We are using artificial intelligence (AI) to solve real-world challenges: more accurate forecasts, more timely predictions of hazardous events (tornadoes, lightning, avalanches, etc), and providing detailed snow condition descriptions.
In short, we are deploying AI to help our customers, not to cut costs.
When most people hear of AI, they immediately think about ChatGPT and similar LLMs (large language models) that provide a chat interface, but AI’s impact on weather forecasting goes far beyond a chatbot.
The following notes discuss why we are excited about AI, and address questions and concerns that we’ve heard from our customers and our internal team.
- Joel Gratz
CEO & Founding Meteorologist
What is AI?
Artificial Intelligence (AI) essentially means "pattern matching". Human brains learn patterns, and now computers can learn patterns as well or better than humans.
How is OpenSnow using AI?
Our PEAKS model uses AI to learn from past storms to more accurately predict future storms. Our focus for PEAKS is to improve forecast accuracy for snow/precipitation, temperature, and winds in complex mountain environments.
For example, the PEAKS model should identify that a northwest wind with the appropriate temperature and moisture will cause storms to drop higher-than-forecast amounts of snow in Alta, Utah, and conversely, the model should identify that a southwest wind will usually cause storms to drop lower-than-forecast amounts of snow in Vail. Based on testing, PEAKS may improve the forecast accuracy up to or above 50% in many situations. See the full PEAKS FAQ for more.
Our StormNet model uses AI to forecast lightning, tornadoes, hail, and severe winds based on real-time radar and satellite data. Forecast information is updated every 2 minutes and extends 7 days into the future.
For example, the StormNet model has identified the timing of a thunderstorm's first lightning strike with up to 30 minutes of advanced notice, and it has also identified the likelihood of a tornado tens of minutes and even hours in advance. Based on testing, StormNet's skill scores are similar to or sometimes exceed those of the National Weather Service. See the full StormNet FAQ for more.
Our AI Overviews use AI to simplify tables and charts into understandable bullet points discussing the snow conditions and forecast for the upcoming 10 days. This is available for about 200 locations globally as of December 2025 and will be expanded in the future.
Lastly, our PEAKS Avy model is an attempt to create avalanche forecasts which can be updated frequently and zoomed to specific mountain ranges. We are running a prototype of this model and expect to work with avalanche forecasting centers to verify and improve the system.
Did OpenSnow work with NOAA or other organizations to implement AI forecasts?
Our team is made up of meteorologists, computer programmers, and researchers with decades of experience. We did not undertake these efforts flippantly.
Our PEAKS model was trained using historical weather data compiled by NOAA, NCAR, and the ECWMF. Also, our researchers used start-of-the-art methods to convert these historical data into the first-of-its-kind mountain-focused weather prediction model.
Our StormNet model was trained using historical weather data compiled by NOAA as well as well as data from SPC and NCEI. Like PEAKS, our researchers used start-of-the-art methods to convert these historical data into the first-of-its-kind severe weather prediction model.
For PEAKS and StormNet, the training data is similar or identical to that used by Google and other organizations. The data has been created over the past 50+ years based on the public's investment in science (tax dollars from citizens across the world). This public investment in science allows governments, for-profit companies, non-profit organizations, and individuals to use the data to create new and potentially massively improved weather forecast techniques. In short, the system is working as intended by encouraging innovation across the world.
Our AI Overviews were created based on six months of research and adjustment, understanding how our human forecasters think about and communicate weather, and dialing this into a system that outputs a weather forecast that makes sense and is something that we, as experienced forecasters, stand behind and want to read. Like humans, AI LLMs can make mistakes, and we watch for these and attempt to prevent them as much as possible.
Will the AI Overviews replace our human forecasters?
The AI Overviews will NOT replace human forecasters - I am one of the local forecasters and I am not going anywhere:-)
OpenSnow is known for their human forecasters, and these human forecasters are still writing!
We implemented AI Overviews for two reasons:
1. To be able to cover more locations with more frequent updates.
2. To simplify tables, charts, and (sometimes) longwinded human commentary into bullet points that can be read quickly and understood more easily.
For the 2025-2026 season, we made a few changes that had nothing to do with AI, such as adding a New York Daily Snow and ending the Southern California Daily Snow due to lower readership.
Otherwise, we made no substantive changes and your favorite local expert forecasters will continue to write as they have in the past.
Does OpenSnow consider AI's environmental concerns?
I think the best context is that we are a very small user of AI and energy, especially compared to larger technology firms.
- We have only a handful of servers, similar to other small technology companies, and only one or two of these servers handles AI-based tasks. Other servers process weather data, power our website and mobile apps, and run our backend admin systems.
- Our AI model training runs use the equivalent energy of about 1-3 gallons of gasoline. For all of our training of PEAKS and StormNet, total energy consumption is likely on the order of a few full tanks of gasoline.
- To keep up with more people using OpenSnow, we will likely need to add additional servers, and the power required for these additional servers will likely dwarf the small amount of incremental power that we are using to train and create our AI features.
AI technology created important improvements in weather forecasting technology in just the last few years, and we expect these improvements to continue. We think that the energy usage to enable these improvements is justified. Finding and employing clean energy solutions is likely more important than throttling AI improvements, at least as it relates to weather and other science applications.
That said, we acknowledge that we use energy and that hundreds of millions of skiers and snowboarders around the world use energy to drive and fly to enjoy the sport of sliding on snow.
This is why we partner with Stripe to send 1% of our subscription sales to innovative projects that directly remove carbon dioxide from the atmosphere and ensure its sequestration in secure, long-term storage. As of November 2025, we have sent over $72,000 to Stripe Climate. This is a drop in the bucket of the funding needed to enable clean energy and carbon removal, but we think that it's a drop in a good direction.
Questions?
Send an email to hello@opensnow.com — a real human will get back to you within 24 hours.