What is StormNet?


StormNet predicts lightning, hail, damaging thunderstorm winds, and tornadoes using an AI-powered severe weather forecasting system that we believe to be the first of its kind in the world.


How do I interact with StormNet?

  • COMING SOON: You can receive an alert when lightning, hail, damaging thunderstorm wind, or a tornado is likely. Go to OpenSnow > My Location > Alerts.

    Image Placeholder

  • View a color-coded map of lightning, hail, damaging thunderstorm wind, or tornado probability, extending from near real-time up to seven days in the future. Go to OpenSnow > Maps > StormNet

    Image Placeholder


How do I use the StormNet map?


Tap the layers button (highlighted in blue in the above animation) to choose from three map-based views:

  • Radar: The highest-resolution radar available for the United States from each radar station.
  • Radar + Risk: High-res radar alongside StormNet risk probabilities from the past 1.5 hours through the next 4 hours. 
  • Risk Forecast: StormNet risk probabilities from real-time to 7 days into the future

Tap the lightning button (bottom right in the above animation) to choose from four weather hazards: lightning, hail, damaging thunderstorm wind, or tornado probability.

Tap the radar button (next to the lightning icon in the above animation) to choose a radar site and the type of radar data. Radar sites will also be automatically selected when you move around the map.


Find StormNet at OpenSnow.com/StormNet as well as on the OpenSnow iOS app. It is available for free for a limited time and will eventually be part of an annual subscription.


How can StormNet help me?

  • Make real-time decisions to seek safety if lightning or severe weather is heading in your direction. This is particularly useful when hiking, driving, and participating in most outdoor events.
  • Plan travel and activities based on severe weather risk, with forecasts extending out to seven days at one-hour intervals.
  • Hiking: A heightened risk of lightning in the 30-60-minute time frame may necessitate turning around early or seeking shelter / going to a lower elevation.
  • Driving: An elevated risk of hail in an area that you plan on driving to in the next 30-90 minutes may affect your plans and help you avoid vehicle damage.
  • Safety: A heightened risk of a tornado in the 30-60-minute time frame will give you the necessary 'heads up' to pay close attention to the weather over the coming hour, and take cover if needed, potentially giving you a life-saving alert.

Do any organizations use Stormnet?
  • StormNet is currently used by the United States Department of Defense, the News Media, and various Emergency Management organizations. If your organization would like to license StormNet for use in aviation, insurance, logistics, or other areas, please contact us at stormnet@opensnow.com

Why did OpenSnow create a severe weather forecasting system?

  • In 2011, OpenSnow began predicting snowfall for skiers and snowboarders.
  • We then expanded to create forecasts to support many types of outdoor activities, with forecasts available globally.
  • Since lightning is a risk to outdoor activities, especially on longer mountain-based adventures, we endeavored to create a system that would provide both (a) real-time lightning forecasts to support in-the-moment decisions, and (b) multiple-day outlooks to support trip planning.
  • Meteorologist Andrew Brady built StormNet to predict the risk of tornadoes, hail, and damaging thunderstorm winds, and OpenSnow acquired his technology and his expertise to expand the AI system to predict lightning and other weather variables to help people stay safe and enjoy the outdoors.
  • The acquisition of StormNet is a sign that OpenSnow is committed to innovation, and we have multiple world-first innovations on the way, mostly focused on mountain weather.

OpenSnow already shows a “Lightning Risk” map layer. Is StormNet different?
  • The “Lightning Risk” map layer on OpenSnow displays lightning risk data from the “LightningCast” product, developed at the University of Wisconsin. This product updates every 5 minutes, predicts the lightning risk out to 1 hour, and is produced for both the eastern and western parts of the United States.
  • The “StormNet” map shows the StormNet lightning risk, which is updated every 2 minutes, predicts the lightning risk out to 7 days, and is one unified system across the United States. While we think that StormNet is the better choice (more accurate and more timely), we still show both models, as more forecast data often leads to better decisions.


How accurate is StormNet?


Based on over 250 severe weather events from 2024, StormNet provides more accurate tornado forecasts than the National Weather Service and is more accurate than other lightning prediction systems.

We are unable to compare forecasts for hail and damaging thunderstorm winds as we are not aware of similar forecasting systems.


Overall accuracy

StormNet has an average accuracy percentage of 98.8% across all hazards. (1)


Accuracy by StormNet Version

StormNet’s accuracy improves with each new version across all hazards (Lightning, Wind, Hail, and Tornado).

The accuracy for each hazard is shown for both 30-minute and 60-minute predictions. (2)

Image Placeholder


Accuracy for Lightning

StormNet produces the most accurate lightning forecast compared to three other models for which accuracy statistics are published (3).

Image Placeholder


Accuracy for Tornadoes

Probability of Detection (POD) at the fifty percent risk probability:
  • 67% for StormNet vs 62% for National Weather Service
  • Higher is better
  • Source: (4) & (5)

False Alarm Rate (FAR) at the fifty percent risk probability:
  • 35% for StormNet vs 70% for National Weather Service
  • Lower is better
  • Source: (4) & (6)

StormNet shows a clear signal for tornadoes 48 hours (two days) in advance:
  • The 48-hour forecast had a tornado risk of 22% in cases when there WAS a tornado.
  • The 48-hour forecast had a tornado risk of 0.007% in cases when there was NO tornado.
StormNet shows a clear signal for tornadoes 3 hours in advance:
  • The 3-hour forecast had a tornado risk of 35% in cases when there WAS a tornado.
  • The 3-hour forecast had a tornado risk of 0.0002% in cases when there was NO tornado.

The University of Oklahoma evaluated a very early version of StormNet and found, “At short lead times before tornadogenesis, this study found that StormNet outperforms the National Weather Service in tornado detection percentage at lower tornado probability thresholds, while even possessing the ability to detect potential for tornadogenesis up to an hour before tornadogenesis.” (7)




How does StormNet compare to existing forecasts?


Severe weather forecasts have existed for decades, and StormNet’s technology offers many improvements over the normal forecast products from the U.S. Government's Storm Prediction Center (SPC).

Frequent Updates
  • StormNet: Every 2 minutes
  • SPC: A few times per day
  • More updates is better
  • StormNet provides 720 updates per day (every 2 minutes) while existing severe weather forecasts from the Storm Prediction Center (SPC) are updated, at most,  a few times per day (varies based on the forecast time frame), and existing weather models are updated at most once per hour. The weather is constantly changing, and StormNet re-evaluates and updates forecasts, taking into account the rapid evolution of weather.

Detailed Timing

  • StormNet: 30 min & 60 min intervals
  • SPC: 12-24 hour intervals
  • Shorter intervals is better
  • StormNet provides forecasts with 30-minute intervals to one hour in the future, and then provides 1-hour intervals to 7 days in the future. Existing severe weather forecasts from SPC are provided for 12 and 24-hour intervals.

Hyperlocal
  • StormNet: 2-mile & 7-mile resolution
  • SPC: 25-500+ mile resolution
  • Smaller resolution areas is better
  • StormNet provides a unique forecast every 2 miles through two days and every 7 miles from two to seven days into the future, while existing severe weather forecasts from SPC cover areas from 25 miles to hundreds of miles.

Ease of Use

  • StormNet: All in one place, from real-time alerts to 7-day forecasts
  • SPC: Multiple sources
  • StormNet is a unified system for four impactful severe weather hazards, from real-time to one week out. Conversely, the National Weather Service issues Warnings for real-time information while the Storm Prediction Center issues broad multi-day outlooks.

Should StormNet be used for official watches and warnings?

The National Weather Service, the Storm Prediction Center, and other government agencies are the official sources for watches, warnings, and severe weather information.

StormNet should serve as a trusted and automated guide for severe weather risk, but StormNet should NOT be treated as an official source.


Don’t “weather models” already forecast severe weather?

Existing weather forecast models, of which there are tens to hundreds, are incredible feats of engineering and predict the ingredients for severe weather, but they are not designed to excel at severe-weather forecasting and are not real-time tools.

StormNet was trained to do ONE thing - forecast severe weather from real-time through 7 days into the future.


Don’t existing weather apps and companies forecast severe weather?

Existing weather apps and companies usually pass along official government watches and warnings, and often communicate if lightning or hail is already close by.

While there are valuable features, we are not aware of another weather app or company that matches StormNet’s ability to provide 2-minute updates for the risk of lightning, hail, damaging thunderstorm winds, and tornadoes from real-time through 7 days in the future.



How does StormNet work?


StormNet analyzes Super-Res Radar, satellites, ground-based sensors, and weather forecast models to find complex patterns that lead to lightning, hail, damaging winds, and tornadoes.


What are some details about the AI model?

  • Learned by evaluating about 7.5 trillion weather data points from past severe weather events.
  • Ingests 125 million weather data points every 2 minutes to update its forecasts.
  • Contains about 90 million training parameters.
  • Consists of a specialized system that learns from weather patterns, just like a meteorologist would learn by watching the weather over many, many years.
  • Architected with elements from convolutional neural networks, transformers, and graph neural networks.

Is StormNet the only system of its kind in the world?

As of August 2025, when this FAQ was published, we think StormNet is the first system of its kind in the world that predicts multiple components of severe weather (lightning, hail, thunderstorm winds, and tornadoes) from real-time through 7 days in the future.

There is a lot of work involved to transition a technology from research to real-time, reliable operations, and our team is proud to bring this technology to the public. 

We know of two other AI-based systems that predict lightning during the next hour, but we are not aware of another system (whether AI-based or a conventional physics-based model) that forecasts four types of severe weather with 2-minute updates from near-real-time to seven days (168 hours) in the future.

Governments, universities, private companies, and individuals are rapidly creating new technology, so there could be and likely are additional systems in private use, or systems that will soon be operationalized for public use.

We welcome the competition, and we are excited that weather prediction is getting more accurate and that innovation is being democratized with AI-based technology allowing organizations outside of academia and governments to create first-of-their-kind products.


What’s next for StormNet?

We will produce updated versions of StormNet, which should improve the accuracy of the model. We also anticipate adding forecasts for weather events beyond lightning, hail, damaging thunderstorm winds, and tornadoes.



How does StormNet fit into the greater meteorological community?



StormNet builds on 50+ years of work performed by scientists and engineers across the globe, and most of this work was supported through the collective tax dollars of the world’s citizens.

While StormNet and many other recent weather forecasting innovations are being privately funded, we acknowledge and appreciate that our work would not have been possible without the public’s past and ongoing support for science, the collection of past weather data, and real-time weather monitoring systems across the globe.

StormNet was initially created by a single individual on a piece of plywood (see the story below).

OpenSnow’s acquisition of the technology and work to create an operational system was funded through the profits of our business – we believe in investing in scientific advances that will benefit our customers and push forward the bounds of weather and environmental prediction.

We are thankful to be working on this technology in 2025, as the money and people needed to produce world-first technology are now within the grasp of small companies like ours (we have 7 full-time technical employees, and two people were primarily responsible for the development and operations of StormNet).

Aren’t human forecasters important?

Human forecasters are important, and OpenSnow employs many humans that write daily forecasts throughout the winter.

That said, AI-powered technology like StormNet is starting to help computers to ‘think’ like humans and provide forecasts that previously could have only been provided by humans.

The automation of many aspects of weather forecasts is an exciting development, as this should lead to more frequent updates and eventually more accurate predictions.

As stated above, the human forecasters at the National Weather Service, the Storm Prediction Center, and other government agencies are the official sources for watches, warnings, and severe weather information.

What is the benefit of an automated system?

Humans, weather stations, radar, and satellite systems are NOT able to fully observe the conditions at every point, but StormNet, as an automated AI-based system, works to 'fill in the gaps' through state-of-the-art weather pattern recognition.




Can you show me examples of StormNet and Super-Res radar?



Super-Res Radar with the risk of damaging thunderstorm winds.

The grey to blue colors ahead of the storm indicate elevated damaging wind probabilities. In this example, damaging wind probabilities are 40% to 50% ahead of the main core of the storm.

Image Placeholder

To make this forecast, StormNet considers:

  • Where are the storms?
  • Will this storm produce damaging winds?
  • Where is damaging wind most likely to occur within this storm?
  • Which direction is the storm moving? Is there any possibility that it may switch directions?
  • How long will the damaging wind threat persist?
  • Is it only for the next couple of minutes, or will it persist during a 30-60 minute period?

Super-res radar with the risk of lightning.

The pink color over and ahead of the storm indicates an 80-100% probability for lightning. Notice that the final frames show the high probability of lightning continuing into the next four hours beyond the end of the radar animation.

Image Placeholder


Below, the radar is turned off, and we are looking at hail probabilities for the next day.

The blues indicate lower probabilities, whereas the yellows are localized areas where there is increased confidence in hail occurring.

Image Placeholder


What is Super-Res Radar Reflectivity?

Reflectivity is what we normally think of as “radar”. It measures the intensity of precipitation based on how much energy a radar signal bounces back from objects in the atmosphere, like raindrops or snowflakes.

Super-Res reflectivity is the highest-resolution radar data.

Image Placeholder


What is Super-Res Radar Velocity?

Velocity measures the speed at which precipitation (like raindrops or snowflakes) is moving toward or away from the radar site. Using the Doppler effect, the radar detects changes in the frequency of the returned signal caused by motion.

Green = precipitation moving toward the radar with an increased frequency

Red = precipitation moving away from the radar with a decreased frequency

This information helps meteorologists understand wind patterns inside storms, which can reveal important details like rotation in a thunderstorm or wind shear that might indicate severe weather.

Image Placeholder


What is Super-Res Radar Spectrum Width?
Spectrum width shows if the speeds of precipitation particles are similar or different.

Green = a small spectrum width, which means the particles are all moving at nearly the same speed.

Red = a large spectrum width, which means the particles are moving at different speeds, suggesting turbulence, wind shear, or other chaotic motion. This helps meteorologists identify areas of atmospheric instability, like strong gust fronts or tornadoes.

Image Placeholder




Who created StormNet?


StormNet was created by Andrew Brady, a meteorologist and technologist with a passion for life-saving weather prediction. As of August 2025, Andrew resides in Georgia and is 25 years old.


What was Andrew’s path to creating StormNet?

  • Andrew's passion for meteorology started when he was a young child. After school, he would read meteorology books and, later, would read online articles on the subject. One highlight of his childhood was begging his mother for a megaphone so he could make severe weather announcements to the neighborhood where he grew up.
  • In the year 2020, Andrew began working to improve the prediction and communication around impactful weather and, eventually, save lives. He experimented with the WRF weather model, adjusting configurations and source code, and eventually built a real-time system to run WRF.
  • In the year 2021, Andrew wanted to produce more specific severe weather forecasts, so he used data from the WRF weather model as the basis for a machine learning algorithm that would output probabilistic severe weather predictions. This was the first step toward StormNet.
  • In the year 2023, Andrew built a complex deep learning model that would take various data sources, as well as current conditions, and would produce hyper-local severe weather forecasts. This was the first version of StormNet, which stands for Severe storm and TOrnado Real-time Monitoring NETwork.
  • After working on StormNet, Andrew realized that such a complex model wouldn't be able to run on his Dell PowerEdge server cluster. It needed graphics processing units (GPUs), but Andrew did not have enough money to rent high-end GPUs from third-party server companies.
  • To realize his dream of an advanced severe weather prediction system, Andrew had two options:

    1. Make StormNet simple enough to work without GPUs

    2. Build a GPU server at home
  • Andrew embraced idea #2 and decided to build his system completely from scratch. He ordered a motherboard that had the correct adapters, ordered the GPUs second-hand, and ordered all of the remaining supplies to build this computing system. Then, he went to the local home improvement store to purchase a piece of plywood to put the system on.

    Image Placeholder
  • Andrew had no idea how this home-grown system was going to work – or if it would work – but it was worth a shot if he could build something special. Over the next several weeks, the parts came in, and he built the system. It didn't work at first, but after lots of trial and error and rebuilds, the green light on the motherboard finally came on – it was working.

How did Andrew operationalize StormNet?

  • Andrew collected the training data that he would use, and since this had never been done before, he had to figure it out on the fly.
  • Andrew tried various architectures (graph attention networks, convolutional neural networks, multi-layer perceptron, etc) before eventually deciding that he would need to build something custom for this task.
  • After weeks of trial-and-error, StormNet v0.1 was trained. This was cool, very cool.
  • He ran the model on various past events, such as the Mayfield, KY, 2021 EF4 tornado. StormNet was able to predict that a tornado would form and move into Mayfield with 30min+ lead time.
  • Trial-and-error continued through the end of 2023, and finally, in January 2024, he announced StormNet to the world.
  • At the time, StormNet was producing 1-hour tornado predictions across the eastern CONUS every 5 minutes. It gained popularity in the weather community, and Andrew quickly iterated updates and upgrades, allowing the system to learn from its performance.
  • By March 2024, after a lot of R&D effort, Andrew released StormNet v1.0.
  • Through Spring 2024, StormNet's popularity continued to increase due to its accuracy and usefulness.
  • New versions of StormNet were able to predict more than just tornadoes – hail and damaging wind outputs were also added.
  • By summer 2024, Andrew met with meteorologists from Fox Weather, The Weather Channel, RadarOmega, and others to gather valuable feedback on StormNet.
  • In fall 2024, OpenSnow acquired AtmoSphere Analytics with a vision to:
  • Make this groundbreaking technology available to many more people
  • Develop additional AI-based weather models focused on lightning, and for specific use cases to help people stay safe and enjoy the outdoors.
  • From late 2024 into 2025, Andrew and the OpenSnow team have worked hard to improve StormNet and to build a robust operational system (moving from research to operations involves a lot of work, more than most researchers realize).




Is there an API available for StormNet?


For individuals, find StormNet at OpenSnow.com/StormNet, the OpenSnow iOS app, and by going to OpenSnow > Maps. StormNet is available for free for a limited time and will eventually be part of an annual subscription.

For businesses in aviation, insurance, logistics, and other areas that want this forecasting for internal use and for their clients, please contact us at stormnet@opensnow.com





Summary



How should I interpret the risk forecast colors?

For lightning, hail, damaging thunderstorm winds, and tornadoes:

< 25%.........Blue..........Low risk
25-50%......Green.......Possible
50-75%......Yellow.......Likely
75%+..........Red...........Very likely – stay alert and seek NWS warnings


How often does StormNet update?

Every 2 minutes, with continuously refreshed data and forecasts.


What is Radar + Risk?

Radar + Risk overlays current radar with StormNet’s forecast risk maps, showing where hazards are happening – and where they’re headed.


Are there any limitations or downsides to StormNet?

  • StormNet is only available for the contiguous (lower 48) United States and adjacent areas.
  • StormNet is a state-of-the-art AI weather forecasting system, though like all weather forecasts, it is not guaranteed to be accurate. Always consult official sources of weather information for life and property protection.
  • StormNet is architected to be reliable 24 hours a day, 7 days a week, 365 days a year, though like all computer systems, it may be unavailable due to software issues, hardware issues, or a delay in the incoming weather data used to make its predictions.
  • Super-res radar data is sourced from NOAA. Occasionally, some radar data is delayed or missing due to maintenance or unanticipated issues. NOAA often fixes these issues quickly. Radars with delayed data are colored red on the StormNet map, and NOAA’s real-time radar status is available here.

Still have questions?

Email us at hello@opensnow.com — a real human will get back to you within 24 hours.






Press Background



Origin & Creator

StormNet was created by Andrew Brady, a meteorologist and technologist with a passion for life-saving weather prediction. Andrew began experimenting with weather models in 2020 and built a foundation of science and engineering with a singular goal: to improve life-saving weather prediction. StormNet evolved into a powerful AI system capable of hyper-local forecasts with lead times up to seven days.


From Garage to GPUs

Unable to afford enterprise GPUs, Andrew built a custom home GPU server from scratch – assembling parts, rigging components on plywood, and debugging it into a working system that would later train StormNet v0.1. The model gained traction in 2023 and 2024, particularly after successfully predicting real-world events like the Mayfield, KY EF4 tornado.


Acquisition by OpenSnow

In Fall 2024, OpenSnow acquired AtmoSphere Analytics to scale StormNet’s technology and impact. Since then, the team has continued improving StormNet’s speed, accuracy, and capabilities.


How It Works

StormNet is a deep learning system that analyzes radar, satellite, and surface sensor data to detect patterns leading to severe weather. It produces real-time, high-resolution forecasts for lightning, hail, damaging winds, and tornadoes. The system updates every 2 minutes and offers hourly forecasts up to 168 hours ahead.


Performance & Evaluation

  • Accuracy: 98.8% across all hazards
  • Tornado Detection (short-range): 67% vs 62% for NWS (higher is better)
  • Tornado False Alarm Rate: 35% vs 70% for NWS (lower is better)
  • Probabilities 3 hours before tornado events are up to 175,000x higher than baseline
Independent evaluations by a University of Oklahoma researcher praised StormNet’s detection ability, especially at short lead times.


Differences from SPC Outlooks

While the U.S. Government’s Storm Prediction Center issues forecasts that are human-generated and update once or twice daily, StormNet is AI-driven and updates every 2 minutes, offering hourly predictions for specific hazards rather than broad regional outlooks.


Data Visualization

StormNet integrates with Super-Res Radar to offer intuitive, layered maps that show current radar imagery alongside probabilistic risk contours for severe weather.


How does StormNet help with planning?

StormNet can help you avoid risks when hiking, driving, or planning outdoor events, with real-time alerts and multi-day forecasts.


Is StormNet the only system of its kind in the world?

As of August 2025, when this FAQ was published, we think StormNet is the first system of its kind in the world that predicts multiple components of severe weather (lightning, hail, thunderstorm winds, and tornadoes) from real-time through 7 days in the future.


How does StormNet fit into the greater meteorological community?

StormNet builds on 50+ years of work performed by scientists and engineers across the globe, and most of this work was supported through the collective tax dollars of the world’s citizens. While OpenSnow is not publicly funded, we acknowledge and appreciate that our work would have been impossible without the public’s past and ongoing investment in science.


For media interviews, visuals, or data access, reach out to hello@opensnow.com





Sources & Notes


(1) Accuracy in machine learning statistics is defined as the proportion of grid points in a test set where the prediction is correct. This includes both true positives and true negatives. Since true negatives dominate these datasets, a high accuracy percentage is expected.

(2) This plot is based on internal evaluations of a 250+ event benchmark. This benchmark consists of a diverse set of severe weather events (or non-events) from 2024. These events were held out of training, so the model has never seen these events.

(3) This plot compares StormNet with other state-of-the-art lightning prediction models. Sources for these models, including published evaluation results, are found below.

(4) National Weather Service Statistics:


(5) Probability of Detection (POD) is defined as the proportion of grid-points in a test set where the hazard is forecast (>50%) and occurs (within n miles of the point). Higher is better.


(6) False Alarm Rate (FAR) is defined as the proportion of grid-points in a test set where the hazard is forecast (>50%) but does not occur. Lower is better.


(7) Barton, Brayden. (2024). Evaluation of STORM-Net 1-hour Tornado Forecast Detection Prior to Tornadogenesis of Significant Tornadoes during February -Early May 2024