New GFS Model – implemented on March 2022 – Technical article


What are the improvements of the new GFS Model?

As mentioned in the title this is going to be a technical, actually only a little technical, article with several technical terms.

GFS is the acronym for Global Forecast System, that is a global numerical weather prediction system containing a global computer model and a variational analysis, run by the U.S. National Weather Service (NWS).

GFS is run 4 times a day, producing forecasts up to 16 days by the use of FV3 Model with a resolution of 13 km, vertically divided in 127 levels, in the first 120 hours outputs are made every hour, then every 3 hours.

In this article we are going to summarize pros and cons of the new GFS Model, GFSv16 implemented on 22 March 2022, with respect to GFSv15, based on the review work of the National Oceanic and Atmospheric Administration (NOAA) and, at the end, we will add the link to the official document.

Images, tables and graphs are all from the official document.

How many and what were the GFSv16’s goals?

NOAA’s review resume them in 6 points:

1) Reduce low-level cold bias during cold season

2) Improve inversion handling in the planetary boundary layer (PBL)

3) Reduce near-surface temperature biases in all seasons

4) Improve forecast of the upper atmosphere

5) Replace the operational Global Wave Deterministic model (Multi-1) with a wave component coupled to the atmosphere

6) Transition GFS evaluation material from VSDB to MET/METplus.

Major updates of GFSv16

We can sum them up to 13 points: 2 in the model resolution, 4 in the Physics, 1 in the Wave Coupling and 6 in the Data Assimilation.

Model resolution:

  • Increased vertical resolution from 64 to 127 layers and model top raised from 54 km to 80 km

Physics updates:

  • PBL/turbulence: K-EDMF => sa-TKE-EDMF
  • Orographic Gravity Wave Drag => Orographic + non-orographic GWDs
  • Radiation: Updates to cloud-overlap assumptions
  • Microphysics: Improvements to GFDL MP

Coupling to Wave:

  • One-way coupling of atmospheric model with Global Wave Model (GWM)

Major Data Assimilation Upgrades:

  • Local Ensemble Kalman Filter (LETKF)
  • 4-Dimensional Incremental Analysis Update (4DIAU)
  • Height increments are added, and specific humidity increments are reduced in the stratosphere and mesosphere
  • Improved Near Surface Sea Temperature(NSST) analysis
  • Land Data Assimilation (GLDAS) for spinning up soil moisture

GFSv16 was evaluated by NOAA and NWS through statistics, case studies and representative examples through Verification and Post-Processing teams within the Verification, Post-Processing and Product Generation Branch.

What are the strengths of GFSv16?

There are 7 points highlighted by the Model Evaluation Group:

  • Improved 500-hPa AC scores in the medium range (better with synoptic pattern)

GFSv16 had higher AC scores than GFSv15 at the majority of forecast lead times (Days 1–8)

GFSv16 had statistically significantly higher AC scores at Days 2–6

GFSv16 typically forecasted the location of cutoff lows earlier and more consistently than GFSv15

  • Some indication that GFSv15 progressive issue has been improved

GFSv16 forecasted the position of the cold front more correctly and consistently than GFSv15, even in the short range

  • Improved position of relevant frontal boundaries
  • Mitigated the low-level cold bias seen in GFSv15 during the cool season

GFSv16 has less of a cold bias at longer lead times

  • Identifies TC threats more often and at longer lead times

On average, GFSv16 is more cyclogenetic than GFSv15

  • GFSv16 forecasted a hurricane making landfall along the Gulf of Mexico days earlier than GFSv15

  • GFSv16 has lower track error than GFSv15 for strong TCs (≥65 kt)during most of the medium range in both the North Atlantic and East Pacific
  • GFSv16 forecasted Dorian to track north of Puerto Rico more than 24 h earlier than GFSv15(not shown)
  • GFSv16 forecasted Dorian to turn right and skim the Florida coast 36 h earlier than GFSv15
  • Overall, positive impact on HWRF (Hurricane Weather Research and Forecasting) forecasts
  • Improved QPF Equitable Threat Scores (ETS) and bias in the medium range
  • 24-h QPF improvements appear the most pronounced in the medium range, which is consistent w/ improved 500-hPa AC scores
  • F144: Statistically significant improvement at 0.2–35 mm thresholds
  • 24-h QPF bias improvements also the most pronounced in the medium range
  • Reduction of the high bias at lower QPF thresholds is statistically significant
  • Reduction of the low bias at medium-to-high QPF thresholds is statistically significant
  • Overall, improved snowfall location and amounts at longer lead times
  • Resolved low-level warming issue seen in a few GFSv15 cases
  • Improved ability to capture the temperature profile in shallow, cold air masses .
  • Cold air damming events have always been a major weakness of the operational GFS, likely connected to systematic problems with developing and maintaining inversions
  • Cold air damming events have always been a known strength of the NAM (especially the 3-km NAM nest)
  • There has always been some thought that higher vertical resolution in the boundary layer aids in modeling shallow, cold air masses, so the additional vertical layers in GFSv16 may help
  • Improved 2-m T forecasts in shallow, cold air masses may be tied to a better handling of low-level clouds
  • GFS Wave has lower globally-averaged RMSE and bias for Significant Wave Height

What are then the concerns of GFSv16?

They are 8 , according to the NOAA document :

  • 1) Increased right-of-track bias at longer lead times for North Atlantic Tcs
  • A slower and right-of-track bias at longer lead times suggests that GFSv16 may be recurving TCs earlier than GFSv15
  • GFSv16has a larger slow bias than GFSv15 that grows with forecast length in the N Atlantic
  • GFSv16 has a larger right-of-track bias than GFSv15 that is largest at longer lead times
  • 2) Larger TC False Alarm Ratio (FAR) in the western North Atlantic (70°W–50°W)
  • 3) Tendency to strengthen all TCs in the long range (pre-formation, not in stats)
  • 4) Degradation of HMON (Hurricanes in a Multi-scale Ocean-coupled Non-hydrostatic model) performance
  • Atlantic Basin track forecasts were degraded when initializing with GFSv16
  • EPAC track forecasts were improved, and intensity forecasts were overall improved, especially in the North Atlantic basin
  • 5) Some regional degradation of waves forecasts
  • The GFS atmospheric component has been one-way coupled to a global wave model (WAVEWATCH III), whereby GFSv16 now produces global wave forecasts for up to 16 days each cycle. With this upgrade, the current (stand-alone) Multi_1 global wave model will be decommissioned from NCEP operations
  • During the wave component of the GFSv16 science evaluation, it was determined that:

– there was consistent improvement of the winds, wave periods, and directions

– consistent degradation of significant wave heights was seen over both the North Pacific and North Atlantic

– the 95th percentile winds and seas aggregated over all buoys seemed to be in favor of Multi-1 forecasts in the North Atlantic Basin

  • A joint plan from EMC and OPC was created to identify, ameliorate, and re-evaluate the low bias in large-amplitude wave heights (Hs > 4m, 7m) in GFSv16

Waves mitigation plans are:

  • Re-optimize the physics options in WAVEWATCH III for GFSv16
  • Re-evaluate the GFSv16 low bias in large-amplitude wave heights for 5 selected cases
  • If physics optimization alone does not sufficiently improve the high-seas results, introduce additional high-resolution (4 arc-min) grids for US Atlantic and Pacific coasts

Caveats :

  • These improvements will not be part of the GFSv16 implementation and will need to be included in a potential intermediate (v16.1) upgrade
  • The introduction of additional high resolution coastal grids will lead to an increase in CPU resources needed to run GFSv16-Wave in operations. Appropriate High-Performance Computing Resource Allocation Committee approvals will need to be sought
  • 6) Exacerbation of low instability (i.e., CAPE) bias that already existed in GFSv15, driven largely by dry soil moisture
  • Instability guidance was degraded relative to GFSv15 in the heart of the retrospective warm season
  • GFSv16 surface-based CAPE magnitudes were consistently lower at all valid times and at all forecast lead times
  • Drier top-layer soil moisture initial conditions promoted over-mixing of the PBL, leading to a reduction in available low-level moisture
  • The retro runs showed a possible additional issue with occasional signs of early decoupling of the surface layer
  • Focusing on 00Z/12Z valid times since special 06Z/18Z soundings counts are limited :

  • Lower GFSv16 CAPE magnitudes were most pronounced at 00Z valid times
  • Operational GFSv15 CAPE analyses/forecasts were consistently lower than obs
  • CAPE magnitudes in GFSv16 analyses/forecasts were consistently lower than those from GFSv15
  • GFSv16 top-level soil moisture is considerably drier than in v15
  • Good alignment between lower 2-m dew points and largest areas of reduced CAPE
  • GFSv16 PBL was drier/warmer/deeper than GFSv15 and obs in the unstable air
  • The largest differences are late in the day. Is early surface decoupling playing a role in maximizing CAPE reductions at 00Z?
  • 7) Colder low-level temperature analyses, especially in the cool season
  • The analyses and very short range v16 forecasts are colder than those in v15
  • This low-level analysis cold bias appears to be primarily a cool season issue

  • 8) Lack of considerable improvement in forecasting radiation inversions
  • GFSv15 and v16 both fail to capture the strength of the low-level inversion and end up way too warm at the lowest levels
  • GFSv16 shows very modest improvement over v15
  • Observed winds are weak at the lowest level; both GFS versions have winds that are too strong

Finally, the NOAA document ends with a summary:

  • There are several notable improvements in GFSv16: substantial mitigation of the low-level cold bias at longer forecast ranges and improved medium range synoptic performance (objective AND subjective), including improved QPF scores, are especially prominent
  • Tropical performance is a bit mixed, with improved intensity, some improvement with hurricane tracks, and better probability of detection. These are somewhat offset by a larger right-of-track bias and a higher false alarm rate, as well as the tendency to over-intensify modest systems. HWRF runs with GFSv16 initialization appear to be improved, but it’s less clear for HMON
  • Low instability was already a major problem in the GFS, and it’s worse in this version in the warm season; dry soil appears to be the major factor, leading to over-mixed warm season PBLs and reduced low-level moisture
  • Replacing the deterministic Multi-1 global wave model with GFSv16 having the wave component coupled to the atmosphere overall improves global wave predictions, but the impact is mixed regionally. The loss of the 4 arc-min grids degrades performance in the coastal U.S. regions, as expected. A mitigation plan has been enacted
  • There appears to be some improvement in the handling of shallow low-level cold air masses in GFSv16, and an issue in GFSv15 with odd profiles in some precipitation type events has been resolved, but the targeted significant improvement in the handling of radiation inversions did not materialize and remains a priority target for improvement
  • Finally, low-level temperature analyses and short-range forecasts are overall notably too cold in the cool season, which has implications for downstream applications.

This was a summary of a National Weather Service article

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: