The ECMWF IFS 4DVAR data assimilation system uses surface pressure from SYNOP observations globally.

Large Observation-minus-Background (O-B) departures may lead to rejection of these observations either at runtime by the data assimilation system or when systematic (bias), at the level of the monthly data selection filter.

There may be several factors leading to large departures. Finding out the cause of the departures is not a trivial task from NWP centre point of view. Such an investigation also requires collaboration with the observation data provider or OSCAR (WMO Observing Systems Capability Analysis and Review Tool) meta data authors.

Here we list points to be considered in causal investigations of significant departures.

1. Diverse types of patterns in surface pressure departures, causes and troubleshooting with the data providers

1.1 Pattern: Bell curve

1.1.1 Cause

A low pressure system passage not captured by the model.

1.1.1.1 Action

No action is necessary.

1.2 Pattern: Step-wise jump

1.2.1 Cause

Observation meta data change (barometer height, position) or instrument malfunction within the evaluation period.

1.2.1.1 Action

Check with the data provider.

1.3 Pattern: Systematic large departures over several weeks

1.3.1 Causes

Metadata issues at reporting station level (BUFR SYNOP observations) for Lat/Lon/Elevation. GPS/GNSS based measurement of station elevation requires GPS/GNSS Undulation Of The Geoid Correction. [1] If this is not done correctly at the station, there will be elevation errors with the original BUFR SYNOP report or its OSCAR meta entry. Each will lead to departures.

1.3.1.1 Action

Check with the data provider, where applicable, Latitude Longitude and elevation are computed and encoded correctly.

1.3.1.2 Action

Check GPS/GNSS Undulation Of The Geoid Correction is applied on data provider side.

1.3.2 Causes

Sensor or encoding error, TAC to BUFR conversion errors.

1.3.2.1 Action

Check with the data provider.

1.3.3 Cause

OSCAR meta data update for NWP model. Erroneous manual entries for position or elevation.

1.3.3.1 Action

Check with the data provider.

1.3.4 Cause

In some rare cases, stations report only PMSL (Mean sea level pressure) and not PSTN (station level pressure). In the 4DVAR data assimilation system, use of PSTN is preferred and potentially should lead to smaller biases compared to PMSL.

1.3.4.1 Action

Communicate to the data provider.

1.4 Pattern: Spikes of large departures over several weeks

1.4.1 Cause

Occasional human error in measurement or encoding (for manual stations).

1.4.1.1 Action

Investigate and correct at data provider side.

2. Modelling challenges

  • A station on steep orography may have large biases due to model representatives mismatch.

  • Land sea mask filter sometimes rejects coastal stations bordering model sea points.

IFS 4DVAR may be able to bias correct systematic errors however addressing the reasons for systematic errors is preferable.

Resolution of bias issues will potentially lead to greater quality and usability of observations and is well worth a pursuit.


Definitions of the terms and acronyms:

An evaluation period of one month is assumed.

Step-wise jump: Step up or down over a period of a month. This happens once or twice within the period.

Spike: A short lived jump of O-B values. The jump in O-B values may only last a day to few days. There may be many spikes within the one month period.

Systematic large departures over several weeks: No jumps, and a rather consistent bias over a month. 

References:

[1] Pauley, P.M., Ingleby, B. (2022). Assimilation of In-Situ Observations. 29,78