Series Recap

Let’s recap our journey through the series “Geospatial AI Agents & AI Models Demystified,“ Starting with our first article, and its restaurant analogy through to this final article, where we pull all the pieces together to provide a real example.

Article 1: Geospatial AI Agents & AI Models Demystified

Complexity Level: Beginner-friendly, conceptual.

Focus: We walked into a futuristic restaurant where the AI agent (waiter) takes your order and coordinates with specialized AI models (chefs) to deliver tailored solutions. We introduced the idea of geospatial AI, imagining a city planner asking, “Identify suitable land for expansion, avoiding flood zones.” The agent breaks this into tasks for vision, geospatial, analytics, and language models, but we kept it theoretical – no code or real-world data yet.

Restaurant Analogy: The waiter (agent) listens to vague requests, and chefs (models) prepare dishes like satellite imagery analysis or zoning checks.

Article 2: Geospatial AI Agents & AI Models: A Simple Example

Complexity Level: Basic, hands-on.

Focus: We stepped into the kitchen with a simple geospatial agent (geospatial_agent) that geocodes two landmarks in Paris (Eiffel Tower and Louvre Museum), calculates their distance, and creates a Folium map with markers and a red line. No satellite data or AI models here – just geopy and folium for straightforward mapping.

Restaurant Analogy: The waiter (agent) takes a basic order (“Show me the distance between two Paris landmarks”), and a simple chef (basic geospatial tools) prepares an easy dish (a map and distance report).

Article 3: Geospatial AI Agents & AI Models: Introducing Prithvi – A Real Geospatial Model

Complexity Level: Intermediate, introducing real AI models. Article 3:

Focus: We upgraded our order to assess flood risk in New Orleans, using a pre-trained geospatial model called Prithvi via terratorch. The flood_risk_agent geocodes New Orleans, attempts to load Prithvi, and simulates a flood risk score (0–100%) on a Folium map. As we went through the process, we hit complexities; the Prithvi’s setup is tricky, requiring PyTorch and facing potential dependency issues or reliability problems, so we kept the output simulated rather than real.

Restaurant Analogy: The waiter (agent) handles a more complex order (“Assess flood risk in New Orleans”), calling on a gourmet chef (Prithvi) who’s skilled but finicky – requiring special ingredients (dependencies) and sometimes delivering inconsistent results. We acknowledged the challenge and chose a simpler path.

This brings us to this article; our final in the series. Time to dive in a little deeper.

Real Example: AI Agents and Models Working Together

Let’s continue with our futuristic restaurant analogy; where AI agents (our trusty waiters) and models (expert chefs) will serve up a groundbreaking solution for New Orleans: mapping flood risks with precision and speed. Let’s see how the workflow we’ve built in this series transforms a real-world challenge, building on everything we’ve learned.

Please note, I have included the source code for this example at the end of the article.

The Order: A City Planner’s Request

Imagine a city planner in New Orleans walking into our restaurant and saying, “I need to identify areas at risk of flooding in the city”. Just like a waiter, the AI agent listens, interprets, and breaks this complex request into manageable tasks, coordinating with specialized models behind the scenes.

The Chefs at Work: Models in Action

In Article 2, we cooked up a simple geospatial dish, using basic geocoding and mapping to calculate distances and create a map of landmarks in Paris, like the Eiffel Tower and Louvre Museum. Now, let’s step up to a real-world challenge: detecting water and potential floods in New Orleans. Here’s how our AI agent and models tackle it, building on the lessons from Article 3’s exploration of the Prithvi model:

• Satellite Imagery Analysis (Vision Chef): Our vision model scans Sentinel-2 images from May 2024, focusing on green (B03) and near-infrared (B08) bands. Using the Normalized Difference Water Index (NDWI)—calculated as (Green – NIR) / (Green + NIR) – we identify water where the index is greater than 0. It’s like shining a flashlight to spot water on a map, keeping it simple and reliable.
• Risk Assessment (Geospatial Chef): A geospatial model layers this water data with historical flood patterns, marking blue polygons on a New Orleans map to highlight flood-prone zones.

This streamlined approach builds on Article 3, where we explored the Prithvi model – a more advanced geospatial AI – but ran into challenges with its complexity. Instead of wrestling with Prithvi’s “gourmet recipe,” we opted for NDWI’s straightforward method.

The Meal: A Transformative Output

As we show in the animation below, the AI agent after interaction with the AI Model, delivers an interactive map showing flood risk areas. Work that once took weeks for experts is now done quickly, and precisely.

Why this is Disruptive

Geospatial AI’s is about to revolutionize many industries. It’s many advantages include:

• Speed: Tasks shrink from months to hours, as our demo showed.
• Cost: Reduces reliance on expensive surveys, democratizing access for smaller cities or NGOs.
• Precision: AI spots patterns humans might miss, like subtle flood risks in satellite feeds.
• Democratization: Non-experts can leverage insights once limited to big firms or agencies.

From farmers analyzing soil health via drone imagery to disaster responders, being able to predict flood risk will prove to be a game-changer.

Limitations and the Future

Thanks to geospatial, AI, we have an exciting future ahead. But it is worth remembering, just like like human chefs, AI models lack true creativity – they remix existing data but don’t invent new paradigms. And while our waiter follows scripts, advanced agents are evolving to anticipate needs proactively, like suggesting flood mitigation strategies before a storm hits. As the “kitchen” (infrastructure) improves, expect every industry to experience this disruption.

Conclusion

This series has taken us from the basics of AI agents and models (Article 1) to practical geospatial examples (Articles 2 and 3), culminating in this real-world application for New Orleans.

The collaboration between agents and models isn’t just innovative – it’s revolutionary, reshaping how we solve problems. Geospatial AI is about to make every industry smarter, faster, and more accessible.

Matt Sheehan is a Geospatial 2.0 expert. He publishes a weekly Spatial-Next Newsletter which dives deeper into advances in the geospatial world, providing important news, opinions, new research and spotlights innovators. Subscribe to the newsletter here.

Source Code

Note, to use this code .. right-click on image .. save as .. and save the image locally .. next drop this into ChatGPT and use this prompt: “Can you pull the python from this image and give me that code.”

Sign up here for your Sentinel AP ID and Secret Key: https://services.sentinel-hub.com/ims/subscribe/emailSignup