accurately prove this.

Discrepancy. There’s a huge difference between modulz and obs—
☞ Decadal variability is tough to model. Very noisy. Things are getting better.

ENSO. ENSO is based on REAL data. AGW models are based on nonsense
☞ Climate models are based on physics. They also include REAL data.

Hot Spot. A Tropical Hot Spot forecasted by modulz doesn’t exist in the real
☞ The troposphere is warming around 70-80% faster than the surface.

Impacts. Modulz embed significant climate impacts
☞ False. Modulz embed observed physical properties of GHGs and the rest of the atmosphere; climate impact emerges as a consequence (PaulS).

Impose. Don’t impose modulz on us—
☞ Nobody does. They’re mostly useful tools to explore the future.

Natural. Modelz can’t generate observed events from natural processes—
☞ Incorrect. They routinely produce substantial climate effects from designated natural processes, e.g. Tambora, TSI increases, etc (PaulS).

Observation. Should we trust modulz or observations—
☞ Unfortunately observations of the future are not available at this time.

Parameters. The stoopid modulz that have been fit to data—
☞ Although climate models contain parameters that may be tuned, climate models are not really fit to observations. If that were the case, the models would all reproduce perfectly the observed global trend. We all know this is not the case, and that the spread is quite large (Eduardo).

Purpose. Climate modulz are not fit to tell us how much fossil fuel—
☞ Researchers use Shared Socioeconomic Pathways for that.

Propagate. At each step there are errors and these errors propagate—
☞ By that logic there is no correlation between age and height. Also, it’s the net error that matters for the energy imbalance will prevail in the end {1}.


Real. A model is not real—
☞ Are you suggesting it’s some kind of spiritual entity? Besides, a model isn’t meant to replace what it models. Scientists are not seeking replicas or prosthetics. When was the last trip to the center of the Earth, BTW?

Refuted. We should discard models that produced falsified outcomes—
☞ The MET won’t ditch its models because it did not rain the day they said there were 80% chances of rain. Economists won’t ditch their models because inflation did not turn out to follow the path they forecasted. Same for epidemiologists. Modelling is a bit like democracy: it’s the worst way to do science, except for all the others.

Science. Modulz are not science—
☞ Have you ever used a thermometer? No model, no measurement. No model, no data. No model, no implementation of any theory whatsoever. Unuseable science.

Stoopid. Since modulz are stoopid shows that AGW is stoopid too—
☞ We don’t need no damn models to support AGW.

Sensitivity. Establishing the true climate sensitivity will save us trillions—
The ratio of temperature change to cumulative carbon emissions, is approximately independent of both the atmospheric CO2 concentration and its rate of change on these timescales {2}.

Tweaked. Modulz are tweaked to get the results scientists want
☞ If only it was that easy, we would not need models in the first place. Many properties emerge from runs. They still need to hindcast properly.

Until. We need better modulz until
☞ They’re what we got {3}. I’ll keep my eyes open even if hindsight isn’t 20/20.

Variability. It is impossible to state that natural variability has had no effect—
☞ The IPCC might not ignore what one can read on an introductory entry.

Volcanoes. Volcanic eruptions can only be included tuning—
☞ Indeed. Do you have a point?


{1} Imbalance. This is the crucial idea. However lousy are our models, in the end if more energy comes in that comes out, there will be warming.

{2} Sensitivity squirrel. Arguing about sensitivity is as relevant as trying to decide under which conditions plateaux exist. Don’t be suckered in by curiosity, keep the eye on the ball.

{3} Skeletons. All modellers have skeletons of spherical cows in their closets.


Climate and Carbon Cycle Models; nice compendium by the U of Chicago.

SSP Database; SSP stands for Shared Socioeconomic Pathways.

2022-03; How not to interpret the emissions scenarios in the IPCC report.

2020-02; But RCPs.

2020-02; Models in Science; a philosophical explanation as to why models are here to stay.

2020-01; Modeling Myths of Climate Change; DICE is far from being perfect.

2019-09; Computer Simulation in Science; Eric haz a good entry to the philosophical problems.

2019-02; CMIP6: the next generation of climate models explained.

2018-04; Explainer: How ‘Shared Socioeconomic Pathways’ explore future climate change.

2018-01; How do climate models work?

2018-01; Timeline: The History of Climate Modelling.

2016-06; Climate meme debunked as the ‘tropospheric hot spot’ is found

2014-08; What Is Business As Usual?; allow NG to explain.

2014-05; Can we trust climate models?

2013-08; Is the climate sensitivity less than 2°C?

2008-11; FAQ on climate models.

Further Readings

2020-10; Variability in historical emissions trends suggests a need for a wide range of global scenarios and regional analyses; https://doi.org/10.1038/s43247-020-00045-y

2020-08; RCP8.5 tracks cumulative CO2 emissions; https://www.pnas.org/content/117/33/19656

2020-07. An Assessment of Earth’s Climate Sensitivity Using Multiple Lines of Evidence. https://doi.org/10.1029/2019RG000678

2019-12; Evaluating the Performance of Past Climate Model Projections; https://doi.org/10.1029/2019GL085378

2019-05; Robust skill of decadal climate predictions; https://doi.org/10.1038/s41612-019-0071-y

2016-09; The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6; https://doi.org/10.5194/gmd-9-3461-2016

2011-10; Cloud variations and the Earth’s energy budget; https://doi.org/10.1029/2011GL049236

2016-05; Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization; https://doi.org/10.5194/gmd-9-1937-2016.

1988-08; Global climate changes as forecast by Goddard Institute for Space Studies three-dimensional model; https://doi.org/10.1029/JD093iD08p09341.

Climateball Episodes

2018-01; We’re climate researchers and our work was turned into fake news

2011-08; How Scientific Debate Should Be Conducted; Roger Senior and John NG discuss.

2015-05; Lukewarmers – a follow up; genealogy of the luckwarm brand.

Engineerily Deriving; a series of comments on the Auditor’s quest for an “engineer-level formal derivation.”