Log of Changes to Pacific Northwest WRF Ensemble Forecasts Forecasts

Ensemble Member Descriptions | Return to Main Ensemble Page | Return to Main WRF Page

8/4/2026

The MPAS ensemble member (which does not have the same set of graphics,
due to its completely different grid structure) has limited graphics
available here.
Beginning with the 2026080500 run, the MPAS ensemble members will have
initial and boundary conditions provided by our WRF-GFS 12-km
real-time runs.

6/22/2026

Beginning with the 2026062300 run, the nofdda member will now more closely
match the physics options of our test mpas runs.

3/5/2026

Beginning with the 2026030600 run, the default graphics size has been
increased from 900x900 to 2025x2025.

3/4/2026

After seeing some statistics, now trying a few things:
1) turning off spec_bdy_final_mu option for all ensemble members
2) for all GEFS, setting use_theta_m=0 -- use dry theta in dynamics
3) for all GEFS, setting use_mp_re=0 -- do not use effective radii from MP in RRTMG
All to take effect with 2026030500 runs and forward.

2/26/2026

The ngps ensemble member (from the FNMOC NAVGEM model) has been
discontinued since reliable access to the grids has been lost.
Another GEFS ensemble member has taken its place.

1/13/2026

Beginning with the 2026011400 runs, the ecmwf and ecaifs members are replacing the
gefs09 and gefs10 ensemble members.

1/12/2026

Beginning with the 2026011212 run, we have begun running a new ensemble member:
ecaifs - initial and boundary conditions from European Centre for Medium-Range Weather Forecasts
AIFS Machine Learning Open Data. This is a standard WRF model run but the
initial and boundary conditions come from an AI version of the ECMWF's IFS model.
The AIFS is a robust model, but has a few drawbacks compared to the IFS:
1) it is output only every 6 hours instead of every 3 hours,
2) it only has 2 soil layers instead of 4.

1/9/2026

Beginning with the 2026010912 runs have added ecwmf ensemble member.
ecmwf - initial and boundary conditions from European Centre for Medium-Range Weather Forecasts
IFS Model Open Data

2/13/2023

Beginning with the 2023021312 run, 6 new stochastic ensemble members
have been added: stoc01 - stoc06.
Members stoc01-stoc03 use stochastic kinetic energy backscatter
techniques along with the same phyiscs setup as the nofdda run.
Members stoc04-stoc06 use stochastic kinetic energy backscatter
techniques along with the same phyiscs setup as the hrrr run.
For full details refer to the full descriptions.

2/10/2023

Beginning with the 2023021000 run, the 75-vertical level ensemble
member has been permanently dropped. It was computationally
expensive, occasionally numerically unstable, and did not
verify any better than the standard 38-level run.
Beginning with the 2023021000 run, we have added the nofdda
member which is identical to the wrfrt (WRF-GFS real-time) run
except that there is no FDDA grid nudging on the 36-km domain.

5/3/2022

Beginning with the 2022050300 run, the ensemble calibration
machine-learning training data has been reset to 2022010512 in order
to incorporate more of the new ensemble members. With this reset,
only the tsno member has insufficient data availability, though this
is expected to change within a few months.

3/17/2022

Beginning with the 2022031712 run, the hrrr ensemble member is now using
the RUC land surface model instead of the Noah LSM. This means the
hrrr member now more closely follows the official HRRR model runs.
Our hrrr namelist file is here and
can be compared to the official HRRR here.

3/4/2022

Beginning with the 2022030500 run, returned to using the Noah-MP LSM
as the default for most ensemble runs.

2/23/2022

Beginning with the 2022022312 run, added a TSNO ensemble member experimentally
which was tuned to improve temperature bias over snow. This member
has these characteristics:
mp_physics = 28 Thompson aerosol-aware microphysics
ra_lw_physics = 4 RRTMG longwave radiation
ra_sw_physics = 2 Goddard shortwave radiation
sf_sfclay_physics = 91 old MM5 surface-layer scheme
sf_surface_physics = 4 Noah-MP Land Surface Model
bl_pbl_physics = 5 MYNN 2.5 level TKE PBL
cu_physics = 3 Grell-Freitas ensemble cumulus scheme
ishallow = 0 no shallow convection for Grell 3D ensemble scheme

1/5/2022

Beginning with the 2022010512 run, the WRF4MYNN ensemble member has
been updated to use the CONUS physics suite
mp_physics = 8,
cu_physics = 6,
ra_lw_physics = 4,
ra_sw_physics = 4,
bl_pbl_physics = 2,
sf_sfclay_physics = 2,
sf_surface_physics = 2,

which results in the following changes:

scheme old new
------- ----------------------------- --------------------------------
cumulus Grell-Freitas ensemble scheme Modifed Tiedtke
PBL MYNN 2.5 level TKE scheme Mellor-Yamada-Janjic TKE scheme
sfc layer MYNN Monin-Obukhov (Janjic) scheme
LSM same as real-time Unified Noah land-surface model
(Noah or Noah-MP)

10/7/2021

Beginning with the 2021100800 run, all ensemble runs are no longer
using aerosol optical depth and the aerosol angstrom exponent fields from
NASA's GEOS system.
These were turned off about 10 days earlier than normal due to
recent problems acquiring the GEOS data in a timely manner.
They are scheduled to be turned back on for the July 1, 2022 run.

8/7/2021

Beginning with the 2021080700 run, we are now running RCON -- a WRF-GFS
ensemble member that uses the Robert Conrick warm-rain modifications
for the Thompson microphysics schemes and also has mp_physics set to
28 which has climatological values of aerosols. This configuration
has been shown to increase precipitation on the windward slopes of
Pacific Northwest mountain ranges, decrease the rain over the crest
and leeward slopes, and improve the rain in the Puget Sound region
between the Olympic and Cascade ranges.

7/27/2021

Beginning with the 2021072712 run, the German Weather Bureau (Deutscher
Wetterdienst) ICON model is added as the dwdi ensemble member.

7/1/2021

Beginning with the 2021070100 run, all WRF ensembles are using
aerosol optical depth and the aerosol angstrom exponent fields from
NASA's GEOS system.
Three-hourly values of these parameters will be used in WRF to
interact with the radiation scheme in an effort to incorporate the
temperature effects of smoke. See this page for more information.
These aerosol effects are scheduled to be turned back off with the 2021100100
run though they may be turned off earlier if smoke conditions over
the Pacific Northwest improve substantially.

1/26/2021

Upgraded our disk storage, preprocessing, and postprocessing computer.

7/7/2021

Dropped the wrf4acm2 ensemble member since its verification scores
were so bad. Its last case was the 2021070712 forecast run.

12/11/2020

Beginning with the 2020121112 run we are now splitting the ensembles
into a 0-36 hour run and a 36-72 hour run. This delays the final
products by about 30 minutes, but does now give access to plumes and
mean and spread graphics (to hour 36) about 2 hours earlier.

11/19/2020

Beginning with the 2020112000 run, fixed a bug in the wrf4acm2
initialization that was causing bad sea-level pressures in the first
3-4 hours of the simulation.

11/18/2020

Beginning with the 2020111900 run the hrrr ensemble member more
closely matches HRRRv4 with the continued exception of having to use
the Noah LSM instead of the RUC LSM since we do not have 72 hours of
9-layer soil data available. Changes include mp_physics=28
(aerosol-aware Thompson scheme), no grid nuding on 36-km domain, and
diff_opt=1 instead of the diff_opt=2 that we use elsewhere.

10/30/2020

After continued CFL errors in with and without adaptive time step
options for the wrfv75 physics member for these cases:
2020100812, 2020101000, 2020101012, 2020101112, 2020101200,
2020101300, 2020101400, 2020101500, 2020101712, 2020102000,
2020102100, 2020102200, 2020102312, 2020102500, 2020102612,
2020102700, 2020102900, 2020102912, 2020103000
(19 failures out of the last 60 runs, 69% success rate)
we have switched to using auto-generated vertical levels. The
namelist.input settings and level heights and thicknesses can be
found here.

10/14/2020

After CFL errors in wrfv75 physics member for cases:
2020100812, 2020101000, 2020101012, 2020101112, 2020101200,
2020101300, 2020101400
we have switched to using an adaptive time step for the wrfv75 member.

8/27/2020

Have activated smoke field initialization for all ensemble members with
the 2020082800 run. See this page for more information.

8/20/2020

Smoke field initialization for all physics ensemble members (hrrr,
gfswrf3, wrf4acm2, wrf4drag, wrf4p3, wrfv75) began with
the 2020082100 run. See this page for more information.

4/10/2020

Australian GASP model has been updated to higher resolution. First
model run to use the new grids is the 2020041000 run.

3/11/2020

Problems with the 2020031100 runs:
wrfv75 - CFL errors near forecast hour 24, case abandoned

Problems with the 2020031112 runs:
wrfv75 - CFL errors near forecast hour 9, case abandoned

1/28/2020

Started using WRF version 4.1.3 as default with the 2020012812 case.
Looking to see if the Noah-MP has improved cold bias over snow.

11/6/2019

Added WRF4ACM2 physics member with the 2019110612 case. This member
uses WRF version 4 with the Pleim-Xiu land surface model and surface
layer along with the Pleim planetary boundary layer scheme.

11/1/2019

Changed CMCG grid preparation and retrieval to use specific humidity
field and to process grids in GRIB2 format in an effort to eliminate
the cold temperature bias of this member. Changes to take effect
with the 2019110200 forecast.

10/14/2019

Beginning with the 2019101412 run, we have added the wrf4p3 ensemble
member. This uses WRF 4.0.3 and the new P3 microphysics scheme
(option 51) which includes one-category ice plus double-moment
cloud water.

5/26/2019

Missed gefs02, gefs03, gefs04 ensemble runs for 2019052600 and
2019052612 due to CFL problems with 2019052600 WRFV75 case that
caused the WRF code to not exit properly.

5/24/2019

Beginning with 2019052400 have added the Météo-France 0.5°: Action de
Recherche Petite Échelle Grande Échelle (ARPEGE) Modèle Atmosphérique
Global

5/17/2019

Added WRF4DRAG ensemble member with the 2019051700 case. This is
identical to GFSWRF4 except with the topo_wind=2 drag
parameterization turned on. The drag is applied as a scaling factor
on ustar over land points. The scaling ranges from 1.175 to 1.575 based
on the terrain variance. This plot shows
the scaling factor labeled as "ustar71". Here are plots of terrain
variance over the 36-km domain and 12-km domain.

5/12/2019

Missed cmcg, cent, gefs05-gefs08, and tcwb ensemble members for
2019051200 due to computer problems.

5/10/2019

Note that there are problems running gfswrf4 on multiple machines.
Could not run 2019050700 case on 4x24 cores (it crashes at hour
10.125) and on 24 cores it crashes just past hour 10.5.

5/9/2019

Beginning with the 2019051000 run, will change the HRRR PHYS ensemble
member to be using the Noah LSM instead of the RUC. The RUC was
producing too much moisture flux at the surface and resulting in
unrealistically cold daytime high temperatures.

5/1/2019

Beginning with the 2019050100 run, added 3 new members:
GFSWRF4 - same physics and initialization as real-time WRF-GFS, but
using WRF version 4.0.3
HRRR PHYS - WRF version 4.0.3 with HRRR physics: MYNN PBL, RUC LSM
and surface layer
WRFV75 - same physics and initialization as real-time WRF-GFS, but
using 75 vertical levels instead of 37.

4/30/2019

Beginning with the 2019043000 run:
1) the wind direction and wind speed means displayed in the plumes
are now vector means, not scalar means.
2) the mean displayed in the plumes now includes the control
(WRF-GFS) run. The horizontal, mapped plots (not plumes) continue
to compute means from just the ensemble members, however.

4/1/2019

Forecasts missed at the end of March:
2019033012 - missed JMA
2019033100 - missed JMA
2019033112 - missed JMA
2019040100 - missed JMA and TCWB

3/11/2019

Beginning with the 2019031112 forecast, have updated ngps retrieval
method and preprocessing.

3/5/2019

Forecasts missed at beginning of March:
2019030200 - missed CMC, GASP, JMA, TCWB, NGPS, and UKMO
2019030212 - missed CMC, GASP, JMA, TCWB, NGPS, and UKMO
2019030300 - missed CMC, GASP, JMA, TCWB, NGPS, and UKMO
2019030312 - missed CMC, GASP, JMA, TCWB, NGPS, and UKMO
2019030400 - missed GEFS01-GEFS10 (attempting to reproduce this)
2019030412 - missed GEFS01-GEFS10

11/26/2018

Beginning with the 2018112612 forecast, added a 12Z cycle.

10/31/2017

Began 0Z daily runs at 4-km resolution for 10 GEFS members, plus
CMC, GASP, JMA, NGPS, TCWB, UKMO.

Back to UW WRF Home Page