top of page
Solar Eclipse

Building a Solar Eclipse Forecast

On April 8th, 2024, a total solar eclipse was visible along a thin line arcing through North America. This rare astronomical event presented an opportunity to Excarta. We knew that many would be making plans to view the eclipse; we could help by leveraging our existing AI forecasting technology to let them find the best spot to view the eclipse 


To successfully view a total solar eclipse, a few conditions must be met: first, you must be within the path of  totality; second, the weather must be conducive. Bad weather can be a hindrance, especially cloud cover. Given that eclipses occur in a brief window of time,  it only takes a few clouds to completely block your view. Knowing the forecast ahead of time can help people prepare. This applies especially for those who plan to travel for the event. Many of our expected users would be booking flights or planning road trips to catch a glimpse of this phenomenon, and it would benefit them to take weather into account.  For example, Dallas and Austin were both inside the path of the eclipse, so those deciding between the two might prefer traveling to the one with clearer skies. 


One challenge we identified early on was that we were working with an immovable release deadline. If we missed it, we would have to wait until the next one– in 2044. 


In a few weeks, we planned, built, and released a full stack web application that would provide users with AI Weather Forecast Data. 


Backend


The Eclipse Tracker served data of two types to our end user: eclipse and weather forecast data. For the eclipse data, we implemented custom logic to calculate the eclipse’s magnitude–that is, the percent of the sun that was obfuscated by the moon– as well as the time of peak magnitude. To validate the output of these functions, we relied on NASA’s open source data set. 


For weather data, we relied on our AI weather ensemble model. This model provided the total cloud coverage at the peak eclipse time, as well as a confidence interval for our estimate. This data came from the forecasts generated by our global weather model. This model would be constantly rerun and reupdated as the event approached, improving the accuracy of its predictions. Each location would get the forecast for cloud cover at precisely the time of peak eclipse for that location.


AI weather forecasts produce different outputs from traditional forecasts, which would prove helpful for predicting weather 10 days out.  For a forecast this far out, a single deterministic forecast is not very useful -- in fact, if someone says they can tell you the exact cloud cover 12 days and 2 hours from now, that claim should be viewed with skepticism. The forecast we showed came from an ensemble forecast -- several runs of the weather model, combined to produce the likely cloud cover, and the uncertainty in the cloud cover.


Frontend


There was another challenge we faced from the outset— no other applications had provided this exact data set, at least not in a user-friendly way. This was perhaps due to the rarity of total solar eclipses. The lack of precedent was particularly challenging for UX design. What was the best way to display the location-specific weather and eclipse data for an entire continent without overwhelming the end user? 


We created a custom UI in our existing web portal to display the forecast and eclipse data. This UI required a custom map component, and this map component required two data overlays. The first represented the full path of the eclipse, arcing from Mexico through Texas and the midwest into Maine. The second was a gradient tileset to represent cloud cover, generated by our forecast API. 

When users clicked on a geographic location, the app would display weather and eclipse data specific to that area. Users could also use a search component, yielding the same effect. 


Excarta's solar eclipse forecast UI
Excarta's solar eclipse forecast UI

Release


The Eclipse Tracker launched three weeks before the cosmological event. In the time leading up to the eclipse, we experienced massive growth in user engagement.  The Washington Post reported on our tool during its official coverage of the rare solar phenomenon, driving significant traffic to our site.  We were also featured on social media platforms such as X and Reddit. 


Finally, the day of the eclipse arrived. The weather conditions confirmed what our model predicted. For some major cities along the path of totality, like Austin, TX, or Syracuse, NY, clouds blanketed the sky, partially blocking the view. Other cities, like Houlton, Maine, were lucky enough to have clear skies. Those who found pockets of visibility could gaze at this rare celestial marvel. 



Post written by Excarta engineer Rylan Lorance, April 2024

10 views0 comments

Comments


bottom of page