Open Data from space

ODCamp Day One_5

WARNING – liveblogging. Prone to error, inaccuracy and howling affronts to grammar and syntax. Posts will be improved over the next 48 hours

There’s increasing amounts of data available from remote sensing – satellite, aerial – and more existing data sets are being opened up. Some of it is about land covers, some about surface elevations and some about pollution modelling.

But the data is challenging. There’s a lot of it, and it often requires several steps to get to the information you actually want. You have to identify the area you want – and draw out the data you want.

Snapshots of places over time are useful – like seeing a forest shrinking over time. It’s a powerful image, but hard to show. It needs to be in an animated file format like GIF. Many satellites have regular coverage of the same areas, opening up to those sorts of portrayals of urbanisation or other landscape changes.

What’s the resolution of these images? What’s the smallest object you can resolve? Many common satellites have different sensors. Spacial resolution can come down to one or two metres. Others have it about about 30 metres. One picture every few days is the common frequency of capture. The cloudiness of the weather condition have an image on what is captured, too.

Getting land use information from Lidar

Using satellite data for flooding monitoring is a challenge – the frequency of capture, and the tendency of rain to go hand-in-hand with clouds make it hard. Aerial Lidar surveys are probably better – but are hard for ordinary people to make use of. It needs specialists – and specialist tools. It’s great that the Environment Agency are doing this, though. They tend to come in TIFF or geoTIFF. It’s been used to analyse crop growth across the country. Strengths of reflection data from Lidar could be used to automatically analyse crops, theoretically. Satellites actually measure radiance – but the sun’s location varies, clearly, so that complicates matters, but the strength of reflection can be analyses against known values for different plant types.

Qgis will extract a lot of information from Landsat images, and they provide a lot of bands at a 30m resolution. It will do some correction based on date and time. And you can use topographic data to correct for mountain shadows.

It also has some useful tools for classification. NextGis is good for identifying old growth forests.

The high-resolution data sets are less used, because they require so processing power to analyse. You can outsource that, but that’s very expensive, too. You’re talking about large amounts of money to access information that is freely given.

A lot of the work on the released data is down by academic or research scientist, who tend to be fairly open about their processes, but not necessarily the code used to do.

On the other hand, depending on your use, going to too high resolution just creates noise that complicates what you are trying to do.

Other uses of data

Are conservation organisations using the data? Yes, to some degree. JNCC is, for example. The CEH, too.

Open Street Map have imported some data – but that might not be useful, because it’s not as detailed as surveying data would be. And some automatically generated results – like vineyards at too high altitudes – do not inspire trust.

Rather than importing data directly, it would make sense to use major changes in the satellite data to trigger a focus investigation from on-the-ground surveyors.

Drones and disaster damage assessment

The World Bank paid drone pilots to overfly the damage from a recent disaster, giving an incredibly rapid damage assessment. The top down images were used for Open Street Map, the side-on images were used to assess damage on a buildit-by-building basis. This experiment proved that the tech has potential applications – but is a very tedious job. Can that be more automated?

In summary: loads of potential, but it’s too hard right now. There’s work needed in tools and cataloging and discovery.

Google docs notes for this session