Assessing the impact of observations using observation-minus-forecast residuals
Dahoui, Mohamed ; Isaksen, Lars ; Radnoti, Gabor
Année de publication
ECMWF assimilates a wide range of observations to help define the initial conditions at the start of a forecast run. It uses a complex data assimilation scheme (4DVAR) to make the best possible use of the available observations. Given the importance of accurate initial conditions for the quality of forecasts, it is useful to monitor and understand the relative impacts of different parts of the observing system on the analysis as well as on forecasts. ECMWF routinely assesses Forecast Sensitivity to Observation Impact (FSOI) using an ‘adjoint-based’ approach where forecast skill is evaluated with respect to analyses. An alternative, observation-based measure of impact called ‘observation-minus-forecast (OMF) residuals’ has been implemented and found to provide complementary results. Results using the OMF residuals approach differ from FSOI but confirm the strong influence of satellite observations, which dominate the observing system in terms of volume. Both measures show that in-situ measurements remain an essential component of the observing system despite their relatively low numbers compared to satellite observations. The overall impact of observations on the analysis and on forecasts depends on the quality of the assimilation system and the forecasting model and locally on the characteristics of the Earth’s surface and dominant weather regimes. The relative impact of each component of the observing system depends on its quality, spatial and temporal distribution, prescribed observation errors (derived from a long-term statistical evaluation of the observing system) and inherent redundancies with other components of the observing system. To estimate the impact of observations, different methods are used. The results obtained depend on the verification measures employed and the atmospheric structures targeted.