The Earth is awash in data about itself. Every day, satellites capture around 100 terabytes of imagery.
In today’s digital age, data is generated at an unprecedented pace, particularly when it comes to our planet. With around 100 terabytes of imagery being captured daily by satellites, the sheer volume of information can be both a blessing and a curse. It holds the potential for enhanced insights into environmental changes, yet it also presents significant challenges in how to analyze and interpret that data effectively.
Take California’s wildfire management, for example. A critical question facing officials and researchers is: How many fire breaks does the state maintain, and how have these measures changed over the past season? The answer is complex, enmeshed in intricate datasets and requiring sophisticated analysis techniques.
Traditionally, assessing this information involved manual inspection of images. As Nathaniel Manning, co-founder and CEO of LGND, notes, “Originally, you’d have a person look at pictures. And that only scales so far.” This method, while useful, does not have the capacity to keep up with the massive influx of satellite data.
In recent years, advancements in neural networks have changed the landscape. Machine learning now allows data scientists to train algorithms to recognize fire breaks in satellite images. Nevertheless, creating an accurate dataset through these traditional methodologies can be costly, often requiring hundreds of thousands of dollars for data that serves just one specific purpose. “You probably sink, you know, a couple hundred thousand dollars — if not multiple hundred thousand dollars — to try to create that data set,” Manning observes.
LGND aims to drastically reduce these costs and complexities by leveraging innovative technology. Bruno Sánchez-Andrade Nuño, co-founder and chief scientist, emphasizes their goal: “We are not looking to replace people doing these things; we’re looking to make them 10 times more efficient, one hundred times more efficient.” This efficiency is critical for organizations needing timely and accurate data to tackle pressing issues like wildfire management.
Recently, LGND secured a $9 million seed funding round led by Javelin Venture Partners, with participation from AENU, Clocktower Ventures, Coalition Operators, and other notable investors. This financial backing positions the startup to expand its offering of geographic data solutions.
The cornerstone of LGND’s technology lies in vector embeddings of geographic data. Unlike traditional approaches that render geographic information in pixels or basic vectors, LGND’s embeddings simplify the process of analyzing vast landscapes. While traditional vectors are flexible, interpreting them can require a deep understanding of spatial concepts or significant computing power. Embeddings compactly summarize spatial data, making it easier to uncover relationships between various geographical points.
Nuño explains, “Embeddings get you 90% of all the undifferentiated compute up front.” Essentially, these embeddings form ultra-concise summaries that streamline computation, ultimately expediting the analysis process.
When considering fire breaks, for instance, they may appear as roads, rivers, or lakes on a map. Each type has unique characteristics, such as a lack of vegetation and minimum width requirements depending on surrounding plant life. By employing embeddings, it becomes significantly easier to locate areas on a map that fit these descriptions accurately and swiftly.
LGND’s enterprise application is designed for large organizations needing to analyze spatial data efficiently. For more specialized requirements, the startup also offers an API that allows targeted queries. This dual approach ensures that various user needs are met effectively.
Manning envisions companies leveraging LGND’s embeddings to explore geospatial data in innovative ways. Imagine querying an AI travel agent for a vacation rental that has specific features — three bedrooms, proximity to excellent snorkeling, a white sandy beach, minimal seaweed during your trip, and no construction work nearby. Crafting a traditional geospatial model for such a query would be a daunting task, cumbersome and time-intensive. However, with LGND’s embeddings, the possibilities for efficient analysis expand exponentially.
If LGND can successfully democratize access to these powerful tools, it stands to capture a considerable share of a market valued at nearly $400 billion. This envisioning of data analysis could fundamentally reshape how businesses and individuals interact with geographical data. As Manning succinctly puts it, “We’re trying to be the Standard Oil for this data.”
Inspired by: Source
