thank you very much for your detailed analysis, which Cristina Prates and I reviewed today. Please find our feedback below and in color.
1. Period definition and data download
When can we expect the implementation of a monthly “Type of period” for Timeliness?
Monthly timeliness aggregations are already available for SYNOP. Timeliness for TEMP will be implemented once a second center provides the required information.
For practical analysis purposes, would it be possible for downloaded Timeliness data to be provided directly in minutes rather than seconds, since the legend and thresholds are defined in minutes?
2. Centres included in Timeliness monitoring
Currently, Timeliness is available only for DWD and ECMWF.
Timeliness for other centers will become available in the webtool as soon as other centers report timeliness.
Timeliness is a global metric and the fact that DWD and ECMWF are the first to report this metric is purely coincidental. Other centers will provide this metric eventually.
To avoid overlapping category boundaries, we would like to propose a clearer definition of thresholds:
Blue: 0–14:59
Purple: 15:00–29:59
Pink: 30:00–119:59
Burgundy: ≥120:00
This would be consistent with the current definition of “less than 15 minutes” for the blue category.
We will consider an update to how the intervals are represented in the legend to make this more clear for the user.
4. Lack of Timeliness-related metadata in OSCAR
Currently, OSCAR does not provide metadata directly related to Timeliness.
Would it be possible to include information such as a scheduled report time, indicating when reports are expected to be transmitted to NWP centres?
In principle the WMDR element “reporting schedule” in OSCAR/Surface documents the expected time of reporting. It does not cover the expected amount of time between when an observation is made and when it reaches a NWP center (the definition of timeliness used by WDQMS), as this varies over time and would therefore be difficult to include as a metadata element.
On the other hand, OSCAR/Requirements includes a threshold for timeliness, which indeed is the basis of the timeliness legend in WDQMS.
5. Documentation in WMO-No. 1224
In WMO-No. 1224 – Technical Guidelines for Regional WIGOS Centres on the WIGOS Data Quality Monitoring System, the concept of Timeliness appears to be described only briefly.
We would welcome clarification or expansion of this section, particularly regarding interpretation, cut-off times, and the relationship to NWP usage.
1224 defines timeliness as reception time - observation time, which is the criteria used by WDQMS. In practice each NWP center has its own way to calculate the reception time. However, these differences are minor compared to the rest of the transmission time. Observations exposed to WDQMS by definition met the cutoff time, since the system only considers observations made available to data assimilation.
6. Interpretation of N/A values
We assume that N/A values in Timeliness often indicate that data were received, but too late to be useful for NWP purposes.
Is there a clearly defined upper cut-off time for Timeliness?
Is there a maximum delay (e.g. +6 hours) after which SYNOP data are no longer considered?
How is the distinction made between “not received” and “received but too late”?
N/A values generally mean that the center does not provide timeliness, even though for internal reasons sometimes the information cannot be computed, in which case N/A (or -9999) is used. This should not be interpreted as the observation not meeting requirements. Different centes have different assimilation windows, ranging from 3h-12h. WDQMS NWP cannot distinguish between not received and received but too late, indeed the latter category is not defined in WDMQS at all. The Transition Monitoring module does have access to all observations regardless of cutoff time and could in principle be used to compile statistics beyond cutoff windows. This is limited to ECMWF data at the moment.
7. Observed national and centre-specific patterns
Based on our analysis:
Stations in ALB are almost entirely blue,
BGR stations are blue except for Musala,
TUR shows approximately 45% blue and 55% purple.
This suggests that reporting configuration (e.g. hourly vs. 10-minute transmission) has a direct impact on Timeliness performance.
Across all analysed countries and all surface land variables (pressure, temperature, humidity, and both wind components), we observe very similar and clearly clustered Timeliness patterns.
We also note systematic delays at ECMWF (up to approximately 50 minutes), while the same data typically reach DWD within 7–11 minutes.
This leads to several learning-oriented questions:
Does this imply that if data do not reach DWD within a short time window, they may still reach ECMWF later?
In case of duplicates ECMWF uses the latest, whereas DWD may use the first, explaining once possible reason for the timeliness difference noted. ECWMF also does not have direct GTS access, to the contrary DWD has, which may also be an explanation.
In rare cases where data appear at neither centre, does this indicate a failure earlier in the data exchange chain?The exact reason for missing data is not readily available to the system and ranges from the observation not having been made to transmission or quality/encoding issues. RWC need to investigate and use their knowledge of the observing system to determine the root cause. If WDQMS can implement tools facilitating understanding of issues, the team is happy to receive feedback and suggestions.
Is there a concept of a primary monitoring centre for Timeliness, or are centres treated independently?WDQMS shows whichever center received the observation first (the “fastest center) in combined mode.
8. Variable-level Timeliness differences
For the same station and the same day, it is sometimes observed that:
pressure appears as purple,
while humidity or meridional wind appear as blue.
This raises the question:
Are all variables always transmitted within the same SYNOP message and processed simultaneously, or
is Timeliness evaluated per variable due to different message sections, validation steps, or ingest pipelines?
Relevant factors to this phenomenon are the schedule of the observation in OSCAR/Surface and possible duplicates as explained above. Timeliness is evaluated at the level of the variable and a hypothetical partial duplicate could lead to one variable having different timeliness from another once which it was reported together with initially.
9. Station–center association
We also observe that some stations consistently appear under ECMWF, while others appear under DWD, for example:
in BGR, Kurdjali and Musala under ECMWF, while other stations appear under DWD;
in ALB, Gjirokastër and Kukës under ECMWF, while most other stations appear under DWD.
Could this be related to:
station representativeness,
altitude or complexity (e.g. mountain stations),
network design,
or is it primarily driven by center-specific ingest configurations?
Our working assumption is that DWD functions as a high-volume ingest center, while ECMWF processes a more selective subset of stations, but clarification would be very welcome.
Regarding the two stations in Albania referenced, the relatively large timeliness of the stations may be the reason DWD cannot consider them, this NWP center using a 3h window consequently shorter cutoff time.
If this issue persists, more detailed information, including the WSI of affected stations and the approximate time-span of the phenomena will facilitate further investigation of this issue.