Leveraging Geospatial Artificial Intelligence in Emergency Management

Artificial intelligence (AI) is often misunderstood in emergency management. On one end, practitioners are eager and ready to embrace AI’s capabilities that can help increase productivity. On the opposite end, the uncertainty of AI causes anxiety, resulting in a delayed adoption of the technology.

To better ensure that AI’s capabilities contribute positively to our work, however, we can establish boundaries and parameters to integrate into well-architected information technology (IT) systems. When adopting AI, its outputs become more reliable by enacting and incorporating standard IT practices (e.g., security, performance, scalability, and reliability), thus making the technology a trusted source to aid decision-making. While incorporating tried-and-true IT security parameters may help alleviate concerns, understanding AI’s capabilities and limitations remains the most significant barrier to full industry adoption.

Many emergency management organizations utilize AI’s text-generating capabilities—like large language models (LLMs) accessed via chatbots—for planning and preparedness tasks. However, extending AI’s use to more advanced applications feels more like the industry’s “final frontier.” Market research unveils one likely reason—there is a void or absence of accessible and easy-to-understand information on how AI could be safely applied throughout emergency management programs. Furthermore, we are often misled into thinking that AI is “new” when, in fact, emergency managers have had access to AI-enabled systems for nearly a decade.

It is true that technology is advancing rapidly, but with the right constraints, training, and application, AI can be used safely and effectively.

Esri has been working on integrating AI throughout ArcGIS for years. This blog article intends to share information and knowledge in an easy-to-understand manner, on ways emergency managers can leverage ArcGIS—an AI-enabled system—for pre- and post-disaster workflows.

ArcGIS—an AI-Enabled System—Designed for Modern Enterprises

In emergency management, AI can support preparedness, mitigation, response, and recovery. While generative AI models have grown in popularity, primarily through the use of LLMs to create new data and text, like natural language, emergency managers also have access to machine learning and deep learning models for things like predictive modeling, object tracking, and imagery analysis.

ArcGIS is a foundational enterprise technology used by both business and government organizations. As a geographic information system (GIS), it provides location intelligence, contributes to the digital transformation of an organization, integrates with the Internet of Things (IoT), helps create digital twins, and enables teams and organizations to collaborate, make decisions, and act. Emergency managers have been using this system for decades to map and understand the world around them.

However, what may surprise some emergency managers is that AI in ArcGIS started with the inclusion of machine learning tools in 2008. In recent years, ArcGIS has been integrating smart assistants, the Text Segment Anything Model, vision language foundational models, and most recently, new generative AI assistants.

In nearly all of these cases, AI was architected directly into ArcGIS without the need for any additional configuration.

This means that emergency managers who have utilized GIS for data analysis, imagery analysis, and prediction modeling, have been successfully integrating AI into their programs for nearly a decade.

Two Types of AI within ArcGIS

ArcGIS is enhanced with AI in two ways: Geospatial AI (or GeoAI) and AI assistants.

GeoAI advances the science of GIS by using AI tools and models to extract information from imagery, video, and data for analysis. It can analyze, make predictions, and extract features from unstructured text or imagery sourced from satellite imagery, drones, aircraft, video feeds, and even mobile phones. GeoAI can also track, detect, and classify objects for additional analysis. To get started and to deploy ArcGIS deep learning packages, check out this blog article: Deep Learning with ArcGIS Pro Tips and Tricks.

AI assistants enhance the ArcGIS experience by using intelligent agents and assistants that use natural language to understand the user’s intent and perform GIS tasks. These assistants employ LLMs that have been trained and bound to focus on ArcGIS, and can be likened to chatbots. AI assistants retrieve knowledge from your data, collaborate with the user by sharing their understanding and clarifying things as necessary, and then suggest actions relevant to your intent.

Let’s break down some scenarios where emergency managers would leverage these capabilities within their programs.

Applying GeoAI and AI Assistants into Emergency Management Workflows

Pre-disaster

One of the more common ways to leverage AI pre-disaster is for predictive modeling and simulations. In emergency management terms, leveraging ArcGIS to simulate coastal, riverine, or flash flooding uses GeoAI capabilities to conduct predictive analysis. Or, when organizations work to identify structures in a floodplain that should be considered for mitigation projects, the building footprints visualized in map layers can leverage GeoAI’s extraction capabilities. These extraction capabilities are pre-trained models to find and identify objects that look similar to one another.

For emergency managers interested in modeling how the climate may change in the future, check out the National Oceanic and Atmospheric Administration’s (NOAA’s) Sea Level Rise Viewer, which is based on ArcGIS.

For non-GIS experts who still need to identify and predict areas of risk, ArcGIS includes a library of pre-trained AI models in ArcGIS Living Atlas of the World. ArcGIS Living Atlas is the world’s largest collection of geographic information, including maps, applications, and data layers. Emergency managers can access these data layers and maps specific to their jurisdiction as well as tap into the system’s ability to do time-series forecasting, weather simulations, and more. Check out this site to access ArcGIS Living Atlas deep learning packages and get started.

GeoAI can also assist emergency managers with situational awareness. For organizations with watch centers or watch desks, GeoAI pre-trained models can be used to track moving vehicles or humans visible by CCTV. This capability is often utilized in security operations centers, where vehicles and VIP protective detail personnel are tracked, and their movements visualized on a screen.

Post-disaster

Imagery analysis for damage assessments is a GeoAI capability that emergency managers use to expedite the recovery process. This capability was successfully applied in the aftermath of Hurricane Ian, where geospatial damage assessments were conducted using artificial intelligence, crowdsourcing, and high-resolution imagery from satellite, air, and ground. This process resulted in $78.3 million in assistance to survivors without requiring an in-person inspection.?

Often, satellite, drone, or aircraft lidar images are collected after an emergency, showing aerial photographs of the extent and scope of damage. Through AI’s feature extraction and segmentation processes, viewers can classify and identify an object, like a building or structure, and the machine can be trained to identify similar structures throughout the image. These pre-trained models can then identify trends and patterns in the images of the disaster area and provide estimates of structure damage, extent, and debris amounts. Check out this Learn Lesson on Classifying Objects Using Deep Learning in ArcGIS Pro.

Using pre-trained models for object classification in ArcGIS Pro, GeoAI can detect patterns from imagery. These capabilities are frequently seen and used as change detection in images, like “swipe maps,” which show damage pre- and post-disaster. The models can also learn to quantify and classify the extent of building damage (i.e., minor, major, destroyed, etc.), estimate debris volumes, and blur objects that may be sensitive.

For a more interactive experience with post-disaster imagery, generative AI can be used via a chatbot in ArcGIS Pro. In this, users can prompt a map model with a question or statement in the prompt box, and the model will describe the image and trends for analysis in natural language. These capabilities are available through this Vision Language Context-Based Classification deep learning package that bridges ArcGIS Pro and OpenAI’s vision language classification models.

Lastly, ArcGIS Survey123 can be configured to extract information from a photo and used to populate data in a form, using an AI assistant and computer vision AI. Check out this video showcasing how AI assistants can be prompted to help generate a Survey123 form, post-disaster. Auto Translate also helps to translate survey components into many different languages (see the section below on Text Translation for more details).

Search and Rescue

Search and rescue missions can be streamlined using many of the same AI capabilities as damage assessments. As consumers of upstream data—like a building footprint data layer and aerial imagery collected pre- and post-disaster—search and rescue teams can quickly identify and prioritize search areas using deep learning pre-trained models. These models search for trends and indicators in the images where damage is most extreme and suggest a prioritized list of geographic areas that should be searched first as an output.

As search and rescue teams are deployed, mobile operators can then utilize the mobile capabilities of Survey123, ArcGIS QuickCapture, and ArcGIS Field Maps to feed real-time data from the field to an operations center, capturing geolocation, text, images, and video. As this information is fed back into a common operating picture, search and rescue teams can swiftly and methodically ensure all affected areas are searched accordingly.

To read more about how GeoAI assists in search and rescue operations, check out this blog article from Hurricane Ian and this animated map showing the mission’s progress over time.

Text Translation

Emergency managers frequently need to translate information into a variety of foreign languages. Preparedness materials, such as educational pamphlets, brochures, and videos explaining family and individual preparedness techniques, are often translated. Emergency alerts, messages, emails, and broadcasts are also translated to reach the community’s broadest number of people.

Recently, the Federal Communications Commission adopted requirements that participating commercial mobile service providers support template Wireless Emergency Alerts (WEA) messages in 13 of the most commonly spoken languages in the US.

One of Survey123’s latest features within ArcGIS is Auto Translate, which allows respondents to easily translate question labels, choices, hints, and all text into the languages of their choice using machine translation. The Auto Translate feature has been implemented into ArcGIS following strict privacy and security guidelines, ensuring no data is ever shared with third parties, complying with ArcGIS Online privacy and security requirements.

An End-to-End System

ArcGIS is an end-to-end AI-enabled system designed and architected to enhance productivity and accelerate decision-making. It provides capabilities for creating, managing, analyzing, mapping, and sharing all types of data, and delivers a decision-making edge through a geographic approach.

As an AI-enabled system, ArcGIS is designed for modern emergency management enterprises. It provides tools for automated data creation/extraction and processing, advanced spatial analysis and visualization, predictive analytics and forecasting, and real-time data integration and processing. AI in ArcGIS is charting a new course for the future of GIS, helping to reshape the world in unprecedented ways.

Want to get started using GeoAI or AI assistants? Here are two things emergency managers can do today: 1) Access the Living Atlas to begin using pre-trained models in ArcGIS. Seeing these models in action may help emergency managers become familiar with how GeoAI works and how its outputs can help inform their programs; and 2) Emergency managers should challenge their GIS professionals to explore the learning packages within ArcGIS Pro and begin incorporating GeoAI for all data, imagery, and modeling tasks.