This website uses cookies. Read about how we use cookies.

Close window

Optimising the use of observations in data assimilation

Surface data assimilated for a single global forecast.

Developing methods to detect, quantify and allow for bias and errors in observations, and to estimate their value for weather prediction.

NWP makes extensive use of terrestrial and satellite observations. This vast network of equipment is funded by national meteorological services, space agencies and private enterprise. It is used to estimate the state of the atmosphere in a semi-continuous way. The vast majority of observations received these days are satellite observations . This page focuses on the traditional conventional network that includes vertical soundings by radiosonde, surface land observations at weather stations and airports, marine observations on-board ships and from buoys, aircraft measurements and radar observations.

Information from observations is used to correct forecast model trajectories, through a process called data assimilation. This works by using the difference (innovation) between observations and the background (prior) forecast from the NWP system to update the background forecast. As this becomes more accurate and detailed, it is increasingly important to estimate and allow for expected bias and errors in both the model forecast and the observations. Therefore, there is some investment to develop methods to detect, quantify and allow for bias and errors in observations, and to estimate their value for weather prediction. The latter is useful for maintaining existing observation networks and planning for upgrades since we are frequently asked questions about the value of proposed observing systems to future NWP systems - a reasonable question from those responsible for running and funding them. To answer them requires experience on a broad range of measured current impacts of observations, from our own experiments and others worldwide, interpreted using judgement and theoretical insight.

Forecast sensitivity to observations

The sensitivity of NWP forecasts to individual observations (or observation types) can be estimated through the use of the adjoint of the linearised approximation of the forecast model and data assimilation system. This is known as FSO and provides a scalar measure of the proportion of forecast error due to the assimilation of some set of observations in terms of the 24 hour total energy norm. Through the adjoint procedure, this can provide a value of sensitivity for each individual observation.

Monitoring, bias correction and quality control

Conventional observations (surface, radiosonde, aircraft) are continuously monitored and their use in the assimilation is controlled by station lists which are updated each month. The updates are informed by a three-stage process:

  1. Produce monthly observation minus background statistics profiles for each station
  2. Obtain suspect stations/layers for each analysis hour
  3. Decide which stations/layers to reject and produce station list records

Key aims

  • Methods for identifying and correcting bias in observations and the model.

  • Methods for detecting gross errors in observations and for monitoring their overall accuracy.

  • Methods for objectively measuring the impact and value of observations.

  • Research into the use of observations not simply related to the main forecast variables, such as cloud and precipitation.

Current projects

  • Development of a system to measure adjoint-based observation sensitivities and impacts from our operational 4D-Var system.

  • Development of a system to improve the bias correction of radiosondes.

  • Development of a system to estimate observing system bias correction coefficients within the variational data assimilation - VarBC.

  • Monitoring and determining the cause of occasional poor forecasts, particularly when due to the use of observations.

Last updated:

Follow us on

Facebook, Twitter, YouTube, Instagram, Snapchat, or LinkedIn Facebook Follow @metoffice on Twitter YouTube Instagram Snapchat LinkedIn