What is this?

We use machines that learn to forecast weather.

We start with numerical weather models that produce 10s of billions of forecasted values every day. A machine looks at these values and compares them to the output of a single sensor at a specfic weather station. Over years of looking at the models output and the sensor, the machine learns to forecast for that sensor. This learning is then scaled to many sensors at many weather stations.

On the Viz page you will see the real-time output from a group of weather stations and the forecasted weather using the above technique. If there is not enough history for certain sensors there may be no forecast as the machine needs more time to learn. At times weather stations go down or have issues, and therefore pages may not return data.

On the Brass Tacks page you will see this same data transformed into a text forecast. Snow totals are also provided using algorithms to transform temperature gradients into snowfall totals using emprical data. The text forecast updates in real time with the sensor readings from the field.

On the Wx Maps page you will see the raw output from our WRF model that is running up to 4x a day. We run a 9km, 3km, and 1km resolution model. The 9km has a point every 9km, the 1km a point every 1km. Therefore the 1km is the highest resolution, but the 9km covers the most area.

Once we start building more of a snowpack you'll also see a tab for "available snowpack models". These are snowpack simulations for available weather stations. They combine the measured weather with variables from model output to create an expected snowpack within the vicinity of each weather station. It includes both a historical and future snowpack forecast. It is updated hourly, and the UI is not mobile friendly. As with all avalanche observations, it should be used with caution. Creating accurate snowpacks from these types of data is extremly difficult, and if you rely too much on it's accuracy when traveling through avalanche terrain you may die.

It takes a lot of hardware to crunch these numbers. Crystal Mountain Resort has stepped up to be a partner and help fund this work. If you would like to see your local mountain have these same capabilities, let them know it's possible if they want to join our group.

Contact here: chris.c.morin@gmail.com