WRF · EXTREME

Flagship Technology from Meteo Centar · 2026

WRF
EXTREME

A frontier, proprietary fork of the Weather Research and Forecasting model — re-engineered from the inside out for best-in-class accuracy in complex terrain and severe weather environments.

Best in class
PBL & surface-layer accuracy
In complex terrain. Code-level modifications across every major physics layer of the WRF core.
Resolution
Tested below 1 km
Hydrometeors
7 species
PBL schemes
YSU + MYNN-EDMF
Status
Production
FLAGSHIP / 01 — TITLE
WRF-EXTREME · ATMOSPHERIC MODEL
WRF · EXTREME
02 / 09 — OVERVIEW
We don't configure. We rewrite.

Years of code-level modification to the WRF core.

WRF-EXTREME is the result of years of deep, hands-on modification of core WRF components. Pioneering modifications span every major physics layer — boundary-layer parameterizations with built-in cold air drainage, a breakthrough sigma-coordinate correction that eliminates spurious warming in valleys, full 7-species microphysics-radiation coupling, terrain-aware snow initialization, and completely rewritten 2-metre diagnostics.

Every change targets a known, specific failure mode of the standard WRF — and together they deliver a step-change in forecast accuracy for mountainous and coastal regions.

Main Modifications

01Coordinate Correction Enginecore
02Cold Air Drainage Flowscore
03Full-Spectrum Radiation Couplingcore
04Terrain-Aware Snow Initinit
05Coastal SST Decontaminationinit
06Reworked PBL Physicscore
07Enhanced T2 & SST Diagnosticscore
08Model State MonitorCORE
OVERVIEW · MODIFICATION SURFACE
STD WRF → WRF-EXTREME
WRF · EXTREME
03 / 09 — Sigma-coordinate corrections
Vertical Coordinate Correction Engine

A budget-based fix for spurious valley warming.

A breakthrough budget-based module that diagnoses and corrects the spurious warming caused by sigma-coordinate terrain-following levels over stable nocturnal boundary layers in valleys and basins — a long-standing, fundamental WRF weakness.

  • GateTopographic Position Index (TPI)basin geometry
  • GateRichardson numberstability check
  • AutoSelf-deactivates on flat terrain & high windsno-op
module_diag_coord_correction.F
! Terrain modulation
grad_term = MIN(terrain_grad_mag(i,j) &
            / grad_ref, 1.0)
basin_term = 0.0
IF (terrain_tpi(i,j) .GT. 0.0) THEN
  basin_term = MIN(terrain_tpi(i,j) &
              / tpi_ref, 1.5)
ENDIF
terrain_factor = MAX(grad_term, &
                       basin_term)

! Stability ramp
stab_factor = MIN((br(i,j)-ri_crit) &
            / (ri_strong-ri_crit), 1.0)

! Apply compensating correction
correction = -alpha * excess &
   * terrain_factor * stab_factor * safety
rthbten(i,k,j) = rthbten(i,k,j) &
                  + correction
VERTICAL COORDINATE CORRECTION
EXAMPLE SOURCE · FORTRAN
WRF · EXTREME
04 / 09 — PHYSICS
Modification surface · core physics

Every major physics layer, rewritten.

6 core enhancements
Simulation-level corrections
Not post-processing
L · 01
Coordinate Correction
Budget-based fix for sigma-coordinate spurious warming over nocturnal valleys. TPI + Ri gated.
Core / Dynamics
L · 02
Cold Air Drainage
Katabatic drainage parameterization embedded into YSU PBL scheme. A first.
Core / PBL
L · 03
Radiation Coupling
Effective radii for all 7 hydrometeors fed into Goddard radiation. Optical-depth floor + scale factors.
Core / Radiation
L · 04
Reworked PBL
Subgrid terrain drag ported to MYNN. Friction-velocity floor over snow ridges. Entrainment reworked.
Core / PBL
L · 05
T2 & SST Diagnostics
Physics-based 2-m temperature with stability mixing & smoothing. Cool-skin / warm-layer SST enhancements of the diurnal energy fluxes at the sea surface.
Core / Diagnostic
L · 06
Model State Monitor
Continuously tracks internal fields and corrects detected deviations before they propagate into large forecast errors.
Autocorrection
PHYSICS LAYERS · CORE
NOT CONFIG · CODE
WRF · EXTREME
05 / 09 — DRAINAGE & RADIATION
More PBL and radiation improvements

Capabilities no stock WRF has.

YSU drainage
Full 7-species coupling
02 / Cold Air Drainage Flows

Katabatic flow, embedded in the PBL.

Pioneering katabatic drainage parameterization embedded directly into both YSU PBL scheme. Models nocturnal cold air pooling in valleys that standard WRF completely ignores — gated by inversion strength, wind speed, surface temperature, and TPI-derived basin geometry.

Gate · 01
Inversion strength
Gate · 02
Wind speed
Gate · 03
Surface temperature
Gate · 04
TPI basin geometry
03 / Full-Spectrum Radiation Coupling

All 7 hydrometeors. One radiation field.

Extended WSM7 microphysics to compute effective radii for all 7 hydrometeor species — rain, graupel, and hail now feed directly into the Goddard radiation scheme alongside cloud water, ice, and snow. Cloud optical depth floor and pressure-level scale factors eliminate thin-cloud radiation biases.

Cloud
Ice
Snow
Rain+
Grpl+
Hail+
+1
+ NEW · NOW FEEDS RADIATION
Drastically reduces temperature biases
EMBEDDED · COUPLED
WRF · EXTREME
06 / 09 — INITIALIZATION
Snow cover reworked. Terrain-Aware Snow Init.

Glacier artifacts removed. Snow shifted to peaks.

Fixes a known problem with global model initialization: IFS/ECMWF encodes glacier ice as snow depth, and coarse-resolution snow fields miss valley-to-peak gradients. Our seasonal snow cap removes glacier artifacts hemisphere-aware, while TPI + elevation-based redistribution shifts snow from valleys to peaks — critical for surface energy balance and near-surface temperature accuracy from hour zero. No awkward preprocessing of the ECMWF data needed. WRF EXTREME handles that automagically. And improves spatial distribution of the snow cover at the same time. Yes, automagically, again.

Step 01 · Strip glacier artifacts

Hemisphere-aware seasonal snow cap

Removes glacier ice that IFS/ECMWF encodes as snow depth — before the model touches it.

Step 02 · Redistribute

TPI + elevation redistribution

Shifts snow mass from valleys to peaks based on Topographic Position Index and elevation — restoring the true valley-to-peak gradient even with low-res data inputs (like 0.25° grids from GFS or ECMWF Open Data).

Outcome

Hour-zero accuracy on T2 & surface energy

Critical for surface energy balance and near-surface temperature accuracy from the very first integration step.

INIT time improvements - Snow cover
HEMISPHERE-AWARE
WRF · EXTREME
07 / 09 — INITIALIZATION
Coastal SST Decontamination

Zero contamination, hour zero.

A dedicated pre-processor that eliminates a systematic coastal SST bias before the model even starts. When WRF's grid (eg. 1 km) is finer than the driving model (e.g. ICON-EU ~6.5 km), metgrid's bilinear interpolation bleeds land skin temperature into coastal water points — shifting the sea surface temperature by several degrees. This causes huge coastal temperature biases deep into simulation.

A KD-tree-based spatial correction detects contaminated points via the LANDSEA fraction field, replaces them with the nearest clean open-water SST, and automatically skips inland lakes. The result? Bias is gone.

WPS / Real.exe pipeline · in-place

01
ungrib · metgrid
Standard interpolation from driving model. Coastal contamination introduced here.
02
sst_decon · KD-tree
Detect via LANDSEA fraction → replace with nearest open-water SST → skip inland lakes. Runs in-place on met_em files.
03
real.exe
Initialization proceeds with clean SST field. No contamination propagates into the run.
04
wrf.exe
Forecast integration begins on a physically consistent ocean boundary.
More init time improvements - SST
KD-TREE · IN-PLACE
WRF · EXTREME
08 / 09 — RUNTIME
And, much more

Inline corrections that feed back into the model state.

Not post-processing
Continuous · self-Improving
06 / Reworked PBL Physics

Topo drag in MYNN. A first.

Deep modifications to YSU and MYNN-EDMF. Subgrid terrain drag (topo_wind) ported to MYNN for the first time. Dynamic friction velocity floor over snow-covered ridges prevents surface decoupling. Moisture-heat entrainment coupling reworked. Both schemes now share the same terrain-aware capabilities.

YSU·MYNN-EDMF
07 / T2 & SST Diagnostics

Physics-based, not post-processed.

Physics-based 2-metre temperature diagnostic with wind-speed and stability-dependent mixing, adjustable bias, and temporal smoothing to eliminate timestep-to-timestep noise. Tunable cool-skin and warm-layer SST coefficients for accurate sea surface representation. All that gives total control over how the model calculates 2m temperature. No more weird noise in 2D fields within extremely stable airmass, unavoidable from the vanilla WRF code.

T2·COOL-SKIN·WARM-LAYER
08 / Model State Monitor

Catches errors before they propagate.

A specifically developed state-monitoring module that continuously tracks the model's internal fields during integration, detects deviations from a physically consistent state, and corrects errors before they have a chance to propagate and degrade the modelled solution. Result: no unphysical model state, which means, much better forecasts.

● MONITORING·SELF-CONSISTENT, SELF-IMPROVING
Subgrid drag, T2 diags, self-improving module
RUNTIME · INLINE
WRF · EXTREME
09 / 09 — STATEMENT
Statement

Not a configuration.
A re-engineered model.

WRF-EXTREME is not a configuration — it is a fundamentally re-engineered atmospheric model built on thousands of hours of code-level modifications to the WRF core. Every patch targets a specific, documented failure mode of the standard model.

Net effect

Spurious valley warming
CORRECTED
Cold air drainage in valleys
MODELLED
7-species radiation coupling
COMPLETE
Glacier artifacts in snow init
REMOVED
Coastal SST contamination
ELIMINATED
Topo drag in MYNN
PORTED
Runtime state drift
MONITORED
PBL accuracy in complex terrain
BEST IN CLASS
END · WRF-EXTREME
FLAGSHIP TECHNOLOGY DEVELOPED AND OPERATED BY METEO CENTAR D.O.O. · 2026