Can We Better Quantify the "Time of Emergence" of Regional Climate Change Signals using CMIP6?

Can We Better Quantify the “Time of Emergence” of Regional Climate Change Signals Using CMIP6?

Scientific Motivation

Understanding where, when, and under what conditions climate-force trends will become “measurable” (that is, distinguishable from background climate variability or noise) is extremely helpful for both validating our understanding of large scale climate dynamics and framing climate risk assessments. Prior works such as Hawkins and Sutton (2012) have analyzed ensembles of model simulations from inter-comparison projects and worked out some important details, such as the idea that, despite experience a lower magnitude of warming, the lower latitudes will likely experience a measurable “climate change signal” before higher latitudes - a result which has been replicated by works such as Mahlstein et al (2011).

Figure from Mahlstein et al (2011) showing how project signal/noise in warming varies by region of the globe
Figure from Mahlstein et al (2011) showing how project signal/noise in warming varies by region of the globe

To help understand and break down the time of emergence of climate-forced signals, it’s often helpful to perform spatial analyses, breaking up the globe into larger regions which we might expect to experience similar changes on similar timescales. Understanding what sets the spatial patterns of these regions is often then helpful in elucidating the dynamics at play which sets different regional climates apart from each other - and more interestingly, how those differences may play out in a warming world.

Proposed Hacking

We will replicate some of the foundational work by Hawkins and Sutton (2012) and Mahlstein and Rutti (2010) but with an emphasis on CMIP6 models. This will entail:

  1. Performing a spatial clustering analysis on different CMIP6 models to identify regions with similar baseline and climate trends; ideally we will explore using machine-learning techniques to “learn” these different regions across many different model simulations
  2. Analyzing climate trends by aggregating them across each region (a) model-by-model and (b) across models
  3. Develop visualizations and dashboards for exploring our results
  4. Create reproducible workflows that automate the entire analyses we develop

Based on this work we would hope to answer a few scientific questions:

  • Has our understanding of the time of emergency of regional warming/drying/wetting/etc trends changed with the data from CMIP6?
  • What regions and what trends might we expect to “emerge” first?

Anticipated Data Needs

TBD - we will use standard dynamic/thermodynamic diagnostics (air temperature, humidity, precipitation, wind components, 500mb geopotential heights etc); a list will be forthcoming with the Data Request thread.

Although it would be interesting to break down these analyses for many combinations of CMIP6 experiments, we will focus on pre-industrial runs (for quantifying the background noise/climate variability in the simulations) and future warming scenarios.

Anticipated Software Tools

We’ll try to use as much of the standard suite of scientific Python tools as possible. In particular, we expect to use xarray, dask, pandas, and scikit-learn. We will probably build some interactive visualization tools for helping to understand our results, and for that we will likely use bokeh and Panel.

Ideally the entire research chain will be automated with a few Jupyter notebooks and a Snakemake build (or similar workflow management tool) to enhance reproducibility. Any core modules will be developed as an open source package.

Desired Collaborators

Anyone! If you can bring some timeseries statistical analysis expertise that would be sweet - we’ll definitely need that to solve some of the core science problems!


This is a great proposal, and certainly the kind of big-picture science question I’d like to see addressed in CMIP6. It also caught my attention since I teach a Data Science/Data Analysis course where I’ve had students calculate time of emergence using a simplified version of this approach. The science questions are reasonably outside my expertise as an oceanographer, though one could ask similar questions about time of emergence in ocean variables as well.

Depending on what direction you go with this, I’d potentially be interested in being involved in the project itself, or in thinking about how reproducible versions of the workflow could be adapted for students (mine and others) to work with.

1 Like

This proposal definitely caught my eye. I think it’s interesting, thorough, and asks questions that the general public and policy makers are interested in. I’d probably be interested in working on a project like this. Not sure exactly what kind of timeseries / statistical analytical skills you are looking for, but depending on that, I might be able to contribute on that front. I am also interested in someday developing / teaching a statistics or analysis class, so anything that would improve instruction or dissemination of this kind of analysis would be interesting to me as well

1 Like

I’m interested in working on this proposal as well! I think time of emergence is an interesting topic to explore with the new CMIP6 models and I’m interested in developing some scripts that will help us answer those questions. I have some spatial clustering experience in R and some class experience with machine learning in python, and same as Marie, depending on what statistical time series analyses you would like to apply, I’d also hopefully be able to contribute there as well.

1 Like

To add to this, are you planning to use ScenarioMIP for future projections (and do you have specific scenarios in mind)?

Thanks for the feedback everyone! I’ve created a channel “#time-of-emergence” on the workshop Slack if folks want to iterate ideas there.

@feliciachiang that’s a really good question… it’s super helpful to study this question thinking about how different emissions pathways influence the time of emergence estimates, so we should absolutely look into any DECK experiments like ScenarioMIP which could provide useful data to study!

1 Like

Thanks for creating the channel, Daniel! Do you have other historical / future projection datasets in mind for the project? I’m not completely familiar with what’s available for CMIP6 at this point, and my current impression is that ScenarioMIP is replacing the RCPs that were used for CMIP5. We can also chat more on the slack channel!

1 Like

Awesome proposal! The scientific question has piqued my interest but I am also very interested in exploring the reproducible science aspect of this proposal. I don’t have the stats background you are looking for but am working on reproducible workflows and data provenance at the UK Met Office, so this project could be very relevant to my research! I’ll join the Slack channel to continue the discussion there.

1 Like

Hi all, I think it would great if people who are interested in this project could briefly get together this week and start formulating a high-level project plan/research outline. I’ve created a Doodle poll ->

Please feel free to indicate your availability! I’ll try to keep an eye on this and send out a potential time early Monday morning.