This is a very big deal for Snowology, but first, how did we get here...

About 4 years ago I was chatting with an NWS meteorologist about my plans for automating our forecasts. He was skeptical that it could be done accurately and even the NWS has their own forecasters manually tweak the blends twice a day. There's admittedly a ton of bad weather forecasting automation out there when it comes East Coast ski areas and this region is by far the most difficult of all to forecast due to how dynamic our weather is and how marginal temperatures can be. Most forecasting you see is simply an automated blend targeted at towns and cities in valleys, not the mountains.

So I asked him if we were to program in how to handle model biases and add in our own special treatments for certain circumstances and even individual mountains, and have it update as frequently as every hour, and tweak things manually when we needed, maybe we could improve the accuracy even more while also adjusting to changes in modeling in real time. He agreed, that would be pretty awesome, but he thought it would be really, really difficult and possibly out of reach, but I could tell that his interest was piqued.

Since our start I've forecasted entirely from mental blends of models since no single model or blend gets the majority of it right all over. It's a skill that takes years of practice by forecasting every winter storm and then doing verification across the entire region, and it has set Snowology apart because I can adjust for biases that are unique to different types of models and storms and even individual ski areas. That's how you create more accurate forecasts. Now I don't tend to toot my own horn often, but the fact is that there's no one else dumb enough to put in 10 years of learning these things for just one region before trying to automate it. I also have decades of experience in design, programming, and internet architecture, and now I have a team of very capable fullstack programmers bringing their own skills to the table and we can in fact automate much of this while also allowing for manual adjustments when needed.

After over 6 months of focused effort involving a team of 5 developers including myself, Luis, Cole, Deane, and Jeff, Snowology is in the final steps of readying our weather automation for launch, covering over 180 ski areas on the East Coast from Quebec to Alabama and this is exactly what it will look like at launch:

I'll go over the how the other parts of this work when we launch, but for now I want to focus on just the top section involving both future and past precipitation, something we're calling The Powculator: The Snow Forecast Calculator.

How The Powculator Works

The Powculator is Snowology's own custom automated adjusted blend of weather models that corrects for biases related to both the models and terrain. For precipitation we will be targeting an elevation that is 2/3 up the vertical rise of ski areas for what falls on average at that elevation and we will be updating our forecasts every 3 hours in order to take advantage of the most recent modeling rather than once a day fully manual forecasts.

We will also soon start work on a system to allow for manual adjustments to the blends, so that for instance if we think the GFS has a better handle on a storm 5 days out, we'll blend the GFS in more heavily. I know AI and machine learning is all the rage these days, but the fact remains that there is no reliable data for East Coast mountains that can be used to validate precipitation forecasts against. Basically that's hogwash. They can match airports and other weather stations using machine learning, but not the mountains which are an entirely different beast. Experienced humans still have the edge in tweaking the blends from day to day, but the repeatable biases will be programmatically adjusted and that functionality will be implemented at launch and it will handle most of the repeatable sources of error that can be addressed, many of the same things I do when I forecast manually.

Although we're calling this "The Powculator", given that it is the East we also need to accommodate slick pow (ice) and underdeveloped pow (r@!n). The goal here is to give everyone a solid yet simplified view of what has happened and what will happen over recent past and future. In the off-season this will become "The Rainculator" of course and it can be used for hiking, mountain biking, and other off-season activities at ski areas.

The Powculator will also keep track of past precipitation for us rather than relying on ski areas who only provide snowfall. For over 2 years I've been trying to find a good way to merge ski area reported snowfall with estimates for non-reporting ski areas but between the over-reporting, under-reporting, mistakes, non-updating, lack of standards, etc., it's just not possible to do accurately at all. Now if we can estimate accurately what fell then that would be superior, and if we can also add in the r@!n and ice; even better! Well, I'm here to tell you that we most definitely can and sizably so.

I did a test of our methods and a review of the results from the last snowstorm which included large areas of sleet, and after careful review of NWS spotter reports, reports from our own followers, snow stakes, photos, etc., I believe that at least 85% of the estimates were likely within 1" of the actual snowfall, and only 1 ski area was likely off by over 2". That's far better than a situation where in this same storm one resort reported 23" on their site yet 10 miles away another reported 6+" when they both received almost the same amount of snow.

You'll also no longer need to wait for 8AM the next morning for snow reports to update because this data will be updated every 3 hours with both past amounts and near future amounts being hyper accurate. So when it is snowing at midnight and you are wondering if the storm is performing to expectations, our site will be updated with these hyper-accurate past estimates along with the most recent modeling for the snow that has yet to fall in order to give you the best estimate possible outside of standing on the mountain yourself.

I'm confident that no one will beat our accuracy overall, both past and future, and for a ski weather service that's the absolutely most important metric for judging quality. This is also why we're sticking to the East Coast. This is where I am experienced and can add the most value, plus it's nice to have locals forecasting your weather, similar to how it is nice to have local ownership of your mountains. Think about that for a moment. Snowology is all about the East, and we've fostered a large and vibrant community online for the past 8 years, and we value our connections to the region, the people, and the ski areas we cover. We understand what is both unique and important to skiers and riders in this region.

When Will it Launch?

We have the individual resort weather pages already complete but we're working on our navigational interface which will include maps and lists, and we're also using this time to address those repeatable model biases with automation which will take our forecasting data into the 2nd generation. I expect this to launch within 10 days, but we will still have plenty of work left to do once we do.

After launch we plan to focus our attention on the following 5 things:

1. Adding hourly forecast data to ski area weather pages.
2. Adding "favorites" functionality so that you don't have to search for ski areas that you are mostly interested in.
3. Creating a system for manual adjustments to automated blends.
4. Creating maps with automation that show our forecast data.
5. Launching a Snowology app for both Apple and Android.

The work on these systems will surely never end, but we'll be working hard even over the next off-season to dial everything in with a focus on presentation, ease of use, comprehensiveness, and accuracy.

Knowledge Is Powder!

-- Matthew Scott