The Copernicus Arctic Regional reanalysis (CARRA) dataset is already in use for couple of years and hereafter we gathered some user questions, which were primarily collected from the registration forms of the forthcoming CARRA User Workshop (21 September, 2023, and also some user questions submitted to the Copernicus User Support. Hereafter we list these questions (some of the questions are edited for clarity) classified into some major categories and we try to give brief answers to them.


Would you tell me more, please if CARRA is based on AROME-Arctic or the same HARMONIE-AROME setup of MetCoOp (​​​​​​​MetCoOp is the Meteorological Cooperation on Operational Numerical Weather Prediction (NWP) between Estonia Environment Agency, Finnish Meteorological Institute, MET Norway and Swedish Meteorological and Hydrological Institute)?

CARRA1 and CARRA2  have no direct links to MetCoOp model setup, although these systems are all based on the HARMONIE forecast system. CARRA1 as a reanalysis system is an extension to the HARMONIE forecast system and as such has similarities in configurations to the Danish-Icelandic forecast system (over Greenland and Iceland) and the Norwegian forecast system (AROME-Arctic) over Svalbard-North Scandinavia. However, the main code ingredients of the CARRA system are the same as used in both MetCoOp and AROME-Arctic as well as the Danish-Icelandic "IGB" operational NWP suite: main parameterisations, the non-hydrostatic dynamics and the 3D-Var data assimilation scheme. The differences between the three operational NWP systems CARRA builds upon are minor, but with some notable differences for instance on observation usage. For CARRA there was a testing phase for finding the best choices for the elements where the operational systems differed. On top of that extensive developments was done for the CARRA system to exploit and adapt to enable good use of many types of historical observations not used in the operational NWP systems. This also included adaptations for high-resolution climate consistent data sets for surface properties such as snow cover, sea surface temperature, sea ice concentrations and glacier albedo.

Regarding the sea ice parametrisation (SICE): is that implemented with a constant sea ice thickness with number of layers or can the ice column actually evolve?

The SICE routine utilises thermodynamic relations to model the evolution of the sea ice thickness. The ice thickness is initialised once per production stream (different streams covering different time periods) with a one year spin up period each. Then the ice cover evolves freely as modelled by SICE without applying any external constraints. There is no consistency for perennial ice across the streams, and there is indeed a jump between the end of one stream and the beginning of a new one. This is only a problem in regions where there is ice all the time. For parts of the domain which are open some times of the year, the thickness will develop anew. Figure 1 below illustrates this. Here we see clear jumps between one stream and the next. These jumps would be more pronounced if only multi-year ice would be included.

Figure 1. Monthly ice thickness anomalies in the CARRA product and fitted ice thickness anomaly trend. Multiyear monthly means for the time period from 2000 to 2020 are used as a reference when computing anomalies. (a) Western CARRA model domain; (b) eastern CARRA model domain. Also on the panels, the 95% confidence interval of the CARRA anomaly trend is shown, and separate production streams S1–S5 of the CARRA system and the back extension stream BE are marked. Batrak et. al (2023; Preprint)

Figure 2 (Batrak, pers. comm) shows probably the most extreme discontinuity of the ice thickness evolution in the CARRA data set, which happens at the transition from the back extension stream to the first production stream.

FIgure 2. Ice thickness series for a number of points which include perennial ice to the north of Greenland, annual ice cover in the Baffin Bay and some kind of multiyear ice to the east of Greenland (disappears at times but mostly stays there). (Batrak, pers. comm).

How significant is the (spin up) error from initialization for the 1H, 2H, 3H forecasts? Could you make the data access in CDS consistent with e.g. the ERA5-Land data?

In Numerical Weather Prediction (NWP) the spin-up effect is associated to the initial imbalance between some physical quantities (typically the hydrological cycle-related quantities), which results in some noise in the ensuing forecasts. In practise, we consider spin-up deficiency mainly for accumulated properties involving moist processes such as precipitation and evaporation.

Note that CARRA analyses are performed at every 3h. However we don't analyse all variables and therefore there are cumulative quantities, which are provided as forecasts. Generally, we don't recommend to combine analyses and forecasts, because they might result in some jumps between analysis and forecast steps.  See more information about this issue at Copernicus Arctic Regional Reanalysis (CARRA): Added value to the ERA5 global reanalysis#AnalysisversusForecasts?

For instantaneous quantities if you absolutely need hourly frequency then you can use the 1h and 2h forecasts in intermediate times since the spin-up effects for these instantaneous quantities are considered to be minimal. 

For accumulated quantities though we recommend to use forecasts always longer than 6h range while constructing the various time windows (for instance the difference between the 12h and 6h forecasts from the same initial condition can give a spin up free 6h accumulation). In general the guidance provided at The use of precipitation information from the Copernicus Arctic Regional Reanalysis (CARRA) should be followed. 

Some general description about the spin-up in CARRA can be found at Copernicus Arctic Regional Reanalysis (CARRA): Data User Guide#Singlelevelvariables

How do the heat flux calculations differ in CARRA from ERA5?

In both cases sensible and latent heat fluxes (H and LE, respectively) formulations are based on aerodynamic resistances. But there are some differences. We describe some of them below.

The expressions of H and LE in ERA5 and CARRA can be described as:

H=rho*Ch*U*Ft  and LE=rho*Lv*Ch*U*Fq

where rho is the air density, Ch the turbulent exchange coefficient for heat and vapour, Lv is the latent heat of vaporization, U is the wind speed, Ft is a function of air and surface temperature, and Fq is a function of air and surface specific humidity, respectively. Ft and Fq represent the temperature and specific humidity gradients between surface and the environmental air, respectively.

In ERA5 and CARRA Ch, U, and Ft are different.

  • In ERA5 Ch is calculated from the Obukhov length based on an iterative process to compute the Richardson number whereas CARRA uses directly the Richardson number (Louis 1979 formulation).
  • In ERA5 U (wind modulus) accounts for free convection velocity w* but not in CARRA, which uses only the wind components u and v.
  • In ERA5 Ft uses enthalpy where CARRA uses surface temperature.

Please note that the list of differences above might not be exhaustive. 

What are major differences between the ERA5 reanalysis and CARRA?

First of all ERA5 is a global reanalysis based on the IFS model and CARRA is a regional reanalysis targeting the European part of the Arctic based on the HARMONIE-AROME model. They are different in terms of horizontal resolution (31 km in ERA5 vs. 2.5 km in CARRA), data assimilation algorithm (4D-Var in ERA5 vs. 3D-Var in CARRA), observation usage etc. In CARRA more emphasis was put on spatial resolution and on extended use of in-situ observation data. CARRA is providing added value with respect to ERA5. This is explained in details at the Copernicus Arctic Regional Reanalysis (CARRA): Added value to the ERA5 global reanalysis page. 

What is the accuracy of snow depth estimates?

The snow is an analysed quantity in CARRA, which means that the snow analysis corrects the modelled snow amount using the observed snow.

The following observations are used in the snow analysis:

  1. Snow depths at synoptic stations and at additional networks of "climate observing stations".
  2. "Pseudo observations" based on satellite products giving probability of snow in cloud free regions.

The performance of such analysis is good in regions with representative observations. Lack of conventional observations in parts of the domain, e.g. in Greenland and Svalbard, necessitated the use of satellite snow data for CARRA. The use of satellite snow extent in addition to snow depth observations helps to distinguish between snow free and snow covered ground in the melting season and leads also to improved T2m scores.

The snow analysis in CARRA uses (for technical reasons) climatological mean values of snow density in the conversion of observed snow depths to Snow Water Equivalent (SWE), leading to small systematic errors in the snow amounts during the winter.

Validation and qualitative considerations of the CARRA snow on Svalbard is published at Maniktala, D. (2022). Analysing seasonal snow cover trends and patterns on Svalbard,

How is the solar radiation downward data obtained? What is the basic principle behind it?

The solar downward radiation is computed from detailed 2-stream radiative transfer computations for all 66 model half-levels including the surface. For the gas radiative transfer 112 RRTM-SW spectral g-points are included, and for clouds and aerosols 6 spectral bands. See more details at 

Iacono, M.J., Mlawer, E.J. and Clough, S.A., 2001, March. Validation of the RRTM shortwave radiation model and comparison to GCM shortwave models. In Proc. 11th Atmospheric Radiation Measurement (ARM) Science Team Meeting.

Iacono, M.J., Delamere, J.S., Mlawer, E.J., Clough, S.A. and Morcrette, J.J., 2002, April. Cloudy sky RRTM shortwave radiative transfer and comparison to the revised ECMWF shortwave model. In Twelfth ARM Science Team Meeting Proceedings, St. Petersburg, Florida (pp. 8-12)

Fu & Liou (1992):<2139:OTCDMF>2.0.CO;2

Mlawer et al. (1997):

I would love to see comparisons of surface winds, temperatures and precipitations to ERA5.

General comparisons between ERA5 and CARRA are provided at Copernicus Arctic Regional Reanalysis (CARRA): Added value to the ERA5 global reanalysis

List of references where such comparisons might be available are at Copernicus Arctic Regional Reanalysis (CARRA): list of references

Particularly, see more details on

What would be your guess why CARRA snow thickness (on sea ice) is usually much larger than from altimetry?

It is not clear which altimetry product the question is referring to and how accurate it is. However, a comparison study of such data with CARRA data would be of interest. The modelling of snow on sea ice in CARRA applies the SICE scheme (Batrak et al, 2018, This scheme has been shown to give advances over previous schemes, which don't attempt to take the snow on sea ice into account, however it still does not give a perfect representation of the snow on sea ice. To speculate about sources for possible biases, one could mention the absence of ice dynamics in the scheme: If there is an area of intensive snow accumulation, ice has no chance to transport this snow and it keeps piling up in the same grid cells. Also the absence of snow-ice formation in the model (even though it isn't as important in the Arctic as in the Antarctic). Additionally, the CARRA system doesn't represent the process of snow precipitation loss to the open sea portion of a grid cell when ice concentration is below 100%.


Can you give a tutorial, please on how to obtain hourly precipitation timeseries? Why is it left to the user to produce this?

It is difficult to download and calculate daily accumulated precipitation, would this be simplified? What are the interpretation of the precipitation forecast "leadtime" and "time" variables?

The precipitation amount is stored as accumulated precipitation since the forecast start time. This is a standard procedure in Numerical Weather Prediction (NWP) although some centres re-compute the precipitation amounts for shorter time intervals (like ERA5 for 1h intervals). In CARRA this was not considered, instead it was thoroughly explained in the CARRA Data User Guide (Copernicus Arctic Regional Reanalysis (CARRA): Data User Guide#Singlelevelvariables) and even a designated page was created to explain all the details for the users ( The latter document also includes some recommendations and examples. 

Have you considered any user-friendly web tools that can be used to visualize CARRA data on a map?

This request can be considered in the context of the entire C3S CDS (Climate Data Store) infrastructure. At the moment we don't provide visualisation tools for the users (for any CDS datasets, but only for applications), however it might be easier to have once the new CDS (CADS as its new name) will be deployed (probably in Q1 2024). In the CARRA User Workshop ( we plan to show some (Python) tools where such data visualisation will be possible. 

I couldn’t download specific humidity data for CARRA (West-domain) from the Copernicus server for June and July 1991. Would you check this, please?

We have tried some random time steps and are able to download this. If your problem persists, please open a ticket at our User Support (

Do you have CARRA specific humidity data available somewhere? Or could you provide a calculation how specific and relative humidity depend on each other in the CARRA reanalysis?

We have 2m relative and 2m specific humidity for the CARRA single level CDS entry at!/dataset/reanalysis-carra-single-levels?tab=form.

For pressure and height levels we have only relative humidity and for model levels only specific humidity. See at 

The conversion between relative and specific humidity is therefore relevant for the pressure, height and model level CDS entries.  

The thermodynamic conversion information between these two quantities can be found in the literature and even it is easy to find having a web search.

How many CARRA in-situ observations are used in the reanalysis but not presently available to e.g. CDS or the public?

Generally, we don't provide the observations used in CARRA for the public. This is due to data policy issues.

Extensive work have been done to collect and use local in-situ observations in CARRA. Traditionally  the European Arctic region (Greenland, Island, Svalbard etc) is rather observation-sparse. By collecting non-GTS in-situ observations, the amount of assimilated surface observation data increases by multi-fold. Over Greenland, observation data collected over non-populated ice sheets by the Greenland Climate Network (GCnet), PROMICE (PROgramme for Monitoring  Greenland ice Sheet) have been used in CARRA. Further, satellite derived albedo products over Greenland ice sheets have been used, making the reanalysis system unique in providing observation improvement for description of the vast Greenland ice sheet.

The short description of the CARRA system is available at Copernicus Arctic Regional Reanalysis (CARRA): Data User Guide#Annex:BriefoutlineoftheArcticreanalysissystem

Detailed information on the CARRA system and its data usage can be found at

WIll CARRA reanalysis data for radiation parameters be published?

We have several radiation parameters in the CDS at the single level CARRA entry:!/dataset/reanalysis-carra-single-levels

These are as follows:

  • Direct solar radiation
  • Surface latent heat flux
  • Surface net solar radiation
  • Surface net solar radiation clear sky
  • Surface net thermal radiation,
  • Surface net thermal radiation clear sky
  • Surface solar radiation downwards
  • Surface thermal radiation downwards
  • Time-integrated surface direct short wave radiation flux
  • Top net solar radiation
  • Top net thermal radiation


What is the status of the overview article “Schyberg et al.: The Copernicus Arctic regional reanalysis”? (Announced at: Copernicus Arctic Regional Reanalysis (CARRA): list of references as in preparation)

Unfortunately, this was delayed. The manuscript is still in preparation and will hopefully be ready for review by the end of this year.

Would you give some information on the progress of the pan-Arctic CARRA (CARRA2) system? When this system will be available for users and what are the updates of the new CARRA2 reanalysis with respect to CARRA1?

Some information about the CARRA2 pan-Arctic system can be found at See section: "Plans for the next generation Arctic reanalysis: pan-Arctic extension" and particularly the itemised list of improvements of CARRA2 with respect to CARRA1.

In the CARRA User Workshop (21 September, 2023; we will give a short introduction to CARRA2 including its timeline. The CARRA2 production is expected to start in the spring of 2024 and finish by autumn of 2026. The CARRA2 reanalysis data will be gradually loaded into to CDS. The public release of the first batch of CARRA2 data in CDS is expected around March 2025. As such the complete datasets are likely to be available towards the end of 2026. CARRA2 will be covering the pan-Arctic area north of 60 degree with comparable spatial and temporal details as in CARRA1, although the latter covers only the European side of the Arctic.

What is the main difference between the pan-Arctic CARRA product compared to ERA5, and when CARRA2 is accessible for a user test?

The release of the first batch of CARRA2 data is scheduled for March 2025.  Main differences: CARRA2 will be produced with a different NWP system, a regional system which has higher horizontal resolution (better than 3 km grid size) than ERA5 and ERA6, and will be adapted for using higher-resolution data sets describing the surface. See also our answer to the previous question.

Will there be any improvements in the representation of sea ice conditions in the arctic fjords in CARRA2?

This is now under discussion in the CARRA team for the preparations of the CARRA2 system therefore we can provide meaningful answer only in a later stage.

Is there any attempt to produce wave variables in CARRA in the future

The CARRA reanalysis system is an atmosphere-land system, which does not include ocean component. Consequently, there are no wave variables available. Those data are available from global coupled reanalysis systems like ERA5 (also available in the CDS) or from the Copernicus Marine Service (see at There is also a HARMONIE-AROME based high resolution wave hindcast dataset available at

Are there plans to make data extraction more accessible to improve user uptake (e.g. targeted extraction of data from smaller regions/points; availability of data through AWS/Google Earth Engine)? Is there a way to download CARRA data from the Copernicus server for a single grid cell without the need to download the complete domain first?

This is a valid question what we are aware of. First of all we tried to speed up the access to the data, i.e. faster download speeds for the full (whole domain) data files. We think that with the migration of our archiving system to Bologna this gave some improvements. We are also planning to put the most used parts of the CARRA datasets into spinning discs instead of tapes. The next step will be to make possible for the users to have any geographic subset of the CARRA data. This is work in progress (the complications are related to the distorted Lambert projection geometry towards the Pole and the fact that the netcdf files don't have a good data compression capabilities).

When will monthly and daily means be available?

The software computing CARRA daily and monthly mean values are ready and will be deployed soon for the entire CARRA dataset. Since we need to get through the entire more than 30-year sub-daily CARRA dataset for the daily/monthly mean computations it will take a bit of time to complete the process and open the dataset for public use. Our latest estimate is that the means will be available for the users around mid-2024. We will give news on our progress in the CARRA User Forum (see at Copernicus Arctic Regional Reanalysis (CARRA)).