CMIP6 multi-model ocean datasets technical notes

This page provides the technical documentation for the gridded ocean datasets of Coupled Model Intercomparison Project Phase 6 (CMIP6) global climate models (GCMs) made available on the Canadian Climate Data and Scenarios (CCDS) website. The documentation includes a description of the CMIP6 ocean multi-model datasets, processing methodology, and a list of models included in the ensembles.

On this page

• Overview
• Data and processing
• Data download and model inclusion criteria
• Programs and languages
• Ocean level selection and interpolation
• Additional preprocessing steps for individual models
• Regridding
• Missing values
• Data corrections
• Data masks for ACCESS-ESM1-5
• Anomalies and time periods
• Ensembles calculations
• Dataset licence
• References

Overview

Climate projections vary across GCMs due to differences in the representation and approximation of earth systems and processes, and natural variability and uncertainty regarding future climate drivers. As a result, it is common practice to use a multi-model ensemble, or average across many GCMs, as opposed to using a single GCM in isolation. For this purpose, the Canadian Climate Data and Scenarios (CCDS) site has global ocean multi-model ensembles for a suite of ocean variables, scenarios, future time periods, and percentiles. Only one realization member from each model was included in the multi-model ensembles, giving each model equal weight.

Provided on CCDS are gridded ocean datasets, based on CMIP6 GCMs projections, on a common 1x1 degree resolution grid. Both multi-model ensembles as well as individual model simulations are available. Multi-model output is available for historical simulations and six Shared Socioeconomic Pathways (SSPs), four future periods, and up to eight percentiles of the CMIP6 ensemble distribution. Datasets are available as both actual and anomaly values. Anomalies of projected changes are expressed with respect to a historical reference period of 1995-2014. See Table 1 for additional details on the datasets and the ocean variables.

Data and processing

Data download and model inclusion criteria

Monthly CMIP6 ocean GCM datasets were downloaded from the Earth System Grid Federation (ESGF) online database in NetCDF format. GCMs were selected for inclusion into the CMIP6 ocean surface dataset based on several model criteria:

1. Available for at least one of the five ocean variables: dissic, ph, sos, tos, and talk.
2. Available for least one of the six SSPs (SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP4-6.0, SSP5-8.5) and the historical experiment (models with only the historical experiment and no SSPS, or models without the historical experiment were excluded).
3. Available on a monthly time scale.
4. Available for variant r1i1p1f1.
5. SSP datasets extend until at least the 2090s.

The cut-off date for downloading available model simulations from ESGF was June 22^nd, 2023. If multiple versions of the dataset were available on ESGF, the latest one was selected.

Programs and languages

Climate Data Operators (CDO) version 2.0.3 was used with Python version 3.10.4 to produce the CMIP6 ocean datasets. NetCDF Operators (NCO) version 5.0.5 was used to format metadata and coordinate bounds of the datasets.

All scripts were executed on a Unix system in a Bourne-again shell (bash) environment. The number of models in each ensemble may differ according to model data availability for each SSP and variable. See Table 2 for additional details.

Ocean level selection and interpolation

All ocean variables currently available contain data for the top level of the ocean. Sea surface variables (temperature and salinity) already contain data isolated to the sea surface, while dissolved inorganic carbon concentration, pH, and total alkalinity still have values for many levels of depth extending from the surface down to the seafloor. For the latter variables, the highest level of depth, i.e. closest to the surface, was isolated to represent sea surface values. Sea surface salinity and temperature have already undergone this same process and been isolated to the top-level of the ocean before distribution as salinity and temperature at the surface are frequently used in the climate science community.

The top level of the data was selected with the CDO function ‘sellevidx’. The top-level ocean depth, or sea surface, varied among models and modelling organizations. Please see Table 2 for each model’s topmost ocean depth.

Individual datasets were then combined, and the years of interest were extracted (using CDO functions ‘mergetime’ and ‘selyear’). The data was then regridded using a nearest neighbour remapping method (‘remapnn’) to a regular 1 by 1 degree grid (180 degrees of latitude by 360 degrees of longitude). A land-sea mask was applied to the data following regridding. The mask was downloaded from the IPCC WGI Interactive Atlas GitHub and converted to mask land instead of sea, and from a -180 - 180 longitude grid to a 0 - 360 longitude grid. The Intergovernmental Panel on Climate Change’s (IPCC) Interactive Atlas is a tool for visualizing CMIP5, CMIP6, and CORDEX projections for multiple variables, future scenarios, time periods, and warming levels.

Additional preprocessing steps for individual models

Regridding

Due to the complex native grids of the CMIP6 ocean models, some select GCMs required extra preprocessing steps. When preprocessing Institut Pierre-Simon Laplace (IPSL) models (IPSL-CM5A2-INCA or IPSL-CM6A-LR), the variable was selected (using CDO function ‘selvar’) before applying the level selection function to prevent errors in variable identification. Additionally, several models contained duplicate lines which appear as a vertical line of missing values between Asia and Antarctica through the Indian Ocean once regridded. To rectify this, those models with a duplicate line (IPSL-CM6A-LR, CMCC-CM2-SR5, CMCC-ESM2) had the extra data removed prior to regridding by using the function ‘selindexbox’ and inputting the second, and second-last latitudes and longitudes.

Missing values

Some models had projections that did not extend to 2100 (CAMS-CSM1-0, E3SM-1-0, and IITM-ESM). These models had additional years of missing values added to the datasets to ensure the length of time in all datasets was the same. Therefore, the years 2099 and 2100 (depending on the model and SSP) for the three models listed above were filled with missing values. The CDO functions ‘settunits’, ‘settaxis’, ‘setvrange’, and ‘selyear’ were used on masked regridded datasets to create an extension that was then merged with the original dataset using CDO function ‘mergetime’.

Data corrections

Some models for sea surface salinity contained values less than zero. Salinity is measured in parts per thousand and therefore cannot be less than zero. To fix this, all datasets were run through a correction method using the CDO function ‘setrtoc’ to change any negative value to zero.

Data masks for ACCESS-ESM1-5

Preprocessing quality checks of the ocean datasets revealed that dissolved inorganic carbon concentration and total alkalinity simulations produced by the ACCESS-ESM1-5 model had values that were considerably higher than any other participating model. However, such values were contained to grid cells within the Red Sea. This pattern of higher than expected values was found across all experiments and times in the ACCESS-ESM1-5 data for both variables. The modelling organization of ACCESS-ESM1-5, The Commonwealth Scientific and Industrial Research Organisation (CSIRO), identified this issue and indicated that it was due to poor mixing in the Red Sea.^Reference1 They also indicated that such issues were local to the Red Sea and did not, in any way, impact values for the rest of the world's oceans. There were some fixes implemented, but some unrealistic ocean values within the Red Sea still remain.

Since CSIRO stated that values within the Red Sea may be unrealistic, if the spread of values from the ACCESS-ESM1-5 model differed from other models, then the Red Sea region was masked out for that variable for ACCESS-ESM1-5 specifically for multi-model ensemble calculations. Gridded data within the region 30 to 45 degrees East and 13 to 30 degrees North were masked out with the functions ‘setclonlatbox’ and ‘setctomiss’. Note that this was only performed for inclusion in multi-model ensembles. This was done to reduce biases in the multi-model ensemble from unrealistically high values. Individual ACCESS-ESM1-5 model simulations of the two variables that are available for download remain unaltered.

Anomalies and time periods

Datasets are also provided in the form of anomalies according to a historical reference period of 1995-2014. Anomalies (n) are calculated by subtracting the mean over the historical reference period (µ) from original values at the time of interest (x) using the following expression:

$n_i=x_i-{\mu }_{\mathrm{1995-2014}}$

Historical baseline datasets were created using the CDO commands ‘timmean’, ‘selseas’ (for seasonal baselines), and ‘selyear’. The annual anomaly datasets were then produced with ‘sub’ and ‘yearmean’, and seasonal anomaly datasets with ‘sub’, ‘seasmean’, and ’selseas’.

Time period datasets were calculated from the anomaly datasets as means over a particular future time period using ‘selyear’ to select the years of interest and ‘timmean’ to calculate the time period average. See Table 1 for selected future time periods.

Ensembles calculations

Ensemble statistics, specifically the minimum, maximum, and mean, and the 5^th, 25^th, 50^th (median), 75^th, and 95^th percentiles, are provided as a statistical measure to demonstrate some of the range of uncertainty across model projections. The ensembles were calculated with CDO ensemble statistics functions, ‘ensmin’, ‘ensmax’, ‘ensmean’, and ‘enspctl’. See Table 1 for more details.

Dataset licence

Open Government Licence - Canada (http://open.canada.ca/en/open-government-licence-canada)

Individual model datasets and all related derived products, including the multi-model ensembles, are subject to the terms of use of the source organization.