Statistically downscaled climate scenarios and indices from CMIP6 global climate models

On this page

• Overview
• Downscaling methodology and data
• CanDCS-U6
• CanDCS-M6
• Climate indices
• Data updates and corrections
• Updates
• Corrections
• Mean temperature
• Seasonal climate indices
• Ensembles
• File formats
• Daily
• Indices
• Monthly
• Ensembles and seasons
• Dataset licence
• References

Overview

Environment and Climate Change Canada has a mandate to provide Canadians with past and future climate change information for climate impact assessments, adaptation planning and mitigation policy development. There is a growing demand in Canada and internationally for global climate model (GCM) projections to be downscaled at high spatial resolutions. ECCC’s Climate Research Division (CRD) and the Pacific Climate Impacts Consortium (PCIC) previously produced statistically downscaled climate scenarios based on simulations from climate models that participated in the Coupled Model Intercomparison Project phase 5 (CMIP5) in 2015. These have been widely used and are referenced in Canada’s Changing Climate Report.^Reference1 The CMIP5 downscaled scenarios are available on ECCC’s Canadian Climate Data and Scenarios website and PCIC’s climate change scenarios data portal.

ECCC and PCIC have updated the CMIP5-based downscaled scenarios with two new sets of downscaled scenarios based on the next generation of climate projections from the Coupled Model Intercomparison Project phase 6 (CMIP6). CMIP6 climate projections are based on both updated global climate models and new emissions scenarios called “Shared Socioeconomic Pathways” (SSPs).^Reference9 Statistically downscaled datasets have been produced from 26 CMIP6 GCMs (Table 2) and under three different emission scenarios (i.e., SSP1-2.6, SSP2-4.5, and SSP5-8.5). The first set of scenarios was produced using the same downscaling method (BCCAQv2)^{Reference3Reference12} and downscaling target data (NRCANmet)^Reference8 as the CMIP5-based downscaled scenarios. The second scenario dataset uses a new multivariate method (MBCn) and a new target dataset (ANUSPLIN and PNWNAmet blended dataset). Downscaled daily maximum and minimum temperatures and daily precipitation are available across Canada at 10km grid spatial resolution for the 1950-2014 historical period and for the 2015-2100 period following each of the three emission scenarios.

Mean temperature and ensemble datasets are additional products produced by ECCC, with PCIC adding SSP (SSP3-7.0) to the CanDCS-M6 dataset in 2024 (only available for a smaller subset of models, see updates section).

Downscaling methodology and data

Proper usage of statistically downscaled data requires a basic understanding of how climate models simulate the climate system and how outputs from climate models are combined with historical observations to create downscaled climate scenarios. Uncertainties are present in each step of the climate modelling and downscaling chain and it is recommended that users familiarize themselves with the basic principles of climate modelling and downscaling before using these data.^Reference4

Statistically downscaled datasets have been constructed using output from 26 CMIP6 GCMs that are available at the Earth System Grid Federation (ESGF) Data Nodes, including 10 initial-condition members of the CanESM5 model produced by the Canadian Centre for Climate Modelling and Analysis^Reference10 and one run from each of 25 other CMIP6 models (Table 2).

CanDCS-U6

Bias Correction/Constructed Analogues with Quantile mapping (BCCAQ) is a hybrid downscaling algorithm that combines downscaling by BCCA (Bias Correction/Constructed Analogues)^Reference7 with quantile mapping. The use of Quantile Delta Mapping (QDM)^Reference3, a change-preserving form of quantile mapping, distinguishes BCCAQv2 from the previous version of the algorithm. Details are provided by Werner and Cannon (2016).^{Reference 12} Code implementing the BCCAQv2 method is also available from the Comprehensive R Archive Network (CRAN) repository via the R package “ClimDown”.^Reference5

Daily minimum temperature (°C), daily maximum temperature (°C), and daily precipitation (mm/day) outputs from 26 CMIP6 GCMs were downscaled using the BCCAQv2 algorithm. Historical 1/12° (~10km) gridded NRCANmet dataset (Australian National University thin-plate smoothing Splines methodology (ANUSPLINv1)) of daily minimum temperature, maximum temperature and precipitation for Canada^Reference8 were used as the respective downscaling “targets” (training data used to calibrate BCCAQv2) so that the historical downscaled climate model output has statistical characteristics that resemble those of NRCANmet as closely as possible during the 1951-2010 period. Once calibrated for a particular climate model and variable for the historical 1951-2010 reference period, BCCAQv2 is applied to the climate simulations of that variable for the selected model under three emission scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5). This process is repeated separately for each variable and model, and hence is termed a “univariate” downscaling scheme.

CanDCS-M6

A univariate downscaling method, which downscales each variable independently, can omit or poorly represent important phenomena that are characterized by combinations of temperature and precipitation. A multivariate approach to downscaling, while more computationally expensive, can maintain inter-variable dependencies and allows for analyses of multi-variable phenomena. N-dimensional Multivariate Bias Correction (MBCn)^Reference2 is a multivariate downscaling and bias correction methodology that preservers inter-variable dependencies and the statistical characteristics of a continuous multivariate distribution. The MBCn method uses a modified form of quantile delta mapping (QDM) that is able to work with multiple variables and preserve the statistical properties between each variable.

The CanDCS-M6 dataset includes the same variables, GCMs, and grid resolution as described for CanDCS-U6 but implements a new target dataset. ANUSPLINv1 has a known dry precipitation bias on the west coast of Canada, to address this, the CanDCS-M6 target dataset blends several new observational gridded datasets. The new blended downscaling target dataset (hereafter referred to as PCIC-Blend) combines daily values of precipitation, and maximum and minimum temperature from the PCIC meteorology for Northwest North America dataset (PNWNAmet)^Reference11 with daily precipitation from the new Adjusted Daily Precipitation gridded dataset (ANUSPLIN Adj)^Reference6, and daily maximum and minimum temperatures from the version 2 ANUSPLIN dataset (ANUSPLIN v2) ^Reference8.

The datasets are blended using a weighted average applied over a limited common region, with a diagonal along the eastern side Rocky Mountains acting as the blending region to combine the PNWNAmet (representing western Canada) and the two ANUAPLIN datasets (representing central and eastern Canada). The PCIC-Blend dataset therefore makes use of the updates and improved performances of multiple new products, improving the representation of precipitation in western Canada. The incorporation of PNWNAmet provided the opportunity to extend data into the western US down to 41°N. A continuous dataset across the US border along the west coast allows the CanDCS-M6 dataset to cover the extent of the Columbia river basin, identified as a key watershed in the hydrologic impacts theme.

Unlike univariate indices, which are derived exclusively from either daily precipitation, maximum temperature, or minimum temperature, multivariate indices are calculated using some combination of the three variables. Phenomena which result from the combination of temperature and precipitation conditions such as drought, wildfires, and snow, require effective simulation of the dependencies between temperature and precipitation, which a univariate approach cannot accomplish.

Climate indices

For each downscaled GCM, 31 climate indices have been calculated (Table 3). The climate indices include 27 Climdex indices established by the Expert Team on Climate Change Detection and Indices (ETCCDI)^Reference13 and 4 additional indices that are slightly modified from the Climdex indices. These indices are calculated from daily precipitation and temperature values from the downscaled simulations and are available at annual or monthly temporal resolution, depending on the indices.

Data updates and corrections

Updates

Multiple additions were made to the CanDCS6 datasets in September 2024.

PCIC added SSP3-7.0 to CanDCS-M6 which previously contained SSPs 1-2.6, 2-4.5, and 5-8.5. The SSP3-7.0 dataset only includes the first run (r1i1p2f1) of the CanESM5, unlike other CanDCS6 datasets which provide 10 (r1-10). Furthermore, three models of the 26 in the model list were unavailable for SSP3-7.0, and are therefore excluded from the dataset (HadGEM3-GC31-LL, KACE-1-0-G, and KIOST-ESM).

ECCC expanded the CanDCS6 datasets which previously included individual model simulations of minimum and maximum temperature and total precipitation by adding mean temperature and a variety of ensembles. Mean and percentile ensembles are available for all CanDCS6 datasets including indices.

Corrections

During the creation of the CanDCS-M6 dataset, PCIC identified sporadic anomalously large daily maximum temperature values in three of the contributing 26 GCMs. PCIC determined that the anomalous values were caused by errors in the land surface and atmospheric components shared by models. To prevent the errors from propagating into the downscaled dataset and avoid introducing missing values by omitting the anomalous values, PCIC implemented an outlier detection and correction methodology which replaces anomalous temperatures with values robustly interpolated from surrounding grid cells. The three models with known errors are, the two Met Office Hadley Centre (MOHC) models HadGEM3-GC31-LL and UKESM1-0-LL, and KACE-1-0-G from the Korean Meteorological Administration (KMA). PCIC has re-run simulations for the three models using the CanDCS-U6 methodology and corrected the original datasets. The updated CanDCS-U6 datasets replaced the ones already on CCDS on August 25^th, 2023.

PCIC later made corrections to the CanDCS-M6 precipitation dataset, and a small portion of the CanDCS-M6 CanESM5 dataset (CanESM5, r3i1p2f1, SSP1-2.6, 2060-2100). Both corrected datasets used the same correction method for detecting and replacing anomalous values and were updated on CCDS on September 25^th, 2024. Note that these corrections did not impact the associated Climdex indices, unlike the former maximum temperature corrections for which all indices were updated at the same time.

Mean temperature

CanDSC datasets include maximum and minimum temperature and total precipitation. From these, ECCC has calculated mean temperature datasets for all models, scenarios, and time periods. Python (v3.10.4) and the Xarray package (v2022.3.0) were used to calculate daily mean temperature. Python, Climate Data Operators (CDO) (v2.0.3), and NetCDF Operators (NCO) (v5.0.5) were used to create the monthly datasets (CDO function ‘monmean’), all in a Bourne Again SHell (Bash) environment.

Seasonal climate indices

12 of the 31 Climdex indices produced by PCIC are distributed on a monthly, as opposed to an annual time scale. For these indices, ECCC has created seasonal and annual versions calculated as seasonal/annual maximums, means, or minimums depending on the index (Table 4). The calculations were carried out with CDO (v2.0.3) in a Bash environment.

Ensembles

ECCC has calculated an ensemble mean and five ensemble percentiles (5^th, 25^th, 50^th, 75^th, 95^th) across the CanDCS6 model distribution. Ensembles are available for all datasets, variables, indices, SSPs, and periods. The ensembles were derived from a singular realization from each model, including CanESM5 model (r1i1p2f1), to prevent bias towards a singular model. CDO (v2.0.3) was used with Python (v3.10.4) and NCO (v5.0.5) in a Bash environment to produce the ensembles which were calculated with CDO ensemble statistics functions ‘ensmean’, and ‘enspctl’.

Daily, seasonal, and annual ensembles required extra preprocessing steps before calculating the ensembles. Seasonal and annual versions of the contributing model datasets were produced with CDO (temperature: 'seasmean', 'yearmean'; precipitation: 'seasum', 'yearsum'). The Xarray package (v2022.3.0) for Python was used to convert the calendar (‘convert_calendar’) of daily datasets to Proleptic Gregorian for any model with a different calendar type. Proleptic Gregorian was selected as it was the most prevalent calendar type of the contributing models. The Xarray function adds days containing missing values, no values were artificially created with interpolation through this method.

File formats

Daily

Downscaled simulations have been organized into separate directories containing one simulation for each GCM, variable, realization, and shared socioeconomic pathway (add “MBCn” tag for CanDCS-M6 datasets). For example, the downscaled simulation of CanDCS-U6 daily precipitation from the first realization (r1i1p1f1) of ACCESS-CM2 following the SSP1-2.6 pathway is contained within the directory:

ACCESS-CM2_pr_r1i1p1f1_ssp126{_MBCn}_5_year_files

In view of the large file size of the downscaled simulations (57 Gb per file), each complete simulation has been split into 5-year subsets. Each 5-year file is labelled following the same filename format to describe the simulation and time interval contained within the file. For example, the first file within the directory listed above is the following:

pr_day_BCCAQv2+ANUSPLIN300_ACCESS-CM2_historical+ssp126_r1i1p1f1_gn_19500101-19551231.nc

Components of the filename format include:

• Variable: "pr"
• Timestep: "day"
• Downscaling method and target: "BCCAQv2+ANUSPLIN300" (or “MBCn+PCIC-Blend”)
• GCM: "ACCESS-CM2"
• Past and future pathway: "historical+ssp126"
• Realization: "r1i1p1f1"
• Grid Label: "gn"
• Time interval: "19500101-19551231"

Each simulation directory contains one 6-year file (1950-1955, 2.3 Gb) and 29 5-year files (1.9 Gb each) that together comprise the 151-year Canada-wide downscaled simulation for that GCM, variable, realization, and socioeconomic pathway. The grid label identifier denotes whether GCM simulations are provided using the native grid of the model ("gn") or have been regridded ("gr") to another primary grid by the modelling centre. Instances where simulations from a GCM are provided at more than one grid resolution are indicated using an integer designation (e.g. "gr2").

Indices

Files of climate indices are labelled following the same filename format. For example, the first file within the Climdex directory of ACCESS-CM2 is the CDD at annual (ann) time resolution is the following:

ACCESS-CM2_ssp126_r1i1p1f1/cddETCCDI_ann_BCCAQv2+ANUSPLIN300_ACCESS-CM2_historical+ssp126_ r1i1p1f1_1950-2100.nc

Components of the filename format for the climate indices include:

• ETCCDI climate index: "cddETCCDI"
• Time resolution: “ann”
• Downscaling method and target: "BCCAQv2+ANUSPLIN300" (or “MBCn+PCIC-Blend”)
• Time interval: “1950-2100”

Monthly

Files of monthly individual GCMs differ slightly between CanDCS-U6 and CanDCS-M6. The monthly sum of pr for the same GCM and SSP is the following:

pr_monthly_total_BCCAQv2+ANUSPLIN300_ACCESS-CM2_historical+ssp126_r1i1p1f1_1950-2100.nc

ACCESS-CM2/pr_day_MBCn+PCIC-Blend_ACCESS-CM2_historical+ssp126_r1i1p1f1_195001-210012_monthsum.nc

Components of the filename format include:

• Variable : "pr"
• Monthly calculation:
• U6 : “monthly_total” (pr) or “monthly_average” (tasmin, tasmax, and tas)
• M6 : “monthsum” (pr) or “monthmean” (tasmin, tasmax, and tas)
• Downscaling method and target: "BCCAQv2+ANUSPLIN300" (or “MBCn+PCIC-Blend”)
• GCM: "ACCESS-CM2"
• Past and future pathway: "historical+ssp126"
• Realization: "r1i1p1f1"
• Time interval: “195001-210012”

Ensembles and seasons

Ensembles and seasonal datasets follow a similar filename format to the original files. The ensemble type replaces the GCM, and time resolution is dependent on selection. For example, the first file within the seasonal tas directory is the following:

seasonal{monthly}/tas/tas_annual_avg_BCCAQv2+ANUSPLIN300_ensemblemean_historical+ssp126_1950-2100.nc

Components of the filename format for the ensembles include:

• Time resolution: "annual" (or “monthly_avg{sum}” (sum only for pr)) or season: “DJF”, “MAM”, “JJA”, “SON”)
• Operation: “avg” (or “sum” for pr)
• Downscaling method and target: "BCCAQv2+ANUSPLIN300" (or “MBCn+PCIC-Blend”)
• Ensemble type: “ensemblemean” (or percentile: “enspctl5”, “enspctl25”, “enspctl50”, “enspctl75”, “enspctl95”)
• Time interval: “1950-2100”

The first file within the daily pr directory for SSP1-2.6 is the following:

daily/ssp126/pr/pr_day_BCCAQv2+ANUSPLIN300_ensemblemean_historical+ssp126_19500101-19551231.nc

Components of the filename format include:

• Variable : "pr"
• Timestep: "day"
• Downscaling method and target: "BCCAQv2+ANUSPLIN300" (or “MBCn+PCIC-Blend”)
• Ensemble type: “ensemblemean” or percentile (“enspctl5”, “enspctl25”, “enspctl50”, “enspctl75”, “enspctl95”)
• Past and future pathway: "historical+ssp126"
• Realization: "r1i1p1f1"
• Grid Label: "gn"
• Time interval: “195001-210012”

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.