Hacker News Reader

Comment by topherhaddad | Hacker News Reader

It is rather ambiguous, apologies. Imaging is a 2-dimensional measurement, so spatial resolution improvements follow a squared nature. Our GSD will be 10cm, so we arrive at 9x by (30^2/10^2). Said another way, nine 10cm pixels will fit in one 30cm pixel. The 9x refers to the visible imagery, whereas our thermal (LWIR) GSD will be 2m, and the best available today is 70m.

For thermal, we're only doing broadband 7.5um - 13.5um. You're correct in that direct emissions measurements is best made with SWIR. However, there are applications where heat signatures can be used to calculate emissions. For example, if you know the surface material and fuel type of a power plant, you can calculate carbon emission from the heat signature in the thermal image. Climate Trace will use our data for this.

View on HN

Thanks for the answer! It was clarified in the website, but the "9x" sentence was still ambiguous. Very often, image processing people will understand "resolution" as a linear unit, like dpi or gsd. Using its square seems strange, likely only useful for marketing purposes, and thus a red flag for technically minded people.

That said, 10cm per pixel is impressive if you manage to pull it through! Cannot wait to buy these images!

I totally agree the 9x is somewhat marketing speak, but I do think that the way GSD has historically been characterized by the 1-dimensional measurement does not do spatial resolution improvements enough justice. Electro-optical imaging systems are analogous to any digital sampling system. If you were sampling a sine wave at 90 Hz instead of 10 Hz, that's clearly a 9x improvement as you end up with 9x more samples. The same is true for imaging, you're digitally sampling an object with discrete pixels (of course there's more to overall image quality than just spatial resolution). Another side note is that the government NIIRS imagery rating system captures this squared effect through the log2(GSD) term in the equation. Photographers also think more about it in 2D, as they characterize resolution in total Megapixels in the image. Ok I'm done, I'll get off my resolution soapbox :)

I would agree that resolution is usually considered a linear unit (eg X cm/pixel). The improvement still looks great, but I also find the 9x a bit misleading.

You mentioned agriculture as an avenue, but I'm already aware of farmers using drones for such tasks. Also importantly, it can be done on the farmers' timescale, rather than the last time a satellite was overhead.

They monitor all sorts of things, soil wetness, crop/weed growth, weed identification, cattle location (they used to use helicopters here in Oz for such things)...

You also want to know your neighbor's data which you can't easily fly a drone over. And not just for competition.

Just imagine if you could get that as a service where you just open your phone and check on progress as needed. Better yet, you might get notifications when changes are detected so you can spend more time elsewhere. There is no reason that a set of satellites couldn't cover a single spot of the Earth once/more per day.

Hopefully they have good ML talent on hand that can help sort through the imagery and provide some form of automatic analysis for their customers.

The post specifically says that they're an imagery pure-play. That kind of ML automated analysis is something they envisage another business layering on top.

> provide some form of automatic analysis for their customers.

These automatic analyses are much better when they integrate data from several satellites. This seems a job best done by focused third parties that choose the optimal combination of satellites. A single satellite company would be inevitably biased towards using preferrably their sensors, to a detriment of analysis quality.

Reply on news.ycombinator.com