...
- ERA5T: for the months September, October, November and at least a part of December 2021, the final ERA5 product is and will be different from ERA5T due to correction of the assimilation of incorrect snow observations in central Asia. Although the differences are mostly limited to that region and mainly to surface parameters, in particular soil moisture and to a lesser extent , 2m temperature and 2m dewpoint temperature, all the resulting reanalysis fields do can differ slightly over the whole globe within their uncertainty (which is estimated by the ensemble spread). On the CDS disks, the initial, ERA5T, fields have been and will be overwritten, i.e., for these months, access to the original CDS disk, ERA5T product is not possible after it has been overwritten. Potentially incorrect snow observations have been assimilated in ERA5 up to this time, when the effects became noticeable. The quality control of snow observations has been improved in ERA5 from September 2021 and from 15 November 2021 in ERA5T.
- ERA5 uncertainty: although small values of ensemble spread correctly mark more confident estimates than large values, numerical values are over confident. The spread does give an indication of the relative, random uncertainty in space and time.
- ERA5 suffers from an overly strong equatorial mesospheric jet, particularly in the transition seasons.
- From 2000 to 2006, ERA5 has a poor fit to radiosonde temperatures in the stratosphere, with a cold bias in the lower stratosphere. In addition, a warm bias higher up persists for much of the period from 1979. The lower stratospheric cold bias was rectified in a re-run for the years 2000 to 2006, called ERA5.1, see "Resolved issues" below.
- Discontinuities in ERA5: ERA5 is produced by several parallel experiments, each for a different period, which are then appended together to create the final product. This can create discontinuities at the transition points.
- The analysed "2 metre temperature" can be larger than the forecast "Maximum temperature at 2 metres since previous post-processing".
- The analysed 10 metre wind speed (derived from the 10 metre wind components) can be larger than the forecast "10 metre wind gust since previous post-processing".
- ERA5 diurnal cycle for near surface winds: the hourly data reveals a mismatch in the analysed near surface wind speed between the end of one assimilation cycle and the beginning of the next (which occurs at 9:00 - 10:00 and 21:00 - 22:00 UTC). This problem mostly occurs in low latitude oceanic regions, though it can also be seen over Europe and the USA. We cannot rectify this problem in the analyses. The forecast near surface winds show much better agreement between the assimilation cycles, at least on average, so if this mismatch is problematic for a particular application, our advice would be to use the forecast winds. The forecast near surface winds are available from MARS, see the section, Data organisation and how to download ERA5.
- ERA5 diurnal cycle for near surface temperature and humidity: some locations do suffer from a mismatch in the analysed values between the end of one assimilation cycle and the beginning of the next, in a similar fashion to that for the near surface winds (see above), but this problem is thought not to be so widespread as that for the near surface winds. The forecast values for near surface temperature and humidity are usually smoother than the analyses, but the forecast low level temperatures suffer from a cold bias over most parts of the globe. The forecast near surface temperature and humidity are available from MARS, see the section Data organisation and how to download ERA5.
- ERA5: large 10m winds: up to a few times per year, the analysed low level winds, eg 10m winds, become very large in a particular location, which varies amongst a few apparently preferred locations. The largest values seen so far are about 300 ms-1.
- ERA5 rain bombs: up to a few times per year, the rainfall (precipitation) can become extremely large in small areas. This problem occurs mostly over Africa, in regions of high orography.
- Large values of CAPE: occasionally, the Convective available potential energy in ERA5 is unrealistically large.
- Ship tracks in the SST: prior to September 2007, in the period when HadISST2 was used, ship tracks can be visible in the SST.
- Prior to 2014, the SST was not used over the Great Lakes to nudge the lake model. Consequently, the 2 metre temperature has an annual cycle that is too strong, with temperatures being too cold in winter and too warm in summer.
- The Potential Evaporation field (pev, parameter Id 228251) is largely underestimated over deserts and high-forested areas. This is due to a bug in the code that does not allow transpiration to occur in the situation where there is no low vegetation.
- Wave parameters (Table 7 above) for the three swell partitions: these parameters have been calculated incorrectly. The problem is most evident in the swell partition parameters involving the mean wave period: Mean wave period of first swell partition, Mean wave period of second swell partition and Mean wave period of third swell partition, where the periods are far too long.
- Surface photosynthetically available radiation (PAR) is too low in the version (CY41R2) of the ECMWF Integrated Forecasting System (IFS) used to produce ERA5, so PAR and clear sky PAR have not been published in ERA5. There is a bug in the calculation of PAR, with it being taken from the wrong parts of the spectrum. The shortwave bands include 0.442-0.625 micron, 0.625-0.778 micron and 0.778-1.24 micron. PAR should be coded to be the sum of the radiation in the first of these bands and 0.42 of the second (to account for the fact that PAR is normally defined to stop at 0.7 microns). However, in CY41R2, PAR is in fact calculated from the sum of the second band plus 0.42 of the third. We will try to fix this in a future cycle.
Expand title The instantaneous turbulent surface stress components (eastward and northward) and friction velocity tend to be too small The ERA5 analysed and forecast step=0, instantaneous surface stress components and surface roughness and the forecast step=0, friction velocity (friction velocity is not available from the analyses in ERA5) tend to suffer from values that are too low over the oceans.
The analysis for such parameters is obtained by running the surface module to connect the surface with the model level analysed variables.
However, at that stage, the surface aero-dynamical roughness length scale (z0) over the oceans is not initialised from its actual value but a constant value of 0.0001 is used instead.
This initial value of z0 is needed to determine the initial value of u* and the surface stress based on solving for a simple logarithmic wind profile between the surface and the lowest model level. This initial u* is in turn used to determine an updated value of z0 based on the input Charnock parameter and then the value of the exchange coefficients needed to determine the output 10m winds (normal and neutral) and u* (see (3.91) to (3.94) with (3.26) in the IFS documentation). The surface stress is output as initialised.
This initial value for z0 is generally too low ( by one order of magnitude or more):
Over the oceans, for winds above few m/s, z0 is modelled using the Charnock relation:
z0 ~ (alpha/g) u*2
where alpha is the Charnock parameter, g is gravity, and u* is the friction velocity
with typical values of
alpha ~ 0.018
g=9.81
u*2 = Cd U102
where Cd is the drag coefficient
Cd ~ 0.008 + 0.0008 U10
for U10=10m/s => z0 ~ 0.003
As a consequence, the analysed instantaneous surface stress components will tend to be too low and even the updated value of z0 (surface roughness) will also tend to be too low.
For forecast, instantaneous surface stress components, surface roughness and friction velocity, the same problem affects step 0. However, this problem will not affect the accumulated surface stress parameters (recall the accumulated parameters are produced by running short range forecasts), because the accumulation starts from the first time step (i.e. at time step 0 all accumulated variables are initialised to 0).
This problem can easily be fixed, by using the initial value of Charnock that is available at the initial time.
Note, in ERA5 the parameter for surface roughness is called "forecast surface roughness", even when it's analysed.
Expand title ERA5 forecast parameters are missing on 1st January 1979 from 00 UTC to 06 UTC ERA5 forecast parameters are missing for the validity times of 1st January 1979 from 00 UTC to 06 UTC. This problem has occurred because the forecast producing these data started from 18 UTC on the last day of 1978. This gap can be filled by using forecast data from the ERA5 back extension (preliminary version), with date=19781231, time=18 and step=6/to/12:
Code Block language py title Request for total precipitation forecast hourly data for 1st January 00UTC-06UTC #!/usr/bin/env python3 import cdsapi c = cdsapi.Client() c.retrieve('reanalysis-era5-complete-preliminary-back-extension', { 'date': '1978-12-31', 'levtype': 'sfc', 'param': '228.128', 'time':'18:00:00', 'step':'6/7/8/9/10/11/12', 'stream': 'oper', 'type': 'fc', 'grid': '0.25/0.25', 'format': 'netcdf', }, 'era5.preliminary-back-extension-temperature-tp.nc')Eventually, the data gap will be filled by the re-run of the ERA5 back extension.
Maximum temperature at 2 metres since previous post-processing: in a small region over Peru, at 19 UTC, 2 August 2013, this forecast parameter exhibited erroneous values, which were greater than 50C. This occurrence is under investigation. Note, in general, we recommend that the hourly (analysed) "2 metre temperature" be used to construct the minimum and maximum over longer periods, such as a day.
Expand title Four reasons why hourly data might not be consistent with their monthly mean The ERA5 monthly means are calculated from the hourly (3 hourly for the EDA) data, on the native grid (including spherical harmonics) from the GRIB data, in each production "stream" or experiment. This can give rise to inconsistencies between the sub-daily data and their monthly mean, particularly in the CDS. In general, the inconsistencies will be small.
- In the CDS, the ERA5 data (sub-daily and monthly mean) has been interpolated to a regular latitude/longitude grid. This interpolated sub-daily data will be slightly different to the native sub-daily data used in the production of the ERA5 monthly means.
- The netCDF data available in the CDS has been packed, see What are NetCDF files and how can I read them, which states "unpacked_data_value = (packed_data_value * scale_factor) + add_offset" and "packed_data_value = nint((unpacked_data_value - add_offset) / scale_factor)". This netCDF packing will change the sub-daily values slightly, compared with the native sub-daily data used in the production of the ERA5 monthly means.
- The GRIB data in the ERA5 monthly means (and sub-daily data) has been packed using a binning algorithm (which is different to the netCDF packing algorithm). Monthly means produced in other formats, such as netCDF, will differ from the ERA5 monthly means because of this packing.
- Finally, there is a further reason why monthly mean values might be different to the mean of the sub-daily values, which even occurs in MARS. This cause only affects forecast parameters (the CDS provides analysed parameters unless the parameter is only available from the forecasts), such as the Total precipitation, and only occurs sporadically. In order to speed up production, ERA5 is produced in several parallel "streams" or experiments, which are then spliced together to produce the final product. Consider, the "stream" change at the beginning of 2015. The ERA5 forecast monthly means for January 2015 have been produced from the sub-daily data from that "stream", the first few hours of which (up until 06 UTC on 1st January 2015) come from the 18 UTC forecast on 31 December 2014. However, the sub-daily forecast data published in ERA5, is based on the date of the start of the forecast, so these first few hours of 2015 originate from the "stream" that produced December 2014. These two "streams" are different experiments, with different data values. The resulting inconsistencies might be larger than for the other three causes, above, depending on how consistent the two streams are.
- ERA5 sea-ice cover and 2 metre temperature: in the period 1979-1989, in a region just to the north of Greenland, the sea-ice cover outside of the melt season is too low and hence the 2 metre temperature is too high. For more information, see Section 3.5.4 of Low frequency variability and trends in surface air temperature and humidity from ERA5 and other datasets
- ERA5 CDS: wind values are far too low on pressure levels at the poles in the CDS
- ERA5 back extension 1950-1978 (Preliminary version): tropical cyclones are too intense
- ERA5 back extension 1950-1978 (Preliminary version): large bias in surface analysis over Australia prior to 1970
- ERA5 back extension 1950-1978 (Preliminary version): the deep soil moisture tends to be too dry
...