GeoAI Artificial Intelligence in GIS (Ismael Chivite, Nicolas Giner, Matt Artz (eds.)) (Z-Library)

Esri Press, 380 New York Street, Redlands, California 92373-8100 Copyright © 2025 Esri All rights reserved. e-ISBN: 9781589488458 The Library of Congress has cataloged the print edition as follows: The information contained in this document is the exclusive property of Esri or its licensors. This work is protected under United States copyright law and other international copyright treaties and conventions. No part of this work may be used, used to train, or used to prepare derivative works, copied, reproduced, redistributed, publicly displayed or performed, or transmitted in any form or by any means, electronic or mechanical, including photocopying, digital scanning, and recording, or by any machine learning, articial intelligence, information storage, or retrieval systems, except as expressly permitted in writing by Esri. Any unauthorized use of this work to train articial intelligence (AI) technologies is expressly prohibited. All requests should be sent to Attention: Director, Contracts and Legal Department, Esri, 380 New York Street, Redlands, California 92373-8100, USA. The information contained in this document is subject to change without notice. US Government Restricted/Limited Rights: Any software, documentation, and/or data delivered hereunder is subject to the terms of the License Agreement. The commercial license rights in the License Agreement strictly govern Licensee’s use, reproduction, or disclosure of the software, data, and documentation. In no event shall the US Government acquire greater than RESTRICTED/LIMITED RIGHTS. At a minimum, use, duplication, or disclosure by the US Government is subject to restrictions as set forth in FAR §52.227-14 Alternates I, II, and III (DEC 2007); FAR §52.227-19(b) (DEC 2007) and/or FAR §12.211/12.212 (Commercial Technical Data/Computer Software); and DFARS §252.227-7015 (DEC 2011) (Technical Data–Commercial Items) and/or DFARS §227.7202 (Commercial Computer Software and Commercial Computer Software Documentation), as applicable. Contractor/Manufacturer is Esri, 380 New York Street, Redlands, California 92373-8100, USA. Esri products or services referenced in this publication are trademarks, service marks, or registered marks of Esri in the United States, the European Community, or certain other jurisdictions. To learn more about Esri marks, go to: links.esri.com/EsriProductNamingGuide. Other companies and products or services mentioned herein may be trademarks, service marks, or registered marks of their respective mark owners. For purchasing and distribution options (both domestic and international), please visit esripress.esri.com. 194977

Contents Introduction ix Part 1: GeoAI Technology Overview 1 Part 2: Public Sector Applications 11 GeoAI Helps Stave Off Pest Infestation of Hemlock Trees 12 Forest Pest Branch, Urban Forest Management Division, Fairfax County, Virginia Using GeoAI to Inventory ADA Curb Ramps Saves Significant Time and Money 18 Douglas County and the City of Omaha, Nebraska Improving Roadways Using Drone Imagery and Machine Learning 23 Utah Department of Transportation Drones, Maps, and AI Unite City Functions 29 City of Vilnius, Lithuania As Sea Level Rise Threatens, Safeguarding a Sense of Place with a Digital Twin 34 Government of Tuvalu

Deep Learning Helps Automate Map Updates to Better Serve Citizens 39 Government of Kuwait GIS and AI for Precise Damage Assessments 46 Esri Getting the Most of GeoAI in Emergency Management 51 Esri Part 3: Private Sector Applications 57 Deep Learning Model Unlocks Potential of Solar Energy Development 58 Pivot Energy CEOs May Be Underusing this AI Capability 62 Bouwinvest Another AI Capability that Business Leaders May Be Overlooking 65 Esri Mapping New Possibilities for Business Success 68 Esri GeoAI, Reality Capture, and the Future of Digital Twins 72 Esri

Mapping the What-Ifs: Fertile Ground for AI-Powered Simulations 77 Esri Part 4: NGO/Nonprofit Applications 81 Drone Mapping Helps Find Flood Victims, with AI Assistance 82 United Nations World Food Programme GeoAI, Corporate Responsibility, and the Vigilance of a Climate Watchdog 91 Amazon Conservation Mapping Land Mines and Explosive Remnants of War 97 The HALO Trust Part 5: Next Steps 103 Contributors 109

Dedication “GeoAI—the integration of spatial analysis, AI, and big data—is creating new insights that promise to transform our understanding of the world.” Jack Dangermond Esri cofounder and president

Introduction Organizations across the globe have long relied on geographic information system (GIS) technology to manage and analyze data through the powerful lens of location, helping them tackle some of the toughest business and societal challenges. GeoAI—GIS enriched with articial intelligence (AI)—is already helping organizations get better, faster answers. What is AI? Simply stated, AI is the simulation of human intelligence in machines, training computers to recognize and detect patterns, extract information and meaning from data, and to solve problems and help make decisions through articial learning. Machine learning, deep learning, and generative AI are the categories of AI that are most relevant to GIS. Machine learning is an application of AI that allows machines to learn without being specically programmed to do so. It uncovers insights from data through methods incorporating decision trees or cluster analysis. Deep learning is a type of machine learning based on articial neural networks, which progressively uses multiple layers of information to extract higher-level features from the raw input. It uses more advanced methods and helps solve complex problems across large data volumes, with a focus on automated data extraction and pattern recognition.

Generative AI is a subset of deep learning that creates new data, such as text, images, videos, audio, and 3D models. Generative AI models learn patterns from existing data and use this knowledge to generate new and possibly unique outputs. GeoAI: AI in GIS GeoAI advances the science of GIS by using AI tools and models to automate data extraction and perform analysis on imagery, video, text, 3D, vector/tabular, time series, and other data. It extracts and classies features from unstructured text or imagery sourced from satellites, drones, aircraft, video feeds, and even mobile phones. GeoAI is also used to detect patterns, clusters, and anomalies in data, and to make predictions and forecasts. A key impact of GeoAI is making feature extraction and spatial analysis more widely available. This means better-informed decisions, more ecient operations, and the ability to tackle complex spatial problems that could have been previously out of reach for some organizations. Also, organizations that already use GIS extensively will benet from the ability to tackle complex problems by combining human GIS expertise with AI capabilities. This could lead to entirely new applications and insights that weren’t possible before. How can GeoAI help GIS professionals? Automated data extraction: AI helps GIS professionals by automating processes to extract useful GIS information from data. Object detection in aerial imagery and named-entity

recognition from unstructured text are two examples. Object detection in imagery helps emergency response teams quickly map locations of debris. Named-entity recognition helps law enforcement ocers process text documents in search references to events, people, and the like. Such tasks involve repetitive work; AI lets the machine do this work so that humans can focus their energy and expertise on more complex problem solving that the machine can’t do. And AI does this work more quickly and at scale. Deeper insights: ArcGIS includes many tools to perform analysis of geospatial information. With AI techniques, Esri is adding even more tools, providing GIS analysts with new options to identify patterns, make predictions, and ultimately gain better insights from data. AI brings together more powerful tools that, when used correctly, allow GIS users to do things we could not do before. Good examples include state-of-the-art machine learning tools for creating predictions using large, multivariate datasets and for making forecasts based on complex time series patterns. Multimodal analysis is also enhanced with AI, enabling analysis across unstructured text, images, and other data modalities not supported by traditional tools. Using GeoAI The combination of AI and GIS has already changed how leading organizations manage operations. GeoAI enables new levels of sustainability, eciency, and growth. The next section of this book presents an overview of current GeoAI capabilities, followed by stories illustrating how organizations are already using GeoAI in the public, private, and NGO/nonprot sectors. The book concludes with a ®

section about the next steps you can take to learn more about getting started with GeoAI. Learn more about GeoAI by visiting: go.esri.com/geoai_book

Part 1 GeoAI Technology Overview GeoAI integrates articial intelligence, or AI, with geospatial data, science, and technology to increase understanding and solve spatial problems. AI is the ability of computers to perform tasks that typically require some level of human intelligence and reasoning—through programming that continually adapts, infers patterns, generalizes, and improves output over time. We can use GeoAI for applications such as detecting and categorizing objects in imagery and lidar, identifying clusters and anomalies in data, and making predictions and forecasts. The intersection of GIS, AI, machine learning, and deep learning creates opportunities that weren’t available before. GeoAI oers new ways to evaluate numerous solutions to dicult spatial problems. Spatially explicit models incorporate an aspect of geography, such as location, shape, or proximity, into an algorithm, making the models more ecient, accurate, and representative of the reality we want to model. With these techniques, we can allocate resources based on meaningful spatial patterns and relationships, nd trends and anomalies in space and time, and incorporate spatial relationships into predictions and forecasts. Machine learning in ArcGIS Machine learning is a branch of AI in which computers learn patterns within data, and then use what they’ve learned to predict outcomes or make decisions. Machine learning algorithms are data-driven and

operate with minimal human intervention. As machines process more and more data, they are trained so that they automatically “learn” how to adjust their behavior and improve their performance based on previous experience. Machine learning shows up everywhere in our daily lives and across many industries. Product recommendations, trac alerts, social media ads, health-care diagnoses, fraud detection, predictive maintenance— all use machine learning in some way, shape, or form. Irrespective of the specic industry or application, the types of problems solved by machine learning generally fall into three main categories: clustering, prediction (which includes regression and classication problems), and forecasting. Machine learning in ArcGIS Pro In the context of ArcGIS technology, machine learning is far from new. In fact, machine learning algorithms have been incorporated within ArcGIS and used in geographic applications of these three categories for many years. For example, you can classify pixels within remotely sensed data using K-Nearest Neighbor or Support Vector Machine algorithms. Or you can apply decision tree ensembles, machine learning techniques that combine multiple decision trees to improve predictive accuracy, to classication problems with vector and tabular data using the Forest-Based and Boosted Classication and Regression tool. You can also take advantage of logistic regression and maximum entropy (Presence-Only Prediction—MaxEnt) approaches for predicting binary classication outcomes. For clustering problems, you have access to algorithms that group spatial data based solely on the data attributes (Multivariate

Clustering), their locations (Density-Based Clustering), or based on both the attributes and locations of the data (Build Balanced Zones). You also have access to a family of global regression models and decision tree ensembles for regression tasks, as well as the ability to apply decision tree ensembles to time series forecasts. Lastly, you can use Causal Inference Analysis to go beyond prediction and uncover the true causal relationships between variables. Behind the scenes, the algorithm uses machine learning to isolate the eect of a true exposure on an outcome from other confounding variables. An example would be isolating the eect of fertilizer (cause) on corn yield (eect) in the presence of other related variables, such as soil type, farming techniques, and environmental variables. Although all these algorithms are considered machine learning, there is a fundamental dierence between applying a traditional, nonspatial machine learning method to spatial data (such as, Forest-Based and Boosted Classication and Regression) and using true spatial machine learning. In the latter case, geography is incorporated directly into the mathematics of the machine learning algorithm through notions of shape, adjacency, orientation, contiguity, proximity, density, spatial distribution, and so on. Examples of spatially explicit machine learning algorithms include spatial autoregression and dierent types of geographically weighted regression, as well as spatially constrained multivariate clustering. Your choice of machine learning algorithm should always depend on the underlying problem you are trying to solve, the structure of your data, and your desired goals and deliverables. AutoML

In the past decade, machine learning has experienced rapid growth in both the range of applications it is used for and the amount of new research produced. Some of the driving forces behind this growth are the maturity of the machine learning algorithms and methods, the generation and proliferation of volumes of data for the algorithms to learn from, the inexpensive computers to run the algorithms, and the increasing awareness among businesses that machine learning algorithms can address complex data structures and problems. Many organizations want to use machine learning to take advantage of their data and derive insights, but there is an imbalance between the number of potential machine learning applications and the number of trained, expert machine learning practitioners to address them. As a result, there is an increasing demand to standardize machine learning across organizations by creating tools that make machine learning widely accessible throughout and can be used o the shelf by nonexperts in machine learning, as well as by domain experts. Recently, automated machine learning (AutoML) has emerged as an approach to address the demand for machine learning in organizations across all experience and skill levels. AutoML aims to create a single system to automate (in other words, remove human input from) as much of the machine learning workow as possible, including data preparation, data engineering, model selection, hyperparameter tuning, and model evaluation. In doing so, it can be benecial to nonexperts by lowering the barrier of entry into machine learning but also to trained machine learning practitioners by eliminating some of the most tedious and time-consuming steps in the machine learning workow. Deep learning in ArcGIS

Deep learning is available in dierent formats across the ArcGIS ecosystem, making it increasingly accessible for users of varying skill sets. Whether you’re interested in using ArcGIS Online to test a pretrained model or using the ArcGIS API for Python to create a custom model, there is an ArcGIS option for you. Whichever platform you choose, ArcGIS has the tools to help you accomplish your deep learning tasks. Pretrained deep learning models Training AI models is a time- and resource-intensive process, but ready-made pretrained GeoAI models automate the task of digitizing and extracting geographic features from imagery, point cloud, and text datasets. Manually extracting features from raw data, such as digitizing building footprints or generating land-cover maps, is time consuming. Deep learning automates the process and minimizes the manual interaction necessary to complete these tasks. However, training a deep learning model can be complicated because it requires large quantities of data, computing resources, and knowledge of how deep learning works. With pretrained models, analysts do not need to invest time and eort in training a deep learning model. The models have been trained on data from a variety of geographies. As new imagery becomes available, we can extract features and produce layers of GIS datasets for mapping, visualization, and analysis. Pretrained models can be accessed from ArcGIS Living Atlas of the World and other online repositories.

More than 100 pretrained models are already available, and even more are being developed every day, including the following: Image feature extraction and detection: extract features, such as buildings, vehicles, swimming pools, and solar panels, from aerial and satellite imagery. Pixel classification: classify land-cover satellite imagery. Point cloud classification: classify power lines and tree points using point cloud data. Image redaction: blur sensitive areas from imagery to comply with privacy policies. Object tracking: track moving objects, such as vehicles, in motion imagery. Named-entity recognition: identify or categorize entities from text. Additionally, foundational models such as Prithvi have been trained on geospatial data, and vision-language models (such as OpenAI’s GPT-4 and GPT-4o, as well as Llama) have been integrated and now bring generative AI capabilities to GeoAI. Deep learning geoprocessing tools and wizards in ArcGIS Pro ArcGIS Pro is a desktop application that includes deep learning tools with the ArcGIS Image Analyst, ArcGIS Spatial Analyst™, and ArcGIS 3D Analyst™ extensions, as well as in the GeoAI toolbox. ArcGIS Pro has deep learning capabilities within a suite of geoprocessing tools. This familiar environment provides an intuitive