The atmosphere and oceans are being routinely observed by a myriad of instruments. These instruments are positioned on board orbiting satellites, aircraft and ships, surface weather stations, and even balloons. The information collected by these instruments can be used to ensure that modelled weather forecasts adhere to reality using a process known as data assimilation.
Figure 1: Data coverage of the AMSU-A instrument, on board 6 different satellites, within a 6 hour time window (copyright ECMWF)
For the observations to be useful it is necessary that:
- The observations can be compared to the forecast variables (e.g. temperature, humidity and wind)
- We know the uncertainty in those observations
- We know the uncertainty in the weather forecast model itself (so we know how much to trust the forecast vs how much to trust the observations)
These fundamentals of data assimilation are continually evolving, as the weather models become more sophisticated and are addressing new societal needs, new instruments are developed and computational resources and mathematical techniques advance.
These different aspects of data assimilation were addressed at the fifth annual international symposium on data assimilation held at the University of Reading during a very hot week in July 2016. This symposium brought together 200 scientists from 15 countries spread across four different continents and received sponsorship from NCEO, the Met Office and ECMWF.
Figure 2: Participants of the Fifth annual international symposium on data assimilation (photograph copyright (C) Stephen Burt).
This symposium comprised of 10 different sessions, one of which focused on the particular problem of assessing the value of observations. This is important not only for evaluating which (of the very many) observations are most important for providing an accurate weather forecast but also for designing instruments which are able further to reduce the uncertainty in the forecast. This latter problem is particularly difficult due to the fast pace at which data assimilation systems are changing, which means that by the time the instrument is operational (possibly in a few decades time) its value may be very different than if the data could be assimilated today.
There are many possible metrics for assessing the value of observations. Some are based on how sensitive the forecast skill is to the value of the observations, others try to quantify the amount of information in the observations for reducing the uncertainty in our knowledge of the current state of the atmosphere. Computing these metrics before the instrument is built and the data is available relies on accurate estimates of the error characteristics of the instrument and its relationship to the model variables and, hence, is very challenging.
It is clearly difficult to describe the value of a future observation unequivocally by a single figure. Instead we need to provide insight, through on-going research, as to how the value of observations are sensitive to changes in the ever evolving data assimilation system. There will be much to discuss at the next symposium!