Every high-resolution satellite in orbit is trying to get pictures of critical parts of Nepal. On Tuesday, we posted images of Kathmandu from WorldView-3, and the hills north and west of the city are filling in day by day. But ongoing collection is a tough job, and it will stay tough for three big reasons: orbits, weather, and terrain.

When you pan around a satellite map, it’s easy to imagine that our planet’s entire surface can be seen at once in high resolution. But it’s really a very patchy process. The only satellites that can cover the whole world in a day are low-resolution weather and environmental satellites. NASA’s Terra satellite, for example, carries a MODIS sensor whose data we use often:

Central Nepal (centered on Kathmandu) as it appeared from space on the 25th, about 35 minutes before the earthquake: almost completely cloudy.

Like most surface-imaging satellites, Terra’s orbit follows a cycle that carries it over each part of Earth in the local mid-morning. Practically every high-resolution satellite photo you’ve ever seen was taken between about 10 and 11 a.m., local time.

Here’s each morning between the earthquake and today. I’ve overlaid the footprints of the high-res images that DigitalGlobe is providing for Humanitarian OpenStreetMap Team efforts:

Clouds obscured almost the entire country on the 26th. The satellites were given a high-priority task to collect images anyway, trying to pick up anything useful through clear patches. (And one did – read on.) This much high-resolution coverage of a single region on a single day is usually only seen during wars.

The sky was relatively clear early on the 27th, more typical of the April–May hot season before the mid-year monsoon. The satellites were trying for the same general areas as on the 26th, and had better success.

Clouds returned on the 28th, making high-res collection impossible. The black gap is the edge of the low-res Terra satellite’s imagery swath.

Partly clear skies returned to the west on the 29th, and the satellites picked up large areas of the hills there.

It’s often surprising how much useful information a high-res satellite can pick up from what look like impossible weather conditions. The very first picture we processed after the earthquake, from WorldView-3, was almost completely overcast. It seemed like a picture that only a meteorologist could love. But when you zoomed in on Kathmandu, there were a few small gaps:

We processed the image and brightened it to account for the cloud shadows:

It was the first post-event imagery, and it was good enough for damage assessment in a few neighborhoods, like Sundhara:

Displaced people have gathered in Tundikhel and set up tents. In the lower left corner is the collapsed Bhimsen Tower.

Weather brings us to the reason why imaging satellites like the mid-morning window: there’s full sunlight, but water evaporating from the surface hasn’t yet formed afternoon clouds. In Kathmandu at this time of year (like in the hot season of many relatively humid climates), water vapor rises during the heat of the day, cools at altitude, and often forms rainclouds by afternoon.

Even in the clear weather on the 27th, some images pushed the envelope in another way. One of the pictures was from 42° off-nadir – looking way out to one side of the orbital track. From that angle it saw the sides of hills and buildings, not just the tops:

Imagery processing pipelines do terrain correction by taking a digital topo map of the area that an image covers and removing the perspective – “uncrumpling” the landscape. (You can also see the effect in the MODIS view from the 28th, in the jagged artifacts where the image edge drapes across the Himalaya.)

Terrain correction is especially difficult over Nepal, one of the world’s most rugged countries:

Nepal meets the edge of the Ganges river’s floodplains in the south, but virtually the only other flat land is the sediment-filled valley of Kathmandu. (Same view as the MODIS images above.) © Mapbox © OpenStreetMap.

Kathmandu’s elevation is about 1,300 m (4,250 ft). Only 60 km (40 mi) to the south are mango farms as close to sea level as Pittsburgh is, while the same distance to the north are Himalayan peaks higher than any mountain in North America.

Perfect terrain correction would need a perfectly accurate image location and a perfectly accurate topo map. Since you can’t have either in practice, an image from a large off-nadir angle – more than about 20° to one side – will never be quite as good as one from directly above. But WorldView-3 is uniquely good at knowing exactly where it’s pointing, to within a tiny fraction of a degree. That means more accurate “uncrumpling”, and is one of the special features that let it take useful pictures where no other satellite could. Its 42° off-nadir picture was usable – though in the end we didn’t publish it, because it collected an even better one of the same area, just a few seconds later in its orbital pass.

What’s next

Over the next few days, at every opportunity, satellites will collect as much imagery as possible over the affected areas of Nepal. We will continue to send the pictures through our pipeline to make them useful for the Humanitarian OpenStreetMap Team and its thousands of volunteers. You can contribute!

Header image: “Nagarkot Countryside” by Andreas Kollegger on Flickr.