Fuqing Zhang, Professor, Department of Meteorology and Department of Statistics; Director, Penn State Center for Advanced Data Assimilation and Predictability Techniques; Pennsylvania State University - November 2017 UCAR Congressional Briefing

1. Big Science, Big Data and Big Computing for
Advanced Hurricane Monitoring & Prediction
Fuqing Zhang
Director, Center for Advanced Data Assimilation and Predictability Techniques
Professor, Department of Meteorology and Atmospheric Science
Pennsylvania State University
Research sponsored by NSF, ONR, NOAA and NASA
FROM RESEARCH TO INDUSTRY:
How Earth system science enables private sector innovation

2. Track forecasts have improved drastically over past 25 years: a 3-
day forecast today is as accurate as a 1-day forecast was in 1989.
Intensity forecast accuracy has remained generally stagnant over
that same period, except for the last few years, thanks to the
Hurricane Forecast Improvement Program (HFIP) led by NOAA.
National Hurricane Center Official TC Forecast Errors

3. NSF HPC Computing
At Texas Advanced
Computing Center
System Name: Ranger
Operating System: Linux
Number of Cores: 62,976
Total Memory: 123TB
Peak Performance: 579.4TFlops
Total Disk: 1.73PB (shared)
HFIP Allocation : 30M SUs
(July 1, 2008- 31 March 2009)
Goals of NOAA’s Hurricane Forecast
Improvement Project (HFIP)
• Reduce average track and intensity forecast error by
half for days 1 through 5
• Significantly increase the probability of detection for
rapid intensity change and decrease false alarm ratio
• Extend lead time for hurricane forecasts out to Day 7
My team’s HFIP effort co-funded by ONR, NSF and NASA

4. How to make better input to the hurricane models?
High-resolution observations from Hurricane Hunters and UAVs: Provide
crucial airborne inflight measurements, dropsondes, Doppler Radar Winds, …
50°
~ 3 km
Data assimilation: The process of generating initial conditions for weather prediction models
through combining the background model estimate, and all applicable up-to-date observations.
NOAA P3 NASA
Global
Hawk

5. PSU WRF-EnKF Hurricane Analysis & Prediction System
with advanced assimilation of airborne Doppler Radar Vr
Evaluated for all 100+ P3 TDR missions during 2008-2012
The PSU system uses the NCAR’s WRF model; TDR Methodology now adopted by NOAA .
(F. Zhang and Y. Weng 2015, Bulletin of the American Meteorological Society)
PSU WRF-EnKF Hurricane Intensity error (knots)

6. Penn State hurricane research in NSF “Big Data” rollout

7. ext Frontier: Geostationary Satellite GOES-R
from NASA to NOAA

8. Assimilating All-sky GOES-R Radiances: Harvey (2017)
Independent observations vs. EnKF analysis of channel 10
PSU WRF-EnKF assimilates channel 8 radiances every 1 hour
Research Funding Provided by ONR, NASA, NOAA and NSF

9. Independent Observations vs. EnKF analysis
Assimilating All-sky Satellite Radiances: Harvey (2017)

10. PSU WRF-EnKF Harvey Forecast with GOES-R Assimilation
in comparison with WRF(NoDA), operational HWRF & best track
Research Supported by ONR, NASA, NOAA and NSF

11. Research Supported by ONR, NASA, NOAA and NSF
FV3 Prediction of Hurricane Harvey with a 3-km nested domain
PSU WRF-EnKF GOES-R assimilation used for FV3 initial vortex
FV3 3km-nest simulated radar reflectivity (left) vs. observations (right) at Landfall
KCRPKCRP
KHGXKHGXKEWX KEWX
KGRKKGRK

12. Research Supported by ONR, NASA, NOAA and NSF
FV3 Prediction of Hurricane Harvey with a 3-km nested domain
PSU WRF-EnKF GOES-R assimilation used for FV3 initial vortex
FV3-forecasted (left) vs. observed event total rainfall (right) with 6-day lead time
KCRP
KHGX
KEWX
KGRK
KLCH
KPOE
KSHV
KLIX
KCRP
KHGX
KEWX
KGRK
KLCH
KPOE
KSHV
KLIX
Point maximum of 40+ inches Point maximum of 50+ inches

13. Promises of US’s Next-Generation Global Prediction System: FV3
Comparison with EC model on 1-year-mean 10-day-forecast 500mb anomaly correlations
Forecast lead time (h)
FV3 model with EC model initial condition (IC) comparable to EC day 1-7 but better thereafter;
FV3 model with current GFS initial condition is considerably worse than either run with EC IC;
US forecast is inferior due mostly to poorer IC, inferior data assimilation ingesting less data
Courtesy of Linus Magnusson at ECMWF
and SJ Lin at NOAA/GFDL
AnomalyCorrelationCoefficient(ACC)

14. From Research to Industry
Concluding Remarks: Invest on big data hurricane science
• There are great potentials and needs for improving hurricane prediction through
collaborative research and sustained federal investment on data, science and personnel
• Future hurricane prediction advances may come from the following areas:
• Observations: advanced observing systems such as those from airborne dropsondes,
Doppler radar and weather satellites
• Model: cutting-edge higher-resolution weather prediction models with more accurate
numerics and physics
• Data assimilation: comprehensive algorithms and methodologies that that can more
effectively ingest existing and future observations into state-of-science models
• Computing: high-performance computing facilities that can perform advanced analysis
and forecasting in a timely manner.

15. + 3 kts initial intensity error
Shear only
Initial V + shear
Track
Initial + track
Initial inner core
moisture error
Predictability and Error Sources of Hurricane Intensity Forecast
(Emanuel and Zhang 2016, Journal of Atmospheric Sciences)

16. With realtime EnKF assimilation of airborne Doppler winds
(Zhang and Weng, 2015 BAMS)
What “big data” means for hurricane research: ensembles

17. Research Supported by ONR, NASA, NOAA and NSF
FV3 Prediction of Hurricane Harvey with a 3-km nested domain
PSU WRF-EnKF GOES-R assimilation used for FV3 initial vortex
in comparison with FV3(GFS), operational GFS, EC & best track
Ongoing PSU collaboration with NOAA/GFDL