the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 Apr 2022
Climate projections from IPCC models and regression models: A comparison
Abstract. In this paper, we show that the projected global temperature change from CMIP5 models is remarkably similar to that obtained using simple regression models relating the global temperature to the atmospheric concentrations of CO2. This result is strengthened when we consider the projections obtained using the CMIP6 models.
- Preprint
(861 KB) - Metadata XML
- BibTeX
- EndNote
Interactive discussion
Status: closed
- RC1: 'Comment on egusphere-2022-134', Anonymous Referee #1, 16 May 2022
-
RC2: 'Comment on egusphere-2022-134', Anonymous Referee #2, 20 May 2022
Arrhenius found that the global temperature is linked to the logarithm of CO2 concentration. Starting from this assumption, the author builds two linear models based on data of the last decades and then tries to make predictions for the future in various scenarios. He claims that his results are very similar to the temperatures projections coming from GCMs.
In my opinion, this paper could be acceptable only after major revisions.
In fact, first of all the results by linear models are similar to those by GCMs only in the highest emission scenarios, while in the low emission ones there are notable differences, so that the conclusions by the author are wrong. Secondly, in my opinion the importance of this paper could be just in the discussion of these differences.
For instance, it is very impressive to see how in RCP2.6 and also in SSP2.6 the last decades of the century show a decrease of temperature in the linear models and an almost stationary trend in GCMs. In my opinion, this is a clear hint that the linear model fails to catch the inertia of the climate system, for instance this “hysteresis” effect on the global temperature. Why? Is this failure related only to the absence of nonlinear terms in the model? The author could develop a nonlinear model and test this hypothesis. Or he needs other variables, even only through a decomposition between land and sea temperatures (the inertia of the oceans is well known). Studying these differences in performance could be the adjoint value of this paper and it should revised in this perspective.
In doing so, the author should also stress more clearly that, in any case, his very simple model cannot substitute GCMs. On the contrary, it can highlight the differences to be discussed from a physical point of view.
Citation: https://doi.org/10.5194/egusphere-2022-134-RC2 -
EC1: 'Comment on egusphere-2022-134', Riccardo Farneti, 24 Jun 2022
Dear Author,
thank you for submitting your work to GMD.
Your manuscript has now been seen by 2 referees. Considering their advice, and my own evaluation, I regret to inform you that we cannot publish your manuscript in GMD. The Reviewers revealed major deficiencies and flaws that appear very unlikely to be fixed upon revision and I therefore discourage submission of a revised manuscript.
I am sorry that we cannot be more positive on this occasion and thank you for the opportunity to consider your work.
Citation: https://doi.org/10.5194/egusphere-2022-134-EC1
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-134', Anonymous Referee #1, 16 May 2022
The author shows that (according to his assessment) the results of the multi-model mean (MMM) Global Average Temperature from CMIP5 and CMIP6 models according to the different scenarios (RCPs and SSPs) is well approximated by applying a simple linear regression fitted to historical observations of temperature and CO2 emissions, since 1976 and through 2005 or 2014.
I’m sorry to say I don’t see the value of this work and I recommend rejection.
First, GCMs produce hundreds of self-consistent variables and phenomena, so to propose that this linear regression on global temperature has value in substituting for complex GCMs is preposterous.
The comparison to make is with energy balance models, that through a simple function, as simple as a linear regression but accounting for physics and, importantly, feedbacks can indeed replicate global temperature and not only multi-model means but individual models that differ at a basic level by different equilibrium climate sensitivity. Some of this simple models importantly account for possible changes in the carbon cycle strength with warming, something a linear regression fitted on a short observational period can never do. The community has also, for many years now, recognized a strong linear dependence between global temperature and cumulative emissions. So, the idea that simple relationships can be found is not new, except here it is applied to the wrong pair.
The way the author shows agreement is extremely superficial and I question it at a fundamental level which is appreciable to the naked eye. The presentation of results is actually troubling in its hand-waviness. I don’t see good agreement in rates and even behavior for important scenarios like the highly mitigated (which are more and more the focus of the community). RCP2.6 stabilizes for CMIP5 MMM while it declines for the results of the regression, and strongly so. RCP4.5 MMM keeps increasing, even if at a slower rate, while the blue line stabilizes. The rate under RCP6.0 is completely different, with the two lines starting apart and ending close to one another. SSP1-2.6 rates of decrease are significantly different. SSP2-4.5 trajectories diverge at the end. I could go on. Why is the author not even bothering commenting on these discrepancies, let alone showing the behavior of residuals between the red and blue line, which would be obviously far from i.i.d., undermining the whole proposition of applying a linear model? The starting date of 1976 is also arbitrary and would demand some sensitivity analysis to the choice of the start date…but that is a detail that I don’t think is worth pursuing given the fundamental flaws of philosophy and performance one sees.
Citation: https://doi.org/10.5194/egusphere-2022-134-RC1 -
RC2: 'Comment on egusphere-2022-134', Anonymous Referee #2, 20 May 2022
Arrhenius found that the global temperature is linked to the logarithm of CO2 concentration. Starting from this assumption, the author builds two linear models based on data of the last decades and then tries to make predictions for the future in various scenarios. He claims that his results are very similar to the temperatures projections coming from GCMs.
In my opinion, this paper could be acceptable only after major revisions.
In fact, first of all the results by linear models are similar to those by GCMs only in the highest emission scenarios, while in the low emission ones there are notable differences, so that the conclusions by the author are wrong. Secondly, in my opinion the importance of this paper could be just in the discussion of these differences.
For instance, it is very impressive to see how in RCP2.6 and also in SSP2.6 the last decades of the century show a decrease of temperature in the linear models and an almost stationary trend in GCMs. In my opinion, this is a clear hint that the linear model fails to catch the inertia of the climate system, for instance this “hysteresis” effect on the global temperature. Why? Is this failure related only to the absence of nonlinear terms in the model? The author could develop a nonlinear model and test this hypothesis. Or he needs other variables, even only through a decomposition between land and sea temperatures (the inertia of the oceans is well known). Studying these differences in performance could be the adjoint value of this paper and it should revised in this perspective.
In doing so, the author should also stress more clearly that, in any case, his very simple model cannot substitute GCMs. On the contrary, it can highlight the differences to be discussed from a physical point of view.
Citation: https://doi.org/10.5194/egusphere-2022-134-RC2 -
EC1: 'Comment on egusphere-2022-134', Riccardo Farneti, 24 Jun 2022
Dear Author,
thank you for submitting your work to GMD.
Your manuscript has now been seen by 2 referees. Considering their advice, and my own evaluation, I regret to inform you that we cannot publish your manuscript in GMD. The Reviewers revealed major deficiencies and flaws that appear very unlikely to be fixed upon revision and I therefore discourage submission of a revised manuscript.
I am sorry that we cannot be more positive on this occasion and thank you for the opportunity to consider your work.
Citation: https://doi.org/10.5194/egusphere-2022-134-EC1
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 827 | 289 | 25 | 1,141 | 21 | 44 |
- HTML: 827
- PDF: 289
- XML: 25
- Total: 1,141
- BibTeX: 21
- EndNote: 44
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The author shows that (according to his assessment) the results of the multi-model mean (MMM) Global Average Temperature from CMIP5 and CMIP6 models according to the different scenarios (RCPs and SSPs) is well approximated by applying a simple linear regression fitted to historical observations of temperature and CO2 emissions, since 1976 and through 2005 or 2014.
I’m sorry to say I don’t see the value of this work and I recommend rejection.
First, GCMs produce hundreds of self-consistent variables and phenomena, so to propose that this linear regression on global temperature has value in substituting for complex GCMs is preposterous.
The comparison to make is with energy balance models, that through a simple function, as simple as a linear regression but accounting for physics and, importantly, feedbacks can indeed replicate global temperature and not only multi-model means but individual models that differ at a basic level by different equilibrium climate sensitivity. Some of this simple models importantly account for possible changes in the carbon cycle strength with warming, something a linear regression fitted on a short observational period can never do. The community has also, for many years now, recognized a strong linear dependence between global temperature and cumulative emissions. So, the idea that simple relationships can be found is not new, except here it is applied to the wrong pair.
The way the author shows agreement is extremely superficial and I question it at a fundamental level which is appreciable to the naked eye. The presentation of results is actually troubling in its hand-waviness. I don’t see good agreement in rates and even behavior for important scenarios like the highly mitigated (which are more and more the focus of the community). RCP2.6 stabilizes for CMIP5 MMM while it declines for the results of the regression, and strongly so. RCP4.5 MMM keeps increasing, even if at a slower rate, while the blue line stabilizes. The rate under RCP6.0 is completely different, with the two lines starting apart and ending close to one another. SSP1-2.6 rates of decrease are significantly different. SSP2-4.5 trajectories diverge at the end. I could go on. Why is the author not even bothering commenting on these discrepancies, let alone showing the behavior of residuals between the red and blue line, which would be obviously far from i.i.d., undermining the whole proposition of applying a linear model? The starting date of 1976 is also arbitrary and would demand some sensitivity analysis to the choice of the start date…but that is a detail that I don’t think is worth pursuing given the fundamental flaws of philosophy and performance one sees.