...

Introduction

This user guide describes the datasets released from the Arctic Regional Reanalysis service, which is part of the Copernicus Climate Change Service (C3S). The datasets will include the actual grid point reanalysis information on different levels (atmospheric vertical levels, surface including soil). This version also provides details on the uncertainty information provided. We will refer to the dataset as the CARRA (Copernicus Arctic Regional ReAnalysis) dataset.

First in this user guide you will find sections on the data availability, formats and data content followed by the information on the uncertainty of the data. For a short description of the principles and methods behind this reanalysis, numerical weather prediction and data assimilation, see the Annex at the end of this document.

Data availability

The Arctic regional reanalysis data can be downloaded from the Copernicus Climate Data Store (CDS) at the following locations:

...

The scripts in section 2.1 are also available for download via the open GitHub athttps://github.com/metno/carra_cds

CDS API example, C3S Arctic reanalysis

The example below shows a Python script for retrieving a subset of the C3S Arctic Reanalysis data from CDS for the month of July 2012 for the West domain into an output file named: "CARRA-West_2012-07.grib". Such script can be derived from the download form (see "Show API request" at the bottom of the form).

...

In the example above single level data are retrieved from the western domain. If instead pressure level data from the eastern domain is required the following command should be used in the script: c.retrieve('reanalysis-carra-pressure-levels', 'domain': 'east_domain'...)

Data formats

The reanalysis data is available in two formats: GRIB and NetCDF.

GRIB

The GRIB data format is a standard binary format for modelled meteorological data governed by the World Meteorological Organisation (WMO). This data format stored the data in a very compact way and the metadata is contained within the files. It can be read with the tool eccodes that is available from the European Centre for Medium-range Weather Forecasts (ECMWF): https://confluence.ecmwf.int/display/ECC/GRIB+tools. National weather services and weather companies use a wide array of graphical tools to display GRIB data. The GRIB data format is recommended for users comfortable and familiar with this format.

NetCDF

NetCDF (Network Common Data Form, https://www.unidata.ucar.edu/software/netcdf/docs/index.html ) is an open standard data format developed and supported by the American University Corporation for Atmospheric Research (UCAR). It is a widely used format for geoscientific data. A large array of access libraries and applications for reading and plotting NetCDF data exist: https://www.unidata.ucar.edu/software/netcdf/software.html.

Data content

Single level variables

Diagnostic single level output is available in three hourly intervals at 00, 03, 06, 09, 12, 15, 18 and 21 UTC.

...

Soil level variables

Soil level variables are given for two model depths, where the first depth is the soil surface and the second depth is the so-called root depth. The root depth varies with the cover type climatology. Please note that the soil level variables are accommodated in the Arctic Regional Reanalysis single level variables catalogue entry.

...

Model level variables

Model level variables are output at 65 hybrid model levels of the HARMONIE-AROME model. These follow the surface at the lowest levels and are gradually evolved into pure pressure levels at the highest levels. These are the levels at which the model computations are done. The height level and pressure level variables are interpolated from these data.

...

Pressure level variables

Pressure level variables are interpolated to 23 specific pressure levels: 1000, 950, 925, 875, 850, 800, 750, 700, 600, 500, 400, 300, 200, 100, 70, 50, 30, 20 and 10 hPa. Thus, they are on isobaric surfaces.

...

Height level variables

Height level variables are interpolated to 11 specific height levels: 15, 30, 50, 75, 100, 150, 200, 250, 300, 400 and 500 metres above the surface.

...

Details about the data fields

All data fields are model grid box (2.5 km X 2.5 km = 6.25 km²) averages. When output data are compared to local data near the coast or close to a glacier boundary, it should be remembered that the model variables represent such averages. For instance, to compute the surface forcing at specific glacial sites near a glacier boundary, it will be most representative to choose output from a model grid box that is fully on the glacier, rather than the closest model grid box if this only partially covers the glacier. Also, very local rain shower intensity can be higher than the modelled average grid box intensity.

Output variables can be either instantaneous, accumulated, or maxima/minima from a given period. This is specified for each of the variables listed below.

Precipitation and water fluxes

Precipitation at the surface is output as two separate types: Rain and total solid precipitation. These have the unit kg/m², which for rain with a density of 1000 kg/m³ is the same unit as mm. For the solid precipitation, using this unit removes confusion between mm water equivalent and mm thickness. Total solid precipitation is the sum of the model variables snow and graupel. Total precipitation is the sum of all three precipitation types. The precipitation species include both convective and stratiform precipitation and are available only for the forecast time steps. They are accumulated variables meaning that they are accumulated from the beginning of the forecast. For instance, the 24h-forecast includes accumulated precipitation over 24 hours. Hourly precipitation can be retrieved by subtracting two accumulated precipitation fields with one-hour separation.

...

Water fluxes at the surface are output as surface runoff, percolation (drainage), water evaporation and snow sublimation. All these variables have units of kg/m². Surface runoff occurs, when the model soil is saturated with water and more precipitation comes. Percolation is drainage of water below the deep soil level in the model. The water evaporation can be both positive and negative, where negative values signify condensation. This is also true for the snow sublimation.

Cloud and water vapour integrated variables

All cloud and water vapour variables are instantaneous, i.e. they are given for the time step at which they are output. Vertically integrated water vapour is given in units of kg/m². It is vertically integrated from the surface to the top of the atmosphere. In practice it is computed from the specific water vapour on the 65 model levels (see section 4.4). Likewise, integrated cloud liquid water, integrated cloud ice water, and integrated graupel are computed from the specific cloud liquid water, cloud ice and graupel on the 65 model levels.

...

Visibility is given with the unit m and is calculated from the cloud water, cloud ice, rain, snow and graupel present at the lowest model level. For cloud and precipitation free conditions mist is calculated from the lowest model level relative humidity and the concentration of cloud condensation nuclei. Empirical relations are used (pers. comm. Esbjörn Olsson, SMHI, 2013). Cloud base height and cloud top height are output in units of m above the surface. They are defined for the highest and lowest model level with more than 4/8 cloud cover.

Variables at 10-metre height

All 10-metre height wind variables are instantaneous, i.e. they are given for the time step at which they are output. Diagnosed u and v wind components (u and v) are output at the WMO standard height of 10 m above the surface with the unit m/s. The u- and v-components follow the direction of the Lambertian model grid with the u-component being directed 90 degrees clockwise relative to the v-component. The diagnosed winds are computed from the winds at the lowest full model level (see section 4.3). The 10-metre u wind component (u_10m) in the case of a stable or neutral boundary layer is calculated as

...

10-metre u and v wind gust components are also output. These are computed from the diagnosed 10-metre winds and the turbulent kinetic energy (pers. comm. Gwenaëlle Hello, Meteo France, 2007).

Variables at 2-metre height

2-metre temperature (T_2m) is diagnosed from the so-called skin temperature at 0 metre height (T_0m) Since the 2-metre temperature is often referred to as the surface temperature we here, in order to avoid misunderstandings, explicitly use 2-metre and 0-metre subscripts. and the temperature at the lowest model level (T_lml). For stable and neutral boundary layers it is calculated as

...

2-metre specific humidity in units of kg/kg is diagnosed in the same way as the 2-metre temperature, only from the specific humidity at 0 metre height and at the lowest model level rather than the temperatures at these levels. From this and the saturation specific humidity the 2-metre relative humidity in % is computed.

Variables at 0-metre height

The surface pressure is given in units of Pascal (Pa). From this the mean sea level pressure [Pa] is computed by reducing the surface pressure to the mean sea level. To avoid confusion with the 2-metre “surface” temperature, the 0-metre temperature is referred to as the skin temperature. It is given in units of K. Both the pressure and temperature variables are instantaneous, i.e. they are given for the time step at which they are output. The surface is the lowest model half-level. Thus, the variables at this level are explicitly calculated on each model iteration.

...

The u and v components of the accumulated surface momentum flux are given as output variables in units of kg m/s. The momentum roughness length and the heat roughness length as used in the model are given as output in units of m. Note that the roughness lengths should only be used from forecast data, as the analysis time values are incorrect – and not used in the model.

Variables at the top of the atmosphere

At the top of the atmosphere (TOA) the accumulated solar net TOA radiation and the accumulated thermal net TOA radiation are output variables in units of J/m². These are both considered on a horizontal surface and are both positive in the downward direction. Since the downward solar TOA radiation is always larger than the upward solar TOA radiation, the solar net TOA radiation is always positive. Since there is virtually no downward thermal TOA radiation, the thermal net TOA radiation is always negative. The TOA is the highest model half-level. Thus, the variables at this level are explicitly calculated.

Snow variables

The snow variables are given as instantaneous values from the most recent model time step relative to the output time. The snow density output unit is kg/m³, the snow water equivalent (SWE) output is in units of kg/m², and the snow fraction output has fractional units in the range 0-1. The snow depth can be derived from these variables.

Sea and sea-ice variables

For the sea the sea surface temperature is output in units of K. For areas partially or completely covered with sea-ice, the following variables are output: Sea-ice area fraction [-], upper layer sea ice temperature [K], sea-ice thickness [m], and sea-ice snow thickness [m]. Here the sea-ice fraction and sea surface temperatures are in fact interpolated input data and are only updated once every day. During the course of a forecast they are kept constant. All other sea and sea-ice variables are given as instantaneous values from the most recent model time step relative to the output time.

Soil level variables

The soil level variables are computed at the skin depth immediately at the surface and at the "root depth", that is the average depth from which the vegetation retrieves its water. This level varies depending on the surface type and is also defined for surface types without vegetation - for instance rock. The volumetric soil moisture content is the volume concentration of liquid water at root depth. The volumetric soil frozen water content is the volume concentration of ice at root depth. Please note that the soil level variables are listed in the "single level" catalogue entry. The root depth variables are used to account for effect of the evapotranspiration on the temperature and humidity at and immediately above the surface. They are a means to an end rather than a detailed model of what the humidity is at specific depths in the soil. For anyone who wishes to perform detailed modelling of processes below the surface we recommend to use surface fluxes from our reanalysis dataset as forcing for more advanced sub-surface models.

Model level variables

The model level variables are computed at the full model levels, and are given as instantaneous values from the most recent model time step relative to the output time. There are 65 vertical model levels in HARMONIE-AROME.  These full model levels are hybrid-sigma coordinates that are counted from the model top toward the surface. They go from being pure pressure levels, i.e. levels with constant pressure starting at 10 hPa, 30 hPa, etc. to being relative to the surface topography in height. Level 64 is at approximately 30 m height and level 65 is at approximately 12 m height above the surface. The following thermodynamic variables are the output variables at model levels: Temperature [K], u-component of wind [m/s], v-component of wind [m/s], turbulent kinetic energy [J/kg]. Here turbulent kinetic energy is the mean kinetic energy per unit mass from eddies in turbulent flow. Note that the HARMONIE-AROME weather forecasting model with 2.5 x 2.5 km² resolution does not explicitly resolve this turbulent energy. The u- and v- wind components follow the direction of the Lambertian model grid with the u-component being directed 90 degrees clockwise relative to the v-component. From these model level wind components, the model level wind speed and wind direction relative north can be calculated with Equations 3 and 4. The grid rotation angle 𝛼 can be computed with this script: https://github.com/metno/NWPdocs/wiki/From-x-y-wind-to-wind-direction.

The following moisture, cloud and precipitation variables are output at model levels: Specific humidity, specific cloud liquid water, specific cloud ice water content, specific cloud rain water content, specific cloud snow water content, and graupel. All of these variables are given in units of kg/kg. Additionally, the cloud cover in % is output on model levels. The model level cloud covers can be used to compute the total cloud cover based on other cloud cover overlap assumptions than the one described in section 5.2.

Pressure level variables

Pressure level variables are output at the pressure levels: 1000, 950, 925, 900, 875, 850, 825, 800, 750, 700, 600, 500, 400, 300, 250, 200, 150, 100, 70, 50, 30, 20, and 10 hPa, and are given as instantaneous values from the most recent model time step relative to the output time. The model level variables described in section 4.3 are vertically interpolated to fixed pressure levels. The following thermodynamic variables are output at pressure levels: Temperature [K], u-component of wind [m/s], v-component of wind [m/s], geometric vertical velocity [m/s], geopotential [m²/s²], pseudo-adiabatic potential temperature [K], and potential vorticity [(K m²)/(kg s)]. Here the geopotential is the work required to lift an air parcel of unit mass from mean sea level to the given pressure level. The pseudo-adiabatic potential temperature is the temperature that an air parcel would have if it were first expanded through a pseudo-adiabatic process to 0 hPa pressure and thereafter compressed to a pressure of 1000 hPa through a dry-adiabatic process. The potential vorticity is a measure of the capacity for air to rotate in the atmosphere. The u- and v- wind components follow the direction of the Lambertian model grid with the u-component being directed 90 degrees clockwise relative to the v-component.  From these pressure level wind components, the pressure level wind speed and wind direction relative north can be calculated from Equations 3 and 4. The grid rotation angle 𝛼 can be computed with this script: https://github.com/metno/NWPdocs/wiki/From-x-y-wind-to-wind-direction.

The following moisture, cloud and precipitation variables are output at pressure levels: Relative humidity [%], cloud cover [%], specific cloud liquid water content [kg/kg], specific cloud ice water content [kg/kg], specific cloud rain water content [kg/kg], specific cloud snow water content [kg/kg], and graupel [kg/kg]. These pressure level variables are interpolated from the full model level variables described in section 4.3.

Height level variables

Height level variables are output at the following fixed heights above the surface: 15 m, 30 m, 50 m, 75 m, 100 m, 150 m, 200 m, 250 m, 300 m, 400 m and 500 m. They are interpolated to these heights from the full model level variables described in section 4.3. The following thermodynamic variables are output at height levels: Temperature [K], pressure [Pa], wind speed [m/s], and wind direction [degrees from North]. Note here that the wind direction is defined as the direction from which the wind comes!

The following moisture, cloud and precipitation variables are output at height levels: Relative humidity [%], specific cloud liquid water [kg/kg], and specific cloud ice water content [kg/kg]. These height level variables are interpolated from the full model level variables described in section 4.3.

Static fields

Static fields are output variables that do not change depending on the model initial time or the forecast length (in other words they are time-independent). These include the land-sea mask, that is the fraction of land in a given model grid box of 2.5 x 2.5 km² in units of %, and the orography in units of m. For each model grid box in HARMONIE-AROME 4 tile fractions are defined in units of %. These are: The fraction of sea, the fraction of inland water (lakes and rivers), the fraction of urban areas, and the fraction of nature, i.e. land areas that are not inland water or urban. The fraction of glaciers is also output. This is assumed to be a constant field with glacier extents representative of the middle of the full reanalysis period (1997-2021). Glacier extent in remote Arctic locations is not available as accurately mapped yearly datasets. The official maps are outdated due to major calving events in the recent decades. HARMONIE-AROME has not yet been designed to deal with changing land-sea masks or other surface classifications. Thus, these are static fields.

What are the uncertainties of the data fields?

A user of the reanalysis fields may be interested in knowing the accuracy or uncertainty of the provided data. Uncertainty is an expression of the expected or average deviations from what is considered the "true" value, and several approaches to estimate uncertainties are possible.

...

In section 6.5 we present uncertainties measured as statistics of actual deviations from observations (also known as verification statistics). Note that observations are a reference not identical to the actual truth, as they also will have uncertainties as well as representativeness issues. This verification statistics is provided for a set of near-surface quantities which are covered by the meteorological observation network, including 2m temperature, 10m winds and precipitation.

Field based atmospheric uncertainty estimation

The Earth system observations are unevenly distributed in space and time and the largest part of the Arctic Regional Reanalysis (CARRA) domains has sparse data coverage. Additional information in the form of numerical simulations is needed in order to construct the gridded dataset. Reanalysis employs a data assimilation system for this purpose, the same device that is used to initialize a numerical weather prediction forecasting system with the latest observations. The analysis gridded dataset is obtained as an optimally weighted average of observations and short-range weather forecast (numerical simulations) where weights are dependent on the distance to observations, the forecast error and observation error statistics and the statistical relationships valid between meteorological quantities (model state variables), such as horizontal wind, temperature, specific humidity, surface pressure, geopotential.

...

The methodology does not account for systematic errors or for uncertainties associated with the forecasting model used for numerical simulations, such as for example parameterization of unresolved processes. In addition to the climatological upper bound analysis error standard deviation that we have discussed above, we provide a refined climatological analysis error standard deviation. The refined climatological analysis error standard deviation is the climatological upper bound analysis error standard deviation multiplied by a rescaling factor. The rescaling factor is obtained by confronting the climatological upper bound forecast error variance against observation error variance in observation space. The multiplication by rescaling factor allows compensating for the limitations of the methodology mentioned above to some extent. The refined climatological analysis error standard deviation is a more reliable estimate of the uncertainty of the CARRA datasets than the climatological upper bound analysis error standard deviation in the areas with dense observation coverage. At the same time, the short-range forecast error variance is always lower in the areas covered by observations than in the areas of sparse observation coverage. Thus, the refined climatological analysis error standard deviation can underestimate the actual uncertainty of the Arctic reanalysis data in the areas with poor observation coverage. We propose to use the refined analysis error standard deviation in data rich areas and to use the upper bound standard deviation in data sparse areas as a measure of uncertainty.

Uncertainties for the CARRA-East model domain

In this section we provide the actual estimates of the uncertainty in the form of the "Upper Bound" standard deviation and the "Refined" standard deviation for analysis errors over the CARRA-East domain. We provide both summer and winter statistics. The uncertainty estimation is given both in form of tables for surface pressure and mean sea level pressure (Table 7), u-component of wind (Table 8), v-component of wind (Table 9), temperature (Table 10), geopotential (Table 11) and relative humidity (Table 12) and in form of figures for u-wind component (Figure 2), v-wind component (Figure 3), temperature (Figure 4) geopotential (Figure 5) and relative humidity (Figure 6). All tables contain the name of the meteorological variable, short abbreviation, units and vertical levels. Uncertainty of the surface pressure field is provided at 0 m above ground and at 0 m above mean sea level. The uncertainty of the u- and v- wind components, temperature, geopotential and relative humidity is provided at standard pressure levels: 50hPa, 100hPa, 150 hPa, 200hPa, 250 hPa, 300hPa, 400hPa, 500hPa, 600hPa, 700hPa, 800hPa, 850hPa, 900hPa, 925hPa, 950hPa and 1000hPa. The figures show uncertainty estimates as a function of standard pressure levels. The Upper Bound standard deviation (STDV) of the analysis errors is shown as a solid curve and the Refined Standard Deviation of the analysis error is shown as a dashed curve. Summer statistics are presented in the plots to the left and winter statistics are presented in the plots to the right.

...

Uncertainties for the CARRA-West model domain

In this section we provide the actual estimates of the uncertainty in the form of the Upper Bound Standard Deviation and the Refined Standard Deviation for analysis errors over CARRA-West domain. For this domain, we provide all-year-around averaged statistics. The uncertainty estimation is given both in form of tables for surface pressure and mean sea level pressure (Table 13), u-wind component (Table 14), v-wind component (Table 15), temperature (Table 16) and specific humidity (Table 17), and in form of figures for the u- and v- wind components (Figure 7, left and right panel, respectively), temperature (Figure 8 , left panel) and specific humidity (Figure 8, right panel). All tables contain the name of the meteorological variable, short abbreviation, units and vertical levels. Uncertainty of the surface pressure field is provided at 0 m above ground and at 0 m above mean sea level. The uncertainty of the u- and v- components of wind, temperature and specific humidity is provided at standard pressure levels : 50 hPa, 100 hPa, 150 hPa, 200 hPa, 250 hPa, 300 hPa, 400 hPa, 500 hPa, 600hPa, 700 hPa, 800 hPa, 850 hPa, 900 hPa, 925 hPa, 950 hPa and 1000 hPa. The figures show uncertainty estimates as a function of standard pressure levels. Note that the uncertainty estimates for the CARRA-West domain are not given for exactly the same variables as for the CARRA-East domain due to technical processing challenges. The Upper Bound Standard Deviation of the analysis errors is shown as a solid curve and the Refined standard deviation (STDV) of the analysis error is shown as a dashed curve.

...

Questions & Answers on field uncertainty estimates

For information and "questions and answers" (Q&A) on the uncertainty estimates for the ERA5 host reanalysis used on the lateral boundaries of this Arctic Regional Reanalysis, see this link:
ERA5: uncertainty estimation

...

For advanced users needing more detailed uncertainty information, uncertainties from ERA5 for the actual variable, time and position, might provide an additional indication (upper bound as concerns the near-surface variables) of the actual uncertainty of CARRA.

Uncertainties relative to observations: Verification statistics

Any observation or analysis of a meteorological variable is uncertain. For an observation, the uncertainty may be assessed by comparing measurements performed with similar instruments. In average, these instruments should observe the same values, but they will not get exactly the same results. Often the distribution of such random and independent deviations relative to the average measurement is a Gaussian distribution. This is defined by having a mean value 𝛍 and a standard deviation 𝛔 (STDV) as shown in Figure 9. The numbers on the distribution shown here are arbitrary and are just meant as an illustration.

...

For further details see the test and verification report for CARRA: https://datastore.copernicus-climate.eu/documents/reanalysis-carra/CARRATestVerificationFinal.pdf and the full system documentation: https://datastore.copernicus-climate.eu/documents/reanalysis-carra/CARRAFullSystemDocumentationFinal.pdf

Known Issues

Since the regional reanalysis is run nested into the ERA5 global reanalysis, it is affected by the known issues of ERA5. In addition to those issues, we have found that ERA5 uses incorrect glacier masks for most of the glaciers in the regional Arctic reanalysis domain, and the glaciers in ERA5 always have an analysis albedo of 0.85. This is wrong, since for instance exposed glacier ice albedos during summer are unaccounted for. These areas affect the general circulation and thermodynamic state in ERA5 and can affect the quality of the Arctic regional reanalysis.
Additionally, the Arctic reanalysis has the following known issues:

• https://confluence.ecmwf.int/x/p3ltD#CopernicusArcticRegionalReanalysis(CARRA):knownissuesanduncertaintyinformation-Knownissues

Annex: Brief outline of the Arctic reanalysis system

The details about the Numerical Weather Prediction model system used can be found in the full system documentation: https://datastore.copernicus-climate.eu/documents/reanalysis-carra/CARRAFullSystemDocumentationFinal.pdf

To supplement this we provide in the following a short description of the service as well as a description of basic principles of data assimilation in numerical weather prediction aimed at a non-specialist audience.

The service

The C3S_322_Lot2 contract of the Copernicus Climate Change Service produces and delivers a regional reanalysis (RRA) for the Arctic including long-term datasets of Essential Climate Variables (ECVs) for the 24 year period from 1997 to 2021. The model domains are shown in Figure 1. The modelling system has a resolution of 2.5x2.5 km² and 65 vertical levels in the atmosphere. The produced datasets are freely available and can be used by anyone who wishes detailed long-term atmospheric data, for instance to study typical ranges of meteorological variables, as a reference for climate model runs, or to investigate highly resolved changes in the ECVs during the reanalysis period. An extended model version for the entire Arctic (pan-Arctic) area is run for a period of 1 year as a proof-of-concept.
The reanalysis model is the weather forecasting model HARMONIE-AROME cy40h1.1.1, which has been enhanced with more Arctic input data, and more extensive surface and atmospheric data assimilation. Data assimilation is explained further in section 7.2. Additionally, the model formulation has been improved with a specific focus on processes essential in the Arctic.

Principles of reanalysis systems

Atmospheric reanalysis is a method to reconstruct the atmospheric states by using historical observations (in situ, surface and satellite remote sensing) together with a weather forecasting model. It mainly provides a physically and dynamically coherent description of the state of the atmosphere. Surface variables are included mainly to the extent that they affect the atmosphere. This synthesis is accomplished by assimilating the observational data into a meteorological model and thereby forcing the model to reproduce the observations as closely as possible and make a gridded dataset that covers also locations from which no observations are available. The advantage of a reanalysis is that it provides a spatially and temporally complete, and consistent record of the atmospheric state. This cannot be achieved with an observational dataset alone, since interpolation methods cannot adequately represent the non-linear variability of the atmosphere.
The main advantages of reanalysis dataset are:

...

The Arctic reanalysis system applies the so-called 3D variational data assimilation (3D-VAR) reanalysis method. The 3D-VAR method is depicted schematically in Figure 19. At fixed points in time the model state is adjusted based on the observed state, taking into account the statistics of model and observation errors. The Arctic reanalysis system is run with eight cycles per day performing analyses at 00 UTC, 03 UTC, 06 UTC, 09 UTC, 12 UTC, 15 UTC, 18 UTC and 21 UTC. The forecast lengths vary between 3 and 30 hours.

Related articles