Scaling Sustainability Expertise: Reliable Farm-Level Geospatial Workflows with Agents

Can an AI agent run a multi-stage geospatial workflow end-to-end against cloud-native data? And is it legitimate? Scaling complex geospatial analyses is critical to tackling sustainability challenges. Many of these workflows are site-specific and requires a human expert to address nuances. However, expertise doesn't scale. To begin to tackle this, we ran an experiment on an agricultural sourcing use case: a six-stage geospatial analysis across 117 properties and hundreds of fields, pulling and analyzing five geospatial datasets. We measured the agents' accuracy at each stage, and our main finding is that agent performance improves significantly when agents are given step-wise instructions rather than a single high-level objective. The cloud-native geospatial ecosystem is what makes this possible. Agents can query only the data they need, when they need it, without standing up a pipeline first. This access pattern makes site-specific, farm-level analytics possible. We see this as a promising pilot. With this approach, we can further the use of AI agents for scaling complex workflows, ideal for running site-specific analyses like field-level drought risk, flood risk, and to answer other sustainable sourcing and climate risk questions in hyper-local areas. We'll demo the approach, including where it breaks down. Presented by Nissim Lebovits geospatial software engineer at Radiant & Tristan Grupp agricultural data scientist at WRI. New York Climate Week (NYCW). Hosted by Cloud Native Geospatial Forum (CNG). Learn more here and here.