If there were one central theme to this year’s SatSummit, it would be: “machine learning will find all the things, but we need more training data.” It is clear that many teams are finding new and useful ways to leverage enormous amounts of imagery. Against a backdrop of copious compute and ingenious, learning machines, the remote sensing community is steadily moving towards a feature extraction utopia. That said, its fair to say we are not there yet.

Here are a few trends (in no order) to pay particular attention to:

  1. Analysis Ready Data (ARD) is an expectation. We will enjoy more products with simpler data structures, consistent metadata, common understandings of geography and time. The opportunity to combine disparate sources based on their geography and capture time is hugely compelling. ARD may be more interesting for the commercial sector who want to sell more products. However, the government Earth Observation (EO) sector is moving very quickly and deliberately. Seemingly there is a growing desire to publish data more expediently. A case in point is NASA’s commitment to the open data cube.
  2. Get on board with the Spatio Temporal Asset Catalogue (STAC)!
  3. ARD is removing the need for data preparation activities (snort!). Remote sensing people are now investing that time in creating training image chips for machine learning.
  4. In an interesting twist, those same RS people are creating lots of image chips so in the future they will not have to. Once the machines are “trained” there will be a diminished need to train further. Perhaps, this activity will become evolutionary, supporting ML to discern features as they change? Training is a “now” problem.
  5. Time is as difficult to solve as ever, it sounds so easy, but complexity abounds.
  6. Ethical discussions are beginning to emerge. Measuring assets from space could be an invasion of privacy. Is it fair to remotely assess the value of a farm?
  7. Most groups seem to be pointing in a similar direction. Our community is moving towards the goals of better feature extraction and management of enormous amounts of imagery. The cloud is playing an enormous role in this process, distributed. Elastic computing is essential to modern remote sensing.
  8. Developing in the open is OK. There is a great deal of open source code now readily available. However, even companies that maintain open source code are creating great businesses around deployment and management.
  9. Data pipelines are very popular but are still often custom built.

Thanks to the team at SatSummit (mapbox and development seed) for putting on another excellent event!

Sharing options

Observations • Chloe Papalazarou

Economic Indicator Monitoring using SAR

Variables tied to the distribution of resources and economic productivity in a market are known as economic indicators and colloquially known by names such as the Unemployment Rate…

gray conveyor between glass frames at nighttime

Observations • Will Cadell

Prescient Computing

Our community is perfectly positioned to evolve from analyzing landscapes retrospectively to looking forward. The next ‘killer app’ for geospatial isn’t an app at all, it’s probably the…

body of water under cloudy sky during daytime

Observations • Julien Jacques

Making Use of Geospatial Data

For over a decade, organizations have been heavily investing in GIS tools and capturing petabytes of geospatial data. These massive acquisitions of datasets show that geospatial technologies are…

Need a geospatial partner?

Our team complements organizations like yours—by providing on-tap access to geospatial, analytics, and mapping expertise.

Let’s talk

Join our team?

We’re always looking for skilled technologists to help us build the future of geospatial. Got a minute to find out more about us?

Working Here

Sharing options