Skip to content
SHiELD: System for High-resolution prediction on Earth-to-Local Domains

SHiELD is a Unified Forecast System (UFS) prototype atmosphere model showing the power of a unified prediction system across a variety of time and space scales designed for a wide array of applications. It shows the abilities of the Finite-Volume Cubed-Sphere Dynamical Core (FV3), especially its flexible nonhydrostatic dynamics, variable-resolution capabilities, and integrated physics, coupled with the elegance of the Flexible Modeling System (FMS) framework. Improvements in FV3 or the physics in SHiELD can be easily transferred into other FV3-based models, including the GFDL Modeling Suite and the Unified Forecast System.

SHiELD is an open-source model with code and drivers available on GitHub. A container, SHiELD-in-a-Box, is also available for convenient portable usage of SHiELD.

(Roman case indicates near real-time Tier 1 configurations; Italics indicate preliminary Tier 2 configurations not run in real-time)
Configuration Use Domain Integration Length
SHiELD Medium-range weather prediction Global 13-km with 91 vertical levels 240 hours (10 days)
C-SHiELD Short-to-medium-range contiguous US severe weather prediction Global 13-km, 3-km CONUS nest, 63 vertical levels 126 hours (5.25 days)
T-SHiELD Medium-range Atlantic tropical cyclone and hurricane prediction Global 13-km, 3.5-km tropical Atlantic nest, 63 vertical levels 168 hours (7 days), hurricane season only
S-SHiELD Subseasonal-to-seasonal prediction Global 25-km with 91 vertical levels 30 to 90 days, with up to 10 ensemble members
X-SHiELD eXperimental DYAMOND-class global storm-resolving model (GSRM) Global 3.25-km with 79 vertical levels 40 days
R-SHiELD Regional limited-area model (LAM) configuration Regional 13, 3, or 1 km 54 hours
Tele-SHiELD Short-to-medium-range severe weather prediction, urban-scale impacts, coastal weather and hydrology, sub-grid scale land-use variability Global 13-km, 4.3-km first nest, 1.4-km second nest for Northeast Corridor areas, 91 vertical levels 240 hours (10 days)

 

In-Depth Information for Near Real-Time Configurations

 

2022 SHiELD Global Prediction System

13-km SHiELD is a global medium-range prediction model that acts as the “core” for SHiELD development and the base for the other configurations. We develop this model to demonstrate fundamental advances in FV3 algorithms, model initialization, and in SHiELD’s physical parameterizations. It also is a powerful scientific tool for our research on global variability and predictability, especially for tropical cyclones, on surface-atmosphere coupling, and on cloud-radiation interactions and how they affect prediction skill. SHiELD can also be run with FV3’s stretched and nested grids to locally zoom in on a region of interest.

SHiELD 2022 real-time: https://shield.gfdl.noaa.gov/?model=SHiELD_2022


Updated: September, 2022
Lead: Linjiong Zhou
Operator: Matt Morin
Contributors: Lucas Harris, Kun Gao, Kai-Yuan Cheng, and the GFDL FV3 Team
Forecast Experiment Design:

  • Initialized 00Z and 12Z daily from operational GFS analyses
  • Integrated to 10 days
  • Horizontal Resolution: 13-km global domain (C768)
  • Vertical Resolution: 91 layers, lowest mid-level at ~12 m, model top at 40 Pa
  • Computational Performance: 80 min/10 days using 3072 cores on Gaea C4

FV3 Dynamical Core:

  • Nonhydrostatic Lagrangian dynamics
  • GFDL cloud microphysics v3

SHiELD (GFS) Physics Package:

  • Scale-aware TKE-EDMF planetary boundary layer scheme
  • Scale-aware SAS (shallow and deep) convection scheme
  • RRTM radiation with GFDL cloud-radiation interaction scheme
  • GFS (orographic and convective) gravity wave drag scheme

SHiELD Ocean Component:

  • 1-D mixed-layer ocean model
  • Improved thermal roughness length parameterization under high winds

SHiELD (GFS) Land Component:

  • Noah land surface model

References:

2023 C-SHiELD Global-to-Regional CONUS Prediction System

C-SHiELD is our Contiguous United States nested model used for prediction of high-impact weather events, especially severe thunderstorms and heavy precipitation. This model is capable of simulating individual convective storms without a parameterization, thereby giving better information on the impacts of these events. The global-nested configuration of C-SHiELD allows it to make useful predictions much farther in advance than do limited-area prediction models typically used for convective-scale prediction, allowing for unique medium-range prediction of these events. C-SHiELD is contributed in real-time to the Hazardous Weather Testbed’s Spring Forecasting Experiment every May. We also use a lower-resolution version of C-SHiELD to explore subseasonal prediction of severe thunderstorm outbreaks.

C-SHiELD 2023 real-time: https://shield.gfdl.noaa.gov/?model=C-SHiELD_2023


Updated: May 30, 2023
Lead: Kai-Yuan Cheng
Operator: Matt Morin
Contributors: Linjiong Zhou, Lucas Harris, and the GFDL FV3 Team
Forecast Experiment Design:

  • Initialized 00Z and 12Z daily from operational GFS analyses
  • Integrated to 5.25 days
  • Horizontal Resolution: 13-km global domain (C768), 3-km CONUS nested domain (2160 x 1200 grid cells)
  • Vertical Resolution: 63 layers, lowest mid-level at ~15 m, model top at 64 Pa
  • Computational Performance: 118 min/5.25 days using 3024 cores on Gaea C5

FV3 Dynamical Core:

  • Positive-definite tracer advection
  • Virtually-inviscid advection scheme (hord=5)
  • Two-way grid nesting with simplified two-way update
  • GFDL Inline Cloud Microphysics version 3 with hail on nest
  • Saturation Adjustment Delay (saturation adjustment is done only at the final remapping timestep)

SHiELD (GFS) Physics Package:

  • NCEP 2017 Scale-Aware SAS (shallow and deep) convection tuning to improve convection structure and propagation
  • TKE-EDMF Planetary Boundary Layer with GFDL 2021 updates
  • RRTM Radiation with bugfix and 2021 GFDL cloud-radiation interaction
  • GFS (orographic and convective) Gravity Wave Drag with GFDL 2018 updates

SHiELD Ocean Component:

  • 2019 Mixed-layer Ocean Model

SHiELD (GFS) Land Component:

  • Noah Land Surface Model with GFDL 2020 updates on global domain and Noah-Multiparameterization Land Surface Model with GFDL 2021 updates on nest domain

References:

2024 T-SHiELD Global-to-Regional Tropical Nested Prediction System

T-SHiELD is a nested model for Tropical prediction, especially tropical cyclones. Unique among tropical cyclone models, T-SHiELD uses a 3.25 km nest over the entire tropical North Atlantic. This large convective-scale domain allows us to explore the effects of explicitly-simulated convection on tropical cyclogenesis and synoptic-scale systems, which in turn has a significant impact on tropical cyclone development and track. It can also be configured to run over other ocean basins, and is also useful for predicting strong extratropical storms in the North Atlantic in which intense winds and precipitation may occur over small-scale areas. T-SHiELD is contributed in real-time every year in late summer/early fall to the HFIP Real-time Experiment (HREx).

T-SHiELD 2024 real-time: https://shield.gfdl.noaa.gov/?model=T-SHiELD_2024


Updated: July 26, 2024
Lead: Kun Gao
Operator: Matt Morin
Contributors: GFDL FV3 Team
Forecast Experiment Design:

  • Initialized every six hours (00, 06, 12, and 18Z) from operational GFS analyses
  • TC vortex initialization procedure applied for strong storms (Vmax > 30m/s)
  • Integrated to 6 days
  • Horizontal Resolution: 13 km uniform global domain (C768r10), 3.25 km North Atlantic nested domain (2304 x 1152 grid cells)
  • Vertical Resolution: Coarse grid 63 layers, lowest mid-level at ~15 m, model top at 64 Pa; Vertically-nested grid 75 layers, lowest mid-level at ~10 m, model top at 200 Pa
  • Computational Performance: 250 min/6 days using 3072 cores on Gaea C5

FV3 Dynamical Core:

  • L75 vertical level configuration with enhanced near-surface resolution
  • Hord 6 for horizontal advection of dynamical quantities
  • Hord -5 (positive-definite) for tracers

SHiELD (GFS) Physics Package:

  • TKE-EDMF PBL with shear-dependent vertical diffusion
  • Refined mixing length parameterization in the nested region
  • Refined surface enthalpy flux parameterization in the nested region
  • No deep convection parameterization in the nested region

SHiELD Ocean Component:

  • Mixed-layer Ocean Model
  • Improved thermal roughness length parameterization under high winds

SHiELD (GFS) Land Component:

  • High-resolution Noah Land Model with GFDL updates

References:

 

References for Experimental Configurations

 

2020 S-SHiELD Subseasonal-to-Seasonal Prediction System

S-SHiELD is a lower-resolution configuration of SHiELD intended for Subseasonal-to-Seasonal (S2S) simulation, intended for extended-range prediction with a focus on extreme weather events. S-SHiELD is currently under development. We are also working closely with the developers of SPEAR on S2S product development.

References:

2021 GFDL X-SHiELD eXperimental GSRM

X-SHiELD is an eXperimental Global Storm Resolving Model (GSRM) or kilometer-scale (K-scale) model run at a sufficiently high global resolution to explicitly simulate individual convection cells. X-SHiELD is GFDL’s contribution to the DYAMOND project, an international intercomparison of GSRMs. We are also collaborating with AI2 Climate Modeling to accelerate X-SHiELD using emerging computing platforms, and to train a Machine Learning system on the output to create an efficient emulator of the expensive GSRM.

Animations of a previous version of X-SHiELD can be seen on the Mesoscale Dynamics Data Visualizations page.


Lead: Lucas Harris
Contributors: Linjiong Zhou, Kai-Yuan Cheng, Kun Gao, and the GFDL FV3 Team
Special thanks to Daniel Klocke (MPI)

Our submission of X-SHiELD output to DYAMOND  is available at ftp://data1.gfdl.noaa.gov/4/SHiELD/202011/20200120.00Z.C3072.L79x2/history/. (Please use an up-to-date FTP client with passive FTP mode for access; a web browser, Microsoft Explorer, or macOS Finder will not work.) A guide to the variables can be found at the  DYAMOND data guide  .

PIRE data is available through NOAA Open Data Dissemination (NODD) on Google Cloud.

Forecast Experiment Design:

  • Initialized 00Z on 20 January 2020 (DYAMOND) and 20 October 2019 (PIRE)
  • Integrated to 40 days (DYAMOND) and 456 days (PIRE)
  • Horizontal Resolution: 3.25-km global domain (C3072)
  • Vertical Resolution: 79 layers, lowest mid-level at ~10 m, model top at 3 hPa
  • Computational Performance: 78 min/day (standard diagnostics) and 93 min/day (enhanced 3D diagnostics) with 13824 cores on Gaea C4

2021 FV3 Dynamical Core:

  • Positive-definite Tracer Advection
  • GFDL 2021 Inline Cloud Microphysics
  • “Effectively-inviscid” dynamics (unlimited PPM advection for dynamical quantities, eighth-order divergence damping)

2021 SHiELD (GFS) Physics Package:

  • TKE-EDMF Planetary Boundary Layer with GFDL 2021 updates and GFSv16 stratocumulus enhancements
  • NCEP 2017 Scale-Aware SAS (shallow only) convection; no deep convection
  • RRTM Radiation with bugfix and 2021 GFDL cloud-radiation interaction
  • Orographic GFS Gravity Wave Drag with GFDL 2018 updates

SHiELD Ocean Component:

  • 2019 Mixed-layer Ocean Model nudged to EC analyzed SSTs

SHiELD (GFS) Land Component:

  • Noah-MP Land Surface Model with GFDL 2021 updates and EMC high-resolution surface inputs

References:

2020 R-SHiELD Regional Domain

R-SHiELD is a developmental system for applications of SHiELD on non-global domains. The Environmental Modeling Center of the National Centers for Environmental Prediction (NCEP) has implemented a Limited-Area Modeling (LAM) capability within FV3. This capability permits efficient short-term, high-resolution simulation for problems in which the overhead of a global model is unnecessary or unwanted, including the regional variant of the Hurricane Analysis and Forecast System and the Rapid-Refresh Forecast System. A doubly-periodic domain is also available in FV3 for idealized simulation.

References:

2022 Tele-SHiELD Northeast Corridor Domain

Tele-SHiELD is a developmental system for high-resolution medium-range severe weather prediction with a focus on the major metropolitan areas along the Northeast Corridor.  The use of telescoping nest function achieves 1.4-km high resolution in the inner-most domain for us to investigate urban-scale impacts, coastal weather and hydrology, and sub-grid scale land-use variability in the model.