PEAKS is our in-house weather forecast system that is up to 50% more accurate in mountain terrain.

  • PEAKS learns from past storms and transforms normal weather models (used by other apps) into sharper, more location-specific forecasts.
  • Learning from past storms is what human meteorologists do, but unlike humans, PEAKS can forecast for every location on earth with updates taking only seconds.
The bottom line is that OpenSnow has developed a world-first type of model that evaluates past storms to create more accurate predictions.

This should result in your ability to make higher-confidence decisions when searching for your favorite conditions.

Where is PEAKS available?

We anticipate using PEAKS for precipitation forecasts (amount of snow, rain, and mix) for these locations starting in late November 2025.

“PEAKS

Example at Alta

Alta, Utah, is one of the snowiest locations in the United States.

Typical weather models underpredict the snowfall at Alta (grey line), while the PEAKS model (orange line) closely aligns with the precipitation that Alta receives (black line).

Image Placeholder

How does PEAKS work?

PEAKS transforms normal weather models into higher-resolution models, delivering sharper, more location-specific forecasts.

Image Placeholder

This map compares a normal weather model and our PEAKS weather model using a precipitation forecast in northern Utah - higher amounts of precipitation are shown as orange colors.

The normal weather model is "blob-y" and smooth as it does NOT understand the location of each mountain peak and valley bottom.

The PEAKS weather model transforms this "blob-y" forecast into a higher-resolution forecast that accurately predicts areas of higher precipitation along favored mountains, and areas of lower precipitation in sheltered valleys.

How can I see forecasts from PEAKS?

In November 2025, every precipitation forecast in OpenSnow, globally, will be powered by PEAKS.

By the end of 2025, we anticipate switching all temperature and wind forecasts on OpenSnow, globally, to be powered by PEAKS.

Said another way, we take the best normal weather models from around the world, transform them into high-resolution models using PEAKS, and the resulting data is shown as the OpenSnow forecast.

Accuracy

How much more accurate is PEAKS?

PEAKS predictions are more accurate by:

  • Precipitation: +42%
  • Temperature: +82%
  • Wind: +72% 
Results are based on historical forecasts, computed on unseen data across approximately 6 billion grid points per variable.

These numbers use mean squared error (MSE), which penalizes big misses, and compares PEAKS to normal weather models that provide a 10-day forecast across the globe:

  • Normal model: 10-25km (6-15 mile) resolution
  • PEAKS model: 3km (1.8 mile) resolution
Each storm is different, and in the future we hope to set up a public-facing, real-time accuracy tracking system.

Why mountain weather is hard to forecast

Most weather models predict conditions in large boxes, often 15 to 25 miles wide, and every location within this 15-25 mile wide box gets the same forecast.

That scale can work over flat ground where the weather does not change quickly.

But that scale does NOT work well in mountains, where the weather changes rapidly with elevation.

One part of a mountain range can get hammered while a nearby location receives much less snow, and most weather models cannot accurately predit this.

How we made our forecasts in the past

In the past, we would combine forecasts from many models, some with large boxes (15-25 miles wide) and some with smaller boxes (3-10 miles wide).

We would adjust this data based on each model's limited understanding of elevation.

This was a band-aid solution: We took forecast data that poorly handled the mountain's impact on weather and tried to adjust it the best that we could.

Why we built PEAKS

Most other weather forecast companies are focused on where most people live, which is cities and NOT mountains, so they are NOT focused on making mountain forecasts more accurate.

But OpenSnow is focused on the mountains, so we were uniquely motivated to solve this problem.

In spring 2024, we started an internal research experiment that became PEAKS.

Early results were strong, so we doubled down, and for the last eighteen months we have iterated week after week on data selection, the training pipeline, and the model design.

During the 2024-2025 season, we ran pre-release versions internally for our forecasters. Their feedback was positive and specific, which helped us sharpen the system.

During the 2025 spring and summer, PEAKS took major steps forward in accuracy and stability, and we now feel that PEAKS is ready to be shared with our users.

As a result, starting in November 2025, PEAKS powers every OpenSnow forecast, regardless of your subscription tier.

How PEAKS works

1. PEAKS looked at data from historical storms.

2. PEAKS found the patterns that made storms produce different conditions compared to what normal weather models expected (temperature, wind and rain/snow). For example, PEAKS learned the pattern that resulted in a certain storm producing more (or less) snow than expected.

3. PEAKS uses these learned patterns to transform normal weather models into higher-resolution forecasts that understand how mountains impact the weather.

Why the PEAKS design works

PEAKS treats the storm as one system and solves for everything together.

  • Storm-specific. Instead of static ratios (like "Alta always gets X% more snow than the forecast), PEAKS adjusts to each storm’s temperature, wind, and precipitation structure.
  • Shared context. It reads the broad storm pattern and local terrain together, sharpening favored aspects and elevation bands.
  • Spatial continuity. It forecasts fields, not isolated points, which reduces seams and speckled artifacts and preserves smooth, realistic patterns of temperature, wind, and precipitation.
  • Single model. Learned relationships across variables cut down on cascade errors you get from separate models.
What PEAKS takes in, and what it predicts

Inputs: PEAKS ingests 38 dynamic weather variables from normal models plus 7 static variables that describe the surface, such as topography and land–water mask.

Outputs. PEAKS predicts three variables simultaneously on a high-resolution grid: precipitation, temperature, and wind speed.

What this means for your OpenSnow forecasts

  • More accurate mountain totals. Better alignment between the forecast and what you find on the ground.
  • Fewer surprises. Clearer signals for which elevations and slopes are favored in each storm.
We will continue to train, refine, and release new versions of PEAKS.

Questions?

Send an email to hello@opensnow.com — a real human will get back to you within 24 hours.