Ensemble Sensitivity Analysis (ESA) is applied to select types of observations, in various locations and in advance of forecast convection, to systematically evaluate the effectiveness of ESA-based observation targeting for ten convection forecasts. To facilitate the analysis, Observing System Simulation Experiments and perfect models are utilized to generate synthetic targeted observations of temperature and pressure for future assimilation with an ensemble prediction system. Various observation assimilation experiments are carried out to assess the impacts of nonlinearity, covariance localization, and numerical noise on ESA-based observation-impact predictions.
It is discovered that localization applied during data assimilation restricts targeted-observation increments onto the forecast responses of composite reflectivity and 3-hourly accumulated precipitation making impact-predictions poor. Additionally, numerical noise introduced by non-linear perturbation evolution tends to reduce the correlations between observed and predicted impacts; small, random-perturbation experiments often yielded similar impacts on forecasts as targeted observations. Nonlinearity also manifests in the observation impacts when comparing targeted observations to non-targeted, randomly-chosen observations; random observations have seemingly the same impact on forecasts as targeted observations. The results, under idealized conditions and simplified ensemble configurations, demonstrate that ESA-based targeting for non-linear convection forecasts may be most applicable at short time scales. Important implications for ESA-based targeting methods employed with real-time ensemble systems is also discussed.
Current affiliation: Department of Atmospheric Science, Colorado State University