|Title||Enhancing the impact of IASI observations through an updated observation error covariance matrix|
|Publication Type||Technical memorandum|
|Year of Publication||2015|
|Secondary Title||ECMWF Technical Memorandum|
|Authors||Bormann, N, Bonavita, M, Dragani, R, Eresmaa, R, Matricardi, M, McNally, A|
The present memorandum investigates the use of an updated observation error covariance matrix for IASI in the ECMWF system. The new observation error covariance matrix is based on observation-space diagnostics and includes inter-channel error correlations, but also assigns significantly altered error standard deviations. The use of the new observation error is investigated in detail in assimilation experiments, including an assessment of the role of error inflation and taking inter-channel error correlations into account. The influence of the observation error update on the Ensemble of Data Assimilations (EDA) used for background error specification is also examined. The updated observation error covariance leads to a significant improvement in the use of IASI data, especially in the tropics, the stratosphere, and for humidity. The benefits are particularly strong for the short-range forecasts, whereas the impact in the medium range is less pronounced. The update also has a particularly large positive impact on the ozone analysis, related to especially large modifications in the observation error for ozone-sensitive channels. The observation error update leads to a modified spread in the EDA, with some reductions in spread in areas where improved short-range forecast impact is diagnosed. The study highlights the benefits of taking inter-channel error correlations into account, which allows the use of an observation error covariance for IASI that is overall more consistent with departure statistics. At the same time, the study also demonstrates that error inflation can be used to partially, though not fully, compensate for neglected error correlations. Adjustments such as scaling of the originally diagnosed observation error estimates are found beneficial also when inter-channel error correlations are taken into account.