HAFS Development Activities at AOML

Xuejin Zhang1, William Ramstrom1,2, Andrew Hazelton1,2, Ghassan Alaka1, Jr., S. G. Gopalakrishnan1, & Frank Marks, Jr.1

AOML1 & UM/CIMAS2

Collaborators: NCEP/EMC Hurricane Group (Lead: Avichal Mehra)GFDL Weather and Climate Dynamics Division (Lead: Lucas Harris)

Current Status and Ongoing EffortsCurrent Status● Single moving nest on tile 6 in global domain● Atmospheric prognostic variables shifted● API for shifting surface variables and terrain parameters● Atmospheric diagnostic variables recomputed● C96 and C768 with 3X refinement tested● Code merged in HAFS repository completed● Workflow for moving nest ready● Code optimization in progress

Ongoing Development● Preprocessing high-resolution orography and surface static

fields● Storm tracking algorithm● Merge telescopic and moving nest code into HAFS repository● Data assimilation for moving nest● Physics optimization for variable resolutions● Post-processing and visualization● Track generation for moving nest

Plot Contributed by William Ramstrom at AOML/HRD

Moving Nest DevelopmentLatest development● Performance optimization moving nest

○ Moving nest overhead 2.5% of overall runtime

○ 211 sec of 7841 sec runtime C768, 120hr○ Overall speedup: ~1200 sec to 211 sec

● Shifting of delz field, which has no halo● Shifting of SST/skin temperature

○ Template for other surface variables○ Added 2D procedures to variable motion

interfaces

Performance Analysis – C768 3X Moving Nest• setup_aligned_nest() is bulk of moving nest

time: ~1100 sec; move atmos. vars ~100 sec• spherical_linear_interpolation 230 sec• mid_pt_sphere 516 sec• cell_center2 307 sec

Optimized Results• Save results for parent grid (which

doesn’t change) between calls• spherical_linear_interpolation 8.19 sec• mid_pt_sphere 4.95 sec• cell_center2 3.61 sec

104X85X

28X

C768 3X Nest 120 Hours

HAFS PBL Physics Updates➢ Ongoing evaluation of modified EDMF-TKE continues, using observations and

LES data➢ Set of 2020 hindcasts performed in HAFS-B using Hybrid EDMF to compare with

real-time runs using modified EDMF-TKE➢ MEDMF-TKE tested in a large sample of hafsv0.2a baseline runs for the 2021

version (see results on the next slide)➢ Manuscript submitted (Ping Zhu, FIU lead) documenting the impact of changes in

N2 in the Hybrid-EDMF scheme on PBL structure

HAFS PBL Physics Updates

➢ Modified EDMF-TKE runs shown in blue➢ Large reduction in negative intensity bias

in HAFS-A at Days 3-5➢ Improved intensity skill at most leads➢ Track skill mostly neutral➢ Sample still being finalized

HAFSV0.2A Baseline Physics Tests (2020 TCs) With MEDMF-TKE HAFSV0.1B 2020 Tests Hybrid-EDMF vs. MEDMF-TKE

➢ HAFS-B 2020 used MEDMF-TKE

➢ Ran a set of 2020 cases using Hybrid-EDMF

➢ Very different bias characteristics (high vs. low)

➢ No coupling➢ Inflow structure

leads to more RI in EDMF-TKE

➢ Example shown for Hurricane Zeta

HAFS Post-Processing & Diagnostics● HAFS workflows → Rocoto “hrdgraphics” task has been updated

○ No changes required from the user side○ Recommended that this task be turned on for all 2021 HAFS experiments (ask Gus for help)

● Graphics are created by HRD’s GPLOT software (GitHub)○ Several upgrades and bug fixes have been committed○ Precipitation partitioning by type, structure metrics, potential vorticity maps, etc.

● Graphics publicly available on the AOML Hurricane Model Viewer○ 2020 HAFS graphics will be kept on the site for quick reference

● All HAFS A-Decks will be shared with NHC in near-real-time● Next-gen. graphical production software (Python-based) is in development

2020 Basin-Scale HWRF Retrospectives● 4 seasons (2017-2020) using the same model configuration is available as a

baseline for moving nests in HAFS (2017-2019 complete, 2020 in progress)● Moving nests for up to 5 storms + self-cycled ensemble DA