To discuss the experimental design and the boundary conditions. The following reflects the state of the discussion after the 2014 Namur meeting:
Experimental design “Last Millennium” Transient simulations 850-1849 followed by historical experiments, set of boundary conditions for solar, volcanic, land-cover-change, greenhouse gases to be blended with those for historical (1850-2010) simulations. Updated PMIP3 protocol LM will build on DECK experiments, in particular the pre-industrial control simulation as unforced reference and the historical simulations. Moreover, LM provide initial conditions for historical simulations starting in the 19th century that are considered superior to the PiCtrl state as it includes integrated information from the forcing history (e.g. large volcanic eruptions in the early 19th century).
➢ Prioritization of the suggested experiments, including any rationale LM: one complete last millennium (850-2010) simulation using the standard boundary conditions and forcing. Groups are encouraged to run ensemble simulations as well as alternative forcings.
Boundary conditions: Please start discussion on updates/improvements w.r.t. PMIP3 here:
[ PMIP3 Wiki Home ] - [ Help! ] -
Michael Sigl and I are developing a new volcanic forcing data set for the past 2000 years, perhaps as far as 2.5k. The history is based on the AVS-2k Antarctic volcanic sulfate record (Sigl et al., 2014) and a similar Greenland composite. The forcing will have a realistic 2D (latitude, height) spatial structure, and we aim to provide a full set of self-consistent variables (AOD, extinction, effective radius, SAD, mass).
We plan to have the forcing set and some initial tests completed by the end of 2015. Michael will attend the splinter meeting at the EGU and will be happy to share more details, collect feedback, and learn more about other efforts. Of course, comments here or by email are also more than welcome. Thanks,
Matthew Toohey, GEOMAR, Kiel Germany
Sigl, M., McConnell, J. R., Toohey, M., Curran, M., Das, S. B., Edwards, R., Isaksson, E., Kawamura, K., Kipfstuhl, S., Krüger, K., Layman, L., Maselli, O. J., Motizuki, Y., Motoyama, H., Pasteris, D. R. and Severi, M.: Insights from Antarctica on volcanic forcing during the Common Era, Nat. Clim. Chang., 4, 693–697, doi:10.1038/nclimate2293, 2014.
Hello to those interested in the simulated past2k world,
I have, along with Gudrun Brättstrom at Stockholm University been investigating experimental design in climate model ensembles covering the Last Millennium (LM). We favour a factorial design approach as a basis for an experimental design, principally because have become interested in looking at interaction effects in the climate response to external forcings in LM ensemble simulations (at the moment we’re looking at the GISS LM ensemble).
While this work is ongoing, we thought we would at least share some of our thoughts seeing as we weren’t able to listen in on the EGU meet-up and haven't published anything about this yet. When you are testing for the Earth System’s response to a given set of forcings or boundary condition changes, it perhaps seems logical to conduct a series of independent studies varying those boundary conditions of interest. Such as comparing two simulations which differ only in terms of their land-use forcing.
With a full factorial design (in this case the factors are different forcings/boundary conditions) you can change two or more of these factors in order to observe the effect the changes have on one or more response variables (temperature, precipitation, etc).
Given the ongoing debates regarding not only the magnitude of forcings, but the response to those forcings (the latter particularly being part of CMIP6 scientific question (1)), a full factorial experimental design setup would be ideal to gauge the effects of each forcing in a model simulated ensemble as well as any interactions between forced responses (e.g. in temperature). This is best achieved with replicate simulations of each forcing/boundary condition combination. The other end of an experimental setup like this would be to generate only one or a few climate model simulations that include all the forcings and our best present guess of those forcings.
We realize that due to computer and time demands that is likely unrealistic for 1000 year transient model simulation runs to try every possible combination of factors, let alone replicates, but in this case a fractional factorial design can be used. This would mean focusing on several main effects (key factors/forcings that we expect to have important consequences for a climate (e.g. temperature response) and two-factor interactions rather than entertaining the possibility of many factors being important to climate response and the possibility of three-factor or more interactions.
Just our two cents to the discussion!
Alistair and Gudrun
As far as I can tell, last millenium simulations have all used fairly realistic boundary conditions (for pre-industrial part: solar, volcanic, land use). It would be very informative if for a given model, simulations could be made first with only a single forcing (i.e. solar only, volcanic only, land use only) as well as with combinations i.e. solar+volcanic, solar+land use, volcanic+land-use, and with all three together.
This would enable us to statistically evaluate the time scale ranges over which the boundary conditions act linearly and over which they act nonlinearly.
The GISS E2-R experiment comes close (control runs, solar only and solar plus volcanic), but is missing the volcanic only run needed to answer the question. It might be easy to make an additional run or two from a model that has already been run?
It's great to see that PMIP4 will also feature a Last Millennium experiment. Based on experiences with PMIP3 and subsequent developments, I have some comments and suggestions which I hope people find helpful:
1. Ensemble sizes
The Last Millennium experiment is fundamentally different from the other proposed PMIP4 experiments as (i) it is a transient experiment, and (ii) it is simulating a period when internal variability can dominate over any response to external forcings. One of the aims of the Last Millennium experiment is also to allow the relative contributions of internal variability and externally-forced changes to the studied. Under the circumstances, I think there is relatively little value in each model only being run once. I realise that running a state-of-the-art model for even one 1000-year simulation is a major undertaking and therefore that PMIP4 cannot require groups to conduct multiple simulations. However, I believe that each group should be “strongly encouraged” to run an ensemble of at least three simulations, and that this minimum ensemble size should be stated in the experimental design.
2. Single-forcing experiments
I concur with other comments on this page that single-forcing experiments are extremely valuable, not least so that they can be used as the basis for formal detection and attribution analyses. I would encourage PMIP4 to develop protocols for such experiments, which should form a “second tier” of experiments in addition to the core CMIP6/PMIP4 experiments.
3. Choice of boundary conditions
I presume that PMIP4 will provide multiple reconstructions of solar and volcanic activity, similarly to PMIP3. I would strongly encourage PMIP4 to identify ONE solar and ONE volcanic reconstruction as being the “preferred” choice. Most modelling groups will only be able to choose one combination of forcings. As PMIP is primarily a *model intercomparison* project, it is important that the models be run with consistent forcings. Of course, groups should be encouraged to choose as many different combinations of forcings as possible, so that the PMIP4 multi-model ensemble will span the full range of uncertainty. But, to enable a true intercomparison between models, it would be strongly preferrable for each model to be run at least once with a common set of forcings.
4. 2k experiments
With the volcanic reconstruction that has recently been developed by Sigl et al, all the boundary conditions now exist to enable transient simulations of the last 2k. Furthermore, from the data perspective, Phase 1 of the PAGES 2k Network has already generated continental-scale temperature reconstructions that span this period for at least some regions; between now and the end of 2016, Phase 2 of the PAGES 2k Network will be greatly expanding this effort. Thus all the ingredients are now in place for a genuine, large-scale data-model comparison exercise that spans the whole of the last 2k. I would therefore strongly encourage PMIP4 to develop a protocol for a “2k” experiment (this would be a very appropriate activity for the PAST2k group that already exists within PMIP!). Such an experiment could also form part of a “second tier” of experiments in addition to the core CMIP6/PMIP4 experiments.
Good luck! I would be happy to contribute towards the development of protocols for the additional experiments that I describe above. I also look forward to using the CSIRO Mk3L model to complete at least some of the PMIP4 experiments.