Quantifying CMIP6 Model Uncertainties in Extreme Precipitation Projections

From Weather and Climate Extremes

AmalJohn a c HervéDouville a AurélienRibes a PascalYiou b

a Centre National de Recherches Météorologiques, Météo-France, CNRS, Toulouse, France
b Laboratoire des Sciences du Climat et de l’Environnement, UMR 8212 CEA-CNRS-UVSQ, IPSL & U Paris-Saclay, 91191 Gif-sur-Yvette, France
c Université de Toulouse, France

Received 8 October 2021, Revised 9 February 2022, Accepted 10 March 2022, Available online 21 March 2022, Version of Record 26 March 2022.


https://doi.org/10.1016/j.wace.2022.100435 Get rights and content
Under a Creative Commons license Open access

Abstract

Projected changes in precipitation extremes and their uncertainties are evaluated using an ensemble of global climate models from phase 6 of the Coupled Model Intercomparison Project (CMIP). They are scaled by corresponding changes either in global mean surface temperature (ΔGSAT) or in local surface temperature (ΔT) and are expressed in terms of 20-yr return values (RV20) of annual maximum one-day precipitation. Our main objective is to quantify the model response uncertainty and to highlight the regions where changes may not be consistent with the widely used assumption of a Clausius–Clapeyron (CC) rate of ≈7%/K. When using a single realization for each model, as in the latest report from the Intergovernmental Panel on Climate Change (IPCC), the assessed inter-model spread includes both model uncertainty and internal variability, which can be however assessed separately using a large ensemble. Despite the overestimated inter-model spread, our results show a robust enhancement of extreme precipitation with more than 90% of models simulating an increase of RV20. Moreover, this increase is consistent with the CC rate of ≈7%/K over about 83% of the global land domain when scaled by (ΔGSAT). Our results also advocate for producing multiple single model initial condition ensembles in the next CMIP projections, to better filter internal variability out in estimating the response of extreme events.

1. Introduction

Global climate models provide an increasingly comprehensive representation of the climate system and are used as a primary tool for understanding and projecting changes in climate mean, variability and extremes due to human activities. The Intergovernmental Panel on Climate Change (IPCC) in its sixth assessment report (AR6) has re-estimated an increase in the observed global mean surface temperature of 1.09 °C in 2011–2020 relative to the beginning of the industrial revolution (1850–1900), which can be fully attributed to a human influence (IPCC AR6 SPM Masson-Delmotte et al. (2021)). This anthropogenic global warming is reckoned to have long-term consequences on all components of the climate system, including changes in the daily precipitation distribution. Several generations of multi-model simulations contributing to the Coupled Model Intercomparison Project (CMIP), supported by observational evidence, show that both the frequency and intensity of extreme daily precipitation events have increased over recent decades (Allen and Ingram, 2002Asadieh and Krakauer, 2015Scherrer et al., 2016Karl and Easterling, 1999Kharin et al., 2013Min et al., 2011O’Gorman, 2015). This is also documented in the IPCC special report on Managing the Risks of Extremes Events to Advance Climate Change Adaptation (SREX, Seneviratne (2012)).

In the absence of moisture limitation and of significant dynamical response, the extreme precipitation intensity is expected to increase exponentially with the atmospheric temperature at a rate determined by the Clausius–Clapeyron (CC) relationship. A robust scaling of daily precipitation extremes with global warming across scenarios was confirmed by Li et al. (2020) who found that changes in precipitation extremes follow changes in global warming at roughly the CC rate of ≈7%/°C in the latest-generation CMIP6 models. Several studies based on climate model simulations show a future increase of precipitation extremes with temperature at a rate comparable to or higher than the CC rate (Li et al., 2020Kharin et al., 2007Pall et al., 2007Allan and Soden, 2008Sugiyama et al., 2010Kao and Ganguly, 2011Muller et al., 2011). However, wet extremes are not expected to intensify in all regions (Trenberth, 2011Pfahl et al., 2017).

All these studies either show the multi-model mean or median and have not yet assessed the uncertainties in global CMIP6 projections. A suite of different model projections often exhibits a large spread (Lehner et al., 2020) and can even disagree on a particular region becoming wetter or drier (sign change in the future). Even where there is an overall consensus among the models on the sign of changes in the projected extremes due to a warmer climate, the magnitude of such changes can differ considerably. Though the climate models have improved over recent decades (Wyser et al., 2020Zelinka et al., 2020), these improvements do not necessarily result in a reduced spread among the projections (Douville et al., 2021). Thus, the main focus of this paper is to quantify the model uncertainties in extreme precipitation projections based on CMIP6 models. We also aim to provide a blueprint on using these projections to identify regions where the projected changes in daily precipitation extremes are consistent with the CC rate and those where they are not.

Changes in extreme precipitation against a backdrop of warming climate arise both due to thermodynamic and dynamic effects (Pfahl et al., 2017). A sub-CC relation or even negative dependence on global mean temperature has been found for precipitation extremes over some regions, especially over the climatologically dry oceanic regions in the subtropics, presumably as a result of decreasing moisture availability and enhanced large-scale subsidence (Berg et al., 2009Hardwick Jones et al., 2010Utsumi et al., 2011Pfahl et al., 2017). But the question of an appropriate choice of temperature for scaling extreme precipitation is still an open question and the available studies differ in scope (Zhang et al., 2019Schroeer and Kirchengast, 2018Sun et al., 2021). There is a large-scale warming contrast between the continental landmass and the oceans with certain regions over the ocean experiencing a negligible or limited change in the projected surface temperature. The larger warming observed over land may result in a lower scaling with local mean temperature, which may not be considered as a sub-CC scaling rate (Wang et al., 2017). Any departure from the CC rate can be an indication of a dynamical response which may be either amplified or offset by a thermodynamic response regionally (Pfahl et al., 2017Sherwood et al., 2010O’Gorman, 2015). Thus here we explore changes in extreme precipitation simply scaled by either global mean or local surface air temperature changes.

Several studies (Alexander et al., 2006Tebaldi et al., 2006Sillmann et al., 2013aSillmann et al., 2013b) have used various indices as a proxy for different features of precipitation extremes. Here we focus on extreme events with typical return periods of 20 years (or 20-year return values, RV20) as estimated from the annual maximum one-day precipitation (RX1DAY). Projected long-period RX1DAY return value changes are larger than changes in mean RX1DAY and increase with increasing rarity (Mizuta and Endo, 2020Wehner, 2020). Here we did not explore longer (e.g., 50 or 100 years) return periods since the associated uncertainties would be even stronger than for our RV20 estimations due to the limited sampling.

The goal of this study is to assess the uncertainties of projected changes in extreme precipitation based on the multi-model CMIP6 ensemble, to discuss the limitations of assessing the inter-model spread using such ensembles of opportunity, and to highlight the regions where projected changes may not be consistent with the widely used assumption of a Clausius–Clapeyron rate of ≈7%/K (Kharin et al., 2013Westra et al., 2013Seneviratne et al., 2021). For this purpose, we use the SSP5-8.5 scenario from 35 CMIP6 models. The total spread in this ensemble is therefore a combination of both model response uncertainty and internal variability. Therefore, we also assess the potential contribution of internal variability to the inter-model spread by analyzing the projected changes of the RV20 in the CanESM5 model with 25 realizations, with the same concentration scenario.

The rest of the paper is structured as follows. We start by introducing in Section 2 the models and methods used in this study. Turning to the results in Section 3, we address the uncertainties in the model projections along with a discussion on the role of internal variability using the ensemble simulations from CanESM5. The role of local versus global temperature scaling is also assessed. Section 4 summarizes the main findings. Other supporting figures and tables are available in the online supplementary material.

Read the full paper here.

5 1 vote
Article Rating
29 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Pat from kerbob
April 14, 2022 6:22 pm

I rubbed the old magic 8 ball and got results just as relevant.
Send money

TonyL
Reply to  Pat from kerbob
April 14, 2022 7:20 pm

Why do you need money? Did you you break your magic 8 ball?
Try to be more careful in the future.

Pat from kerbob
Reply to  TonyL
April 14, 2022 7:51 pm

Dropped it due to shocking revelations

H.R.
Reply to  Pat from kerbob
April 14, 2022 9:03 pm

If it was a real Magic 8 Ball, Pat, it would have told you that you were going to drop it.

I think you had one of those fake ones that they use to model climate.

TonyL
April 14, 2022 6:30 pm

Read through the jargon.
What I get is this:
The more the models converge and agree with each other, the more “robust” the projection.
Here, the term “robust” refers not to any purported accuracy of the projection, but rather as to how well the different projections within the ensemble agree with each other.

News Flash:
We all agree with each other, that proves we are right! Bonus: Anybody else who also agrees with us is further proof of how right we are! So take that, you contrarians.

Also note the notorious RCP 8.5 makes its appearance, the choice for the projection set to use.
How about that.

Reply to  TonyL
April 14, 2022 6:45 pm

This is what happens when the trust in authority is greater than the trust in truth, especially when multiple ‘authorities’ agree, although anyone who believes that models are authoritative will never know the truth.

ihfan
Reply to  TonyL
April 14, 2022 6:58 pm

Confirmation bias, expressed as a series of algorithms, all converging on the absurd.

TonyL
Reply to  ihfan
April 14, 2022 7:16 pm

I like it, well put.

Old Man Winter
Reply to  TonyL
April 14, 2022 7:27 pm

Great summary!

Reply to  TonyL
April 16, 2022 1:28 am

Nicely put. The science of the climate is replaced by the study of the models. Modeling is bad enough but now the model outputs are the data! Nothing at all about real climate.

Mike
April 14, 2022 7:57 pm

Projected changes in precipitation extremes and their uncertainties are evaluated using an ensemble of global climate models

Yeah. It hasn’t happened yet but it will happen. You can take that to the bank!

H.R.
Reply to  Mike
April 14, 2022 9:18 pm

I’ll add that to the list of failed predictions.

Lessee here… page 37…., about halfway down… there!

lee
Reply to  Mike
April 14, 2022 11:03 pm

Because we know that ensembles correct for any errors. 🙄

Last edited 2 months ago by lee
MARTIN BRUMBY
April 14, 2022 9:39 pm

So, models all the way down.
As usual.

I think that if they went back to the old study of tea leaves and chicken entrails, their “projections” would be just as accurate and convincing.

And they’d save loads of electricity and be able to have nice chicken suppers.

Why on earth are we paying taxpayer’s good money on this hocus pocus, anyway?

We know the result. “Worse than we thought”. As always. These psyentists are oxygen thieves.

TonyL
Reply to  MARTIN BRUMBY
April 14, 2022 10:34 pm

And they’d save loads of electricity and be able to have nice chicken suppers.

When you put it like that, your proposal makes much good sense.

H.R.
Reply to  MARTIN BRUMBY
April 15, 2022 9:13 am

MARTIN B: “Why on earth are we paying taxpayer’s good money on this hocus pocus, anyway?”


It just struck me why there are so many studies and grants for mostly model-based garbage, and they are certainly not science.

It goes back to “Tell a big lie often enough…”

The purpose it is to crank out a high volume of studies for the “often enough” part of making people believe the big lie.

And that’s why WUWT could post half a dozen “it’s-bad-really-really-bad” studies every day if Anthony and CTM were so inclined. I think the editorial challenge here at WUWT is picking which nonsense ‘study’ to post.

[edited for typo]

Last edited 2 months ago by H.R.
Waza
April 15, 2022 1:42 am

Can anyone direct me to an interactive workshop where I can see climate modellers present this paper on increased extreme precipitation followed by another group of climate modellers present their equivalent paper on the increase of droughts.

Oldseadog
Reply to  Waza
April 15, 2022 1:56 am

Both papers would be presented by one group who would have taken a coffee break between presentations, and the group would have just finished reading Alice In Wonderland and Alice Through The Looking Glass so would be able to believe in two impossible things at the same time.

Reply to  Waza
April 16, 2022 7:33 am

Not a workshop, but an example of submission regarding one model, INMCM5. The paper is:Simulation of Possible Future Climate Changes in the 21st Century in the INM-CM5 Climate Model 
https://link.springer.com/article/10.1134/S0001433820030123

Abstract
Climate changes in 2015–2100 have been simulated with the use of the INM-CM5 climate model following four scenarios: SSP1-2.6, SSP2-4.5, and SSP5-8.5 (single model runs) and SSP3-7.0 (an ensemble of five model runs). Changes in the global mean temperature and spatial distribution of temperature and precipitation are analyzed. The global warming predicted by the INM-CM5 model in the scenarios considered is smaller than that in other CMIP6 models. It is shown that the temperature in the hottest summer month can rise more quickly than the seasonal mean temperature in Russia. An analysis of a change in Arctic sea ice shows no complete Arctic summer ice melting in the 21st century under any model scenario. Changes in the meridional stream function in atmosphere and ocean are studied.

My Synopsis
https://rclutz.com/2020/07/24/best-climate-model-mild-warming-forecasted/

Joao Martins
April 15, 2022 3:00 am

Better repair and improve infrastructures than loosing money and time in guesswork.
Engineering is needed, not voodoo-science.

Last edited 2 months ago by Joao Martins
DMacKenzie
Reply to  Joao Martins
April 15, 2022 8:52 am

I personally countered a few years of sea level rise by putting a 1/4” of stucco on top of the beach wall…Maybe next time I’ll plan on a century worth and use a concrete block.

Petit_Barde
April 15, 2022 3:25 am

Atmospheric CO2 concentration has increased from 360 to 410 ppm with no extreme event increase, including extreme precipitations. There is no trend whatsoever.

How could it be that in the future CO2 concentration increase should magically induce extreme events, be it extreme precipitations, flooding, draughts, heat waves, cold snaps, etc. ?

Doesn’t that show that climate models are wrong ?

duane
April 15, 2022 5:07 am

Now how many angels does your model say can dance on the head of a pin?

Because the answer to that question would be just as valuable to humanity as the rest of your model outputs.

buck smith
April 15, 2022 7:04 am

Increased precipitation is a negative feedback for warming, no?

Gordon A. Dressler
April 15, 2022 7:41 am

Looking at uncertainties in model predictions is, of course, important.

However, I humbly suggest that one should first start by examining the median/mean predicted trend of any given climate model, let alone an ensemble of climate models, expressed as °C/year, or more specific to the above article, an increase in precipitation intensity per year.

If climate model projections can depart from measured data trends by factors of 3 to 8, as is the case for the 25+ climate models used for CMIP-x reports from the IPCC (ref: https://wattsupwiththat.com/2022/03/11/climate-model-democracy/ ), then looking at uncertainties in the model predictions is pretty meaningless.

DMacKenzie
April 15, 2022 8:45 am

A study of studies of models (acorn a little far from tree maybe ?) that says the 7%/degree likely causes 5% more extremes….Is this really anything more statistically significant than picking any number between 0% and 7% ? Not with rainfall “event” variations from -70% to +270% per “record”…

Last edited 2 months ago by DMacKenzie
Mark BLR
April 15, 2022 9:56 am

From the “Introduction” section of the paper :

The Intergovernmental Panel on Climate Change (IPCC) in its sixth assessment report (AR6) has re-estimated an increase in the observed global mean surface temperature of 1.09°C in 2011–2020 relative to the beginning of the industrial revolution (1850–1900), which can be fully attributed to a human influence (IPCC AR6 SPM Masson-Delmotte et al. (2021)).

NB : Technically this is correct, as shown in Figure SPM.2 (at the top of page SPM-8).
“Observed warming” ~= the temperature change attributed to “Total human influence” ~= a 1.1°C increase in GMST (from 1850-1900 to 2010-2019).

One question that comes to (my incredibly cynical) mind, though, is :
“The paper is about precipitation changes, why are you talking about temperature ?”.

Another extract from the paper :

The goal of this study is to assess the uncertainties of projected changes in extreme precipitation based on the multi-model CMIP6 ensemble, to discuss the limitations of assessing the inter-model spread using such ensembles of opportunity, and to highlight the regions where projected changes may not be consistent with the widely used assumption of a Clausius–Clapeyron rate of ≈7%/K (Kharin et al., 2013; Westra et al., 2013; Seneviratne et al., 2021). For this purpose, we use the SSP5-8.5 scenario from 35 CMIP6 models.

Weeeeeeeell, OK.

If you want to check a set of models against each other, you use RCP8.5 because that gives you the biggest “signal” and the largest amount of “internal (model) variability” to analyse.

It still doesn’t explain why you aren’t even attempting to compare the “hindcast / Historical Data” model runs against the precipitation that actually occurred in the past on the real-life planet Earth …

Figure SPM.3 (on page SPM-12), “Green hexagons indicate regions where there is at least medium confidence in an observed increase in heavy precipitation.”

[ Enter “sarcastic” mode” ]

Yes, but it’s a shame that there’s only one hexagon (NEU / Northern Europe) with a 3-star “High” rating when it comes to the IPCC’s “Confidence in human contribution” to that observed precipitation pattern, one (CNA / Central North America) with a 2-star “Medium” rating and all of the (43 ?) others only have a 1-star “Low” rating, isn’t it …

[ Exit “sarcastic” mode” ]

More seriously, what does the IPCC have to say about the CMIP6 model ensemble and comparisons of precipitation “hindcasts” against actual observations ?

The start of the “Box SPM.1.2” paragraph (on page SPM-15) :

This report assesses results from climate models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) of the World Climate Research Programme. These models include new and better representation of physical, chemical and biological processes, as well as higher resolution, compared to climate models considered in previous IPCC assessment reports. This has improved the simulation of the recent mean state of most large-scale indicators of climate change and many other aspects across the climate system. Some differences from observations remain, for example in regional precipitation patterns.

Ah ! OK … forget I asked …

Last edited 2 months ago by Mark BLR
Andy Pattullo
April 15, 2022 10:24 am

90% of our useless models which we made ourselves with no help from Mom and Dad show the exact problem we wanted to warn you about, so be scared, be very scared.

%d bloggers like this: