Leap year 2024 in on-the-fly reforecasts
Hindcast production challenge for 29 Feb 2024
1. Brief description:
The approaches, how to produce on-the-fly reforecasts for Thursday 29 February, differ. The solutions used by 4 centres impacted by 29 Feb 2024 were:
- ECMWF, CMA
- date=20240229
- hdate=YYYY0228
- YYYY=past 15 years for CMA
- YYYY=past 20 years for ECMWF
- ECCC, RHMC (29 Feb replaced by 1 Mar)
- date=20240301
- hdate=YYYY0301
- YYYY=2001-2020
2. Recommendation:
Users should be aware, how 29 Feb 2024 on-the-fly reforecasts were produced by different centres.
For the future (e.g.2028), all impacted centres like CMA, ECCC, ECMWF, RHMC will be recommended to use 1 March, so that the solution for users would be the same.
BoM (ammc)
Wrong maximum values of surface air maximum temperature (Tmax) at some points for BoM data
1. Brief description:
Some coastal grid points may display unrealistic Tmax values. All of the spurious Tmax points are flagged as being ocean points based on the POAMA*'s land-sea mask.
2. Recommendation:
To eliminate the spurious values, use Tmax over land points only, based on POAMA's land-sea mask .
* POAMA stands for Predictive Ocean Atmosphere Model for Australia.
Problem with sea-ice cover in BoM re-forecast data
1. Brief description:
The sea-ice cover data in the BoM re-forecasts is incorrect. The climatological evolution of sea ice in the BoM re-forecats is for 1 January through to early March (62 day re-forecasts) for all start dates. Therefore, the sea ice coverage is wrong in the S2S archive, except for the forecasts that start on or near 1 January
2. Recommendation:
It is recommended to not download the BoM sea-ice cover re-forecast data until the correct data has been fully uploaded in the ECMWF S2S data server. There will be an announcement when the problem has been fixed.
GRIB 2 encoding - incorrect packing type (technical issue not an error) Fixed since 30 Jan 2020
1. Brief description:
All data for S2S is required to be encoded in GRIB 2 format with simple packing (to allow faster data processing by users although it results in bigger data volume). The realtime outputs from BoM were unwantedly using grib_jpeg advanced packing until 30 Jan 2020. In contrary, the reforecast have always been encoded correctly using the required simple packing.
2. Recommendation:
Users should be aware of the changing GRIB 2 packing for given model.
Incorrect values of 3 parameters in BoM real-time data since February 2020
1. Brief description:
Due to technical problem related to data encoding, 3 parameters could have incorrect values (very high maximum):
- Convective precipitation (only step +24h)
- Total precipitation (only step +24h)
- Snow depth water equivalent (not only step +24h)
2. Recommendation:
Users should consider the issue when using affected parameters. The issue will be fixed only in the next planned model version ACCESS-S2 (implemented probably later in 2020)
CMA (babj)
Problem with sea-ice cover in CMA re-forecast data. Fixed 31st March 2016
1. Brief description:
The sea-ice cover data in the CMA re-forecasts was incorrect in the ECMWF S2S database prior to 31st March 2016. CMA confirmed that there was a problem in their data processing procedure which main effect was to have sea-ice values equal to 0 or 1 instead of covering the range 0 to 1.
The correct data was replaced in the ECMWF S2S database on 31st March 2016.
2. Recommendation:
For users that have downloaded CMA's sea-ice cover reforecast prior to 31st March 2016, we recommend to download it again.
Problem with initial conditions CMA real-time data in January 2015 and 2016. Fixed 24th May 2016
1. Brief description:
CMA has informed us that the real-time forecast for 2nd to 5th January 2015 and 2016 had an issue with the initial fields used. They have corrected the initial conditions and re-run those dates. We have validated the dataset and we have replaced the data in the ECMWF S2S database.
2. Recommendation:
We recommend users to download the data above again.Snow density values
1. Brief description:
The data values for snow density are very different to other models. According to CMA the reason is that their model produces very little amount of snow. The checked extremes of the snow density have to been relaxed because of that specifically for CMA data.
2. Recommendation:
Users should consider it when using CMA snow density data
Change of reforecast type (technical issue) Fixed during 4-5.8. 2020
1. Brief description:
The CMA reforecasts after model change in November 2019 were wrongly encoded as the fixed type. Since the run 2020-06-25 the encoding was fixed to the correct on-the-fly type. The period Nov 2019 - Jun 2020 will need to be fixed so that all reforecast data is encoded the same way (on-the-fly type, not fixed one)
2. Recommendation:
Users should be aware of that situation, when CMA reforecast encoding is mixed until the fix is completed for the whole period. The affected reforecast data (2019-11-11....2020-06-22) can be re-downloaded if needed to have consistent encoding once the fix is finished. A simple fix can be applied also on user's side on already downloaded affected data. Please contact us for more details if needed.
CNR-ISAC (isac)
Numerical artifacts in reforecasts with model version date 20231016
1. Brief description:
There are numerical errors affecting only few grid cells here and there in the fixed reforecast set produced with the model version date 20231016. Only pressure level parameters U/V components, geopotential height and temperature are occasionally affected.
2. Recommendation:
Users should consider this when using affected parameters.
ECCC (cwao)
Sea-ice cover values in some re-forecast data (period 20.9.-13.12. 2018).
1. Brief description:
The sea-ice values from two hindcast years 2016 and 2017 are not correct in all Thursday's ECCC hindcasts outputs in the period 20.9.-13.12. 2018. Other hindcast years were not affected.
2. Recommendation:
Users should not use the affected data.
U and V components on pressure levels and at the 10m (period 11.7.2019-9.1.2020). Fixed during 7-18 May 2020
1. Brief description:
See an example of meridional wind at 200 hPa for the ECCC forecast initialized on 7 Nov. 2019 (incorrect contours around 40E and 200E). This interpolation error affects both real-time and reforecast data.
2. Recommendation:
If you downloaded the affected data before 18 May 2020, get the fixed data available in the archive now instead.
ECMWF (ecmf)
issue with accumulated fields at day 16
1. Brief description:
There is an issue in the "accumulated" fields at day 16 which occurs when the IFS model resolution changes. This creates a discontinuity in the data (it can be seen also in the data on the native model grid so it's not specifically a result of the regridding to the 1.5 degree S2S resolution) which manifests itself when one "deaccumulates" the data to create six hourly accumulations for example. Specifically, one will see discontinuities between the (T+360 - T+354) and (T+366 - T+360) accumulations (most people note large negative total precipitation values but it also affects the radiation parameters).
2. Recommendation:
Users should consider that issue when using the accumulated fields.
Regrettably, there is not much it can be do about this for the S2S data which has already been regridded from the native model grid to a 1.5 degree. It would need the deaccumulation to be performed using two different values at T+360, one based on the higher resolution grid which is used to compute (T+360 - T+354) and one on the lower resolution grid which is used to compute (T+366 - T+360).
Read more about the issue:
- MARS FAQ#HowcanIcomputetheaccumulationusingdatafromtheoverlapstream? (how to avoid it in standard IFS outputs, not in S2S ones)
- ECMWF Newsletter No. 108 - Summer 2006 (pp 14-20) (discussion of the problem in The ECMWF Variable Resolution Ensemble Prediction System (VAREPS))
Hindcast data need to be re-computed with the new cycle after the system change (4 runs only on 13., 17., 20. and 24.6. 2019). Fixed 25th June 2019
1. Brief description:
Four hindcast runs on 13., 17., 20. and 24.6. 2019 had to be re-computed and re-archived after ECMWF model upgrade to the new cycle 46r1 on June 11 to have both types of forecasts for S2S archive, real-time and hindcast, produced by the new model version. This problem could happen because the hindcasts are computed well in advance before the actual real-time date.
2. Recommendation:
If ECMWF hindcast data from the given runs was downloaded in the period 30.5.-25.6.2019, it should be deleted and the correct version of the data downloaded again.
IAP-CAS (anso)
2 metre dewpoint temperature is not available in all reforecasts
1. Brief description:
2 metre dewpoint temperature is available only in the period 20060101-20181231 (valid for the fixed reforecast set with Model version date 20190101).2. Recommendation:
Users should consider it when using that parameter from the specified fixed reforecasts.
JMA (rjtd)
The valid times of the maximum/minimum temperature are wrong in the re-forecast data (impacts ALL hindcast data for the model version: 2014-03-04).
1. Brief description:
See an illustration of the impact of the acknowledged issue in this document.
The accumulated parameters except for total precipitation (tp) are wrong in the real-time and re-forecast data. Fixed 15th March 2017 in the real-time forecasts.
1. Brief description:
Periods of bugs:
- Real-time forecasts data from JMA from 1st January 2015 to 15th March 2017
- Full re-forecasts data for the model version: 2014-03-04
See an illustration of the impact of the acknowledged issue in this document.
KMA (rksl)
Wind components fields on pressure levels (might impact ANY KMA outputs). Fixed by the 10th of October 2017 in all KMA data (real-time and reforecast) by re-archiving.
1. Brief description:
The data values for U/V component parameters on the pressure levels are often clearly wrong (e.g. the maximum (global) value of U component is in one case -14.4178 m/s which is not in the allowed interval [1,250] verified by the ECMWF's checking tool). That error was acknowledged by KMA and it was tried to fix it in some forecast outputs.
2. Recommendation:
Users should consider it when using KMA wind component fields on the pressure levels in both real-time and reforecast outputs.
500hPa geopotential height issue (since 29 Mar 2017) Fixed by the 10th of October 2017 in all KMA data (real-time and reforecast) by re-archiving.
- Update on 2020-03-16: according to the latest information there are still some issues remaining in the 2017 reforecasts
1. Brief description:
The data fields of geopotential on pressure levels 500 hPa and higher are wrong as per attached examples since the 29 March, 2017.
2. Recommendation:
Wait for fix in both real-time and reforecast outputs.
Soil initial data issue affecting mainly surface fields in the real-time forecasts (9-17.6.2019). Fixed by the 25th of June 2019.
1. Brief description:
Wrong soil initial data has been used in the real-time forecasts. Please refer to the graphs and maps below illustrating the issue (see soil initial fields and comparison of forecast anomalies).
2. Recommendation:
If KMA real-time data from the given runs was downloaded in the period 9-25.6.2019, it should be deleted and the correct version of the data downloaded again.
Inconsistency in the first real-time and reforecast runs archived after KMA model upgrade to GloSea6-GC3.2.
1. Brief description:
The first real-time run archived with the new model version (GloSea6-GC3.2) is in contrary to the first reforecast one which is only one month later
2. Recommendation:
Users should consider that when using the KMA reforecasts in that intermediate period between the KMA model changes from to .
Wrong values of Time-integrated top net thermal radiation after the model upgrade to GloSea6 Fixed by the 27th of June 2022.
1. Brief description:
The values of Time-integrated top net thermal radiation (paramId=179) were about 100x larger than thee should be because of the time unit conversion error. The affected periods were:
- real-time forecasts: -
- on-the-fly reforecasts: -
Please refer to the maps below illustrating the issue:
2. Recommendation:
If the affected data described above was downloaded in the period until , it should be deleted and the correct version of the data re-downloaded.
Problematic real-time KMA forecasts because of an error in the ocean initial field Fixed on the 26th of July 2022.
1. Brief description:
There was a problem with the production of the forecast data due to an error in the ocean initial field (During the process of changing the KMA Ocean Data Assimilation System, the date setting error occurred. December's ocean initial data were used to run Glosea6, with model results showed DJF season trends). It affected only short period -
2. Recommendation:
If the affected data was downloaded before , it should be deleted and the fixed version of the data re-downloaded.
Issues in the real-time KMA forecasts because of an error in the sea ice initial field Fixed on the 19th of May 2023.
1. Brief description:
Problem was identified in the initial fields for the real-time forecasts (reforecasts were not affected):
https://jira.ecmwf.int/servicedesk/customer/portal/4/SD-78328
"Sea ice initial values were prescribed along the coastline of the Northern Hemisphere's mid-to-high latitudes when the model ran due to problems with pre-processing of the sea ice initial field. (First slide, some of the Southern Hemisphere was also abnormally values prescribed.)
The longer the lead time(t=60), the bigger the difference of ice cover and SST in the results. (2nd to 3rd slides).
In the case of 2m temperature(4th slide), it was confirmed that the temperature was low with a distribution similar to the incorrectly prescribed sea ice values, and the spatial distribution disappeared after 8 days, but there was a difference in the forecast values.
Only some ocean parameters and 2m temperaturewere checked, but it affected other atmospheric variables too."
2. Recommendation:
If the affected data was downloaded, it should be deleted and the fixed version of the data re-downloaded once available.
Three problematic ocean parameters
1. Brief description:
Three of nine newly introduced ocean parameters in KMA model outputs since the 1st of July 2023 seem to be problematic:
Average salinity in the upper 300m
Depth of 20C isotherm
Mean sea water potential temperature in the upper 300 m
It was decided to remove them from the archive since resp in the realtime resp in the reforecast outputs (The reforecasts are produced fairly in advance).
2. Recommendation:
Users are adviced not to use those parameters archived before their discontinuation.
3. Solution:
Parameters discontinued since resp in the realtime resp in the reforecast outputs
Meteo-France (lfpw)
Interpolation error in ALL data between 19 May and 16 June 2016. Fixed 17th June 2016
1. Brief description:
All S2S parameters were affected in that period. The corrected data was re-archived on June 17 2016.
The problem on the provider's side was caused by a bug in the last versions of EMOS and MARS (issued in March) that was avoided by using the previous version of MARS.
2. Recommendation:
If M-F data from the given period was downloaded in the period 19.5.2016-17.6.2016. it should be deleted and the correct version of the data downloaded again.Wrong 10-meter wind fields Fixed since 22nd October 2020 in the upgraded model version CNRM-CM 6.1
1. Brief description:
10-meter wind fields (zonal and meridian) are effectively wrong.
2. Recommendation:
Not to use 10-meter wind data at all until the next Meteo-France forecast system update (estimated in 2019).
The issue has been fixed since 22nd October 2020 in the new model version Meteo-France Model Description CNRM-CM 6.1
Wrong surface solar radiation downwards (ssrd) data
Update on 03-11-2017:
Our 'ssrd' field is actually correct. The issue described last month only concerns our seasonal forecast system, but not our S2S system.
1. Brief description:
All ssrd data is affected (values are much too high; it comes from a post-processing issue where the direct downward solar radiation has been mistakenly added to the total (i.e. direct + diffuse) downward solar radiation). The issue will be neither fixed in the past re-forecasts nor in the past forecasts as it would require excessive work. It has not been fixed in new forecasts either in order to avoid inconsistencies with the re-forecast and previous forecasts.
2. Recommendation:
Higher resolution of sea ice cover and sea surface temperature (not error just a technical exception)
1. Brief description:
Two surface parameters (sea ice cover and sea surface temperature) after the model upgrade to the version CNRM-CM6.1 have higher resolution (1x1) than the other surface parameters (1.5x1.5) in both real-time (since 2020-10-22) and reforecast (model version date 2019-07-01) outputs. That higher resolution is the same as the one agreed for the newly added 9 ocean parameters.
2. Recommendation:
If needed, users can retrieve those parameters with the same resolution as other surface parameters (1.5/1.5 degree regular lat-lon grid) by specifying it explicitly in their Web-API requests.Wrong 2 metre minimum and maximum temperature
1. Brief description:
After the Meteo-France model upgrade to the version CNRM-CM6, there is an issue with 2 metre minimum and maximum temperature. Those two surface fields (parameters mn2t6 and mx2t6) are identical. Basically, instead of the correct values of 2m Tmin/Tmax, the mean 2-meter temperature over the past 6 hours is provided. This error affects both the real-time data since the real-time forecasts since the 22nd October 2020 and also the corresponding set of the fixed reforecasts (1993-2017) archived under the model version date 2019-07-01 in MARS archive.
2. Recommendation:
As this error cannot be easily fixed in the reforecast data, it was decided not to fix for consistency the real-time forecast fields either.
A sort of possible workaround for users needing 2m Tmin/Tmax fields at the daily (24-hourly) time step is to compute the minimum (maximum) over the four 6-hourly values of a given day. The result would be less sharp than it should, thus it is not fully satisfactory and correct solution.
Occasionally high values of Sea-ice thickness
1. Brief description:
The values of Sea-ice thickness after the model upgrade to the version CNRM-CM6.1 can be occasionally (very rarelly) unusualy higher than usually. Their maxima could reach 160. According to French sea-ice specialists it is not a numerical problem, although such values are very unusual. For example in one of the cases, a dynamical forcing has been found, explaining such massive sea-ice thickening located between 2 islands of the Canadian Arctic archipelago (a rapid increase of meridional surface current has led to sea-ice piling up along the shore of a canadian island). The data from the described case were considered valid and thus allowed to be archived.
2. Recommendation:
Users should be aware of this.Model grid encoding change since 22 Oct 2020
1. Brief description:
After the latest CNRM model change in October 2020, the CNRM grid is inverted, i.e. j-dimension scans positively (south to north) instead of negatively (north to south) as per the S2S requirement Encoding#Gridandresolution
As the related GRIB 2 scanningMode changed from 0 to 64, such GRIB 2 encoding is correct in principle. Still it is not desirable for reasons of consistency with the previous encoding of CNRM data, and in addition may cause problems for users.
https://jira.ecmwf.int/servicedesk/customer/portal/4/SD-92894
2. Recommendation:
Users should be aware of this when processing CNRM data from the affected period.NCEP (kwbc)
Missing steps 0 and 0-24
- Step 0 is not available for all surface instantaneous fields, except for mean sea level pressure (msl)
- Step 0-24 (day 1) is not available for all daily-mean fields except 2 metre dewpoint temperature, Total cloud cover and Convective Available Potential Energy (CAPE)
Identical control forecast and ensemble members 1, 2 and 3 for some reforecast dates
For some re-forecast dates, only the 0Z runs could be recovered, which means that the control, ensemble members 1, 2 and 3 are identical and correspond to the 0Z run.
The list of affected dates is:
- 19990101
- 19991110 - 19991012
- 20001110 - 20001119
- 20011110 - 20011119
- 19990210 - 19990221
- 20000210 - 20000221
- 20010210 - 20010221
RHMC (rums)
Errors in the real-time data (3 runs only on 27.8., 3.9. and 10.9. 2020). Fixed 19th October 2020
1. Brief description:
There were some errors in the RHMC real-time outputs for the period 20200827-20200910.
2. Recommendation:
If RHMC real-time data from the given runs was downloaded before 19.10. 2020, it should be deleted and the correct version of the data downloaded again.