EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Heat stress indicators in CMIP6: Estimating future trends and exceedances of critical physiological thresholds

Clemens Schwingshackl, Jana Sillmann, Marit Sandstad, and Kristin Aunan
Clemens Schwingshackl et al.
  • CICERO, Oslo, Norway (

Global warming is leading to increased heat stress in many regions around the world. An extensive number of heat stress indicators has been developed to measure the associated impacts on human health. Here we calculate eight heat stress indicators for global climate models participating in the Coupled Model Intercomparison Project phase 6 (CMIP6) and compare their future trends and exceedances of critical physiological thresholds with particular focus on highly populated regions. The heat stress indicators are selected to represent a range of different applications, such as extreme heat events, heat-related losses in worker productivity, heat warnings, and heat-related morbidity and mortality. Projections of the analyzed heat stress indicators reveal that they increase significantly in all considered regions as function of global mean temperature. Moreover, heat stress indicators reveal a substantial spread ranging from trends close to the rate of global mean temperature up to an amplification of more than a factor of two. Consistently, exceedances of critical physiological thresholds are strongly increasing globally, including in several densely populated regions, but also show substantial spread across the selected heat stress indicators. Additionally, the indicators with the highest exceedance vary for different threshold levels, suggesting that the large indicator spread is associated both to differences in trend magnitude and threshold levels. The usage of heat stress indicators that are suitable for each specific application is thus crucial for reliably assessing impacts of future heat stress, while inappropriate indicators might lead to substantial biases.

How to cite: Schwingshackl, C., Sillmann, J., Sandstad, M., and Aunan, K.: Heat stress indicators in CMIP6: Estimating future trends and exceedances of critical physiological thresholds, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-8701,, 2020

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 1 – uploaded on 03 May 2020
  • CC1: Comment on EGU2020-8701, Dann Mitchell, 04 May 2020

    Thanks for the really nice presentation Clemens, there is some excellent analysis in there. I was wondering why the second part of your work focusses on thresholds as a proxy for health? There has been a fair amount of work over the last 5-10 years showing that health is NOT a threshold behaviour, i.e. you don't see high levels of mortality at 32C WBT, but low levels at 31C WBT; it is continous dsitribution. Just to be clear I think there is lots of value for looking at thresholds, as you do, I'm just wondering on the health-motivation for it? Many thanks

    • AC1: Reply to CC1, Clemens Schwingshackl, 04 May 2020

      Thank you for your comment, Dann, which touches an important point. The large spread of exceedance rates I showed in my short video already indicates that using thresholds is not ideal. However, only looking at trends in heat stress indicators does not give you enough information about health effects, which is also reflected in the fact that the indicators with the strongest trends are not necessarily the ones with the highest threshold exceedance rates. Moreover, heat stress indicators and exceedances of thresholds are still often used in climate studies, although using thresholds can be misleading, as you commented. The thresholds that I applied had all been used in other publications before.

      I did not have time to talk about this point in my presentation, but I agree that using these thresholds can be misleading. In the paper that we wrote about this topic (which is currently in review) we discuss different issues connected to the usage of thresholds and we try to raise awareness that it is not straightforward to use them - again underlined by the large spread across indicators. In the project EXHAUSTION, for which this study was performed, we are also planning to examine the relationship between different heat stress indicators and health data in more detail. I hope that with my study and the results we will obtain in that project we can make a step towards assessing global health impacts from climate change, which do not rely on threshold exceedances.

  • CC2: Comment on EGU2020-8701, Ana Maria Vicedo Cabrera, 05 May 2020

    Congratulations Clemens on the nice and interesting work. I agree with Dann and you that the use of thresholds as a way to define the association between heat stress and health can be misleading, although for public health purposes these are more applicable - meaning that still the assessment of thresholds in heat stress indicators can be of relevance (acknowledging the specific caveats). However, another critical point here that I would like to highlight is that it remains unclear whether the effect of heat on health is better captured by these heat stress indicators, compared to usual temperature metrics (e.g. mean temperature) - based on my preliminary analyses, still average temperature performed better (better predictability). I'm glad to see that this will be assessed within EXHAUSTION (looking forward to the results). It will be particularly relevant for health impact projections, as different geographical patterns would be expected, compared to already published studies using only average temperature.

    • AC2: Reply to CC2, Clemens Schwingshackl, 05 May 2020

      Thank you for your comment, Ana! With the data we used for this study (mainly climate model data) it is not possible to assess which indicators are more adequate to quantify health effects. For this reason, we decided to include some of the heat stress indicators for a more in-depth examination in the epidemiological part of EXHAUSTION. This will hopefully allow us to better evaluate the suitability of the different indicators to represent health effects.

  • CC3: Comment on EGU2020-8701, Ana Casanueva, 05 May 2020

    Thank you very much for the nice presentation! I have a question/comment. I am not surprised that you have such large uncertainty across indices; although they are highly correlated they are calculated differently and surpassing the thresholds, as you said, is what matters. So the indices themselves are not so comparable, I‘d say. Something that could help to disentangle a bit which indices provide a more robust signal could be to have a look at signal-to-noise ratios for each index separately.

    • AC3: Reply to CC3, Clemens Schwingshackl, 05 May 2020

      Thanks for bringing this question up again, as I could not answer it during the chat session. For the trends we used yearly maximum values of the heat stress indicators, and they all show statistically significant trends (p<0.01) in the analysed regions. In some regions there is indeed a higher variability for certain indicators, both year-to-year variability and variability across different climate models. Is that what you mean by signal-to-noise ratio?

      • CC6: Reply to AC3, Ana Casanueva, 05 May 2020

        I mean the signal to noise ratio as the ratio of the climate change signal for a given index (multi-model mean or median) divided by the multi-model spread (e.g. standard deviation of the model ensemble). Then maybe you can see how robust the projections are for each index, relative to their own projected signals, so it is kind of "standardized". Does it make sense to you?  I have in mind Fig.4 from

        • AC6: Reply to CC6, Clemens Schwingshackl, 08 May 2020

          Yes that makes sense to me. So it reflects the signal strength relative to the inter-model variability. I will have a look at it! Thank you very much for the suggestion!

  • CC4: Comment on EGU2020-8701, Audrey Brouillet, 05 May 2020

    Hi Clemens, very interesting work and presentation.

    (1) Why didn't you include the Wet-Bulb Globe Temperature (simplified or not) in your indicators comparison ? Although it's a criticized index, his wide used in the literature would have enable to compare previous corresponding studies to your new brain new study on CMIP6 outputs.

    (2) Did you apply any biases-correction since you work with absolute values ?

    (3) Have your explored other timescales such as the hottest month or annual extremes (e.g. 95th, 99th percentile) for your input of heat stress indicators  ?

    • AC4: Reply to CC4, Clemens Schwingshackl, 05 May 2020

      Hi Audrey, thank you for your questions. Here are my answers:
      (1) We actually did include the wet-bulb globe temperature, both the simplified version and a version calculated as weighted mean between dry-bulb and wet-bulb temperature (using the approach described by Davies-Jones, 2008 (doi:10.1175/2007MWR2224.1) to calculate wet-bulb temperature). In the figures, they are indicated by the variables TWBGs and TWBG.
      (2) For calculating threshold exceedcandes we applied quantile delta mapping (QDM) to the climate model data. We also compared the results to other bias adjustment methods (quantile mapping and multivariate bias correction) and found that QDM gives the most robust results. We apply QDM directly on the heat stress indicators and use ERA5 as reference dataset (1981-2010).
      (3) We have not explored other percentiles or timescales in this study.

      • CC5: Reply to AC4, Audrey Brouillet, 05 May 2020

        Thank you for you quick answers. Did you submit a publication already ? I look forward to read the corresponding paper if so

        • AC5: Reply to CC5, Clemens Schwingshackl, 05 May 2020

          We wrote a paper which is currently in review. I hope that we will get the results published at some time during summer.