We’re working on the next release of Cloudless Atlas, processing thousands upon thousands of huge Landsat scenes. Many techniques carry over smoothly from our cloudless MODIS project, but the number of pixels is on another level entirely – on the order of 20 trillion. We’re planning for a cluster of 400 servers just to manage the petabytes of data downloads. The main pipeline, though: that will be a cloud unto itself, with just over 1000 high-end machines, each applying our tuned processing kernels to millions of pixels per second. It’s been fun crunching the numbers to get this set up, but the best part is seeing the beautiful images we can produce:
Here’s the famously lovely city of Vancouver, Canada. It’s built on the delta of the Fraser river, which carries pale sand produced by the scouring of glaciers far inland. Just north of the city are the mountains that hosted the 2010 Winter Olympics; to the south are fertile valleys and the border with Washington State. (In fact, the peninsula in the lower left is the American town of Point Roberts, cut off from the US mainland.)
I’ve been looking at a lot of images like this one lately, adjusting various parameters. They’re the very first sample swatches for the cloudless Landsat project – areas carefully chosen to be tricky because of seasonal vegetation, mountains, clouds (having grown up just outside this image’s frame, I can assure you there are plenty of clouds here), different kinds of water, agriculture, and so on. Making basemaps is an exercise in balance.