improving tropical cyclone track and intensity in a global ... · [email protected] - a13r-06, agu...

Improving Tropical Cyclone Track and Intensity in a Global Model with Local Mesh Refinement

Colin M. Zarzycki National Center for Atmospheric Research Advance Study Program (ASP)/Climate and Global Dynamics (CGD)

Christiane Jablonowski (University of Michigan) American Geophysical Union Fall Meeting A13R-06 December 15th, 2014

[email protected] - A13R-06, AGU Fall Meeting, San Francisco, CA, December 2014

Numerical weather prediction

•  Generally two frameworks •  Global models

•  Proper representation of global large-scale dynamics

•  Computationally-expensive for high-resolution!

•  Nested/regional models •  Effective way to make

efficient use of computing resources for a regional problem

•  Boundary conditions, no teleconnections

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V,T U,V,T U,V,T U,V,T U,V,T U,V,T U,V,T U,V,T U,V,T U,V,T U,V,T U,V,T



Numerical weather prediction

•  Generally two frameworks •  Global models

•  Proper representation of global large-scale dynamics

•  Computationally-expensive for high-resolution!

•  Nested/regional models •  Effective way to make

efficient use of computing resources for a regional problem

•  Boundary conditions, no teleconnections

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T

U,V

,T



Bridge the gap: Can we apply this construct of regionally distributing

computation load in a global modeling

framework?


•  Community Atmosphere Model Spectral Element (CAM-SE) dynamical core •  New CAM default as of version

5.3 •  High-order numerics •  Solve primitive equations on

individual elements •  Only nearest neighbor

communication •  Highly scalable

CAM-SE

Den

nis

et

al.

, 20

11


•  Unstructured grid -> conforming quads

•  Timestep restricted to finest grid scale

•  Explicit diffusion (hyperviscosity) function of cell resolution

Variable-resolution CAM-SE

Variable-resolution: regionally high-resolution in a global framework

0.5° (~55 km)

0.125° (~13 km)


CAM-SE “forecast mode”

•  Every 12 hours (00Z and 12Z) from August 1st to November 1st for 2012-2013 •  Atmosphere: GFS analysis, forward DFI (Fillion et al., 1995) •  Ocean: Prescribed SSTs/ice (GFS analysis) •  Land: active (initial cycled spinup, then 12 hourly forecast used for init)

•  8 day forecast = ~1.5 hours of wall clock time on 800 cores (NCAR Yellowstone) •  ~10x cheaper than a globally-uniform 13 km forecast

•  Longer integration? •  10x ensemble members? •  Doubling of resolution?

Sandy TPW: INIT 12Z 10/25/12


Resolution comparison Sandy 500 hPa vorticity: INIT 00Z 10/23/12

Var-res (55 km -> 13 km) Uniform 13 km Uniform 55 km


Resolution comparison Sandy 500 hPa vorticity: INIT 00Z 10/23/12

+120 hours

Var-res (55 km -> 13 km) Uniform 13 km Uniform 55 km


CAM-SE “forecast mode” control

Track error

Unrefined 55 km Refined 13 km



Track error Absolute wind error




Track error Absolute wind error Wind bias





•  Refinement improves both track, intensity skill •  Track behavior of TCs looks good… •  … CAM exhibits a high bias in TC intensity, especially as the

solution moves away from initial state



CAM-SE “forecast mode” sensitivity

•  Sensitivity simulations •  Choose 20 worst CAM forecasts (high wind bias)

1.  Turn off deep convection (CAMY) 2.  Decrease physics timestep by 4x (30 min to 7.5

min) (CAMX)


No deep 13 km dtphys/4 13 km Control 13 km


Precipitation rates

Control

No deep

dtime/4 Marks et al., 2002

Hurricane Leslie, INIT: 2012-08-31 00Z, valid +120 h

TOTAL LARGE-SCALE CONVECTIIVE


Precipitation rates

Control

No deep

dtime/4 Marks et al., 2002

312 mm/day

Hurricane Leslie, INIT: 2012-08-31 00Z, valid +120 h

TOTAL LARGE-SCALE CONVECTIIVE

700-800+ mm/day


Condensate loading feedback?

Prognostic (with condensate)

Diagnostic (w/o condensate)

•  In many hydrostatic models, removal of condensate (rain, snow, etc.) occurs “instantaneously”

•  At high loading (i.e., extreme events), the delta between pressure with and without condensate can be large…

Bacm

eist

er e

t al

., 2

012,

GRL

Courtesy Julio Bacmeister Precip rate ->

Pres

sure

dev

iati

on -

>


•  CLUBB (Cloud Layers Unified by Binormals) •  Replace PBL, macrophysics, shallow convection

•  60 initialization times •  Rerun 40 “worst” (both ways) + 20 “best” from

control

Some (potentially) good news?


CLUBB 13 km Control 13 km


Hurricane Sandy forecasts

Observations CAM members CAM ens. mean CAM-CLUBB GFS (oper.) GEFS members GEFS ens. mean


Hurricane Sandy forecasts

Observations CAM members CAM ens. mean CAM-CLUBB GFS (oper.) GEFS members GEFS ens. mean

CAM uses same initial conditions as GFS, correctly forecasts recurvature +8d -> exonerates data assimilation?

Dycore impact? Possibly, but more than likely, choice of physical

parameterizations key in track differences (ex: Bassill 2014)


Summary

•  CAM-SE demonstrates potential for var-res global models to be used for NWP/TCs •  Refinement improves TC track and intensity

forecast •  Reproduces behavior of uniform high-res runs •  Substantially “cheaper” than uniform high-res runs

•  Tropical cyclone intensity problematic beyond 25 km horizontal resolution with default CAM physics •  Physics/dynamics coupling important for TC

forecasting •  Highlights need for scale-aware physical

parameterizations?

improving tropical cyclone track and intensity in a global ... · [email protected] - a13r-06, agu...

Documents