GeoAI Platform for Natural Language Exploration of Multidimensional Flood Risk Scenarios
This paper presents a GeoAI platform that enables floodplain managers to efficiently explore multidimensional flood risk scenarios by integrating disparate data sources into a unified queryable structure, utilizing a deterministic LLM agent to translate natural language questions into geospatial analyses, and providing a bidirectional map-chat interface to maintain spatial context throughout the workflow.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Coastal cities are living through a new reality where the threat of flooding is no longer a single event but a complex web of rising seas, heavier rains, and powerful storm surges. For decades, experts have used computer models to predict how high the water might get in the future. However, these models often exist in isolation, separate from the maps of where hospitals and schools are located, or the data showing which neighborhoods have the most elderly residents or the fewest cars. When a city planner or a community leader needs to ask a complicated question—such as which areas will need the most help evacuating if a massive storm hits in thirty years—they are forced to manually stitch together these disconnected pieces of information. This process is slow, requires specialized training, and often leaves decision-makers without clear answers when they need them most. The challenge is not just having the data, but having a way to ask questions that weave together weather patterns, infrastructure, and human vulnerability all at once.
Researchers at the Massachusetts Institute of Technology have built a new digital tool designed to solve this exact problem. They created a platform that allows anyone to type a natural language question about flood risks and receive an immediate, map-based answer that considers dozens of different future scenarios. Instead of forcing users to learn complex software or navigate through separate databases, this system acts as a bridge between human curiosity and massive amounts of scientific data. The team tested their system using real-world data from Broward County, Florida, a region that has spent years planning for climate change. They found that the platform could successfully answer complex, multi-part questions in seconds, turning a task that usually takes hours of manual work into a simple conversation.
The core of this new system is a massive digital library that holds 141 different flood scenarios. These scenarios are not just guesses; they are precise computer simulations that combine different levels of rainfall, rising sea levels, and storm surges. The simulations also account for two different stages of future city improvements: one stage representing near-term fixes like better drainage, and a later stage representing major upgrades like higher seawalls and advanced pumping systems. Alongside these water models, the system stores detailed records of the county's critical facilities, such as fire stations and hospitals, as well as socioeconomic data that reveals how vulnerable different neighborhoods are based on income, age, and disability rates. All of this information is organized under a single, unified structure, allowing the computer to look at a specific street and instantly know the flood depth, the nearest shelter, and the demographic profile of the people living there.
To make this data accessible to non-experts, the researchers built a smart assistant that listens to questions typed in plain English. When a user asks, for example, "Show me the hospitals that will be underwater during the worst-case flood in 2050," the system does not just search for keywords. Instead, it breaks the question down into a series of logical steps. It first identifies exactly which of the 141 scenarios matches the description of "worst-case" and "2050." It then locates the hospitals within the user's chosen area and calculates how deep the water would be at each location under those specific conditions. Finally, it ranks the hospitals by danger and displays the results on an interactive map. This process happens automatically, with the system checking its own work to ensure it is using the correct data and not making things up.
The researchers tested this platform with nearly 200 different questions to see how well it performed. They found that the system worked correctly more than 90 percent of the time, providing accurate answers in less than twelve seconds on average. When it did fail, it was usually because the computer ran out of memory while trying to process a very large area, or because an outside data source was temporarily unavailable, rather than because the system misunderstood the question. In the rare instances where the system chose the wrong scenario, it was because the language used in the question was ambiguous, such as using the phrase "worst case" which can have different meanings depending on the context. Despite these minor hiccups, the results demonstrated that the tool could reliably handle complex, multi-layered inquiries that would typically require a team of specialists to solve manually.
What makes this work particularly significant is how it changes the conversation around climate adaptation. In places like Broward County, where thirty different municipalities share a single resilience plan, decision-makers often struggle to align their strategies because the data is scattered across different departments and file formats. This platform collapses those barriers, allowing a planner to ask a single question and get a comprehensive answer that includes flood depths, infrastructure status, and social vulnerability all at once. The system does not replace the need for expert judgment, nor does it eliminate the uncertainty inherent in predicting the future. However, it removes the friction of data collection, allowing planners and community organizers to focus on the solutions rather than the struggle to find the numbers. By turning a multidimensional scientific problem into a simple dialogue, the researchers have shown that advanced flood modeling can be made accessible to the people who need it most.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.