Efficient Estimation of the Impact of Observing Systems using Ensemble Forecast Sensitivity to Observations (EFSO)

Tse-Chun Chen
Noon, 18 July in 2155

Massive amounts of observations are being assimilated into operational NWP. New observing systems with high temporal, spatial, and spectral sampling rates are being developed and deployed fairly regularly. The need to evaluate the usefulness of these observations can not be satisfied by the prevailing OSEs, which are computationally expensive and have limited applicability. We demonstrate that Ensemble Forecast Sensitivity to Observations (EFSO), which quantifies the impact of each observation on the forecasts at low cost, could be implemented as an online monitoring tool of the impact of each observation. Thus EFSO can efficiently identify detrimental impact episodes and the associated observations. To avoid such detrimental episodes, Hotta et al. (2017) have shown EFSO-based Proactive Quality Control (PQC) can reduce forecast error in cases of "skill dropout". We further devised two other data denial strategies: THReshold (THR), which rejects observation if the Moist Total Energy error impact is more detrimental than 10-5 J-kg-1, and Beneficial Growing Mode (BGM) that only keeps observations that are beneficial in 6-hr forecasts and continue to be beneficial after 24 hours. We show in the presentation that both THR and BGM outperform the original PQC method, and BGM (useful for reanalyses) performs even better than THR.