As a non-geospatial expert, I was fortunate to have my request to join the kickoff event for the “Generative AI for Geospatial Challenge” accepted. Since attending the mini-conference, I’ve wanted to yell: “Now I get it!”

After years of being on the periphery, it was the first time I really got ‘it’ in my bones. The ‘it’ being something many of us have been hearing for a while: the Geospatial ecosystem holds a lot of promise for St. Louis.

Hosted by the Taylor Geospatial Institute (TGI) and sponsored by Amazon Web Services (AWS), the Generative AI for Geospatial Challenge is now officially launched and, until November 29th, you can compete for a share of the $1,000,000 of AWS credits being awarded.

The most likely winners will emerge from teams of experienced Geospatial data professionals. For obvious reasons, these teams will be better equipped to address the award criteria, weighing factors like novelty, impact, and commercial implications.

Nonetheless, to my mind, the notion that you, I or anyone with sufficient interest, could apply for this challenge became a fully realized possibility rather than mere boosterism. I’ll be the first to admit that, yes, geospatial talks can often be a little dry, but this time I found a number of discussions worth sticking with and persevering through, even when I was out of my depth.

What was really impressive about this event was how it brought a really complex topic down to… well, quite literally, Earth. And, in so doing, I left the venue thinking to myself, “wow, really, anyone could get into this stuff.”

1. Bringing Geospatial Data Back Down to Earth

The presentations were remarkably effective: they made the complex world of geospatial technology accessible to newcomers like me. That’s partly because of the “Generative AI” angle, which has opened up access to the data in the language of the layman.

Generative AI breaches the technical hurdles that formerly prevented customers from accessing geospatial information. Now, farmers, realtors, and other professionals can more easily leverage this trove of data around places and spaces.

Clinton Crosier, AWS Director of Aerospace & Satellite Solutions, demonstrated this accessibility by sharing practical examples of how satellite data serves everyday needs – or as he put it, “Making the World a Better Place from Space.” Instead of focusing on military or telecommunications applications, he highlighted how satellite data could protect whales by monitoring their movement near shipping lanes. 

2. Making the World a Better Place from Space

Crosier shared two stories of companies that showed how this technology helps solve ground-level problems. 

Latitudo40 used geospatial data streams to create a digital twin of Turin, the city in Italy. The company then used generative AI to “game out” different scenarios, helping urban planners make better decisions.

Degas was using geospatial data to help local farmers optimize their crop production for economic growth in Africa. Degas used generative AI technologies to help farmers query the data and match their crop selection planning to market/pricing data to predict both yield and profits.

These real-world applications hint at a key shift in the industry. As Crosier explained, “Gen AI can think outside the boundaries of its initial programming.” This capability is transforming how we interact with geospatial data – moving from complex technical tools to more intuitive interfaces that anyone can use.

3. “What is Where?”, that is the Question 

For those familiar with AI technologies, it’s worth noting that speakers used ‘Generative AI’ broadly to describe various AI capabilities in geospatial work. 

While traditional Machine Learning (ML) handles much of the heavy lifting in analyzing satellite imagery and geospatial data, Large Language Models (LLMs) are adding new capabilities – particularly in making this data accessible through natural language queries. The speakers were especially excited about Multi-modal LLMs, which can process not just text, but also images and video data all at once.

Mason Grimshaw of Ode Partners/Clay cut through these technical distinctions to focus on what matters most: the practical impact. As he explained, geospatial data has always served one fundamental purpose – to answer the question, “what is where?”

4. Queryable Earth

Prior to LLMs, Grimshaw explained that we would need sophisticated GIS programs and considerable technical abilities to answer the question of, “what is where?” Unlike the world wide web which has always been searchable with natural language text prompts, the possibility of a “queryable Earth” has only recently emerged since LLMs.

Previously, traditional Geospatial data used in things like Google maps, couldn’t answer questions like, “where is grass?” Or, similarly, where are there football fields, solar panels or forests? 

Now, with open source platforms like Clay, dubbed an “AI for Earth”, these questions become answerable through a text chat interface. Essentially, Clay can convert millions of geospatial data points into recognizable meta-textual entities, which are then queryable and can be retrieved via search engines and LLMs.

So, for example, a question that cannot be effectively answered by a search engine, such as, “show me where all the marinas are on the California coastline”, is now easily retrievable information. With such plain language information retrieval capabilities on the way, it’s easy to imagine how that will democratize the access to and use of geospatial data.

5. Planetary Simulation

All the speakers at the “Generative AI for Geospatial Challenge” kickoff were excited about the future possibilities of applying generative AI to other geospatial technologies such as a ‘Queryable Earth’ and ‘Digital Twins’. Admittedly, the use case of Digital Twins took awhile for me to get my head around, but some comments from their panel discussions really added context and color to these high-tech concepts.  

Chris Holmes from Planet said that we can think about a queryable earth as the capability to “chat with a map.” Although it sounds simple, Holmes emphasized that there’s still plenty of theoretical space for this functionality to improve and develop. 

Jason Gilman from Element 84 pointed out how the ability to chat with a real time data map, could be incredibly important in crisis response situations. For example emergency service workers could ask questions most of us have never considered, like, “where should I park in a forest fire?”

6. Real-Time Digital Twins

Peter Doucette from USGS EROS, explained that extrapolating future conditions from present data is “generative” and asserted that Digital Twins ability to ingest all kinds of real time data streams will massively open up the field of scenario planning. 

Echoing similar sentiments, David Page from Oak Ridge National Laboratory explained that the difference between Digital Twins and Digital Maps is analogous to the difference between the MapQuest of yesteryear, and how we use Waze nowdays. He said that Mapquest was a Digital Model (i.e. a map is a model for representing the world), whereas Waze is a Digital Twin because it reflects real world traffic and re-routes your journey based on real-time traffic data.

7. Grand Challenges, Emerging Ambitions

Shaliya Dehipawala from Scale AI described how the applications and use cases of Artificial Intelligence have historically progressed in fits and starts, yet they will need less and less supervised training data in future. To illustrate by example, he said AI has provided another way of “seeing through clouds,” which implies that they have an internal model of the world that is not limited by present data.

It must be noted that overhead sensing has been able to penetrate clouds for decades using tools such as synthetic aperture radar and infrared sensors. For example, NASA has mapped the surface of Venus, which is always shrouded in a thick cloud layer.

However, the fact that AI has provided another way of sensing through clouds, illustrates how unexpected use cases emerge from this technology.

Reiterating this idea, Hook Hua from the NASA Jet Propulsion Laboratory pointed to the emergent capabilities of multimodal LLMs (i.e. video AI) to derive an internal model of “physics” just by watching billions of videos. Hua suggested that, in future, the internal physics models seen in AI generated video could eventually replace the supercomputers NASA currently uses.

What this Means for St. Louis 

The Challenge Kickoff was a GREAT start, and conveyed an infectious enthusiasm for the potential of geospatial science, and generative AI, to boost the St. Louis community. It can make our city and innovation communities stronger. 

The talks at Generative AI for Geospatial Challenge kickoff conference introduced exciting ideas that were genuinely easy to grasp. I would love to see more conversation pitched at this “everyman” level, that empowers us to imagine interdisciplinary collaboration in future.

At least to me, the possibilities of St. Louis geospatial’s ecosystem in particular has felt slightly inaccessible to the broader public. Previous local conferences I attended (admittedly, this was a few years ago) were steeped in conversations around classified data and security clearances. 

With their parade of military figures, these events often left me skeptical. I wondered if the geospatial ecosystem could really generate significant opportunities for St. Louisans at large.

I’m not knocking those previous events —we need to convene high-level events to keep attracting ever higher levels of expertise to the region to grow our economy. Moreover, I would be foolish not to recognize that “security clearance to access classified data from satellites” is a foundation of the sector!

Nonetheless, I harbored private concerns that the geospatial industry was reserved for highly trained newcomers. It didn’t seem like a sector open to adjacent moves, so I suspected that most of us would not be able to participate in this future economy.

Now, my position has completely changed: I’m publicly optimistic rather than privately skeptical. 

The generative AI context for the discussion around geospatial challenges greatly expanded the applicability and relevance of professional experience in other sectors. It created a space for curiosity towards a highly technical and somewhat opaque industry.

The format provided was a perfect gateway into the geospatial ecosystem for anyone, from the tech-savvy tinkerer to the ordinary person. In the face of big global challenges, like climate change, urban planning, and conservation of the natural world, concepts like a “Queryable Earth” and “Digital Twin” confer a sense of personal agency and introduce the possibility of responsible design. 

I could easily imagine parents attending future conferences like this, eager to show their high-school aged children that complex global problems aren’t just the domain of experts anymore. They too could help design solutions to the problems they care about most.

This shift —from geospatial technology being the domain of experts to becoming a tool for everyone— represents a crucial moment for the industry and an opportunity for St. Louis. I’d like to congratulate all those involved in the Generative AI for Geospatial Challenge for capturing this moment of transformation and translating complex concepts into accessible ideas.

The democratization of geospatial technology has arrived right on time to make good on St. Louis’s investment in the sector. This kickoff event showed us that whether we’re experts or newcomers, we all could play a role in growing a local industry. I really hope TGI continues to expand their conference programming along these lines, making room for everyone under their ‘big tent’ conversation.