• HFIP must continue to support outreach and engagement with students and younger scientists via colloquiums such as the HFIP/NCAS-M HAFS training colloquium from May 2024

• The need for a HAFS Reanalysis

  • Train AI models, use to retrospective studies

  • Technically feasible, perhaps via the HAFS basin-scale configuration. Resources to run it are a big question.

• How and when do we make the decision to decommission HWRF/HMON if doing so leads to a decrease in skill, particularly for intensity?

  • HWRF/HMON currently in “no man’s land” with no support, but will continue to run (only 3 TCs) in 2025.

• How do we maintain model diversity when the HAFS-A and -B forecasts are typically quite similar for any particular forecast?

  • Can EMC devise a HAFS-C with less forecast overlap with -A and -B, perhaps with support from HRD?

• AI models (data-driven NWP, such as GraphCast GFS and the AIFS), what will their role be in the forecast process?

• R2O: need to begin folding in transition details at the proposal stage, including resources required, to allow for sufficient planning

• Need to think about the “integrated value chain”; what do the public most need to know in order to make decisions? Can we get to the point where we can say: “there is a 60% chance of a 10-day power outage”?

• Know and target your audience: don’t use the same messaging to all audiences, because they will not all respond the same way!

NOAA Leadership Brief - given by Hendrik Tolman

• Do we still need NCO to run 4 cycles per day?

  • Historically these cycles have corresponded to local TV weather broadcasts, but are 2 higher resolution, larger ensemble cycles superior to 4 lower-res, smaller ensemble cycles?

  • Running a given modeling system only 2x per day could also free up core hours to use elsewhere

OAR Leadership Brief - given by John Cortinas

• In order to ensure the future of HFIP, we need to not only communicate what we have accomplished, but also do better communicating “what can we do?” with additional resources

OSTI Modeling Leadership - Kevin Garrett

• Given increasingly strained budgets, how do we prioritize where funding should go, especially given ending supplementals?

• Can the HAFS Ensemble provide forecast guidance for multi-hazard probabilities the same way WFOs do?

NHC Leadership Remarks - Mike Brennan

• Hurricane Helene (2024) “re-wrote” the risk profile of the U.S. Southeast hurricane landfalls.

• We have come a long way in terms of reducing storm surge fatalities due to improved risk communication and public understanding; how do we similarly communicate the inland freshwater flooding risk and inland wind damage from storms like Helene?

• Frank Marks: see Irene service assessment, discusses improved coordination between NHC, WPS, SPC, and local WFOs.

NCEP Leadership Remarks - given by Vijay Tallapragada

• Strengthening HFIP, HAFS, and the UFS community relies on collaborations, strengthening our community, DA and leveraging new observations, and learning from our international partners

AOML/HRD Leadership Remarks - Ghassan Alaka & Sundararaman Gopalakrishnan (Gopal)

• Given the fact that HFIP falls under the modeling program, how do we bring observations back into HFIP?

  • Obs are a critical component of DA, but HFIP has also historically prioritized strategies for targeted observations and supported experiments to determine where and when new observation systems may be beneficial

• HAFS has made great strides in terms of TC structure prediction

  • The asymmetric structure predicted by HAFS-B in particular for Hurricane Milton (2024) at landfall was a particularly good example of this

• HAFS is the new poster child of rapid development and operationalization of an entirely new model.

  • However, in 2025 and beyond, we need to prioritize greater diversity between HAFS A and B, or NHC will continue to rely on legacy HWRF and HMON models; added physics diversity and operationalizing HAFS Ensemble are potential solutions

  • HAFS multi-storm and basin scale have demonstrated significant forecast skill improvements which we can hopefully operationalize in the near future

NHC

• In 2024, we saw record low track errors in the Atlantic at all lead times

• 2024 was not a record for intensity skill but climatological skill models such as OCD5 shows that the TCs in 2024 were particularly challenging for intensity

• HMON was the top model in both the Atlantic and East Pacific overall, demonstrating that additional improvements to HAFS are still needed before NHC would be comfortable decommissioning the legacy hurricane models

WPC

• We need to understand and communicate TCs along “the spectrum”, from ordinary TCs to catastrophic. Our challenge is to determine where along the spectrum a storm will fall ahead of time.

• How do we get HAFS Ensemble gridded data to WPC?

JTWC

• Challenges remain in predicting TCs in shear and dry air

• It is a challenge for HAFS (and other regional hurricane models) to correctly represent the very large eyewall of some West Pacific TCs

• How do we widen the “funnel” of R2O?

• Not enough resources or personnel to implement every great idea into Ops

• The “pool” at the bottom of the R2O funnel is already full, and NHC can only support adding a new product if an existing product is retired

• Can the NWS move towards an “agile approach” the way the private sector often does?

• We should shift our thinking towards “Dev-Ops” as one unified system rather than a handoff from Dev to Ops

NHC

• GFS has had a high false alarm bias for genesis, while the ECMWF has had too many missed events

• HAFS has exhibited intensity prediction issues, both for strong storms and for weak storms

• All models (both global and mesoscale) have had occasional issues with unrealistic analyses for TCs in recent seasons

EMC

• West Pacific MSLP versus wind balance issues

• HAFS-A and -B intensity issues identified in all basins

Google Deepmind

• Currently retrained once per year

• Retraining is expensive and can change the nature of biases

Group Discussion

• AI models have come a long way, but are currently trained on reanalysis which requires reliable models, DA systems, physics, air-ocean coupling, etc

• We currently cannot discontinue our dynamical models and replace them entirely with AI models

• We have positioned ourselves to a good starting point, but we need to keep communication open because we have a long way to go

• Need to nurture HFIP and OSTI-SBES relationship

• Transition to JEDI is in-progress; significant progress has been made in JEDI ocean, land, and wave DA

• Preliminary results for atmosphere JEDI vs GSI are encouraging, show comparable results with HAFS

• Currently JEDI takes much longer to run than GSI for same observations set; needs optimization

• Significant reliability and calibration issues with SFMR have been identified; at least seven sources of bias/error identified, with different error and bias profiles for every single instrument; entire SFMR database is in jeopardy

• Emphasis on the importance of coupled obs, below-surface “gliders”, and saildrones to penetrate the TC at the ocean level

• Unmanned systems / UAVs offer capabilities to sample parts of the storm (near-surface, TC planetary boundary layer) that cannot be sampled by manned aircraft for safety reasons

• Recent Obs studies have significantly improved our understanding and model representation of the TC PBL; next step: microphysics in deep convection?

• Key point: Obs improve model analysis and forecast, but also improve model in longer term in terms of improved state of the science and more realistic parameterizations

HFIP Realtime Experiments (HREX) and the Hurricane Ensemble in Real-time on the Cloud (HERC)

• Lots of promising experiments with potential for implementation into HAFS 2.x

• For 2025 experiments, we need to “think big” with additional core hours (and more efficient cores) on upcoming NOAA HPC machine Ursa

• EMC and HRD need to work closely with NHC to gauge feedback and take suggestions during the experimentation process

Joint Technology Transfer Initiative and the Hurricane and Ocean Testbed

Key takeaways:

• Need to plan ahead of time how to get projects “through the funnel”

• Increasing emphasis on forecaster tools to process large amounts of data, including new observations and AI models

• Leveraging AI to assist in targeted observations to highlight regions of enhanced model sensitivity

• Forecaster tools that allow hurricane specialists to quickly compare model forecasts to observations