Introducing Alpenglow’s Own Sunrise & Sunset Forecast Model
For nearly 10 years, Alpenglow’s sunrise and sunset quality forecasts have been powered by SunsetWX, a service that generates predictions using the NAM and GFS weather models.
SunsetWX has been a major contributor to Alpenglow’s success. It made high-quality forecasts accessible without requiring deep meteorological knowledge, and allowed Alpenglow to grow from a small side project into something people rely on when planning photography, travel, or just a good evening outside.
I’m incredibly thankful for the work SunsetWX has done, and I genuinely hope it continues to be around for a long time.
But over the last few years, we've been working on Alpenglow’s own model behind the scenes.
The Risk of a Single Dependency
Any app that depends on a single external service is only as reliable as that service.
I’ve always known this, and it’s been one of the main motivators for de-coupling Alpenglow from a single dependency. If SunsetWX goes down, Alpenglow users don’t see any forecasts — and they quite reasonably turn to me for answers.
Recently, I’d been putting the final touches on Alpenglow’s model, preparing to test it in a handful of smaller countries before rolling it out more widely.
Then, on Saturday night, just before going to sleep, I noticed forecasts weren’t loading in Alpenglow — and the SunsetWX site was down.
I flipped the switch to start using Alpenglow’s own model and held my breath.
Things went smoothly, and Alpenglow continued working with only a brief period of downtime.
SunsetWX Isn’t Going Away
I’m glad to say, SunsetWX is back online.
I plan to introduce an option in Alpenglow to choose your preferred forecast source, similar to apps like Carrot. If SunsetWX’s model performs better in your location, you’ll be able to continue using it.
The goal here isn’t replacement for the sake of it.
It’s resilience, flexibility, and long-term improvement. And in the future, other models could also be introduced.
Why Build a Model at All?
With control over the model, it can be adjusted and tuned over time. If new weather data sources become available, they can be incorporated without waiting on a single external dependency.
It also unlocks features that simply weren’t possible before. Many of these will start rolling out over the coming months as I celebrate Alpenglow turning 10, some of them have been on my wishlist for years, and others have been common user requests.
I plan to share more about Alpenglow’s model in the future. I wasn’t planning on having it out in the wild this early in the year, but as a friend put it:
“Sometimes we need motivations like these to make us get things live.”
