How Dark Sky Works (Weather prediction using computer vision) (opens in new tab)

(blog.jackadam.net)

13 pointsthegrossman14y ago7 comments

7 comments

7 comments · 2 top-level

thegrossmanOP14y ago· 3 in thread

As promised in an earlier thread (http://news.ycombinator.com/item?id=3186978), I've written a blog post describing how Dark Sky operates under the hood.

lsllc14y ago

Fantastic, can't wait to get the app! I've always wondered about taking weather radar over time and creating a "heat map" of rainfall to prove that during the summer the t-storms seem to split in half, pass us by and rejoin, leaving me to go out and have to water the lawn while ALL the adjoining towns get inches of rain!

maxerickson14y ago

Have you compared the results of your computer vision approach to the velocity data provided by the NWS?

thegrossmanOP14y ago

You mean the wind velocity data directly from the doppler radar? Yeah... it turns out that wind velocity isn't super accurate for predicting where the storm is going.

If you watch a storm building up over time, you'll often see it staying in one place or even moving against the wind. Almost like a traffic jam on a highway.

So that's why we take a computer vision approach rather than basing it on wind-speed.

itsnotvalid14y ago· 2 in thread

I like the idea except it is U.S. only.

Okay back away from trolling, I thought these kind of modeling was something to be done on server side, but you app seems to be doing it on the client side. What's more, there are many more interested in these (including weather stations) things but I am guessing there is no way you guys are giving away the algorithm to make the prediction.

thegrossmanOP14y ago

We definitely want to expand it beyond the US! But dealing with each country's different network of weather radar is a daunting task.

As far as giving away the algorithm: Unfortunately, it isn't a single algorithm that can be packages nice and neatly. So I'm not sure how that would work.

But, we do plan on developing an API that 3rd-parties could use.

itsnotvalid14y ago

Since I know it would be impossible for a small team to develop such thing for every different observatory on the world, you may try federating with them one way to another to have them feed the data in the API.

However from my heart if such use is organized I feel ethically bad if the end result is not free of charge or expandable to other platforms.

j / k navigate · click thread line to collapse

7 comments

7 comments · 2 top-level

thegrossmanOP14y ago· 3 in thread

As promised in an earlier thread (http://news.ycombinator.com/item?id=3186978), I've written a blog post describing how Dark Sky operates under the hood.

lsllc14y ago

maxerickson14y ago

Have you compared the results of your computer vision approach to the velocity data provided by the NWS?

thegrossmanOP14y ago

You mean the wind velocity data directly from the doppler radar? Yeah... it turns out that wind velocity isn't super accurate for predicting where the storm is going.

If you watch a storm building up over time, you'll often see it staying in one place or even moving against the wind. Almost like a traffic jam on a highway.

So that's why we take a computer vision approach rather than basing it on wind-speed.

itsnotvalid14y ago· 2 in thread

I like the idea except it is U.S. only.

thegrossmanOP14y ago

We definitely want to expand it beyond the US! But dealing with each country's different network of weather radar is a daunting task.

As far as giving away the algorithm: Unfortunately, it isn't a single algorithm that can be packages nice and neatly. So I'm not sure how that would work.

But, we do plan on developing an API that 3rd-parties could use.

itsnotvalid14y ago

However from my heart if such use is organized I feel ethically bad if the end result is not free of charge or expandable to other platforms.

j / k navigate · click thread line to collapse