Discover Kayrros ERCOT Impact Monitor at ERCOT Market Summit ➝ Get the most timely and accurate information on capacity additions to the grid, with cutting-edge satellite imagery and AI.


How do satellites know where to look?

Social media bring value to satellite alert systems through natural language processing

Imagine a satellite orbiting the globe, quietly taking snapshots of the Earth’s oceans and cities to cover the entire world, often in less than one hour. Now, almost 2000 satellites are operating at this pace, providing imagery with higher-resolution and faster speed than ever before. As data from space become more varied, and satellite imagery becomes commercially available, new applications are starting to emerge — from 3D mapping for wildfire prevention to realtime natural disaster response. But there’s one challenge. When small disruptions hit without warning, how does a satellite know where to look? Social media has something to do with it.

At the intersection of the human language and cutting-edge technology is natural language processing (NLP), a sub-division of machine-learning that deals with “unstructured data,” better known in this case as the human language. To put it simply, NLP is the process of training an algorithm to understand text-based data — often coming from Twitter or other news streams — and attach meaning to it.

“Natural language processing is about data that are understandable by humans, but not by machines. So, we have to start from a human representation of this concept and move it into a mathematical universe where it means something to a machine. If two sentences have the same meaning but no word in common, how can you make a machine understand that it means the same thing? In other words, you’re trying to build a space where different sentences with one meaning are encoded in the same way,” explained Data Scientist Vincent Chabot.

At Kayrros, data scientists are building NLP algorithms to scrape, categorize and detect breaking news in realtime — often coming from Twitter — giving satellites in orbit a connecting point on the ground. The biggest challenge? Training algorithms to learn all the languages spoken across the energy industry… the largest industry in the world.

“Taking energy as an example, there are countless industry assets spread across the world, filled with people communicating in many different languages. When training an algorithm, first, there’s a language barrier. Then, you have the syntax; even within the same language, there are tons of ways to communicate the same message,” noted Data Scientist Matthieu Mazzolini.

Kayrros uses NLP to gain early insight on emerging energy supply and demand disruptions, tracking initial intelligence that can allow teams to analyze the impact of events as they unfold. Before news hits markets, the first signs of disruptions — like an explosion, for example — show on social media feeds, when people in the surrounding area hear a blast or see smoke. Being the first to catch wind of intelligence on an event allows Kayrros to provide market intelligence to clients ahead of the market.

Apart from guiding the lens of a satellite, the news gathered from social media also provides the story behind satellite imagery and other quantitative or alternative data that wouldn’t be obvious from analyzing the data alone.

“It brings context to data. If you just see a time series of satellite images, you might not understand the context surrounding them. But if you place this alongside relevant events in the news, you’ll be able to make sense of the data derived from the imagery. Context is something that you can’t guess just by looking at quantitative data,” Matthieu explained.

Kayrros is continuing to build on its proprietary technologies, leveraging geospatial data to bring transparency to operations of the world’s energy industry while also taking advantage of opportunities to apply similar techniques in other fields. Natural language processing adds a valuable layer to realtime monitoring by connecting the technology, algorithms and Earth observation with people on the ground.