Canadian Surface Reanalysis - CaSR
Summary Description
The Canadian Surface Reanalysis (CaSR) is a high-resolution atmospheric reanalysis dataset produced by Environment and Climate Change Canada (ECCC). CaSR v3.2 is produced by dynamically downscaling ECMWF’s ERA5 global reanalysis using operational models and configurations of ECCC’s operational weather and environmental prediction system to provide historical, high-resolution climate and weather data over North America, with a focus on Canada. Surface observations of tempererature, humidity, snow depth and precipitation are integrated through the the Canadian Land Data Assimilation System (CaLDAS) coupled with the Canadian Precipitation Analysis (CaPA). CaSR is a consistent and seamless dataset providing the main meteorological variables coherent with in situ surface observations (Gasset et al. (2021)). Note that the CaSR dataset was previously known as RDRS (Regional Deterministic Reforecast System). A detailed description of the dataset is available in Gasset et al. (2021). Update reference to Khedhaouiria & Bulat (2025) if/when available, here and under Dataset Characteristics!
Dataset Characteristics
- Current version: v3.2
- Available variables: temperature, precipitation, wind, humidity, radiation, snow, surface pressure, sea level pressure, geopotential height (see variables section below)
- Temporal coverage: 1980–2024
- Temporal resolution: Hourly and daily outputs
- Spatial coverage: Canada and U.S.
- Spatial resolution: ~10 km grid spacing (0.09˚)
- Data type: A dynamical downscaling of ERA5 over the North and Central America domain using ECCC’s Global Deterministic Reforecast System (GDRS), the global surface model (GEM-Surf) and the Regional Deterministic Reforecast System (RDRS), coupled with the CaLDAS surface assimilation system (Environment & (ECCC) (2025))
- Data format: netCDF
- Web references:
ECCC Canadian Surface Reanalysis (CaSR) Web Site,
Northern Climate Data Report and Inventory (NCDRI) Web Site - Reference:
Gasset et al. (2021) - Contact: CaSR Helpdesk
When to use CaSR
- When you need climate information about the recent past.
- When you need spatial completeness (e.g., basin-wide hydrology forcings).
- When you need variables not consistently available from stations (e.g. radiation, near-surface winds).
- When you need the best precipitation data over a continuous region of Canada.
- When you need to distinguish between different types of precipitation.
- When you need data at hourly timescale.
- When you need a historical dataset that is consistent with future climate projections available on Portraits Climatiques and ESPO-G6.
Strengths and Limitations
Key Strengths of CaSR
| Strength | Description |
|---|---|
| High resolution | ~10 km grid spacing provides detailed spatial variability, better than global reanalyses. |
| Hourly data availability | Useful for high-frequency climate and weather analysis. |
| Assimilation of observed precipition | Unlike other reanalysis products observed precipitation is assimilated. |
| Detailed precipitation | Precipitation types are distiguished, including freezing rain. Two versions of precipitation are provided: one purely modeled and the reanalysis version that assimilates measured precipitation data are provided. |
| Precipitation confidence index | A confidence index for precipitation informs on the weight of observations in the analysis. |
| Consistent historical record | Spans over four decades, allowing for trend analysis and climate studies. |
| Active development | New versions and longer temporal record are in preparation. |
Key Limitations of CaSR
| Limitation | Description |
|---|---|
| Known biases | CaSR is generally too wet and has an overall cold bias. See the Expert Guidance Section below. |
| Uncommon georeference | The data are made available on a rotated grid which may be difficult to handle. The PAVICS platform hosts a tutorial dedicated to the conversion of rotated grids. |
| Limited to the surface or near-surface | CaSR provides no data above 40m above surface. |
Expert Guidance
The Canadian Surface Reanalysis (CaSR) dataset provides high-resolution, gridded climate data with seamless coverage across North America. CaSR has one of the highest spatial resolutions of gridded observational datasets over Canada and North America and it distinguishes itself from other reanalysis products through the assimilation of precipitation observations through the Canadian Precipitation Analysis (CaPA). It is therefore coherent with in situ observations of precipitation, absolute and dew point temperature and snow depth. The model used in the production includes a representation of lakes, which influence local climate. This feature is visible particularly in the shoulder season temperature fields and influences the thermal surface properties.
The CaPA system used to produce the gridded precipitation fields includes strict quality control procedures to avoid the assimilation of biased observations. This includes using wind speed to determine validity of solid precipitation measurements depending on the gauge type and weather station characteristics. Due to the complexity of the prodedure and difficulties with quality control radar data are currently not assimilated. Overall, CaSR was conceptualized for hydrological applications, where high-resolution precipitation and land-surface variables are critical (Gasset et al. (2021)).
Temperature biases in CaSR exhibit an annual cycle with larger errors in winter and spring and smaller errors in summer and autumn. When compared to the NRCANmet gridded observations, CaSR shows slightly warmer mean minimum temperatures and slightly colder mean maximum temperatures. As a result, the daily temperature range in CaSR is narrower than in NRCANmet. Air temperature biases are clearly correlated to elevation differences between nearby stations and the corresponding grid elevation. With improvement in the latest version of CaSR spring temperatures are higher and less biased in version 3.2 than in version 2.1. Pixels near water or mixed water/land pixels tend to exhibit a stronger cold bias. Overall, the magnitude of the biases varies by temperature variable and region, emphasizing the need for validation of the dataset at a particular location.
Precipitation in CaSR is generally positively biased, and this bias tends to increase with distance from the assimilated in-situ stations. Generally, reanalyses data tend to be too wet, with CaSR being wetter than ERA5-Land. Topography effects on precipitation is better resolved in CaSR than in ERA5-Land or NRCANmet. Due to the offline precipitation analysis, small inconsistencies can arise with other modeled fields such as radiation and humidity.
A study over British Columbia watersheds confirms CaSR’s persistent cold bias alongside a wet bias. CaSR version 2.1 also shows drying trends in BC watersheds. The dataset proves skillful in capturing spatial mountain patterns and temporal trends (Goswami et al. (2024)).
Version 3.2 of CaSR represents a major improvement over version 2.1. It shows better agreement with surface observations across all variables, seasons, and domains. Precipitation performance is similar or slightly improved, and the version includes the additional benefit of a new precipitation partitioning into rainfall and snowfall. Note that ECCC usually does these performance tests on the model output of precipitation, not the analysis that also includes the observation assimilation. Version 3.2 also assimilates up to 1,000 more observations per analysis, particularly improving winter performance in Canada (Gasset et al. (2025)).
A comparison of CaSR v3.2 and CaSR v2.1 (RDRS) winter (DJF) precipitation shows large differences over and around some large lakes in Manitoba (see figure below). Preliminary diagnostics indicate that these differences originate primarily from updates to the model background fields. In both versions, data assimilation tends to dry the background, but in v2.1, the assimilation of observations has a much larger (drying) effect, suggesting that v2.1 was likely too wet in these regions. The exact cause of the lake-related signature remains uncertain at this stage and note that these differences could appear over other lakes as well.
Preliminary analysis suggests that the time-dependant biases that were present in the previous version of CaSR (v2.1) have been corrected, increasing the confidence in trends that are detected using this dataset. However, because trends in CaSR 3.2 have not been fully validated, users are advised to validate such trends using a second dataset.
There was a bug in the assimilition of precipitation observations in version 3.1. The effect of the bug is very important over Quebec. This version should not be used for precipitation.
Variables available in CaSR
For details click on variable group to uncollapse
Data Access
CaSR data can be downloaded from the CaSPAr platform, the PAVICS platform or ECCC’s high-performance computer GPSC-C. The CaSR website provides instructions for these different download options.
To avoid downloading very large datasets in their entirety PAVICS allows partial/regional extraction and provides a tutorial to do so. With a free PAVICS user account, the Jupyter notebook with the Python code in the tutorial can be directly used on PAVICS.
ToDo: Which tutorial should we point to? The (older?) one addressing CaSR https://pavics-sdi.readthedocs.io/en/latest/notebooks/CaSR_basic.html or the (better maintained?) version on PAVICS: https://pavics.ouranos.ca/climate_analysis.html#b*