REPS

1. Overview of REPS v5.0

  • Implemented on: June 11, 2024

  • Developed by: Canadian Meteorological Centre (CMC)

  • Forecast Type: Short-range (72-hour) probabilistic forecasts

  • Grid Resolution: ~10 km (rotated latitude-longitude LAM grid)

  • Vertical Levels: Hybrid vertical coordinate system with 84 staggered levels

  • Forecast Frequency: 4 times daily (00, 06, 12, 18 UTC)

  • Coverage: North America and adjacent oceans

  • Number of Ensemble Members: 21 total (1 control + 20 perturbed members)

Key Updates in Version 5.0:

  • Assimilation switch: Now uses GDPS analysis instead of RDPS.

  • Upgraded perturbation scheme: More refined SPP (Stochastic Perturbation of Physics Parameters).

  • Extended uncertainty representation: More diverse initial condition perturbations.

  • New surface initialization method: Directly obtained from GDPS surface analysis.


2. Data Assimilation & Initial Conditions

2.1 Assimilation Strategy

Unlike other deterministic models, REPS does not use direct data assimilation. Instead, it leverages GDPS and GEPS for its initial conditions.

  • Upper-Air Initial Conditions:

    • Derived from GDPS analysis (15 km resolution) interpolated onto the REPS 10 km grid.

    • Uses GEPS ensemble perturbations, modified via Homogeneous Isentropic Perturbations (HIP).

    • Recentered around GDPS mean analysis to maintain physical consistency.

  • Surface Initial Conditions:

    • Directly obtained from the GDPS surface analysis.

    • Not perturbed to ensure forecast consistency.

2.2 Perturbation Techniques

  • HIP (Homogeneous Isentropic Perturbations):

    • Introduced in the boundary layer (variance 0.8) and upper atmosphere (variance 0.5).

    • Ensures realistic spread in the ensemble initialization.

  • SPP (Stochastic Perturbation of Physics Parameters):

    • Adjusts subgrid-scale physical processes randomly.

    • Uses a higher-resolution grid (384 × 192 points vs. previous 16 × 8 points) for better representation of model uncertainty.

  • Boundary Condition Perturbations:

    • Provided by GEPS (perturbed every 1 hour).

    • Helps capture forecast divergence in high-impact weather scenarios.

  • SKEB (Stochastic Kinetic Energy Backscatter):

    • Not active in REPS 5.0 (removed due to performance limitations).


3. Forecast Model Configuration

3.1 Model & Grid Setup

  • Core Model: GEM (Global Environmental Multiscale) v5.2.0

  • Grid Setup:

    • Rotated latitude-longitude grid (1108 × 1082 points)

    • Uniform 10 km horizontal resolution

  • Time Integration:

    • Iterative-implicit semi-Lagrangian scheme (300-second timestep)

  • Lid Nesting:

    • Model top at ~17 hPa with hybrid SLEVE coordinate.

3.2 Prognostic Variables

  • Core Atmospheric Fields:

    • Wind (E-W & N-S components), temperature, surface pressure, specific humidity.

  • Cloud & Precipitation Fields:

    • Cloud condensate, rain, snow, total ice mixing ratio, turbulence (TKE).

  • Derived Forecast Fields:

    • Mean Sea Level Pressure (MSLP), precipitation rate, boundary-layer height.

3.3 Key Model Enhancements

  • Improved Stochastic Parameterization (SPP)

    • Adjusts turbulence, cloud microphysics, and convection rates.

  • Better Land-Surface Coupling

    • Snow depth and soil temperature now evolve dynamically.

  • More Realistic Convective Triggers

    • Updated Kain-Fritsch deep convection scheme.


4. Physics Parameterizations

4.1 Atmospheric Physics

  • Deep Convection: Kain & Fritsch scheme

  • Shallow Convection: Bechtold mass-flux scheme

  • Microphysics: Sundqvist scheme for cloud and precipitation processes

  • Boundary Layer Turbulence: Turbulent Kinetic Energy (TKE) model

  • Radiation Model: Li-Barker correlated k-distribution scheme

4.2 Land-Surface & Ocean Processes

  • Surface Model: ISBA (Interactions between Soil, Biosphere, and Atmosphere)

  • Sea Ice Model: Uses dynamic sea ice thickness & fraction

  • SST Initialization: Updated from GIOPS ocean analysis

  • Subgrid Orography Effects: McFarlane & Lott-Miller parameterization


5. Ensemble Forecast Interpretation & Uncertainty Quantification

5.1 Spread & Skill Representation

  • Ensemble Mean: Averaged forecast from all 21 members.

  • Probability Forecasts:

    • Probability of exceeding a temperature/precipitation threshold.

  • Extreme Member Analysis:

    • Identifies upper and lower bounds of uncertainty.

5.2 Key Forecast Applications

  • Severe Weather Prediction:

    • Hurricane tracks, snowstorms, heatwaves, thunderstorms.

  • Aviation & Transportation:

    • Wind, turbulence, and visibility forecasts.

  • Energy & Infrastructure:

    • Load balancing for power grids during extreme weather.

  • Disaster Management:

    • Probability-based flood, wildfire, and ice storm risk assessments.


6. Summary & Key Advancements in REPS v5.0

  • Higher horizontal resolution (~10 km) for better regional forecasts.

  • More advanced ensemble initialization techniques (HIP & SPP).

  • Improved deep and shallow convection schemes.

  • Enhanced coupling with land, ocean, and ice models.

  • More efficient uncertainty quantification for high-impact weather events.


7. References & Further Reading


8. Conclusion

The Regional Ensemble Prediction System (REPS) v5.0 is a state-of-the-art short-range probabilistic forecasting system, enhancing uncertainty representation and regional weather prediction accuracy. Its high-resolution forecasts, advanced physics, and improved ensemble perturbation schemes make it an essential tool for meteorologists, disaster planners, and climate resilience initiatives.

Last updated

Was this helpful?