Explaining How Dark Sky Works

Caitlin Dempsey


Dark Sky is a short term weather predictor protoype for iOS devices (more about the developers efforts to turn this into an app at the end of this post).  Dark Sky predicts the time and duration of weather at the local level within a window of 30-60 minutes into the future.  From the Dark Sky site:

Using your precise location, it tells you when it will precipitate and for how long. For example: It might tell you that it will start raining in 8 minutes, with the rain lasting for 15 minutes followed by a 25 minute break.

Dark Sky also allows for the exploration of past radar weather imagery.

Adam Grossman, one of the developers of this weather app, explains in a post how Dark Sky works to provide real time predictions of localized weather events.  The post explains the data source, noise cleanup process, and underlying assumptions about weather behavior.

Free weekly newsletter

Fill out your e-mail address to receive our newsletter!

The app pulls real time radar data from the National Oceanic and Atmospheric Administration (NOAA)’s National Weather Service (NWS) which operates 140 doppler radar stations across the United States.  The first step in the process is to clean up the data to remove noise created by bug and bird migration and ground clutter.  The developers use the Fast Artificial Neural Network C-library to remove 90-95% of the noise from NOAA’s data and then clean up the remainder of the noise use what they refer to as a “secret” process.

Dark Sky screen shot.

The rest of the post explains the process by which storm velocity is extracted and the app is able to predict and interpolate localized weather.  The app is very much a short term prediction for changes in the weather, relying on “a numerical & statistical approach, rather than a meteorological one.”  From Adam Grossman, “Well, here’s the thing: while the weather becomes chaotic and unpredictable at large timescales (hours to days), its behavior becomes increasingly linear at smaller and smaller timescales.”  The app won’t predict weather beyond the one hour window so it’s best applied in areas where the weather tends to change frequently and unpredictably.

The animated gif below shows a comparison of predicted versus actual radar image of a storm.

Predicted versus actual radar image of a storm from Dark Sky.
Predicted versus actual radar image of a storm from Dark Sky.

The developers of Dark Sky are currently soliciting funding via the app’s Kickstarter page. Pledges start at $1 and up.  The objective of the funding drive is to turn the working prototype in a app that can be made available to the public.

Photo of author
About the author
Caitlin Dempsey
Caitlin Dempsey is the editor of Geography Realm and holds a master's degree in Geography from UCLA as well as a Master of Library and Information Science (MLIS) from SJSU.