1 / 37

Shian-Jiann Lin

Ensemble Seasonal Predictions and Medium-Range Deterministic Forecasts of Tropical Cyclones with GFDL’s prototype Global Cloud-Resolving Model. Shian-Jiann Lin. with contributions from: M. Zhao, I. Held, and G. Vecchi. NOAA/Geophysical Fluid Dynamics Laboratory. National Central University

rogan-bray
Download Presentation

Shian-Jiann Lin

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Ensemble Seasonal Predictions and Medium-Range Deterministic Forecasts of Tropical Cyclones with GFDL’s prototype Global Cloud-Resolving Model Shian-Jiann Lin with contributions from: M. Zhao, I. Held, and G. Vecchi NOAA/Geophysical Fluid Dynamics Laboratory National Central University March 9, 2009

  2. Outline • The GFDL prototype global cloud-resolving model [aka, High-Resolution Atmosphere Model (HiRam)] • Simulated tropical cyclone climatology with the C180 (~50 km) and C360 (~25 km) HiRam • Skill of the ensemble seasonal predictions (1985-2005) • Deterministic forecasts with the C360 (~25 km) HiRam Initialization: NCEP analysis (large-scale) with 4D vortex-breeding (small-scale) Preliminarily results: • 5-day forecasts for HFIP (Hurricane Forecast Improvement Project ) • 10-day forecast? Zhao, M., I. Held, S.-J. Lin, and G. Vecchi, 2009: Simulations of global hurricane climatology, interannual variability, and response to global warming using a 50km resolution GCM. Submitted to J. Climate.

  3. The GFDL High-Resolution Atmosphere Model (HiRam) is the same as the GFDL AM2.1 used in the IPCC AR4 Except the following: • Highly scalable non-hydrostatic Cubed-sphere FV dynamical core. • 6-category bulk cloud microphysics (based mostly on Lin et al. 1984); computational efficiency significantly improved with time implicit treatment of microphysics processes and vertically Lagrangian terminal fall of all condensates (rain, snow, ice, and graupel) • The deep convective parameterization scheme (Relaxed Arakawa-Schubert) is replaced by an essentially non-precipitating1shallow2 convection scheme (based on Bretherton et al. 2004) • Surface fluxes modified for high-wind (e.g., hurricane) situation over ocean (Moon et al. 2007) ; coupling to wave model under development. • Micro-physics and shallow convection are tuned for different resolutions to achieve a global balance of the radiative fluxes at top of the atmosphere To obtain better mean climate, precipitation from SC is allowed below the c360 resolution The “shallowness” of the convection can be controlled by an entrainment parameter; parameterized convection is allowed to go deeper in lower resolution configurations

  4. An equal-distance Gnomonic Cubed Sphere grid • Defined by intersects of great circles with equal-distance along 12 edges • Maximum local grid aspect ratio ~ 1.061 • Maximum global grid aspect ratio ~ 1.414 Resolutions evaluated: “cloud-resolving: C2000, x ~ 4.5 km Medium-range forecast: C360 , x ~ 25 km AMIP climate runs: C180 , x ~ 50 km IPCC AR5: C48 , x ~ 200 km Can also be used as a regional model Hurricane in a doubly periodic box

  5. The Finite-Volume (FV) Cubed Sphere dynamical core • Key features in the Cubed-Sphere dynamical core • Quasi-uniform resolution over the globe; self-consistent global-regional nesting (to be implemented) • Vertically Lagrangian control-volume discretization (for both hydrostatic & non-hydrostatic) • A Lagrangian (stable for large CFL number) Riemann Solver for sound waves; non-reflective upper boundary condition • Tracer transport is strictly 2D and uses a vertically dependent time stepping, which dramatically enhanced the computational efficiency if many tracers are required (e.g., chemistry, carbon cycle, and cloud microphysics) • Highly scalable: super linear scaling to ~10,000 CPUs at global cloud-resolving resolutions.

  6. Horizontal resolution vs. degree of parameterized deep convection Parameterized convection precipitation 200 km 100 km Resolution Observed 50 km

  7. Climate Model inter-comparisons: GFDL finite-volume models vs. 10 other IPCC models

  8. Observed cyclone tracks: 1981-2005 Simulated tracks: 1981-2005 (C180 model) One realization

  9. Seasonal cycle of hurricanes (1981-2005) (red: 4-member ensemble)

  10. Inter-annual cycle/trend (1981-2005) Number of West Pacific Typhoons Number of East Pacific hurricanes Number of North Atlantic hurricanes Reds: observed Blue: model (4 realizations) Model-obs correlation ~ 0.83

  11. 2005 Hurricane season C180 climate model ensemble vs. observed Observed

  12. C180 (50-km) model intensity distribution:

  13. Hurricanes: intensity vs. resolutionCentral pressure – surface wind correlation (1981-2005) Cat 4-5 C90 C180 C360 The obvious: “low-res. model has a low intensity bias”

  14. C360 simulation (work in progress)

  15. Outline • The GFDL prototype global cloud-resolving model [aka, High-Resolution Atmosphere Model (HiRam)] • Simulated tropical cyclone climatology with the C180 (~50 km) and C360 (~25 km) HiRam • Skill of the ensemble seasonal predictions (1985-2005) • Deterministic forecasts with the C360 (~25 km) HiRam Initialization: NCEP analysis (large-scale) with 4D vortex-breeding (small-scale) Preliminarily results: • 5-day forecasts for HFIP (Hurricane Forecast Improvement Project ) • 10-day forecast? Zhao, M., I. Held, S.-J. Lin, and G. Vecchi, 2009: Simulations of global hurricane climatology, interannual variability, and response to global warming using a 50km resolution GCM. Submitted to J. Climate.

  16. Skill of CSU (statistical) seasonal hurricane forecasts Suzana J. Camargo, Anthony G. Barnston, Philip J. Klotzbach and Christopher W. Landsea, 2007

  17. Dynamical Seasonal PredictionMethodology: • Apply an atmosphere model at the highest affordable resolution that is capable of producing a credible tropical cyclone climatology (as shown previously) • Initialized with ICs from AMIP-type runs; analyzed ICs not required. • SST: climatology + persistent May/June anomaly of the forecast year • Create large enough ensemble members (5 or more) with small perturbations (in ICs or SST)

  18. How do we get a QBO-like oscillation in the GFDL models? The model top must be at 0.1 mb (~60 km) or higher; vertical resolution ~2km or finer Unless ultra-high horizontal resolution, a gravity-wave-drag parameterization is required Low numerical dissipation to prevent suppression of “resolved” vertically propagating gravity waves

  19. Simulated SAO & QBO Model with higher order divergence damping Model with del-2 divergence damping

  20. GFDL Seasonal (Jul-Dec) hurricane predictions 1985-2005 • 5-member C180 (~50 km) model ensemble • SST: climatology plus persistent anomaly from June of the forecast year N. Atlantic Correlation ~ 0.66 E. Pacific Correlation ~ 0.63

  21. GFDL dynamical models vs. CSU statistical hurricane predictions for 1999-2007

  22. Outline • The GFDL prototype global cloud-resolving model [aka, High-Resolution Atmosphere Model (HiRam)] • Simulated tropical cyclone climatology with the C180 (~50 km) and C360 (~25 km) HiRam • Skill of the ensemble seasonal predictions (1985-2005) • Deterministic forecasts with the C360 (~25 km) HiRam Initialization: NCEP analysis (large-scale) with 4D vortex-breeding (small-scale) Preliminarily results: • 5-day forecasts for HFIP (Hurricane Forecast Improvement Project ) • 10-day forecast? Zhao, M., I. Held, S.-J. Lin, and G. Vecchi, 2009: Simulations of global hurricane climatology, interannual variability, and response to global warming using a 50km resolution GCM. Submitted to J. Climate.

  23. Short-term forecast initialization issues: Deficiency in “kinetic energy” at the meso-scales (NCEP GFS) Observed spectrum (Nastrom & Gage 1985) KE spectrum: NCEP T382 analysis KE spectrum: After 5 days of model “spin up”

  24. NCEP/GFS vortex initialization • “Remove” (by filters) the vortex from the GFS first-guess • “Relocate” the vortex to the observed position Main issue: Initial vortex (if existed) location is correct but the intensity is typically very weak

  25. Kurihara’s GFDL hurricane model initialization • “Remove” (by filters) the vortex from the NCEP analysis • “Grow” a balanced vortex offline by an axis-symmetrical model (to achieve internal dynamical, thermodynamic, and micro-physics balance). • “Insert” the fully developed and balanced vortex into the NCEP background • Main issues: • The large-scale “background” is not in dynamical or thermodynamic balance with the vortex – possibly causing degraded track forecasts • Multiple vortices do not interact with themselves nor the environment • Being 3D (time discontinuous), vortex information does not propagate to the next forecast time.

  26. A simple 4D data assimilation fortropical cyclone prediction: Large-scale nudging (using NCEP T382L64 gridded analysis) + storm-scale 4D (time continuous) vortex-breeding 00Z 06Z 12Z 18Z Assimilation cycle forecast

  27. A simple 4D assimilation for hurricane prediction (continued): “storm-scale vortex breeding” Interpolation in time the NHC “best tracks” (latitude, longitude; slp) Construction of two radial (Gaussian) SLP distributions based on the observed center pressure and model environment pressure (blue area); radial distance determined iteratively. If the model’s SLP is outside the two bounds, dry air mass is instantly “tele-ported” from(into) the inner area (red) to the outer ring (green area) Large-scale nudging towards NCEP analysis is gradually masked out (from green to the red region)

  28. Re-assimilation of hurricane Katrina into NCEP analysis

  29. SLP: Katrina 1st US landfall (~ 4 days before 2nd landfall)

  30. Katrina forecasts 2005 operational models www.weatherundergroud.com 00Z 08/25 GFDL C360 HiRam 00Z 08/26

  31. Intensity prediction: NCEP/GFS forecasts GFDL HiRam forecasts

  32. 5-day forecast initialized on 00Z Aug 26, 2005 Gulf of Mexico: Katrina Western Pacific: super Typhoons Talim and Nabi Column Water Vapor

  33. Hurricane Ike (2008) 10-day forecast: 00Z 20080906-20080915 Water vapor 10-day hurricane forecast?

  34. The GFDL HiRam is a global non-hydrostatic model capable of cloud-resolving • We have developed the prototype global cloud-resolving model. Need resources to further advance the development. The non-hydrostatic C2000 (~ 5 km) “cloud-resolving” configuration may be useful for operational hurricane predictions – we estimated that a 5-day forecast needs • ~1 hour with 100,000 CPUs.

  35. Summary: • A 25-km global model can be skillful in hurricane intensity forecasts -- expect major improvement once we reached “cloud-resolving” resolution. • 10-day track forecast for long-lived hurricanes is feasible -- prediction of the timing and the exact location of the cyclone-genesis seems to be the main challenge. • For seasonal hurricane prediction to be skillful, the model must be able to simulate a credible tropical-cyclone climatology Future works: • (Decadal) Projections of hurricane intensity/frequency changes • Seasonal hurricane predictions with coupled “ocean” + “wave” model.

  36. GFDL Seasonal (Jul-Dec) hurricane predictions 1985-2005 • 5-member C180 (~50 km) model ensemble • SST: climatology plus persistent anomaly from June of the forecast year correlation ~ 0.72

  37. C360 (~25 km) model Scaling

More Related