Use of Targeted High-Altitude Dropsonde Observations from Unmanned and Manned Aircraft to Test Tropical Cyclone Operational Forecast Improvement

A new strategy has been developed for intended use with operational reconnaissance of Tropical Cyclones (TCs) that involves deployment of dropsondes from a relatively high altitude within the TC environment and Inner Core (within 2.5-deg radius of the center). This strategy involves determining regions of ensemble global model uncertainty (sensitivity) and targeting these data-void regions with dropsonde deployments, for the purpose of increasing model forecast skill. To test this hypothesis, a series 7 TCs underwent targeted dropsonde sampling from a NASA Global Hawk (GH) Unmanned Aerial Vehicle (UAV), as well as manned NOAA WP-3D, 53WRS WC-130J and NOAA G-IV aircraft deploying dropsondes from lower altitudes during the NOAA SHOUT 2016 and NASA EPOCH 2017 campaigns. An improved new global model, the new NOAA/NCEP FV3GFS, was run for these cases with a series of data denial experiments which were run to asses dropsonde observation impacts on track, intensity and structure in an operational framework. So this study brought together several new model system improvements: 1) new FV3 model core, 2) improved model physics, 3) improved Data Assimilation, 4) new dropsonde instrument design and data transmission technology and 5) the targeted sensitivity strategy for sonde deployment while imposing typical operational constraints.

Despite having only 7 storms with GH flights during the SHOUT 2016 and EPOCH 2017 time period, which imposed constraints on the level of significance of the data denial results, some promising indications were found. Initial results from denial of the GH-only data relative to the control runs with all other aircraft and satellite data still included, showed small positive forecast improvements of 8-10% for Track and 5-8% for intensity prediction at time periods longer than 2 days, with neutral results for the shorter time periods. These results are about half of the improvement shown by research model GH data denial runs. Nevertheless, these results are considered quite significant due to the longer model run times employed, bracketing the GH flights, used to simulate operational procedures. These recent results also confirm earlier results showing some positive impacts on track and intensity forecasts for TCs in other basins such as WPAC and EPAC with forecast periods at the same time as the Atlantic basin sampling.

Additional runs with both the global model (fv3GFS) and regional hurricane model (HWRF) are planned with denial of the WP-3D, WC-130J and GIV dropsondes as well as the WP-3D Tail Doppler Radar (TDR) observations. Further, a final data denial run is planned by denying all dropsonde data outside of high uncertainty regions, the main reason for deploying dropsondes using ensemble model uncertainty analyses in the first place. In addition, when additional computing resources become available, the data denial strategies discussed here will be applied to a larger data set, including dropsondes from the 3-year Hurricane and Severe Storms Sentinel (HS3) campaign. In addition, dropsonde data in the BUFR format will be assimilated, preserving every observation for availability to data thinning strategies employed by the model, rather than employing the coarse and subjective sampling inherent in the data transmitted using the outdated WMO TEMP DROP data coding format containing only a dozen or so points and only the original launch position, a constraint that prohibits dropsonde use within the inner core.

It is the intention to use these results to develop strategies for sonde deployment from the lower altitude NOAA GIV and forthcoming G550 aircraft in the future. So that even though it is unlikely that the GH UAV aircraft will be available in the near future, these results can still be used to improve operational sampling strategies for available high-altitude aircraft.