Current global operational Numerical Weather Prediction (NWP) systems (e.g.the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS)) generally assimilate on the order of 10 million observations every 6 hours. Furthermore, there is substantial diversity in the sampling characteristics and associated error characteristics of the observation types assimilated. In this context, it is not feasible to obtain sufficiently detailed information for determining which available observations or observation types should be assimilated or rejected in NWP systems using traditional Observing System Experiment (OSE) approaches. Forecast sensitivity to observation impact (FSOI) based estimation techniques (Langland and Baker 2004) enable efficient estimation of forecast impacts due to assimilation of individual observations, and as such, represent a solution to this problem.

The ensemble forecast sensitivity to observations (EFSO) (Kalnay et al. 2012)impact estimation technique uses ensembles of forecasts to perform linear mapping of innovations to forecast error changes. This mapping involves application of Kalman gain matrices consistent with the complete sets of observations assimilated during data assimilation cycles. As with the other forecast-sensitivity based observation impact estimation techniques there are two prominent “contextual” limitations for application of EFSO in NWP systems: i) the observation impacts are estimated with respect to simultaneously assimilating all other observations that contributed to an analysis, ii) EFSO calculations are relative to a background that includes information from all previously assimilated observations. To mitigate these “contextual” limitations in application of forecast-sensitivity based observation impact information, a new quantitative framework we call “EFSO-components” is developed by decomposing EFSO employed forecast errors and innovations into random and systematic components. Lorenz ’96 simple model experiments indicate that application of ”EFSO-components” provides potentially significant advantages in detection of specific observation flaws, and in further advancing the utility of EFSO-based PQC (Ota et al. 2013, Hotta et al. 2017a, Chen and Kalnay 2019, Chen and Kalnay 2020). As such, we explore how “EFSO-components” fundamentally addresses the aforementioned contextual limitations of forecast-sensitivity based observation impact estimation in a manner that explains the potential application advantages according to Lorenz ’96 simple model experiments.

Additionally, a new technique we call predicted EFSO (PEFSO), which is astraightforward extension to EFSO, is introduced in this study. PEFSO represents a potential capability for estimating the hypothetical forecast impacts of unassimilated observations. We explore the potential application of PEFSO as a convenient low computational cost approach for comparing the efficiencies of observing systems in reducing forecast error using Lorenz ’96 simple model experiments.