...
The purpose of the reanalysis applications is to reconstruct the past evolution of the atmospheric flow striving for as high as possible level of spatial-temporal consistency of the data. In order to achieve this a latest state-of-the-art numerical weather prediction model is initialized with consistent records of observations collected over (a long) time period. The CARRA reanalysis data set aims to provide a high resolution reconstruction of the atmospheric flow over Arctic over the last 24 years. The CARRA reanalysis system is based on the HARMONIE-AROME convective-scale non-hydrostatic numerical weather prediction system, see details on the system in Yang et al (2020a) and how it has been verified and tested in Yang et al (2020b). The HARMONIE-AROME system is used for operational weather forecasting at several National Weather Centers in Europe, including Danish Meteorological Institute and MetCoOp, the operational collaboration between Norway, Sweden and Finland. The CARRA reanalysis system employs a 3D-Variational data assimilation scheme with three hour update frequency to constrain the model simulations with observations. The observations are combined optimally with the short-range high resolution background field taking into account uncertainty about the background fields and the observations. The analysis is released in a so-called cycled environment, where a short-range HARMONIE-AROME forecast initiated from the produced analysis is used as a background field for the next analysis.
...
The CARRA reanalysis has a quite comprehensive list of output fields to users, see the Data User Guide (Nielsen et al, 2019). We note from the statistical approach for deriving uncertainty information described above, that we have the restriction that we can only assess the uncertainty of the physical variables available and stored within the 3D-Var system, which is a small subset of the full output list from the model system. Also, to obtain robust statistics of uncertainty related parameters from a limited size ensemble, we need a certain aggregation of statistics, so we expect for instance that local gridpoint-to-gridpoint variations will be noisy with variations not reflecting the real variations of uncertainty. In summary, we propose to provide provided uncertainty estimates for a subset of the full output list at a certain aggregation level to be described in the followinghereafter.
As for what uncertainty measures to provide to users, we propose to provide estimates of of standard deviations of errors. The This framework we have will for instance is not be suited for providing biases.
We will not don't provide uncertainty estimates for most of the single level parameters and soil level parameters listed in the output tables of Sections 3.1 and 3.2 of the DAta User Guide (Nielsen et al, 2019; Copernicus Arctic Regional Reanalysis (CARRA): Data User Guide), but will provide uncertainty information uniquely for: Mean sea level pressure (GRIB code 151)
Several of the remaining single level parameters in these tables will be are assessed by verification information at observing stations, as outlined in Section 3 above, which we propose to include in the next update of the Data User Guide. The variables provided in this way and the breakdown on sub-regions/areas will be are as in Tables 3.1 and 3.2 of the full system documentation (Yang et al, 2020a)., Copernicus Arctic Regional Reanalysis (CARRA): Full system documentation).
For model level variables (see Section 3.3 of the Data User Guide), we will provide uncertainty information as outlined in Section 2 above for:
- Specific humidity (GRIB code 133)
- Temperature (GRIB code 130)
- u component of wind (GRIB code 131)
- v component of wind (GRIB code 132)
We will also provide uncertainty information for these physical variables on pressure levels (a subset of the complete list of pressure level variables in Section 3.4 of the Data User Guide).
We will provide uncertainties as horizontally averaged vertical profiles for each physical parameter, not as full 3D fields due to the issues of statistical robustness mentioned above. These estimates will possibly be provided to the MARS archive (and the Climate Data Store) in the next update of the CARRA output for MARS storage. (MARS archiving is already running for our main output variables and there will anyway be a second step with updating of the MARS archiving and later data release, for the parameters which depend on the GRIB approval process.) The feasibility and effectiveness of storing on MARS will be further discussed with the ECMWF MARS team, and this discussion will form the basis for a proposed decision on in which form(s) the uncertainties will be presented to users.
In the derivation of the uncertainty, standard deviation of errors will be computed for both domains and both for winter and summer. We will
In the derivation of the uncertainty, standard deviation of errors are computed for both domains and both for winter and summer. We assess the differences between domains and seasons to decide on whether there are significant and robust differences in the uncertainties. We will probably provide average values between the domains unless we identify large differences, and most likely provide average values over the year and not seasonal values.
The provided uncertainty statistics is available from Copernicus Arctic Regional Reanalysis (CARRA): known issues and uncertainty information#Uncertaintyinformation
...