Ensemble Kalman filter assimilation of Global-Hawk-based data from tropicalcyclones

Jason Sippel, Gerry Heymsfield, Lin Tian, and Scott Braun- NASAs GSFC

Yonghui Weng and Fuqing Zhang – Penn State University

Background: HIWRAP basics• HIWRAP is a conically scanning

Doppler radar mounted upon NASAs Global Hawk UAV

• It was first used to observe Hurricane Karl in GRIP (2010) and is being used in HS3

• Simulated-data results (Sippel et al. 2013) suggest HIWRAP Vr can be assimilated to improve hurricane analyses

Background

HIWRAP Schematic

EnKF analysis from simulated data

Methods: Experiment setup

• Looking at Karl (2010) – only hurricane that HIWRAP data is available for

• WRF-EnKF from Zhang et al.

(2009) with 30 members

• Similar setup as Sippel et al. (2013) OSSEs

Methods

Model domains

3-km nest

Karl’s track

Methods: Problems and solutions• Significant issues for GRIP data

Outer beam unavailable for Karl – inner beam observes more along vertical axis

Unfolding not possible for some legs Heavy QC required for Vr

• Solutions Compare assimilation of Vr with VWP data

(avoids DA issues with w, and less heavy QC required)

Assimilate position & intensity (P/I) in addition to HIWRAP data to help fill in gaps

Methods

VWP Methodology

Methods: Processing Vr obsMethods

• Only keep data from 2-8 km where refl > 25 dBZ

• Bin 15 scans (10 km) into 2 km x 20° grid, keep median as SO

• Only keep 25% of SOs

Methods: Processing VWP obsMethods

• Bin VWP u and v components into 1 km x 1 km bins every 20 km along track and 1 km in altitude

• Select median value from bin as SO, no thinning

Methods: Assimilation detailsMethods

• Assimilate position first, then position + HIWRAP data + SLP

• P/I ROI: ~1200 km horizontal and 35-level vertical

• HIWRAP ROI: Zhang et al. SCL with 900/300/100-km horizontal and 26-level vertical

Karl assimilation schematic

Assumed errors: Minimimum SLP – 4 hPa Vr – 3 m/sPosition – 20 km

VWP – 1.5 m/s

Results: EnKF storm position• All analyses better than NODA,

error of mean similar

• Error generally less than assumed position error (20 km)

Results

Track evolution from EnKF analyses

Results: EnKF max intensity• All experiments far

superior to NODA

• Slight improvement upon PIONLY with HIWRAP data

• Min SLP generally lower in VR experiment

Results

Maximum intensity evolution from EnKF analyses

Results: Wind radii• PIONLY produces

a storm that is much too large, especially for smaller radii

• Analysis with HIWRAP data is in much better agreement with best-track

Results

Wind radii (km) evolution from various EnKF analyses

Results: Vertical structure• PIONLY analysis (not shown)

produces a shallower, broad vortex that looks unrealistic for a major hurricane

• Analyses with HIWRAP data are more realistic with tall, compact core

• VR analysis more intense by about 5 m/s by last cycle

Results

Azimuthal mean wind speeds at last cycle

Results: Deterministic forecasts• All EnKF-initialized forecasts

improve upon NODA

• Hard to tell difference in track forecasts among EnKF experiments

Results

Comparison of best track with NODA and EnKF-initialized track forecasts

Results: Deterministic forecasts• All EnKF-initialized forecasts

improve upon NODA

• VWP-initialized Vmax forecasts generally better than Vr-initialized forecasts

Results

Best track Vmax (m/s) compared with NODA and EnKF-initialized intensity forecasts

Results: Deterministic forecasts• All EnKF-initialized forecasts

improve upon NODA

• HIWRAP-initialized min SLP forecasts better than PI-ONLY but VWP-initialized are best

Results

Best track min SLP (hPa) compared with NODA and EnKF-initialized intensity forecasts

Summary + Future WorkHIWRAP data appears to be useful for TC analysis and forecasting

• Despite difficulties with early HIWRAP data, EnKF analyses with HIWRAP Vr and VWP data produce accurate estimates of maximum intensity, location, and wind radii

• EnKF-initialized forecasts significantly improve upon NODA, but for this case VWP assimilation produces better forecasts (perhaps because horizontal winds are better constrained)

• Future work will examine the impacts of additional Global-Hawk-based data, including dropsondes, surface wind speeds from HIRAD and water vapor and temperature retrievals from S-HIS