Live on Bloomberg TV

00:00The idea of the macro scope is it's a device for letting you see things that are too large for the human eye to see. Much like a microscope lets you see things that are too small to see. So if you imagine studying a problem like deforestation if you look at an individual tree you can very clearly watch the TV cut down but you don't know whether that's an isolated incident whether deforestation is getting worse whether it's a serious problem . On the other extreme if you look at the entire Brazilian rainforest all at once you'll see a few brown fuzzy patches but you don't know whether those are cornfields that have just been harvested managed plantations or true deforestation. So the idea of the macro scope is to be able to use computer vision and big data to actually look across large areas like the whole Brazilian rainforest or the whole of Indonesia and the whole world and understand trends understand major events by counting all the individual trees. So as we say it all inside it lets you actually count the trees to see the forest. You've said that the data evolution is modeling Moore's Law and geospatial is evolving that way meaning costs are coming down very rapidly as the ability to process data is going up very rapidly. Why do you make that analogy Well it's really quite quite a tight analogy because if we look at the price of satellite it's not very long ago satellites were hundreds of millions of dollars each. And nowadays we're seeing them for single millions and that's driven by many of the same trends that are causing cell phones to get cheap and in fact many of the same components that are in cell phones are now being face hardened than used to actually run small sats at the same time we have folks like Space X who are now building reusable launch rockets which radically reduce the cost of getting to space. So the components get cheaper the launch vehicles get cheaper as a result the price of the satellites are going down by over time. Orders of magnitude which is very much like the pattern we see we have seen over the last 20 years with semiconductors. Are you sort of data source agnostic meaning that any source of data coming from from orbit works or do you actually have some specific requirements . So in general or data source agnostic. Although the work the requirements that we have come directly from the problem we're studying and so if we're trying to predict say U.S. corn yield corn fields are very large. So we can use low resolution satellites we like to use many different spectral bands some of which are invisible to the human eye. If we're counting cars in Wal-Mart stores we need width we need satellite imagery where each pixel is less than a meter in size because cars are small enough that you just can't see them well unless you have sufficient resolution. So we we work backwards from the problem we're trying to solve from the event we're trying to occur and observe and understand and work back to figure out this resolution we need and ultimately we pull in as you said in your intro. Hundreds of satellites and pull in the strings of those very satellites into understanding the various trends across the Earth that we're studying oil whether it's cars in parking lots or levels and oil refineries all of that data has got to becoming very interesting to the investment community. What's the appetite on Wall Street for the product we've had it. We've had a huge amount of interest and especially speaking as a startup. We've said some two or three day sales cycles which is a wonderful thing when you're just getting the company going. So we've had interest from across Wall Street from hedge funds mutual funds banks and increasingly from other areas energy companies insurance companies major governments. Where does machine learning then enter the equation for you. James So the machine learning and computer vision is really important to us in two places. One of them is there's just way too many images for people to look at. So we actually train the computer to count cars planes trains ships to determine the difference between a managed forest and a virgin rainforest. So just taking what we call pixels to numbers taking large volumes of imagery and turning it into counts of things and then in the other areas in forecasting so early in the season we can forecast U.S. crop yield Brazilian crop yield in other places and that's also done by training a machine learning model to predict what the end of season harvest will be. Well obviously a lot of potential customers. What's the market size today . What's the growth profile look like so the overall market for fur all the satellite imagery vendors is in the small numbers of billions. We think over time as we as we work out more and more and start to publish more and more these signals we think it's tens of billions in value across all the industries and governments in the world .

No schedule data available.

Spotting Trends With Satellite Imagery

Do you ever get the feeling you're being watched? Well, you are. There are more than 350 satellites in space today observing Earth. San Francisco-based software maker Orbital Insights uses algorithms to spot trends in satellite imagery and draw conclusions on everything from customer traffic at shopping malls to inventory levels at Chinese refineries. In an interview with Bloomberg North’s Anthony Lacavera, James Crawford, Founder and CEO of Orbital Insights talks about the evolution of geospatial data. (Source: Bloomberg)