ORRC v1.0:
Ouranos Reconstruction – RDPS-CaPA (ORRC) version 1.0 is a dataset designed to approximate the Canadian Surface Reanalysis (CaSR) and provide continuity beyond the CaSR period. It is produced from the Regional Deterministic Prediction System (RDPS), with precipitation fields adjusted by blending RDPS with the Regional Deterministic Precipitation Analysis (RDPA) products generated by the Canadian Precipitation Analysis (CaPA) system. The latter will be referred to as the CaPA hereafter.
ORRC v1.0 provides hourly surface and near-surface meteorological fields at 0.09° (~10 km) resolution on a rotated regular latitude–longitude grid covering North America, Central America, and nearly all of Greenland, from 2015 to present. It supports near-real-time climate monitoring and is intended for operational climate services, event monitoring, and the routine update of climate indicators without waiting for future CaSR releases.
More details about ORRC v1.0 are available in the documentation.
ORRC v1.0 and ORRC-a v1.0 are available through the Ouranos THREDDS Data Server.
The datasets can be accessed through NcMLs aggregation files available in the THREDDS catalog. An example workflow for accessing the datasets is provided on the Dataset page of the PAVICS platform.
This repository contains a minimal example of the ORRC workflow using one week of test data. It includes the core processing steps extracted from the broader production pipeline:
- Download RDPS, CaPA-6h, CaPA-24h, and CaSR v3.2 test data from the Ouranos THREDDS Data Server.
- Preprocess RDPS and CaPA data, including consistency and completeness checks of the raw files and interpolation onto the CaSR v3.2 grid.
- Create hourly ORRC data by combining consecutive 6- to 12-hour forecast lead-time segments from each RDPS cycle for all variables and blending RDPS precipitation fields with CaPA precipitation. ORRC output is saved as one NetCDF file per day.
- Convert ORRC variables and metadata to Climate and Forecast (CF) conventions, produce daily aggregates, and save each variable as a continuous
.zarr.ziptime series. - Produce a bias-adjusted variant, ORRC-a, to reduce discrepancies relative to CaSR v3.2.
In production, the workflow is typically run weekly on newly available RDPS and CaPA data obtained from ECCC's high-performance computer GPSC-C. This repository provides a lightweight example for testing and demonstration purposes.
The ORRC workflow runs in a Conda environment and requires several packages. Create the environment using environment.yml, then activate it:
conda env create -f environment.yml
conda activate orrcThe main ORRC workflow configuration file is located at:
code/data/config_orrc.ymlThe bias-adjustment step uses a separate configuration file located at:
code/data/config_biasadj.ymlThe paths in these configuration files are set up for the test workflow and use relative paths so that the example can be run directly within the repository.
Run all commands from the repository root.
RDPS and CaPA files are retrieved from GPSC-C in FST file format via a GPSC-C collaborator account. In the production workflow, these files are first converted to NetCDF using the fst2nc Python utilities. This test workflow downloads the test data hosted on the Ouranos THREDDS Data Server directly in NetCDF format so the conversion step is not included here.
Download one week of RDPS and CaPA test data, as well as CaSR v3.2 target data for regridding:
python code/download_testdata.pyThis script saves 6- to 12-hour forecast lead times from each RDPS cycle at 00, 06, 12, and 18 UTC as separate NetCDF files in the RDPS directory. It also saves CaPA 6-hour precipitation accumulations at 00, 06, 12, and 18 UTC in the CaPA_coarse directory, CaPA 24-hour accumulations at 12 UTC in the CaPA_24h directory, and one day of CaSR v3.2 data in the CaSR_v32 directory for regridding RDPS and CaPA data.
Preprocess the downloaded RDPS and CaPA files:
python code/preprocess_rdps_capa.pyThis step performs completeness checks, identifies missing files and variables, fills missing content with NaN where needed, and interpolates RDPS and CaPA onto the CaSR v3.2 grid.
Create the hourly ORRC dataset:
python code/create_orrc.pyTo generate ORRC, consecutive RDPS forecast segments are concatenated to form a continuous hourly time series. The resulting 24-hour reporting window spans from 13 UTC to 12 UTC of the following calendar day. To produce a CaSR-like precipitation field, hourly RDPS precipitation increments are scaled to the CaPA-6h and CaPA-24h accumulations. A more detailed description of the precipitation blending procedure is available in Section 2.3.2 of the documentation.
Apply Climate and Forecast (CF) variable and metadata conventions using the miranda Python library, compute daily aggregates as well as daily minimum and maximum for the specified variables, and save each variable as a separate .zarr.zip time series in hourly and daily formats in the staging directory defined in the configuration file:
python code/convert_orrc.pyApply mean bias correction to ORRC to improve its continuity with CaSR v3.2 using the xsdba Python library:
python code/adjust_bias.pyThis step produces the Ouranos Reconstruction – RDPS-CaPA adjusted (ORRC-a) v1.0 for a subset of daily variables, according to the specifications in code/data/config_biasadj.yml. The adjustment is trained over the reference period defined in that configuration file, using the ORRC dataset available from THREDDS as the product to be adjusted and CaSR v3.2 as the reference dataset. Each bias-adjusted variable is saved as a separate .zarr.zip time series. Details of the bias-adjustment method are described in Section 2.4 of the documentation.
The bias-adjustment workflow is designed to support two application modes:
apply_on: urlapplies the bias adjustment to the full ORRC time series retrieved from THREDDS.apply_on: stagingapplies the bias adjustment only to ORRC data available in the local staging directory, which is useful for testing the workflow on newly created data.
Users who want to bias-adjust variables beyond those included in the ORRC-a v1.0 dataset should add a new entry for each variable in the bias-adjustment configuration file and set the appropriate apply_on mode depending on whether they want to process the full time series from THREDDS or only the locally generated staging data.
We evaluated ORRC against CaSR v3.2 over the 2020–2024 period. Details of the evaluation are available in Section 2.5 of the documentation. The main conclusions are:
- Over the dataset domain, normalized errors between ORRC and CaSR are generally lower to moderate for temperature, dew-point temperature, radiation, and sea-level pressure. Larger and more spatially heterogeneous errors occur for relative humidity, precipitation, wind speed, and surface pressure, particularly in regions of complex terrain, coastal gradients, or low reference variability. Corresponding error maps are available in documentation/figures/nrmse.
- Across the selected Bukovsky regions, ORRC generally remains close to CaSR, with most variables clustered near unit normalized standard deviation and high correlations. Agreement is strongest for temperature, pressure, radiation, and specific humidity, while precipitation and wind speed show somewhat larger regional spread. The largest mismatches occur for relative humidity, especially minimum relative humidity, with broader regional scatter and higher errors. Corresponding Taylor diagrams are available in documentation/figures/taylor_diagrams.
- The bias-adjustment step substantially improves the magnitude and the spatial patterns of bias relative to CaSR. ORRC-a should therefore be prioritized for applications requiring comparison with the historical CaSR v3.2 baseline when application needs are met (e.g., daily data for the available variables). Corresponding bias maps are available in documentation/figures/bias.
When using ORRC v1.0 or ORRC-a v1.0, please cite:
Bese, A., & Logan, T. (2026). Ouranos Reconstruction – RDPS-CaPA (ORRC) v1.0 / Reconstruction Ouranos – RDPS‑CAPA (RORC) v1.0 (Version v1.0) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21381479