Impact of Targeted Dropsonde Observations on the Predictability of Extratropical Cyclones in a Global Model Using an Observing System Simulation Experiment Framework

Severe weather events associated with extratropical cyclones have a large societal impact due to the accompanying high winds and heavy precipitation. Even with the advances in numerical weather prediction, there are still large errors in the forecast accuracy of extratropical cyclones partially due to errors in the initial conditions. This is particularly true for cyclones that originate in the Northern Pacific Ocean and impact the western United States (US) due to sparse data coverage over the ocean. Observing System Simulation Experiments (OSSEs), which have the advantage of knowing the ‘truth’ or ‘reference’ atmosphere compared to Observing System Experiments (OSEs), are examined as part of the SHOUT (Sensing Hazards with Operational Unmanned Technology) Program to assess the impact of observing systems before they are deployed. For this study, the impact on the forecast accuracy of simulated dropsonde data from the Global Hawk Unmanned Aircraft System (UAS) is evaluated. Three storms are chosen as case studies from the T511 ECMWF Nature Run (NR), defined as our ‘truth’. For the control (CTL), observations used operationally in 2012 by NCEP are simulated from the NR and assimilated in the NCEP’s Global Forecast System (GFS) data assimilation system. Three experiments are also conducted (Idealized, Sensitivity, and Flight) using the same assimilation system, but assimilating perfect dropsonde observations in addition to the observations used in the CTL. Comparisons between the Idealized experiment and CTL show that the 2-3 day forecast error reduces by 10-20% for all three storms when assimilating dropsonde observations over a large idealized domain in the Pacific Ocean. The addition of dropsonde observations over a smaller region, determined using the Ensemble Transform Sensitivity (ETS) technique, also results in a reduction in forecast error by 1-5% for all three storms. Finally, in the last experiment, a flight path was simulated over the sensitive region found in the previous experiment. The assimilation of dropsonde observations over the flight path resulted in a 1-4% reduction in forecast error. The circumstances behind why certain storms improved more than others is examined.