Application of Mesoscale Ensemble-based Sensitivity Analysis to Observation Targeting

Observation targeting techniques based on ensemble statistics utilize covariance relationships between a chosen forecast aspect and initial condition variables to identify locations where the atmosphere can be observed to reduce errors in the forecast. From Ensemble Kalman Filter (EnKF) theory it can be shown that assimilating additional observations will always reduce the variance in a chosen forecast metric, by a value that is calculated from ensemble-based sensitivity (Ancell and Hakim 2007). Both localization and inflation within the ensemble data assimilation procedure can act to modify this expected value. Furthermore, in a convective environment, the non-linear nature of convective forecasts makes targeting difficult, as ensemble sensitivity analysis (ESA) and associated targeting techniques assume a linear relationship between the two variables. This problem can be exacerbated by the binary nature of convection, with only some of the ensemble members producing convection for certain cases. Nonetheless, accurate forecasts of vertical wind shear, convective available potential energy (CAPE), and composite reflectivity (MDBZ) are critical to the issuance of severe weather outlooks. Thus, this study aims to develop useful adaptive observing techniques toward improving forecasts of convection.

In order to determine whether real-time targeting techniques on small scales hold value for a forecast, a set of experiments will be presented for a specific case study to assess the impacts of non-linear relationships between initial conditions and the forecast metric, localization and inflation effects, and the binary nature of convective forecasts. The 4 April 2012 dryline-initiated convective outbreak over Northern Texas will be used to determine impacts that assimilated surface observations from the West Texas Mesonet (WTM) have on forecast error and to validate the usefulness of targeted observing on convective scales. Impacts are determined by running a control simulation that withholds WTM observations, determining which station would have the greatest impact on the forecast metric via variance reduction formulations, and then assimilating that station's surface observations into the model through the EnKF. Expected impacts and actual impacts can then be compared to determine if variance reduction can be accurately predicted under the various assumptions listed above. Lastly, the actual variance reduction from assimilated observations that predict large impacts to those that predict no impact are compared to establish the overall value sensitivity-based targeting might provide with regard to convective forecasts.