SACRUS
-
Posts
14,635 -
Joined
-
Last visited
Content Type
Profiles
Blogs
Forums
American Weather
Media Demo
Store
Gallery
Posts posted by SACRUS
-
-
1 hour ago, SACRUS said:
1/24 00Z
NYC
QPF / Snow (Frz)
SREF (mean): 1.3 / 9.7
NAM: 1.2 / 4.7
ICON: 1.4 / 8.1
RGEM: 1.2 / 9.5
Updated
-
RGEM

Total QPF storm

Total snow (10:1)

-
-
-
-
-
1/24 00z RGEM


-
1
-
-
-
Cold push
-
1
-
1
-
-
The soon to replace NAM (q2?) RRFS - Rapid Refresh Forecast System
1/24 00z


-
14 minutes ago, SACRUS said:
1/24 00Z
NYC
QPF / Snow (Frz)
SREF (mean): 1.3 / 9.7
NAM: 1.2 / 4.7
ICON: 1.4 / 8.1
Updated -
1/24 00z ICON total storm QPF

Total snow / Frz (10:1)

-
1
-
-
-
-
1/24 00z ICON run
v


-
1/24 00Z
NYC
QPF / Snow (Frz)
SREF (mean): 1.3 / 9.7
NAM: 1.2 / 4.7
ICON: 1.4 / 8.1
RGEM: 1.2 / 9.5
GFS: 1/3 / 11.1
GFS AI AIGFS: 1.1 / 9.8
GEFS: 1.5 / 10.3
UKMET: 0.9 / 7.4
GGEM: 1.3 / 9.1
Euro : 1.2 / 10.2
Euro AI AIFS: 1,2 / 11.0
-
1/24 00z NAM total QPF storm
Total snow / sleet (10:1)

-
1
-
-
-
3 minutes ago, EW9616 said:
I’ve seen some TV Mets referring and showing the Graf model that shows little mixing. Anyone know what model that is? Is correlated with one of the main US model?
.GRAF refers to an AI-based global weather forecasting model developed by NOAA (GSL) and collaborators, and the name stands for:
GRAF = Global Regression AI Forecast model
It’s part of the new generation of machine-learning weather models, similar in spirit to GraphCast or Pangu, but developed inside the NOAA ecosystem.
What is the GRAF model?
GRAF is a pure machine-learning global forecast model that:
-
Learns atmospheric evolution from historical reanalysis data
-
Produces global forecasts without explicitly solving physical equations
-
Uses regression-based deep learning to predict future atmospheric states
Think of it as:
“AI learning how the atmosphere usually evolves, then extrapolating forward.”
Key characteristics
-
Global coverage
-
AI / ML-based (no traditional physics core)
-
Predicts large-scale fields (e.g., 500 mb heights, winds, temps, MSLP)
-
Extremely fast compared to physics models
-
Designed primarily for pattern and flow prediction
What GRAF is good at
-
Large-scale synoptic pattern recognition
-
Jet stream placement
-
Ridge / trough evolution
-
Medium-range guidance (Days 3–7)
This makes it useful for:
-
Pattern forecasting
-
Ensemble support
-
Early signal detection
What GRAF is NOT good at
-
Precipitation amounts
-
Snowfall totals
-
Precipitation type
-
Mesoscale features (banding, fronts)
-
Boundary-layer processes
-
1
-
-
-
-
-
1/24 00z NAM


-





















Extreme Cold, Snow & Sleet: SECS 1/25 - 1/26
in New York City Metro
Posted
1/24 00z GFS total QPF storm
Total Snow (10:1)
