Andrew Brady
@Brady_Wx
Followers
3,248
Following
483
Media
1,058
Statuses
3,929
Innovating machine-learning WX prediction @AtmoAnalytics | @CFB_Predictor | Kennesaw State & Miss St | WX, CFB, ML/AI, Crypto, free speech, and storm chasing.
United States
Joined August 2021
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Iran
• 379200 Tweets
Hurricane Helene
• 58276 Tweets
Saka
• 42078 Tweets
#GSvPAOK
• 41524 Tweets
Nwaneri
• 34201 Tweets
Twente
• 33137 Tweets
UNLV
• 29461 Tweets
Sterling
• 28923 Tweets
Bolton
• 28362 Tweets
Coke
• 28279 Tweets
Getafe
• 27840 Tweets
Ubisoft
• 26290 Tweets
Old Trafford
• 26219 Tweets
Porto
• 21779 Tweets
Epigmenio
• 21277 Tweets
Osimhen
• 19807 Tweets
Lewandowski
• 19464 Tweets
Kaizer Chiefs
• 17938 Tweets
Eriksen
• 17708 Tweets
#السوبر_مع_فايق
• 15718 Tweets
Happy Birthday Mark
• 12818 Tweets
Chiesa
• 11875 Tweets
Maguire
• 10957 Tweets
Pinned Tweet
Excited to announce STORM-Net: Severe and Tornado Real-time Monitoring Network.
STORM-Net is a rapidly-updating state-of-the-art deep learning model that predicts tornadoes, significant tornadoes, damaging severe storm winds, large hail, and flash flooding at high accuracy up to
18
35
234
STORM-Net 1hr tornado probabilities on the somewhat unexpected tornado in southern Oklahoma this evening.
Sniffed it out pretty well
6
30
288
4
0
231
I’m gonna say it.. Tuesday looks more robust than yesterday *synoptically*. This doesn’t necessarily mean it’ll be worse, but the potential is absolutely there for Tuesday to be historic.
#wxtwitter
11
42
234
Here's STORM-Net tornado outputs 1 hour and 45 minutes PRIOR to the PDS tornado warning being issued for the
#wiwx
supercell this evening.
The first was 10min prior to ANY tornado warning being issued for that cluster... and well before any rotation.
Green outline in STORM-Net
12
34
226
Bill Bunting? Deputy director of SPC?
Interesting to see that he spearheaded today’s 1630z outlook as opposed to a lead or mesoscale forecaster.
10
11
218
My son, Dion Marvin Brady, entered the world last night at 10:31pm! 01/29/24 10:31pm! Our little blessing. ❤️
29
1
218
Now this is absolutely wild... STORM-Net nearly 3 hours prior to Battle Creek, MI area tornado... forecast issued at 03:10UTC
4
29
212
@JimVorel
@rawsalerts
Texas law states that being the victim of a robbery or attempted robbery is grounds for lethal force.
9
0
197
Introducing HurricaneVision: The latest deep-learning model from
@AtmoAnalytics
that predicts tropical cyclone track and intensity in the Atlantic basin. HurricaneVision updates hourly and produces probability forecasts for tropical cyclone intensity (threshold percentages) as
14
37
194
If you are a TV meteorologist and are interested in using STORM-Net in your MAX storm software, DM me!
We are allowing trial access during this testing period for TV stations. You will be able to directly input a live KML URL into your MAX graphics and it will display live
22
21
191
12z HazCast tornado probabilities valid Friday afternoon/evening
4
32
195
STORM-Net prediction initialized 20 minutes before the EF-4 Keota, IA tornado touched down
Also correctly predicted the LIVINGSTON county IL tornado, NE AR tornado, and POWESHIEK county IA tornado.. all three of which formed during the hour after this initialization.
7
20
185
@DoorlagNate
@JackPosobiec
They’re beating him to death after he’s in handcuffs and not resisting. Are you blind?
3
0
165
Here’s a radar loop with STORM-Net contours overlayed from yesterday’s tornadic Wisconsin storms.
8
31
165
18z HazCast v5 tornado probabilities for tomorrow evening (18z thru 00z and 00z thru 06z).
9
32
158
STORM-Net is accessible via interactive web interface during our testing phase! You are now able to zoom down to street-level on the STORM-Net model outputs.
Follow along today and Friday, as data will be updating every 5 minutes during possible severe wx!
The interactive
7
22
162
@GAFollowers
No way. You can’t put them over Texas or Alabama. You can’t put Alabama over Texas. You can’t really put either over Michigan.
15
4
155
Here's STORM-Net 1hr tornado probabilities 30 minutes leading up to the Hot Springs Village, AR EF2 tornado.
Note: all of these frames were prior to any tornado warning being issued for the storm or for Hot Springs Village.
7
21
153
STORM-Net’s UI continues to be hit by a DDoS attack. We have been able to mitigate it to this point, and we have traced it to a company/start-up that seems to be working on a similar model that uses deep-learning to detect tornadoes.
Not cool!
24
11
151
Interesting note about STORM-Net: it has seemingly learned that tornadogenesis can occur when storms cross perpendicular to rivers. This is likely part of the reason for the big false positive in Memphis yesterday.
I believe
@ReedTimmerAccu
has noticed something similar with
10
15
150
8
25
150
@NateSilver538
People just don’t want to acknowledge that conservatives were right on this issue. People don’t want to acknowledge that this blows a massive hole in the credibility of characterizing content as “misinformation”.
29
4
137
@jimmykimmel
Well, there’s livestock versions of essentially every human drug.. so any, I guess. Maybe start with a NSAID like Advil for inflammation?
7
0
129
I asked GPT-4 to produce a chart of weather models with their biases and a general performance score (0-10). Do you agree with GPT's table?
#wxtwitter
10
18
138
STORM-Net and radar leading up to the Lincoln, NE tornado this afternoon.
Can see that it started highlighting the segment before it was even supercellular!
2
17
136
Latest HurricaneVision forecast w/ 12z spaghettis overlayed.
I’ve toggled on specific models that I’d like to see, as you can see on the legend.
This is a part of WxToolkit — a new weather data interface.
It’s free to the public, link is below.
2
31
135
00z MMFS-AI TornadoVision 1km tornado prediction model, valid at 21z, 22z Monday and 03z Tuesday. These probabilities are quite high! Followed by the raw 0-3km updraft helicity output. Full data is available for free on our site.
#wxtwitter
#lawx
#mswx
7
23
127
Here’s a statistically perfect SPC outlook based on tornado reports from yesterday.. a hindcast. This is from NIU.
7
5
128
Last 90 minutes of STORM-Net tornado probabilities leading up to (and during) the northern Texas PDS tornado
2
26
125
Seems that this verified nicely with 40+ minute lead time
STORM-NET showing high probability for a tornado in this area shortly. There are two tornado warnings in place.
@Brady_Wx
wanted to verify: the green outline signifies STORM-NET indicating SIG potential right?
4
7
53
10
12
122
Some additional updates on this:
-STORM-Net will now update every 4-5 minutes instead of every 12-15 minutes
-There will be an online portal to view the data (currently free to all during testing)
-Additional training will help improve tornado model's accuracy
Stay tuned for link
Will be turning STORM-Net experimental back on Thursday. By that time, should also have the severe wind and severe hail models available.
4
2
60
7
16
120
Decided to do an in-depth test of STORM-NET on the deadly Rolling Fork tornado of last year. The results here are absolutely mind-blowing.
STORM-NET had the Rolling Fork area in an elevated risk of a SigTor (ef2+) 8 minutes before the storm even had a tornado warning on it, and
7
6
119
Wow! For those of you watching the STORM-Net data, it has been continuously highlighting this area for the last 30-45 minutes.. I figured it was erroneous since I wasn't really seeing anything on radar (no rotation signature, no discreet activity, etc). Now there is a tornado
18
13
115
Big news: STORM-Net v2 will be operational in the coming weeks!
This update includes:
-Increased precision (5mi instead of 10mi)
-16x more training data
-600m (0.6km) input grid resolution
-More reliable data sources
See the image below to compare 10mi vs 5mi precision.
6
14
115
This is incredible. Here's a zoomed in look at STORM-Net's tornado probabilities issued at 15:34UTC, along with the actual touchdown location (occurred at 15:45UTC).
STORM-Net is capable of street-level tornado forecasts with advanced lead-time.
5
14
112
Important STORM-Net update: we will be launching STORM-Net v1.0 on Tuesday, March 12th!
STORM-Net has been extensively tested by us as well as third parties, and we are confident in releasing v1.0 at that time.
At that point, we will begin offering subscriptions for basic
3
20
109
@TheInsiderPaper
One of the most protected & surveilled buildings on the planet, and they can’t find the origin of the cocaine? 🤔
6
5
104
@reedstrongalt
@JenCollinsWx
@CBS12
@LoleskyWX
@NWSMiami
What? Tornadoes haven’t increased in frequency or intensity due to climate change & likely won’t increase.
20
1
103
It’s great to see such long lead-time from STORM-Net today! Seen as much as a hour lead-time with the Wisconsin tornado! Nice detection across the board with low false alarm rate.
Thanks for your analysis
@Wicky_dubs_WX
!
I'm so impressed with stormnet verifying a real time setup today with more lead time then the tornado warnings. Game changer this spring.
5
3
92
3
7
104
STORM-Net v0.4 will be released this weekend (prior to the severe wx threat next week).
The update will include the first 3hr STORM-Net data (for tornadoes) along with 1hr hail outputs and improvements among the wind/tornado 1hr branches.
If you have any suggestions, now is a
The potential for severe weather is forecast to increase around the middle of next week. Broad areas from the Ozarks into the Upper Midwest (2/27) shifting eastward into the Mid-South/Ohio Valley (2/28) have been highlighted. Check back for forecast updates as details may change.
8
223
616
8
13
101
STORM-Net had been highlighting this storm for a long time (1hr plus).. after a few failed attempts at tornadogenesis, it seems to have occurred at this point.
Tornado warning now issued and a strong couplet.
4
16
99
STORM-Net 3hr forecast will be coming soon!
STORM-Net will simulate future evolution of convection including convective initiation and will use that simulation to predict severe hazards over the next 3 hours. We plan on running it every 5 or 10 minutes!
10
8
98
These tornadic storms today are so strange.. And they’re occurring in a very marginal environment. This event will need to be studied!
9
6
99
STORM-Net v0.3 will be released tomorrow! This update includes 3x more training data and we should see a reduction in false alarm cases. Other changes include additional model inputs and minor changes to the architecture. Biggest change, though, is the increased training data.
6
6
96
@OliviaGenoGG
@HarryVanDerCam1
@historyinmemes
LLMs aren’t perfect, but neither are these tools. ZeroGPT, for example, says that the US constitution was written with AI/GPT.
1
1
94
The last time I remember seeing CSU-MLP this high was several days prior to March 31st of last year.
Great work by
@DrAaronHill
and others at CSU on this model!
4
15
95
Well would you look at that.. impressive!
Tornado Warning including West Livingston TX, Shepherd TX and Onalaska TX until 2:30 PM CST
7
110
197
5
10
91
Tornado OTG near Ozark, Alabama.. STORM-Net had a very good handle on this one, including on the 3hr forecasts.
4
10
90
CC overlayed with STORM-Net tornado probabilities last night, centered on Sulphur, OK.
Prayers to those affected that lost family members or property last night. Goes to show how quickly things can ramp up with such a robust parameter-space in place.
3
12
92
Upon the release of STORM-Net v1.0, we will begin offering a STORM-Net text alert system. It will allow users to input a zip code or address, and they will receive a text alert whenever STORM-Net reaches a certain threshold for that location.
Will make it clear that STORM-Net is
8
9
88
Exciting news! We at
@AtmoAnalytics
are excited to announce our newest model: NimbusAI.
NimbusAI is a state-of-the-art super resolution meteorological simulator that uses deep learning to deterministically predict the evolution of near-term precipitation & convective phenomena.
6
16
91
STORM-Net tornado probabilities with 45+min lead time for the Huntsville area. Now a large tornado confirmed (PDS warning). Take cover!
3
12
87
STORM-Net v1 will be running in real-time all day long, outputting probabilities for tornadoes, severe hail, and severe wind.
Access will be open to the public during the severe wx threats today and tomorrow.
4
11
88
12z HazCast (v4, not the newer version) tornado probabilities for Monday and Tuesday afternoons.
8
20
87
STORM-Net v0.4 is fully functional today with new features including:
-3-hour tornado probabilities, updated every ~10min
-1-hour severe hail probabilities
Remember that STORM-Net is still in active development and is open to the public for feedback/experimental use!
6
18
86
12z HazCast tornado and severe wind probabilities for Tuesday evening (00z and 03z 2/28).
5
14
83
Rapid destabilization is occurring across the warm sector. This destabilization is expected to continue over the coming hours.
Big question across the enh-mod risk at this point pertains to convective mode. Discreet supercells would be much more efficient at taking advantage of
2
15
85
STORM-Net is back online. The DDoS attack continues but we have mitigated it and have made some hardware upgrades.
We also pushed a bit of an upgrade to v0.2.1 which includes a slight adjustment which should help with the false positive cases with mid-level rotation.
v0.3,
6
7
85
Supercell party today it looks like.
Imagine today’s convective mode on Monday in Oklahoma. Would’ve been wild.
2
2
83
I get a lot of questions asking how exactly STORM-Net works. An easy way to explain it is like this:
Imagine a meteorologist with years of training/experience is trying to nowcast tornadoes. He or she will look at various mesoanalysis tools.. observations.. radar… etc and have
5
6
84
00z HazCast for tomorrow afternoon; tornado and hail threat at 21z and 00z.
3
11
84
Thank you to everyone who gave feedback on the new STORM-Net real-time severe wx prediction model! The real-time uploads will halt at this time and R&D will continue. I'll be sure to put it back up when the next severe wx/tornado threat occurs.
Feel free to leave any other
5
2
82
Tomorrow’s severe weather/tornado threat is definitely looking robust. Wouldn’t be surprised to see a high risk issued at some point if the recent trends continue.
5
4
80
Very precise forecast from STORM-Net at 20min lead-time. Highest probabilities coincide with location of confirmed tornado.
6
6
80
HazCast v5.0 will feature a significant step up in capabilities from previous HazCast versions. The architecture is much more like STORM-Net than previous HazCast versions, allowing the model to fully capture and "understand" long-range and short-range spatial and temporal
5
6
79
@watchtheweathe
An engineer and entrepreneur.
Tim Samaras lacked a degree but is widely accepted as a self-taught engineer.
IMO, If one has a job in meteorology and/or a proven track record of specialized knowledge, they can be considered a meteorologist even if they don’t have a degree.
3
1
76
Here's my interview with CBS news in Raleigh, NC talking about STORM-Net!
3
10
77
STORM-Net continues to perform really well. Here’s the latest forecast output and radar.
To those that followed the outputs through the night and this morning— what kind of feedback do you have?
11
10
74
A STORM-Net verification here on an *unwarned* storm. 40+ minute lead-time.. was the highest tornado probabilities of the day for STORM-Net, resulted in a true tornado.
3
3
75
STORM-Net tornado probabilities 90min leading up to the confirmed tornado near Hot Springs, AR.. probs started very early with this one! Warning issued right before touchdown.
2
12
74
12z HazCast is highlighting parts of the deep south for the potential for tornadoes on Wednesday afternoon.
3
8
73
Worked all night to get STORM-Net v0.3 ready for potential severe weather this afternoon.. and it's here!
This update includes the following:
-RADAR toggle! Now you can press the 'radar' toggle on the UI and the current CONUS radar will overlay!
-Additional training data (3x
5
7
73
The STORM-Net lightning outputs will be able to predict thundersnow.
4
1
73
Should I start a YouTube channel talking about STORM-Net and other AI weather products?
13
1
73
Here's STORM-Net 3hr tornado probabilities starting 3 hours prior to the deadly Wapakoneta, OH tornado.
Included below are the 3hr STORM-Net tornado probabilities at 3hr, 2hr, and 1hr prior to the deadly tornado, which occurred around 23:50 UTC. Extreme precision!
4
10
71
40 minute lead-time on this one.
Tornado Warning including Orangeburg SC, Bamberg SC and Saint Matthews SC until 3:15 PM EST
0
35
47
5
5
68
Soon, the STORM-Net interface will become a powerful all-in-one weather display tool.
Currently, it displays STORM-Net data, lowres radar data, and NWS warnings.
In the near future, it will display:
-Super-res NEXRAD radar data
-Model data (including MMFS and publicly available
4
10
70
Wow! A sizable plume of very high 100m STP values are progged to move through Kansas later this evening, coinciding with decent ECAPE on MMFS!
STORM-Net is running currently and will continue to run tonight and tomorrow during possible severe weather.
6
5
69
STORM-Net will output severe wind probabilities soon. Very similar to the tornado predictions -- it represents the probability of severe wind (50kt+) over the next hour after initialization. It also predicts the probability of sig wind, though that is not depicted below.
Here is
5
4
67