ESPO-G6-R2 : Ensemble de Simulations Post-traites dOuranos - modles Globaux CMIP6 - RDRS v2.1 / Ouranos Ensemble of Bias-adjusted Simulations - Global models CMIP6 - RDRS v2.1

ESPO-G6:

ESPO-G6-R2 v1.0.0:

ESPO-G6-E5L v1.0.0:

The dataset is described in the following article:

Lavoie et al., An ensemble of bias-adjusted CMIP6 climate simulations based on a high-resolution North American reanalysis. Nature Scientific Data. 10.1038/s41597-023-02855-z (2024). https://www.nature.com/articles/s41597-023-02855-z

Context

The need to adapt to climate change is present in a growing number of fields, leading to an increase in the demand for climate scenarios for often interrelated sectors of activity. In order to meet this growing demand and to ensure the availability of climate scenarios responding to numerous vulnerability, impact, and adaptation (VIA) studies, Ouranos is working to create a set of operational multipurpose climate scenarios called "Ensemble de Simulations Post-traites d'Ouranos" (ESPO) covering North America at a resolution of 0.1. In ESPO-G6-R2 v1.0.0, CMIP6 global climate models simulations are bias-adjusted using the RDRS v2.1 reference dataset. The simulation ensemble covers the period for years 1950-2100 and includes the daily minimum temperature (tasmin), the daily maximum temperature (tasmax) and the daily mean precipitation flux (pr). The experiments included are SSP2-4.5 and SSP3-7.0.

To avoid the "hot model problem" (Hausfather et al, 2022), only models with a Transient Climate Response (TCR) in the likely range (1.42.2 C) were kept in the official ensemble (Table 1). Extra "hot models" and SSP5-8.5 are also available even if they are not in the official ensemble.

Table 1. Members of ESPO-G6-R2 v1.0.0

|Institution|Model |Member |License|TCR (degC)|In TCR likely range|Status| |---|---|---|---|---|---|---| |CAS | FGOALS-g3 | r1i1p1f1 |CC BY 4.0|1.50|| Completed| |CMCC |CMCC-ESM2 | r1i1p1f1 |CC BY 4.0|1.92||Completed| | CSIRO-ARCCSS | ACCESS-CM2 | r1i1p1f1 |CC BY 4.0|1.96||Completed| | CSIRO |ACCESS-ESM1-5 | r1i1p1f1 |CC BY 4.0|1.97||Completed| | DKRZ | MPI-ESM1-2-HR | r1i1p1f1 |CC BY 4.0|1,64||Completed| | INM |INM-CM5-0 | r1i1p1f1 |CC BY 4.0|1.41||Completed| | MIROC | MIROC6 | r1i1p1f1 |CC BY 4.0|1.55||Completed| | MPI-M | MPI-ESM1-2-LR | r1i1p1f1 |CC BY 4.0|1.82||Completed| | MRI | MRI-ESM2-0 | r1i1p1f1 |CC BY 4.0|1.67||Completed| | NCC | NorESM2-LM | r1i1p1f1 |CC BY 4.0|1.49||Completed| | CNRM-CERFACS | CNRM-ESM2-1 | r1i1p1f2 |CC BY 4.0|1.83||Completed| | NIMS-KMA | KACE-1-0-G | r1i1p1f1 |CC BY 4.0|2.04||Completed| | NOAA-GFDL | GFDL-ESM4 | r1i1p1f1 |CC BY 4.0|1.63||Completed| | BCC | BCC-CSM2-MR | r1i1p1f1 |CC BY 4.0|1.55||Completed| | CCCma | CanESM5 | r1i1p1f1 |CC BY 4.0|2.71| |Completed| | CNRM-CERFACS | CNRM-CM6-1 |r1i1p1f2 |CC BY 4.0|2.22| |Completed| | EC-Earth-Consortium | EC-Earth3 |r1i1p1f1 |CC BY 4.0|2.30| |Completed| | IPSL | IPSL-CM6A-LR | r1i1p1f1 |CC BY 4.0|2.35| |Completed| | MOHC | UKESM1-0-LL | r1i1p1f2 |CC BY 4.0|2.77| |Completed| | NCC |NorESM2-MM | r1i1p1f1 |CC BY 4.0|1.22| |Completed| | EC-Earth-Consortium | EC-Earth3-CC |r1i1p1f1 |CC BY 4.0|2.63| |Completed (no SSP3-7.0)| | NUIST |NESM3 | r1i1p1f1 |CC BY 4.0|2.72| |Completed (no SSP3-7.0)| | MIROC |MIROC-ES2L | r1i1p1f2 |CC BY 4.0|1.49| |Completed | | EC-Earth-Consortium |EC-Earth3-Veg | r1i1p1f1 |CC BY 4.0|2.66| |Completed | | INM |INM-CM4-8 | r1i1p1f1 |CC BY 4.0|1.30| |Completed | | AS-RCEC |TaiESM1 | r1i1p1f1 |CC BY 4.0|1.30| |Completed |

Licences: https://wcrp-cmip.github.io/CMIP6CVs/docs/CMIP6sourceidlicenses.html

TCR: Hausfather et al. 2022, Climate simulations: recognize the 'hot model' problem, comment in Nature: DOI: 10.5281/zenodo.6476375

Spatial coverage

The dataset has a resolution of 0.1 over a North American domain on a rotated grid with a grid pole latitude of 31.7583 and longitude of 87.5970. The latitude range covered is from 5.7560N to 83.9816N and the longitude range is from 179.9728E to 9.0204W.

:warning: Users should be careful with precipitation data close to the south edge of the domain where there is less trust in the reference data.

Data is only available on or near land. A mask was created by removing all grid cells that had a sea area fraction of 1 in the reference dataset and then putting back a buffer of one grid cell along the coasts.

Some small regions in Alaska and Greenland have been masked out by NaNs for 2 models. More details are available in section 5 of the documentation of the adjustment method.

Temporal coverage

As the bias-adjustment method requires a consistent number of calendar days (no leap days), all members using a standard calendar were converted to the noleap one by dropping any values for February 29th. KACE-1-0-G is the only model simulated with a 360-day calendar, and was kept as is.

The bias-adjustment was calibrated over the years 1989-2018, the most recont 30-year period available, and applied to the full 1950-2100 period.

Reference data

The ESPO-G6-R2 v1.0.0 dataset uses the RDRS v2.1 (Gasset et al., 2021) as reference dataset. This is a product from Environment and Climate Change Canada (ECCC) created by using the Regional Deterministic Reforecast System (RDRS) to downscale the Global Deterministic Reforecast System (GDRS) initialized by ERA-Interim. The system is also coupled with the Canadian Land Data Assimilation System (CaLDAS) and Precipitation Analysis (CaPA). It was downloaded from CaSPAR.

:warning: A different version of ESPO-G6 uses ERA5-Land as a reference. The information for ESPO-G6-E5L can be found in this release with doi:10.5281/zenodo.7764929.

Methodology

The temperature and precipitation data from the simulations in Table 1 were first extracted over North America. Then, all the extracted simulation data are interpolated bilinearly in cascades to the RDRS v2.1 grid. The ESPO-G6-R2 v.1.0.0 bias adjustment procedure then uses xclim's bias adjustment algorithms to adjust simulation bias following a quantile mapping procedure. In particular, the algorithm used is inspired by the "Detrended Quantile Mapping" (DQM) method described by Cannon (2015). The procedure is bipartite; First, the adjustment factors are calculated based on reference data and simulations over a common period (training stage), and second, the entire simulation is corrected with these factors (adjustment step). The reference period chosen here were years 1989-2018. Adjustments are univariate, where corrections are applied separately for each of the 3 variables. Data is adjusted for each day of the year, using a rolling window of 31 days. Although computational more expensive, the rolling window method allows for better adjustment of the annual cycle. Note that this method does not work well with leap years as there is four (4) times fewer data values for day 366. To remedy this problem, all simulations as well as the reference product are converted to this "noleap" calendar. A more detailed explanation of the adjustment process is given in the documentation.

Data processing tools

Production and regular updates of ESPO-R/G operational datasets represent a challenge in terms of computational resources. Ouranos has invested a great deal of effort in the development of powerful tools for this type of data processing via its xclim software package (Logan et al., 2021). Built upon the packages xarray and dask, xclim benefits from simple-to-use parallelization and distributed computing tools and can be easily deployed in High Performance Computing (HPC) environments.

This repository contains the code used to generate and analyze the ESPO-G datasets. In addition to xclim and other freely available python libraries, it also uses xscen, a climate change scenario-building analysis framework, also being developed at Ouranos. This tool relies on data catalogs as handled by intake-esm as well as on YAML configuration files with a simple but specific structure. The catalog files and all paths needed by the configuration are missing from this repository, since they are specific to the data architecture of the computer running the code. To reproduce ESPO-G, one will need:

• simulation.json and simulation.csv: An intake-esm catalog, compatible with xscen, listing the daily simulation datasets to use as inputs.
• reconstruction.json and reconstruction.csv: An intake-esm catalog, compatible with xscen, listing the daily reference datasets to use as inputs.
• paths.yml: A yaml file with the paths needed by the workflows. configuration/template_paths.yml shows an example of such a file, one only needs to replace the placeholders.

To run the workflow, uncomment the tasks wanted at the top of config file. Then, run

python workflow_ESPO-G.py

Description of the tasks: - makeref: Create the reference dataset with the right domain, period and calendar. - extract: Extract the simulation dataset with the right domain and period. - regrid: Regrid the simulation onto the reference grid. - rechunk: Rechunk the regridded dataset to prepare for the bias adjustment (needed on large datasets). - train: Train the bias adjustment algorithm. - adjust: Adjust the simulation dataset with the trained bias adjustment algorithm. - cleanup: Join each individually adjusted variable back in one scenario dataset and clean up other details. - finalzarr: Rechunk the scenario dataset and save it. - diagnostics: Compute simple diagnostics (defined in configuration/propertiesESPO-G.yml) on the whole domain for a quality check. - concat: Concatenate scenario and diagnostics of the three regions into the complete NAM domain.
- official-diag: Compute diagnostics (defined in configuration/off-propertiesESPO-G.yml) on smaller regions to assess the performance. - indicators: Compute indicators (defined in configuration/portraits.yml) on the scenario. - climatological_mean: Compute the climatological mean of the indicators. - delta: Compute the deltas of the climatological means. - ensemble: Compute the ensemble statistics.

Loop structure: ``` makeref for sim in simulations: # iterate over all models and experiments # We split the domain in three in order to have less computationnaly expensive task. for region in regions: # iterate over 3 sub-regions of NAM. extract regrid rechunk for var in variable: train adjust cleanup finalzarr diagnostics concat

tasks below iterate inside the task

official diag indicators climatological_mean delta ensembles ```

Performance

Bias-adjustment of climate simulations is a quest with many traps. In order to assess the improvements and regressions that the process brought to the simulations, we emulated the "VALUE" validation framework (Maraun et al., 2015). While that project aimed to "to validate and compare downscaling methods", we based our approach on its ideas of statistical "properties" and "measures" to measure bias between the simulations, the scenarios, and the reference.

A detailed analysis is given in the documentation. Our general conclusions concerning the quality of ESPO-G6-R2 v1.0.0 are:

• The marginal properties of the simulations (mean, quantiles) are very well-adjusted, by design of the Quantile Mapping algorithm.
• The climate change signal is also conserved from the simulations by design of the algorithm.
• A side effect of adjusting the distributions explicitly is the improvement of the inter-variable correlation, even though the bias correction algorithm does not aim to adjust these aspects.
• Because tasmin is not directly adjusted, but rather computed from the adjusted tasmax and dtr, it seems that our diagnostics show weaker improvements, compared to tasmax.

Data availability and download

The ESPO-G6-R2 v1.0.0 data is currently available under the CC BY 4.0 license.

At the time of publication, the data is stored on Ouranos' THREDDS server, a part of the PAVICS project: https://pavics.ouranos.ca/twitcher/ows/proxy/thredds/catalog/datasets/simulations/bias_adjusted/cmip6/ouranos/ESPO-G/ESPO-G6-R2v1.0.0/catalog.html

When new versions of ESPO-G will be released, previous versions may be pulled from the server. Please contact us if you wish to obtain these.

Acknowledgements

We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF.

References

Cannon, A. J., Sobie, S. R., & Murdock, T. Q. (2015). Bias correction of GCM precipitation by quantile mapping: How well do methods preserve changes in quantiles and extremes? Journal of Climate, 28(17), 69386959. https://doi.org/10.1175/JCLI-D-14-00754.1

Gasset, N., Fortin, V., Dimitrijevic, M., Carrera, M., Bilodeau, B., Muncaster, R., Etienne Gaborit, Roy, G., Pentcheva, N., Bulat, M., Wang, X., Pavlovic, R., Lespinas, F., Khedhaouiria, D., (2021). A 10 km north american precipitation and land surface reanalysis based on the gem atmospheric model. Hydrology and Earth System Sciences. doi:10.5194/hess-2021-41.

Hausfather, Z., Marvel, K., Schmidt, G. A., Nielsen-Gammon, J. W., Zelinka, M. (2022). Climate simulations: recognize the hot model problem. Nature 2022 605:7908, 605(7908), 2629. https://doi.org/10.1038/d41586-022-01192-2

Logan, T., Bourgault, P., Smith, T. J., Huard, D., Biner, S., Labont, M.-P., Rondeau-Genesse, G., Fyke, J., Aoun, A., Roy, P., Ehbrecht, C., Caron, D., Stephens, A., Whelan, C., Low, J.-F., Keel, T., Lavoie, J., Tanguy, M., Barnes, C., Quinn, J. (2022). Ouranosinc/xclim (0.35.0) [Python]. Zenodo. https://doi.org/10.5281/zenodo.6407112

Maraun, D., Widmann, M., Gutirrez, J.M., Kotlarski, S., Chandler, R. E., Hertig, E., Wibig, J., Huth, R., Wilcke, R. A. I. (2015). VALUE: A Framework to Validate Downscaling Approaches for Climate Change Studies. Earths Future 3, 1, 114. https://doi.org/10.1002/2014EF000259.

Zhuang, J., Dussin, R., Huard, D., Bourgault, P., Banihirwe, A., Raynaud, S., Malevich, B., Schupfner, M., Hamman, J., Levang, S., Jling, A., Almansi, M., Fernandes, F., Rondeau-Genesse, G., Rasp, S., & Bell, R. (2021). pangeo-data/xESMF (0.6.2) [Python]. Zenodo. https://doi.org/10.5281/zenodo.5721118