Tropical Cyclone Prediction for HFIP with COAMPS-TC

Richard M. Hodur1, S. Chen2, J. Cummings3, J. Doyle2, T. Holt2, H. Jin2, Y. Jin2, C.-S. Liou2, K. Sashegyi2, J. Schmidt2

1Science Applications International Corporation, Monterey, CA2Naval Research Laboratory, Monterey, CA

3Naval Research Laboratory, Stennis Space Center, MSInterdepartmental Hurricane Conference

Savannah, GA1-4 March 2010

HFIP: Hurricane Forecast Improvement Project

Tropical Cyclone Prediction for HFIP with COAMPS-TC

Introduction

•COAMPS-TC performed well in TCS 2008•HFIP 2009 Goals:

• Run COAMPS-TC in real-time for west Atlantic and east Pacific TCs• Assess the skill of the COAMPS-TC predictions• Follow-on research to improve prediction of TC track, structure, and intensity

•Outline:•COAMPS-TC Description

•Real-Time COAMPS-TC Runs during HFIP

• System configuration

• Discussion of Results/Implications

•Conclusion/Future Research

COAMPS-TCCoupled Ocean/Atmosphere Mesoscale Prediction System

•Analysis:•Atmosphere:

• 3DVAR Analyses of u, v, T, and Heights (NAVDAS)

• Synthetic Observations Used to Incorporate TC Vortex (1000 – 400 mb)

• Relocation of TC in Background on Warm Starts

•Ocean: 2D OI of SST (NCODA)

•Model:•Numerics: Nonhydrostatic, Scheme C, Sigma-z, Flexible Lateral BCs

•Parameterizations: PBL, Convection, Explicit Moist Physics, Radiation, Surface Layer

•TC Tools: Automated Moving Nests and Tracker, Dissipative Heating, Sea-Spray, Shallow Convection

•Features:•Globally Relocatable (5 Map Projections)

•User-Defined Grid Resolutions, Dimensions, and Number of Nested/Parent Grids

•Incremental Data Assimilation

Coupling to ocean and wave models not included in HFIP 2009 runs

• Synthetic Observations Built From:• Modified Rankine Vortex• JTWC Warning Message w/Satellite Data• NOGAPS T20/L15 truncated fields

• Blend Synthetics w/all other observations in 3DVAR (NAVDAS)

Synthetics to represent TC circulation

900

km

900 kmCOAMPS Synthetics

High-Resolution Synthetic Observations for TC Initialization in COAMPS-TC

Case: 0000 UTC 16 August 2009 (Bill; 03L)

Issues/Comments• Some influence of NOGAPS TC circulation

seen in COAMPS analysis fields (cold starts)• For warm-starts, TC circulation is relocated

to warning position

Wind Speed (kts)40200

NOGAPS first-guess fields –

have their own TC structure

NOGAPS First-Guess900 km

900

km

Improved TC representation with synthetics using 3DVAR

COAMPS Analysis900 km

900

km

Procedure for Running COAMPS-TC for HFIP 2009

• 45/15/5 km grids for WATL and EPAC basins

• 45 km grid fixed for all storms

• Inner 2 grids move with the TC

• All runs automatically submitted based on observed TC location/intensity at 0335 of each watch

• Forecasts run to 120 hours

• First run for each TC is a cold start, 12 h warm start for each subsequent run

• Output from each run posted on NRL web site; Forecast tracks sent to FSU and NCAR

WATL

EPAC

COAMPS-TC 2009 TC ForecastsWATL and EPAC

TC Track Forecast Errors in 2009Homogeneous Samples

West Atlantic Results

East Pacific Results

1. Track Error Related to Initial Intensity of TCWeakest storms (< 60 knots) exhibit the largest track errors at 12-24 hours; most often seen as a right-bias, but this can be a speed bias

2. Spotty Convection during Spin-Up of TCModel solutions exhibit spotty convection during first 24-30 hours until TC matures; Large impact on data assimilation

3. Initial Imbalance in TC Vortex for Strong TCsPredicted TC Intensity decreases in the first 6-12 hours when the initial maximum winds > 60 knots

4. Positive Bias in TC Intensity ForecastsThere is a tendency in the model to make the TC too strong over the course of the 120 h forecasts

Issues with COAMPS-TC Forecasts for HFIP 2009

Black line: Warning positions, large white circle with day at 0000 UTC, small white circle at 1200 UTC.Colored lines: COAMPS forecasts starting from different times with a circle every 12 hours.

Bill

First several forecasts of Bill had

significant speed errors (too slow)

Fred

First several forecasts of Fred had a

significant right bias

1. Track Error Related to Initial Intensity of TCWeakest storms (< 60 knots) exhibit the largest track errors; most often seen as a right-bias,

but this can be a speed bias

Other forecasts from the east Pacific and the west Pacific also exhibited a right bias,

particularly in their early stages

Black line: Warning positions, large white circle with day at 0000 UTC, small white circle at 1200 UTC.Colored lines: COAMPS forecasts starting from different times with a circle every 12 hours.

Felicia

Guilliermo

1. Track Error Related to Initial Intensity of TCWeakest storms (< 60 knots) exhibit the largest track errors; most often seen as a right-bias,

but this can be a speed bias

12- and 24-hour track errors are worse when TC initial maximum wind speed is < 60 knotsThis has been found to occur in all basins

1. Track Error Related to Initial Intensity of TCWATL 2009

NOGAPS Synthetics: NOGAPS fields used for LBC, Synthetics used in NAVDAS to initialize TC Vortex structure (Benchmark)

NOGAPS No Synthetics: NOGAPS used for LBC, TC Synthetics not used (only relocation of TC Vortex at initial time, no special DA of TC Vortex)

GFS Synthetics: GFS fields used for LBC, Synthetics used in NAVDAS to initialize TC Vortex structure

GFS No Synthetics: GFS used for LBC, TC Synthetics not used (only relocation of TC Vortex at initial time, no special DA of TC Vortex)

NOGAPS: “early” run, 1-degree fields

GFS: “real-time” run, ½-degree fields

TC Vortex Specification and Lateral Boundary Condition Sensitivity Testing

Hypothesis #1: Right-bias is largely caused by TC Vortex Specification (Synthetics)Hypothesis #2: Large-scale fields play a large role in TC motion

Sensitivity Testing for Bill and FredEffect of TC Vortex Specification and Lateral Boundary Conditions

Tracks are very sensitive to LBC (Ensembles?) and specification of the TC structureTrack forecasts are improved (particularly, short-term) when synthetics are not used

Bill Results

Smallest track errors with use of GFS and no synthetics

Fred ResultsSmallest track errors with use of

NOGAPS and no synthetics

COAMPS-TC Forecast Tracks for TC BillEffect of TC Initial Structure and LBC (Use of NOGAPS/GFS, Synthetics/No Synthetics)

NOGAPSNo Synthetics

NOGAPSSynthetics

GFSSynthetics

GFSNo Synthetics

Largest Track Errors

Smallest Track Errors

COAMPS-TC Forecast Tracks for TC FredEffect of TC Initial Structure and LBC (Use of NOGAPS/GFS, Synthetics/No Synthetics)

NOGAPSNo Synthetics

NOGAPSSynthetics

GFSSynthetics

GFSNo Synthetics

Largest Track Errors

Smallest Track Errors

2. Spotty Convection during Spin-Up of TCBill: 2009081600

dB

z

12 h Radar Reflectivity

dB

z

48 h Radar Reflectivity

Model solutions exhibit spotty convection during first 24-30 hours until TC matures• Can be very chaotic – not necessarily aligned in TC bands• Most noticeable with weak storms• Negatively influences first-guess fields for next analysis

3. Initial Imbalance in TC Vortex for Strong TCs4. Positive Bias in TC Intensity Forecasts

WATL 2009

COAMPS-TC has tendency to weaken systems during first 12 hours, then strengthens them

InitializationUnbalanced Vortex

PhysicsExchange Coefficients? Lack of Coupling? Mixing? . . . . ?

12 h Intensity Change as a Function of Initial IntensityWATL 2009

Weak storms (<60 knots) tend to strengthen in first 12 hours of the

forecast more than observed

Strong storms (>60

knots) tend to weaken in

first 12 hours of the

forecast more than

observed

SST Change (72 h)

Air-Ocean Coupling in

COAMPS-TC Predicts SST Cool Wake of

2-3°C

Coupled Air-Sea Prediction of Bill with COAMPS-TC

Microwave Satellite Derived SST Shows 2-3°C Cool Wake Similar

to the Coupled Model

COAMPS-TC air-sea coupled forecasts for Bill alleviate an over-intensification bias as a result of cool SST wakes

Intensity Error (kts)18-23 Aug

Intensity Error Markedly Improved

using Coupled Model

Coupled

Uncoupled

TC Prediction for HFIP 2009 with COAMPS-TCConclusions/Future

•COAMPS-TC performed well in WPAC in 2008, 2009 (TCS)•COAMPS-TC did not perform as well in WATL and EPAC in 2009 (HFIP)•Problems with 2009 Performance related to:

•TC Vortex Specification:• Largest Track Errors when Initial Intensity < 60 knots (Right Bias, Speed Bias)• Significant Weakening of TCs > 60 knots in First 6-12 hours (Unbalanced TC Vortex)• First few DA cycles of any TC suffers from spin-up (Spotty Convection)

•Positive Bias in Predicted TC Intensity (Physics? Coupling? . . . . ?)•Follow-on Research:

•Eliminating the Use of Synthetics Suggests:• Significant Improvement in Forecast Tracks (Positive Result)• Reduction in the Weakening of Strong TCs in First 6-12 hours of forecast (Positive Result)• Analyzed Intensity is Weaker than with Synthetics (Negative Result)

•LBC Sensitivity: Motivation for Ensembles using different Global LBC•Future:

•DA, DA, DA, and Physics:• Specification of TC Vortex in Analyses (Synthetics?, Model spin-up?, EnKF?, 4DVAR?, ?, ?, ?)• Improved Handling of Convection on High-Resolution Grids (Initialization and Forecast)• Improved Handling of Surface Fluxes (and all other Physical Processes)• Air-Ocean-Wave Coupling (Recent Results are Encouraging)

•Interactions/Exchanges with HFIP, NOPP, and ITOP are Important

Data Assimilation on cloud-resolving grids in a highly-convective environment is a formidable task