Assessment and tuning of data assimilation systems using passive observations

Abstract

Independent observations can be used to diagnose and tune a data assimilation (DA) system. Analysis increments generally improve the model state near assimilated observations but degrade it further away. High‐resolution aircraft observations from Mode‐S Enhanced Surveillance (Mode‐S EHS) are used as an independent data source to verify increment degradation as a function of distance from assimilated observations. An adaptation of the inherently imperfect gain matrix in DA is proposed such that resulting analyses better fit the independent data source and as such draw model simulations closer to the true atmospheric state. It is found that the structure functions of the background‐error covariance matrix of the experimental mesoscale HARMONIE model are appropriate but too much weight is given to observations relative to the model background. The ECMWF model is well tuned with a slight overestimation of temperature information in the upper troposphere.Finally, a caveat is highlighted when comparing model forecasts from different experiments against observations. It is common practice to use the same observing system both in the analysis and for forecast verification. However, forecast verification is prone to sampling errors, yielding less favourable scores when using an independent data source. Avoidance of biased conclusions on the impact of observing systems, e.g. in observing system experiments (OSE), requires an independent data source (best practice) or a data source used in all experiments (best pragmatic practice) for verification of forecasts from different experiments.