Relationships Between Observation Assimilation, Forecast Improvement, and Adjoint-Derived Observation-Impact

Adjoint-derived observation-impact is most often used to evaluate and monitor the impact of observations on the error of the short-range forecast; evaluations can be made between observations, observing-systems, and inter-comparisons can be made between various analysis/forecast systems. When using observation-impact to assess the value of observations to the forecast, one may ask if a “more is better” approach is appropriate – does large observation-impact really mean an observation is valuable to contributing to forecast improvement?  When one analysis/forecast system expresses larger observation-impact than another, which one is performing superior assimilation?

Experiments were carried out with assimilation of novel observations, both in a synthetic-observation framework and a real-data observation framework, to investigate these questions. In a synthetic observation experiment, profiles derived from the ECMWF analysis were assimilated into the Naval Research Laboratory’s NAVGEM model, simulating the assimilation of a global dataset of high-quality observations. In a real-data example, hourly satellite-derived cloud-track winds were assimilated into the WRF-ARW and compared with the assimilation of no cloud-track winds or 6-hourly cloud-track winds. The effect of assimilating these novel observations on forecast accuracy and the adjoint-derived observation-impact for all assimilated observations is evaluated.