Operational Data Checking (auto alerts)
An automatic data checking system has been implemented at ECMWF to trigger warnings in case of anomalies affecting the quality or the availability of the data assimilated by the model. Selected statistical parameters (number of observations, bias correction, and mean bias-corrected background, analysis departures and probability of gross error) are checked against an expected range. An appropriate alert message (including a time series plot) is generated if statistics are outside the specified ranges. A severity level (slight, considerable, severe) is assigned to each message depending on how far statistics are from the expected values. Two kinds of ranges are used by the automatic checking: Soft and Hard limits. Soft limits are updated automatically using statistics from the last twenty days (extremes are excluded during this process). Hard limits are adjusted manually when required.
Currently, the automatic checking is limited to data passing through the minimisation process (including VarBC passive data). It’s being applied twice a day to the delayed cut-off 4D-VAR cycles (00 and 12) and four times a day to the early delivery assimilation system (00,06,12 and 18).
The alarm system results are based on feedback information from the ECMWF data assimilation and consequently reflect the ECMWF data usage. Although most warnings are related to data problems, they may on occasion also be related to model issues. Alerts are provided for information on an as-is basis and may be withdrawn or changed without notice. ECMWF accepts no responsibility for actions taken on the basis of these results.