Recent Advances and Future Challenges in Hurricane Prediction

Yuh-Lang Lin Professor Departments of Physics Department of Energy & Environ. Systems Senior Scientist NCAT ISET Center Outlines

1. The Need for Skillful Hurricane Prediction 2. Origins of Tropical Cyclones 3. Numerical Weather Prediction 4. Hurricane Track Prediction 5. Hurricane Intensity and Rainfall Predictions 6. Seasonal Hurricane Forecasts 7. Effects of Global Warming on Hurricanes 8. Summary

2 1. The Need for Skillful Prediction  More people live along coastal areas – it takes longer time to evacuate.  Emergency managements are very costly: (e.g., it costs ~$1M per mile of coastline evacuation).  Evacuation decision making is very sensitive to the prediction of hurricane track, intensity and size.  More hurricane related fatalities now due to storm surge or inland flooding which depends on accurate TC prediction.  More and stronger hurricanes are coming due to global warming?! 3 2. Origin of Tropical Cyclones Tropical cyclones form over tropical oceans with sufficient sea-surface temperature (> 26.5oC), circulation (vorticity), moisture and instability, and weak vertical wind shear.

Definitions of tropical cyclones: Tropical Tropical Storm Hurricane/ Depression Typhoon 17 m/s 33 m/s (38 mph) (75 mph)

Hurricane Patricia (2015) Major Hurricane [89 m/s (200 mph), 879 mb] 50 m/s (112 mph) Super Typhoon Typhoon Haiyan (2013) 67 m/s [87 m/s (195 mph), 895 mb] (150 mph)

4 About 85% of major hurricanes were initiated by African easterly waves (AEWs) [e.g., pre-Hurricane Alberto (2000) AEW]

5 (From Lin et al. 2005), based on EUMETSAT Some Basic Dynamics of Hurricane Genesis are still not well understood:  Why are there so many easterly waves and so few tropical storms?  What processes ”choose” a particular Easterly Wave?  What are the major formation mechanisms of hurricanes?

. Conditional Instability of the Second Kind (CISK) (Charney & Eliassen 1964) [dominant theory for 1964 - 1990’s.] . Cooperative Intensification (Ooyama 1964, 1969) . Wind-Induced Surface Heat Exchange (Emanuel 1986) . Marsupial Paradigm (Montgomery et al. 2008)

6 3. Numerical Weather Prediction (NWP)

Observations Preprocessing (Initialization)

Analysis

Running Data NWP Model Assimilation

Postprocessing Numerical Model Output Weather Prediction System Forecasters

(Adapted after Users Uccellini 2006) 7 Physically, the Newton’s second law is applied to describe air motion in x, y, and z directions: F du F F  ma  a   a   x m x dt m

du F dv F dw F  x ;  y ;  z dt m dt m dt m

This gives three momentum equations.

The conservation of mass is applied to derive the continuity equation. The conservation of energy and ideal gas law are used to derive the thermodynamics equation.

8 Mathematically, a NWP model solves an initial-value and boundary-value problem (IVP & BVP) in a rotating frame of reference: (Primitive Equations)

du 1 p   fv  F x-momem. eq. (1) dt  x rx dv 1 p    fu  F y-momem. eq. (2) dt  y ry dw 1 p    g  F z-momem. eq. (3) dt  z rz

d  u v w  Continuity eq. (4)      dt  x y z  dT  Q Thermo. energy eq. (5) dt p  RT Eq. of state (6) 9 NWP Model Development: A numerical model based on the above primitive equations may be developed step by step.

For example, an Advection Model can be constructed based on the inviscid nonlinear Burger equation

u' u'  (U  u')  0 t x

Apply a finite difference method at discrete points in x and t

ut1  ut1 ut  ut i i  (U  ut ) i1 i1  0 2t i 2x

t1 Solve for ui t ut1  ut1  (U  ut )ut  ut  i i x i i1 i1 10 The Advection Model may be used to study some basic wave properties and extend to more complicated models.

u' u'  (U  u')  0 t x

11 Sensitivity test can be performed to understand the nonlinear effects

u' u' U  0 t x

12 The advection model can be extended to 2D & 3D shallow-water tank models based on shallow-water systems

2D Tank Model

3D Tank Model

13 The 3D Tank Model can then be further extended to build a simple NWP model for solving the primitive equations (1)–(7).

In 1922, Lewis Richardson, did the very first numerical weather prediction based on a simple primitive equation model. He made a 6-h forecast with hand calculators which took more than 6 weeks.

The first successful NWP was performed using the ENIAC digital computer in 1950 by Charney, Fjotoft, von Neumann et al.

Today’s NWP: http://www.ncdc.noaa.gov/sites/default/files/NAM_20120710_ 0000_refcclm-small.gif (NOAA NCDC)

14 Mathematically, there exist challenges. For example, 1. Lower, upper, and lateral boundary conditions. 2. A need of initialization: initial conditions, i.e., observed data needs to be put on model grid points and consistent with equations. 3. Data assimilation was developed to incorporate new obs into NWP model, such as Nudging, 3DVAR (variational assimilation), 4DVAR, and EnKF (Ensemble Kalman Filter). 4. The need of conservation of mass of global model leads to the development of staggered grids. 5. Ensemble forecasting is developed to generate a representative sample of possible future states of the atmosphere, as a dynamical system. 6. The number of primitive equations grows when more physical processes are involved, such as moist processes. 7. Then, came the big question of the predictability of the atmosphere, as proposed by Lorentz.

15 Physically, there exist many challenges, too. For example, 1. For a fully-compressible system with sound waves included, CFL criterion will require extremely small time interval. Time-splitting scheme has been developed to resolve this problem. 2. Parameterizations of subgrid-scale processes remain challenging, such as planetary boundary layer, cumulus and cloud microphysics, radiation, air-sea interaction, etc. 3. Inclusion of moisture adds 6 – 7 additional equations and faces challenges in how to parameterize the subgrid processes. 4. Need more accurate, frequent and evenly-distributed data for model initialization. 5. Verification of forecasting results require field experiment (campaign) which are very expensive. 6. NWP models rely on global models to provide i.c. and b.c., thus inherit errors from global model simulations. 7. Need more powerful supercomputers for real-time forecasting.

16 Examples of Special Techniques used in NWP Models: Using a moving, nested grid domain with higher resolution to follow a hurricane:

Note that there is not much data over the ocean, which is one major source of forecast errors! 17 A grid mesh moving with hurricanes Gustav (2008) Ike (2008)

Hanna (2008) Kyle (2008)

18 Roop, Lin, Tang (2008) Unstructured Adaptive Grid

OMEGA Model (SAIC) 19 Numerical Weather Prediction using Global Models Lat/Lon Model Icosahedral Model

• Near constant resolution over the globe • Efficient high resolution simulations  lk ,i  NOAA Earth System Research Laboratory - Boulder, Colorado nk,iPage 20 4. Hurricane Track Prediction

A hurricane may move as far as several thousand kilometers away from its origin. Hurricane track prediction has been improved significantly in the last few decades

Hurricane tracks are mainly influenced by: • Environmental flow (e.g., high) • Synoptic systems (e.g., a cold front) • Variation of Earth’s rotational rates (b effect) • Topography (e.g., Appalachians, Hispaniola, etc.) • Vertical wind shear • Convective heating

• Sea surface temperature distribution 21 The tropical cyclone tracks we are dealing with!

westerly H H

b effect easterly

(Neumann 1993, Lin 2007) Genesis locations and tracks of tropical cyclones with wind speeds of at least 17 ms-1 for the period of 1995-2004.

22 Hurricane Dennis’ (1999) track Track Deflection by Appalachians was influenced by frontal system (Liu, Lin & Chen, 2014)

Typhoon Haitang (2005)

Obs. Wu-Fen- 19/00Z (TY) ★ 25N Shan

18/12Z (TY) 24N Hua- ★ Lien 18/00Z (STY) 23N 17/12Z (STY)

22N

(Wang 1980 NSC; Lin 2007) Jian and Wu (2006) 23 Numerical models used for hurricane prediction

 CLIPER (CLImatology and PERsistence) - a statistical-climatological model - being exceeded by numerical models after the 1980s.

 NHC98 - a mixed statistical-numerical model

 Simplified numerical models: BAMS, VICBAR

 GFDL Model – A triply-nested movable mesh numerical model solving partial differential equations

 Hurricane WRF (HWRF)

 Global numerical models: NCEP GFS, MRF, ECMWF, NOGAPS, UKMET

*Yellow-highlighted are currently used by the National Hurricane Center (NHC) 24 NATIONALNATIONAL HURRICANEHURRICANE CENTERCENTER ATLANTICATLANTIC TRACKTRACK FORECASTFORECAST ERRORSERRORS 500 Major upgrade in global & 1964-1973 hurricane models 400 1974-1983

300 1984-1993 2003-2005

200 1994-2003 • Higher quality observations • Advances in data input into models 100 • Better numerics and physics in models

0 Error (nauticalmiles) Error 12 24 36 48 72 96 120 (Uccellini Forecast Period (hours) 2006) 23 May 2006

25 Major Improvements • Major upgrade in global & hurricane models • Higher quality observations • Advances in data input into models • Better numerics and physics in models

26 Example: Track Prediction of Hurricane Katrina (2005) Near Landfall Earlier

forecast observed 8/29 forecast

observed 8/24

Katrina Prediction

hr 27 Simulation of Hurricane Katrina (2005) by NASA Global Model

(Courtesy of Dr. Bo-Wen Shen NASA/GSFC) 28 Many models had missed forecasting the unusual inland track deflection 5 days before Sandy’s (2012) landfall

Forecasts of Sandy (2012) began at 00Z Oct. 23, 24, 25, and 26 for every 12 h by GFDL, HWRF, ECMWF, and GFS (Blake et al. 2013). The NHC best track is denoted by the hurricane symbol. 29 The offshore forecast error may be due to the Omega Block

Gall et al. (2013) 30 Major Hurricane Joaquin forecasts by models Observed track

We still have plenty room for improvement on longer-term track forecast! 31 Track Forecast Skill Comparison

(From NHC 2014 verification report) 32 GFS Forecast versus Reanalysis (Forecast at 1025.00Z)

33 Before landing about 2330 UTC OCT 29 near Brigantine, New Jersey, in addition to the steering of synoptic systems, Sandy seems experiencing a Fujiwhara effect with the inland trough.

34 Interaction of Sandy (2012) and a Trough simulated by the NASA Global Mesoscale Model

(Courtesy of Dr. Bo-Wen Shen, University of Maryland and NASA) 35 At 500 mb, a series of processes can be found:  At 9/29/00Z, an inland trough was deepening and approaching Sandy.  During 9/29/06Z - 9/29/12Z, there seemed having a Fujiwhara effects going on between Sandy and the trough.  Started at 9/29/18Z, Sandy and the trough merged.

500 mb rel. vorticity & geopotential

Small, weak vortex embedded

Lin, Spinks, Smith, & Shen (2014 Hurricane Conference) 36 What is Fujiwhara Effect?  Two nearby TCs rotate around each other cyclonically. Reason: A TC vortex behaves like a solid-body rotation, thus is advected (e.g., DeMaria & Chan 1984) by the outer circulation of a nearby TC.  This is analogous to the mutual induction of a pair of wake vorticies.

Fujiwhara Effect on TS Parma and Mutual Induction happens between two Typhoon Melor on October 6, 2009 between two Wake Vortices, too.

Thus, the key is to check whether they can affect each other. 37 Simulated 500 mb Streamline Fields at 10/29/14Z

CNTL HfSandy NoSandy

Comparing the 500mb vorticity, geopotential, and streamline fields of cases CNTL, HfSandy, and NoSandy at the ending time of the hypothesized mutual rotation, we found that

Sandy played insignificant roles in enhancing the intensity and repositioning the inland trough. In other words, there was no or insignificant Fujiwhara effect.

38 5. Hurricane Intensity and Rainfall Prediction Improvement needed in intensity prediction

(NHC 2015) [http://www.nhc.noaa.gov/verification/verify5.shtml] 39 Intensity Forecast of Hurricane Patricia (2015)

• Most of the models predicted strengthening • Rapid intensification is underpredicted • The NHC explicitly forecast rapid intensification in their advisory at 11 p.m. EDT on Oct. 21 night and in subsequent advisories on Oct. 22. • Very favorable environment (A. A. Wing, 2015) o low vertical wind shear o very humid air o very warm sea surface temperatures (30oC) o high upper ocean heat content).

40 Intensity Forecast for Hurricane Patricia (2015)

Allison A. Wing (2015) 42 Relative Humidity in Eastern Pacific

The weather channel 43 Very high sea surface temperatures of 30 °C (86 °F)

44 Inner core and rainbands need to be well observed and represented in the model

Hurricane Eyewall Rain Bands Core Structure

45 To improve track, intensity, structure, and rainfall forecasts, we need:

• High quality hurricane core and environmental observations and model resolution • Advanced data assimilation techniques for environment and hurricane core • Advanced modeling system in numerical methods and physics representation • Disciplined approach for transition from research to operations

46 6. Seasonal Hurricane Forecast Based on climatology, the tropical cyclone activity in Atlantic basin is strongly related to some factors such as

• Atlantic surface pressure • Sea surface temperature anomaly • General circulation pattern • El Nino • Quasi-Biennial Oscillation • African easterly waves • Sahel rainfall • Saharan dust

Thus, predictors for seasonal Atlantic hurricane activity can be constructed. 47 Example of Seasonal Hurricane Forecasts

2006 Atlantic Hurricane Season Forecasts (May-issued forecasts)* Actual Forecast 25-year CSU NOAA TSR Season Parameter Average Forecast Forecast Forecast Total Named Storms 11.0 17 13-16 14 9 Hurricanes 6.4 9 8-10 8 4 Major Hurricanes 1.2 5 4-6 3 2

2014 Atlantic Hurricane Season Forecasts Actual Forecast 29-year CSU1 NOAA2 TSR3 Season Parameter Average Forecast Forecast Forecast Total Named Storms 12 9 8-13 12 8 Hurricanes 6 3 3-6 6 6 Major Hurricanes 3 1 1-2 3 2

1Issued 10 April 2014 2Issued 22 May 2014 3Issued 7 April 2014

48 Improvements needed

• Seasonal hurricane forecast still needs significant improvement. • Some dynamics are still not quite understood yet. • Dynamical, instead of statistical, model predictions are more promising.

49 7. Effects of Global Warming on Hurricanes Stronger hurricanes are coming due to global warming?!

Total Atlantic hurricane power dissipation index Increase of CO2 results stronger (PDI) more than doubled in the past 30years hurricanes, but not frequency!

(Emanuel 2005, Nature) (Knutson and Tuleya 2004, J. Climate)

Landsea (2005 Nature) commented that: (1) Emanuel’s analysis was biased. (2) There was a lack of accurate observations in earlier years. 50 8. Summary

• Track prediction has been significantly improved in the last few decades. • Longer-term (e.g., 5 days) prediction can be further improved. • Intensity prediction need significant improvement • Significant seasonal hurricane forecasts may be significantly improved by adopting dynamical models. • Impacts of climate change on hurricane intensity and frequency require more studies. • More and higher quality obs, adv. initialization, better numerics and physics in models needed for further improvement. • Some basic dynamics are still not well understood (e.g., TC genesis, 2ndry eyewall formation and replacement, etc.)

51 Thank you!

52 53 54 Forecast of Hurricane Patricia (2015)

• Most of the models predicted strengthening • Rapid intensification is underpredicted • The NHC explicitly forecast rapid intensification in their advisory at 11 p.m. EDT on Oct. 21 night and in subsequent advisories on Oct. 22. • Very favorable environment (A. A. Wing, 2015) o low vertical wind shear o very humid air o very warm sea surface temperatures (30oC) o high upper ocean heat content).

55 Intensity Forecast for Hurricane Patricia (2015)

Allison A. Wing (2015) 56 Satellite image of Patricia early on Friday morning

The weather channel 57 Relative Humidity in Eastern Pacific

The weather channel 58  Exceptionally favorable atmospheric conditions for rapid intensification of Hurricane Patricia: o light wind shear o very high sea surface temperatures of 30 °C (86 °F) o high moisture levels

59 Genesis Potential Index

 Refinement of Gray’s tropical cyclone genesis index using Reanalysis data (Emanuel & Nolan 2004).

5 3/2 3 3 -2 GP= |10 η| (H/50) (Vpot/70) (1+0.1 Vshear)

η = absolute vorticity at 850hPa (s-1) H = relative humidity at 700hPa (%)

Vpot = potential intensity (m/s) Vshear = magnitude of the vertical wind shear between 200 and 850hPa (m/s).

K.A. Emanuel and D. Nolan, BAMS 85, 667-668 (2004). Any global warming?

Annual Average Global Surface Temperature Anomalies 1880-2008 (http://www.epa.gov/climatechange/science/recenttc.html)

61 Hockey Stick Graph of Global Warming?

Red line: rescaled IPCC 1990 (Fig. 7.1c), based on Lamb (1965) showing central England temperatures; compared to central England temperatures to 2007, as shown in Jones et al. 2009 (green dashed line).[15] Also shown, Mann, Bradley & Hughes 1998 40 year average used inIPCC TAR 2001 (blue), and Moberg et al. 2005 low frequency signal (black).

62 Any global warming?

Robinson et al. (2007)

63 Is the global temperature mainly caused by human activities or natural oscillation?

Robinson et al. 2007

64 Orographic Effects on Typhoons over Taiwan’s Central Mountain Range

Continuous Track Discontinuous Track (strong typhoons) (weak typhoons)

(Wang 1980 NSC; Lin 2007)

65 When blocking is strong (small R/a, U/Nh, Vmax/Nh or h/a), typhoon track is deflected significantly .

Haitang (2005) Obs. Wu-Fen- 19/00Z (TY) ★ 25N Shan

18/12Z (TY) 24N Hua- ★ Lien 18/00Z (STY) 23N 17/12Z (STY) 22N 120E 121E 122E 123E

Jian and Wu (2006) The circling tracks of Haitang (2005) and Krosa (2007) are due to strong blocking.

Lin et al. (2005) 66