The Earth is full of data about itself. Every day, the satellite captures approximately 100 terabytes of images.
But it’s not always easy to understand. Apparently simple questions are extremely complicated to answer. Ask this question about the important economic importance for California. How many fires does a state that could stop wildfires in that area have?
Nathaniel Manning, co-founder and CEO of LGND, told TechCrunch: In recent years, neural networks have become a little easier, allowing machine learning experts and data scientists to train algorithms on how to see a fire break in satellite images.
“You’re probably going to sink hundreds of thousands, if not hundreds of thousands of dollars, and try to create that dataset.
LGND wants to cut these numbers by more than a few orders of magnitude.
“We’re not looking to replace those who are doing these things,” said Bruno Sanchez Andredo Nuno, co-founder and chief scientist of LGND. “We’re looking to make them 10 times more efficient and 100 times more efficient.”
LGND recently collected a $9 million seed round led by Javelin Venture Partners, the company told TechCrunch exclusively. AENU, ClockTower Ventures, Coalition Operators, MCJ, Overture, Ridgeline, and Space Capital participated. Many angel investors also participated, including Keyhole founder John Hanke, Ramp co-founder Karim Atiyeh, and Salesforce executive Suzanne Dibianca.
The core product of a startup is vector embedding of geographical data. Today, most geographical information resides in either pixels or traditional vectors (points, lines, areas). They are flexible, easy to distribute and read, but interpreting that information requires a deep understanding of space, a small amount of computing, or both.
Geographic embedding summarises spatial data in a way that makes it easier to find the relationships between different points on the globe.
“Embedments acquire 90% of all non-undifferentiated computing in advance,” Nuno said. “Embedded is a universal, ultra-short summary that embodies 90% of the calculations that must be done anyway.”
Let’s take a look at an example of a fire. They may take the form of roads, rivers, or lakes. They each appear differently on the map, but they all share certain characteristics. For one thing, the pixels that make up the image of the fire have no vegetation. Additionally, the fire extinguisher must be a constant minimum width. This often depends on the height around the vegetation. Embedments make it much easier to find locations on the map that match these descriptions.
LGND has built an enterprise app to help large companies answer questions that contain spatial data, along with APIs that can be hit directly by users with more specific needs.
Manning sees embedding LGNDs encourage businesses to query geospatial data in a whole new way.
Imagine an AI travel agency, he said. Users may ask to find a short-term rental with three rooms close to good snorkeling. “But I also want to be on the white sandy beach. I want to know that there are very few ocean weeds in February, but perhaps most importantly, at this booking, no construction was occurring within a kilometre of the house,” he said.
Building a traditional geospatial model to answer these questions will take time for every query, not just every query.
If LGND can successfully reach such tools to the masses or those using geospatial data for jobs, it could take a sip from the nearly $400 billion market.
“We’re trying to become the standard oil for this data,” Manning said.
