Application of Ensemble-based Forecast Sensitivity to Observations Metric to a Mesoscale Convective Initiation Case using the GSI-based EnKF System

Recent advancements in high-resolution analyses and forecasts have prompted a nationwide effort to improve the coverage of mesoscale observations in the U.S. A report from the National Research Council recommended the development of a “Network of Networks” in which observations from many government and private sector providers are combined to form a single nationwide mesoscale observing network (National Academy of Sciences 2009). The report also recommended the development of research testbeds to experiment with identifying observational systems that are most impactful on a forecast. One such testbed is located over the Dallas – Fort Worth (DFW) Metroplex, known as the DFW Urban Demonstration network. Many data types have been planned for the DFW testbed, including X-band radars from the Center for Collaborative Sensing of the Atmosphere, Earth Networks Weather Bug surface stations, surface stations from the Citizen Weather Observer Program, observations from Global Science and Technology trucks, and various profilers, in addition to available operational data from METARs, mesonet, aircraft, rawindsondes, profilers, and operational radars.

In this study, the GSI-based EnKF data assimilation system is used together with the WRF-ARW core to examine the impacts of observations assimilated for the 15 May 2013 case. There were 7 tornadoes that day, including an EF4 tornado in Hood County, Texas and an EF3 tornado in Johnson County, Texas. The focus of the assimilation study will be on timing and location of convective initiation, which occurred after 2200 UTC in Northern Texas. The Ensemble-based Forecast Sensitivity to Observations (EFSO) metric of Kalnay et al. (2012) is tested for accuracy against the actual impact from various data denial experiments. The EFSO metric is appealing because no adjoint or additional experiments are necessary to evaluate impacts from several different observing platforms. A key issue of this study is the localization required to obtain accurate impact measurements, as the forecast component of EFSO adds a level of complexity to the localization problem. Several adaptive localization methods will be tested, including a ‘group filter' method which evaluates confidence of regression factors between groups of ensembles, as well as a method based on ensemble correlations raised to a power. The focus of evaluating adaptive methods will be in the ability to capture localizations which reflect the evolving ‘errors of the day' – a key component in the potential real-time application of EFSO to high-resolution mesoscale forecasts.