What’s available

What’s available#

We are publishing statistically downscaled, daily climate model output to support the regional evaluation of the potential impacts of stratospheric aerosol injection (SAI). The outputs cover four scenarios, two global climate models (GCMs), all available ensemble members, and five variables. We bias-correct GCM output against ERA5, and downscale it to a global 0.25° grid with two methods.

Property	Value
Release	`v1.0.0`
GCMs	`CESM2-WACCM6`, `UKESM1-1-LL`
Downscaling methods	`bcsd` (bias correction and spatial disaggregation), `qdmsd` (quantile delta mapping and spatial disaggregation)
Observations	ERA5
Variables	`tas`, `tasmax`, `tasmin`, `pr`, `rsds`
Temporal resolution	Daily
Spatial extent	Global

In addition to the historical period (historical) and the baseline climate scenario (ssp245), we downscale two stratospheric aerosol injection scenarios. In the scenario g6_1p5k, greenhouse gas emissions continue at ssp245 levels while sulfate aerosols are injected into the stratosphere to hold warming to 1.5 °C. The termination shock scenario (g6_1p5k_end) extends g6_1p5k ensemble member 002 from CESM2-WACCM6 to 2100, simulating an abrupt end of aerosol injection at the end of 2084.

Scenario	Group name	Years	GCMs
Historical	`historical`	1978 to 2014	`CESM2-WACCM6` and `UKESM1-1-LL`
SSP2-4.5	`ssp245`	2015 to 2099 (some `CESM2-WACCM6` ensemble members end in 2068 or 2069)	`CESM2-WACCM6` and `UKESM1-1-LL`
G6-1.5K	`g6_1p5k`	2035 to 2084	`CESM2-WACCM6` and `UKESM1-1-LL`
G6-1.5K termination	`g6_1p5k_end`	2085 to 2100	`CESM2-WACCM6` only

At a high level, the release includes two output data products, along with the processed GCM input data they were built from. The table below summarizes all three.

Product	Description	Grid	Variables
Downscaled	Bias-corrected and spatially disaggregated. This is the main product.	0.25°	`tas`, `tasmax`, `tasmin`, `pr`, `rsds`
Coarse bias-corrected	Bias-corrected, but not spatially disaggregated	Native GCM grid, about 1° to 2°	The same five, plus `dtr` (diurnal temperature range)
Processed input	Daily GCM output that the pipeline started from, before bias correction	Native GCM grid, about 1° to 2°	The same five

Data location#

Important

By downloading, copying, or using this data, you agree to the Terms of Data Access.

All data lives in CarbonPlan’s Source Cooperative repository. The bucket is public in AWS us-west-2, so you don’t need AWS credentials to read it. Each store is an Icechunk repository under s3://us-west-2.opendata.source.coop/carbonplan/srm-downscaling/, at the paths below:

Data	Path	Branch
`CESM2-WACCM6` output	`output/production/CESM2-WACCM6-ERA5-global.icechunk`	`v1.0.0`
`UKESM1-1-LL` output	`output/production/UKESM1-1-LL-ERA5-global.icechunk`	`v1.0.0`
`CESM2-WACCM6` input	`input/processed/CESM2-WACCM6.icechunk`	`main`
`UKESM1-1-LL` input	`input/processed/UKESM1-1-LL.icechunk`	`main`

A branch is a version of a store. We publish one branch per release, and so far that is only v1.0.0, which is the branch the example below opens. The input stores keep their data on main.

Data access#

We offer two ways to access the data from the Source Cooperative repository. Which one fits best depends on how much data you need, and whether you want a local copy.

Download a local copy. If you want to work with a small amount of data on your own machine, or you prefer netCDF files, we built a set of access utilities. You can use them to subset, transform, and export the downscaled data from the cloud to your local environment without writing code yourself.
Stream data from the cloud. If you’re comfortable working with data in the cloud without keeping a local copy, you can use tools like Zarr-Python, Icechunk, or Xarray. The following example opens one downscaled group from the current release.

import icechunk
import xarray as xr

storage = icechunk.s3_storage(
    bucket="us-west-2.opendata.source.coop",
    prefix="carbonplan/srm-downscaling/output/production/CESM2-WACCM6-ERA5-global.icechunk",
    anonymous=True,
    region="us-west-2",
)
repo = icechunk.Repository.open(storage)
session = repo.readonly_session(branch="v1.0.0")

ds = xr.open_zarr(session.store, group="bcsd/g6_1p5k/tas/001", consolidated=False)

Data shape#

Group layout#

Zarr organizes data in a nested structure that you can think of as a file system. The key building block is the group, a bundle of arrays and metadata at a particular path that you can open as a single labeled dataset.

Each output store holds both data products for one GCM, and each release of that store is a branch. Within a branch, a group’s path is built from the method, scenario, variable, and ensemble member, with an extra debiased_coarse level for the coarse bias-corrected product. Opening one group gives you a dataset with one climate variable on (time, lat, lon), plus any quality flags. Both methods publish the same set of groups.

{method}/{scenario}/{variable}/{member}                    # downscaled
{method}/debiased_coarse/{scenario}/{variable}/{member}    # coarse bias-corrected

Input data stores are organized differently. Each group holds one scenario, so opening a single group gives you an array for each variable with the dimensions (ensemble_member, time, lat, lon).

Ensemble members#

Available ensemble members depend on the GCM, scenario, and variable. Both output data products and both methods publish the same members, and under debiased_coarse, dtr has the same members as tasmax and tasmin.

CESM2-WACCM6

Scenario	Variables	Members	Years
`historical`	`pr`, `rsds`, `tas`	`r1i1p1f1`, `r2i1p1f1`, `r3i1p1f1`	1978 to 2014
`historical`	`tasmax`, `tasmin`	`001`	1978 to 2014
`ssp245`	`pr`, `rsds`, `tas`	`001` to `005`	2015 to 2099
`ssp245`	All five	`006`	2015 to 2068
`ssp245`	All five	`007` to `010`	2015 to 2069
`g6_1p5k`	All five	`001`, `002`, `003`	2035 to 2084
`g6_1p5k_end`	All five	`002`	2085 to 2100

For the model CESM2-WACCM6, no historical ensemble member carries all five variables, and in ssp245 only members 006 to 010 do. To avoid mixing realizations, we advise you to pick an ensemble member that carries every variable you need.

UKESM1-1-LL

Scenario	Variables	Members	Years
`historical`	All five	`u-by791`	1978 to 2014
`ssp245`	All five	`r2i1p1f2`, `r3i1p1f2`, `r12i1p1f2`	2015 to 2099
`g6_1p5k`	All five	`r2i1p1f2`, `r3i1p1f2`, `r12i1p1f2`	2035 to 2084

For UKESM1-1-LL we downscaled one historical member, u-by791. That name is a Met Office suite ID rather than a variant label like r2i1p1f2, and every scenario member above was bias-corrected against it.

Grid, time, and chunks#

The groups described above tell you which part of the data you’re reading. Within a group, each variable is an array that is physically split into chunks: fixed-size blocks that are compressed and read as a single unit. A read fetches whole chunks, so how much data a request moves depends on how many chunks it touches, rather than how many values it returns. For this dataset, a time series at a single point reads about one chunk per year, while a wide region reads many chunks for every year.

Chunks are bundled into larger files called shards. Shards don’t change which chunks a request reads, so you can mostly ignore them when estimating how much a request will move. The table below summarizes the grid, time axis, and chunk layout of both output products.

Property	Downscaled	Coarse bias-corrected
Dimensions	`time`, `lat`, `lon`	`time`, `lat`, `lon`
Grid	0.25°, 721 × 1440 cells	`CESM2-WACCM6`: 192 × 288 cells (about 0.94° × 1.25°); `UKESM1-1-LL`: 144 × 192 cells (1.25° × 1.875°)
Longitude convention	-180 to 180	-180 to 180
Calendar	Proleptic Gregorian	Proleptic Gregorian
Data type	`float32`	`float32`
Chunk size (`time`, `lat`, `lon`)	365 × 36 × 72 (1 year × 9° × 18°)	365 × 30 × 60
Shard size (`time`, `lat`, `lon`)	1095 × 180 × 360 (3 years × 45° × 90°)	1095 × 90 × 180

Quality flags#

Groups carry quality flags alongside their variable, in both products. Each flag is a uint8 array where 0 means no known issue and 1 means a known issue. The dtr groups under debiased_coarse carry no flags.

Flag	Dimensions	Present on	Marks
`qa_flag_time_varying`	`time`, `lat`, `lon`	Every group except `dtr`	Pixel-days that fail a quality check: outlier screening, a variable-specific plausible range, or a physical relationship such as `tasmin` not falling below `tas`.
`trend_distortion_flag`	`lat`, `lon`	Every group except `dtr`	Pixels where debiasing or downscaling distorts how scenarios compare with each other or with historical, relative to raw GCM output. It’s evaluated on the ensemble mean.

The flags summarize checks we run after downscaling. For the details of each check, see the output integrity checks and trend distortion check notebooks.