Register: Register as a ForceSMIP contributor to get access to the data. Register here. Registration is simply to help us keep track of contributors, and ForceSMIP is open to anyone. Registration remains open for participants who want to contribute to Tiers 2 and 3.
Download the training and evaluation data: After registering, you will receive a link to a repository containing the training and evalution data. The training data consists of 9 variables (tos, tas, pr, psl, siconc, zmta, monmaxtasmax, monmintasmin, monmaxpr; see definitions below) for 5 large ensembles (CanESM2, 25 members; CESM2, 50 members; MIROC6, 50 members; MIROC-ES2L, 30 members; MPI-ESM1-2-LR, 30 members) over the period 1880-2100. The evaluation data consists of 10-21 ensemble members (depending on the tier) chosen from 10 different earth system models and from observations/reanalysis. The evaluation data includes different variables for different time periods (tiers):
Tier 1 (1950-2022): tos, tas, pr, psl, zmta, monmaxtasmax, monmintasmin, monmaxpr
Tier 2 (1900-2023): tos, tas, pr (land only), psl
Tier 3 (1979-2023): tos, tas, pr, psl, mrso, siconc, monmaxtasmax
The metadata for the evaluation data is edited so that it doesn't show which model or observational dataset it came from. Please don't try to figure this out (there is no reason to cheat, and it will be clear if you do).
Develop code that estimates the forced component: Codes will be judged on their ability to estimate the forced response (see definition below) for each member and each variable of the evaluation data. In other words, you will estimate the forced response based on individual ensemble members, and the estimate will be judged based on how well it agrees with the ensemble mean of the corresponding large ensemble. You should make 21 forced response estimates - one for each evaluation member - for each variable and time period. You can develop the code in whichever programming language you choose. We encourage you to make use of the training data to train/tune your method.
Submit your contribution (Tier 1 deadline: March 1, 2024; Tiers 2 & 3 deadline: September 15, 2025): Once you have estimated the forced response for all variables in the evaluation data, please fill out the submission form (will be made available to registered participants). This will give you access to space on a Google Drive to upload your forced response estimate (netCDF preferred; please ask the organizers before submitting another data type) and a zip folder with your code. Note that a Tier 1 submission is not required to submit to Tiers 2 and 3. More than one submission per participant is allowed, but please ask the organizers before making more than 5 submissions. Each submission will need a unique name, which will be entered into the submission form and then used to label the forced response estimate and associated code in the Google Drive.
See the full protocol here.
Please email Robb Jnglin Wills (r.jnglinwills@usys.ethz.ch) or one of the other organizers if you have any questions about the process.
Forced response: Spatiotemporal evolution of anomalies in a climate quantity (either a monthly average or a monthly max or min) in response to external forcing (GHG, aerosols, volcanoes, land-use change, solar, etc.) once internal variability in that climate quantity has been removed. In the case of large ensemble data, the forced response will be defined by the ensemble mean.
Variables:
tos: monthly-mean sea-surface temperature
tas: monthly-mean surface air temperature
pr: monthly-mean precipitation
psl: monthly-mean sea-level pressure
mrso: total soil moisture content
siconc: monthly-mean sea-ice concentration
zmta: monthly-mean zonal-mean air temperature
monmaxtasmax: monthly maximum of daily maximum surface air temperature
monmintasmin: monthly minimum of daily minimum surface air temperature
monmaxpr: monthly maximum of daily total precipitation