alt Hacker NewsI believe NASA / EU provide daily satellite imagery for free (which is of relatively high quality too). I wonder if there's a way to take that data, and training some kind of image recognition model that figures out "movement" or something to the same end? Would be cool to see
Replies
The ESA (Sentinel) data is somewhat delayed and has a low revisit period (AFAIK 6 days, although you might get lucky and get more due to overlap the further away from the equator you get) and low resolution. The Sentinel-1 data (SAR, synthetic aperture radar) might be somewhat useful for this as ships should be more easily identifiable on it and you don't have to worry about cloud cover, but probably still less useful than the delayed crossing data.
Ships don't move that quickly; AIS data refreshed once every few hours would probably be more than good enough.
Read somewhere once that trading firms use satellite imagery of shipping to inform trading strategy. Don't know any more about it unfortunately but it sounds interesting.
I think sentinel-1 has a SAR instrument, it's very easy to see ships with that data
Funnily enough, I did find a few satellite sources at the beginning for the map background and noticed that all the ships seemed to be scrubbed from the image. It's an interesting idea, thanks for the comment!
The sources I used were:
- ESRI World Imagery[1] — free satellite tiles, high-res, but ships are stripped out from the imagery
- NASA GIBS - VIIRS[2] — near real-time daily satellite imagery from NASA, but resolution is ~375m so ships aren't visible anyway
- Mapbox Satellite[3] — high-res and looks great, but same deal — ships are scrubbed from the composited imagery
1. https://server.arcgisonline.com/ArcGIS/rest/services/World_I... 2. https://earthdata.nasa.gov/engage/open-data-services-softwar... 3. https://www.mapbox.com