That is awesome to hear! Building this has been a fun side project and I'm happy to hear it sparked interest for you as well.

Yes, there are multiple interfaces.

This is a nice interactive one: http://rammb-slider.cira.colostate.edu/?sat=goes-16.

There is also an S3 bucket where you can find all data as soon as it is available: https://registry.opendata.aws/noaa-goes/.

But the fun part is receiving it in your backyard, of course ;)

Thanks! The S3 bucket is how it should be done.

Depends on the mode the satellite is operating in. Typically: a full disk image every 15 minutes, the continental USA every 5 minutes, and two selected "mesoscale" regions (they will move around based on interesting events) every minute or even 30 seconds. Not all these products are relayed through this downlink though. On HRIT you'll find a full disk image across 7 different spectral bands every 30 minutes, and mesoscale images across 3 different spectral bands every 15 minutes.

Next to this data, it also contains relayed images from GOES-15 (GOES-West), Himawari-8, EMWIN data, and NWS forecasts and plots.

Hi everybody, cool to see this submitted on HN. I'm around for questions so ask away.

Also beware that the guide is a little out of date:

1) the listed antenna is no longer available, but there are comparable ones for comparable or lower price, and

2) the SAWBird has been generally available for a couple of weeks.

Very cool, having some of the parts already, I'd love to build one of these.

A couple (unrelated) questions:

1) GOES receives data from remote terrestrial weather stations too, right?[1] Is that aggregated and rebroadcast in a way we can receive it?

2) Is data from MODIS available similarly to GOES?

[1] https://ftsinc.com/fire-weather/products/axiom-dataloggers/a...

Great, if you end up building one, keep us updated (here or on Twitter).

1) Yes, this is DCS (Data Collection System) data. It's a very low bandwidth signal (300 or 1200 baud) that ground stations uplink to GOES on UHF. I found there are about 20k separate transmitters in some asset database I once found somewhere. Every one (or the set that's still active) transmits a tiny packet every N minutes (some every minute, others every 30 minutes). This data is also broadcast as part of the LRIT (GOES-15) and HRIT (GOES-16 and GOES-17) streams. There is no processing for this data in the goestools package yet, but you can get the originals and walk over the data. When I last looked at it, the format looked different per receiver, it is possible this is a per-vendor type of thing. Anyway, yes, you can get a copy of this data with the receiver described in the OP guide. See https://github.com/pietern/goestools/issues/11 for some more links.

2) AFAICT MODIS is a product derived from the Aqua and Terra satellites. Both are polar orbiting sats. You can look them up here https://www.wmo-sat.info/oscar/satellites/view/81 and it looks like they don't pack a "direct readout" transmitter, so you wouldn't be able to receive anything directly yourself. Compare this to the page for GOES-16 at https://www.wmo-sat.info/oscar/satellites/view/152 and you see there are a bunch of frequencies with details listed.

It's not. The SUVI data is not broadcast on the HRIT feed, only on the GRB (GOES ReBroadcast) feed. Receiving GRB requires more effort: minimum 10ft dish, dual circular polarized feed, DVB-S2 demodulator, and different software. Never tried this, but would make for a fun project.

Re: your second point, it is strictly synchronous, though since there are 8 GPUs per process (thus have 1 process per machine) the gradient reduction is done in 3 phases. First they are reduced within the process, then across processes/machines, and then broadcast within the processes.

I misphrased that point, agreed. It's not classic driver-driven synchronous training, as you would do in tensorflow. It's using all-reduce (not available in tensorflow yet, i think).

Hi Salvatore! :D

And there's even a tiny Redis dependency (optional though) in the code to generate these results. In particular the collective communication library needs a rendezvous phase where all nodes connect to their peers. Using Redis for this is one of the options. See: https://github.com/facebookincubator/gloo/tree/master/gloo/r...

Hey Pieter! Wow cool :-) Thanks for the info. See you soon!

Interconnect between the servers is 50 Gbit Ethernet (see section 4 of the paper).

Or set the OPTIMIZATION environment variable for even more control ;-)

Thanks, gents, I updated the post with that, and Pieter's correction that Redis keys are not Redis objects, but sds strings.

It's called optimistic for a reason. You assume there is low contention and configure your client to have random back off.