Geospatial artificial intelligence (GeoAI) is an interdisciplinary field that has received tremendous attention from both academia and industry in recent years. We recently published an article that reviews the series of GeoAI workshops held at the Association for Computing Machinery (ACM) International Conference on Advances in Geographic Information Systems (SIGSPATIAL) since 2017. These workshops have provided researchers a forum to present GeoAI advances covering a wide range of topics, such as geospatial image processing, transportation modeling, public health, and digital humanities. We provide a summary of these topics and the research articles presented at the 2017, 2018, and 2019 GeoAI workshops. We conclude with a list of open research directions for this rapidly advancing field.
1st ACM SIGSPATIAL International Workshop AI and Deep Learning for Geographic Knowledge Discovery (GeoAI’17). Redondo Beach, CA, USA – November 7, 2017. DOI: 10.1145/3178392.3178408[PDF]
2nd ACM SIGSPATIAL International Workshop AI and Deep Learning for Geographic Knowledge Discovery (GeoAI’18). Seattle, WA, USA – November 6, 2018. DOI: 10.1145/3307599.3307609[PDF]
3rd ACM SIGSPATIAL International Workshop AI and Deep Learning for Geographic Knowledge Discovery (GeoAI’19). Chicago, IL, USA – November 5, 2019. DOI: 10.1145/3356471[PDF]
Fig. 1. The proposed framework for extracting and understanding the cognitive regions of urban metro stations.
Abstract: The significance of urban metro stations extends beyond their roles as transport nodes in a city. Their surroundings are usually well developed and attract a lot of human activities, which make the metro station areas important cognitive places characterized by vague boundaries and rich semantics. Current studies mainly define metro station areas based on an estimation of walking distance to the stations (e.g., 700 m) and investigate these areas from the perspectives of transportation and land use instead of as cognitive places perceived by the crowd. To fill this gap, this study proposes a novel framework for extracting and understanding the cognitive regions of urban metro stations based on points of interest (POIs). First, we extract the cognitive regions of metro stations based on co-occurrence patterns of the stations and their surrounding POIs on web pages by proposing a cohesive approach combined of spatial clustering, web page extraction, knee-point detection, and polygon generation techniques. Second, we identify the semantics of metro stations based on POI types inside the regions using the term frequency-inverse document frequency (TF-IDF) method. In total 166 metro stations along with more than one million POIs in Shenzhen, China are utilized as data sources of the case study. The results indicate that our proposed framework can well detect the place characteristics of urban metro stations, which enriches the place-based GIS research and provides a human-centric perspective for urban planning and location-based-service (LBS) applications.
Implications for urban planning
As Kevin Lynch stated in The Image of the City (Lynch, 1960), the skeleton of individuals’ mental images is formed by five types of elements in the city: paths, edges, nodes, districts and landmarks, which mediates in the interaction between humans and their environment. The first thing we want to emphasize in this study is that urban metro stations are also one type of such cognitive elements (i.e., landmarks) in cities; their properties as cognitive places should be considered in urban planning and design so as to match people’s cognition. In addition, our extracted cognitive regions of urban metro stations show diverse and irregular shapes, which indicates that unified physical distances frequently used in existing studies and planning practices cannot precisely define TOD precincts perceived by humans. To this end, what we suggest in this study is that urban planning practices should attach importance to “cognitive place” and “cognitive distance”, which load human experiences and perceptions toward the environments (Briggs, 1973; Montello, 1991). This is also coincident with the ultimate goal of urban planning, urban design, and smart-city construction, i.e., making better human societies and improving human lives (Shaw & Sui, 2019).
Timothy Prestby, Joseph App, Yuhao Kang, Song Gao. (2019) Understanding Neighborhood Isolation through Spatial Interaction Network Analysis using Location Big Data. Environment and Planning A: Economy and Space. DOI: 10.1177/0308518X19891911
Hidden biases of racial and socioeconomic preferences shape residential neighborhoods throughout the USA. Thereby, these preferences shape neighborhoods composed predominantly of a particular race or income class. However, the assessment of spatial extent and the degree of isolation outside the residential neighborhoods at large scale is challenging, which requires further investigation to understand and identify the magnitude and underlying geospatial processes. With the ubiquitous availability of location-based services, large-scale individual-level location data have been widely collected using numerous mobile phone applications and enable the study of neighborhood isolation at large scale. In this research, we analyze large-scale anonymized smartphone users’ mobility data in Milwaukee, Wisconsin, to understand neighborhood-to-neighborhood spatial interaction patterns of different racial classes. Several isolated neighborhoods are successfully identified through the mobility-based spatial interaction network analysis.
Keywords: Neighborhood isolation, human mobility, big data, spatial interaction
Spatial interactions between Milwaukee communities and their demographic composition. The brighter flows are places where more flows are overlapping/converging together.The community flows are overlaid on top of a cartogram distorted by the percent of the non-white population of the census block groups to promote a more socially just map.
Acknowledgments
The authors would like to thank Safegraph, Inc., for providing the anonymous mobile phone location big data support. T.P. and S.G. thank the UW-Madison Hilldale Undergraduate/Faculty Research Fellowship for their support for this research.
Abstract: The spatiotemporal variability in air pollutant concentrations raises challenges in linking air pollution exposure to individual health outcomes. Thus, understanding the spatiotemporal patterns of human mobility plays an important role in air pollution epidemiology and health studies. With the advantages of massive users, wide spatial coverage and passive acquisition capability, mobile phone data have become an emerging data source for compiling exposure estimates. However, compared with air pollution monitoring data, the temporal granularity of mobile phone data is not high enough, which limits the performance of individual exposure estimation. To mitigate this problem, we present a novel method of estimating dynamic individual air pollution exposure levels using trajectories reconstructed from mobile phone data. Using the city of Shanghai as a case study, we compared three different types of exposure estimates using (1) reconstructed mobile phone trajectories, (2) recorded mobile phone trajectories, and (3) residential locations. The results demonstrate the necessity of trajectory reconstruction in exposure and health risk assessment. Additionally, we measure the potential health effects of air pollution from both individual and geographical perspectives. This helped reveal the temporal variations in individual exposures and the spatial distribution of residential areas with high exposure levels. The proposed method allows us to perform large-area and long-term exposure estimations for a large number of residents at a high spatiotemporal resolution, which helps support policy-driven environmental actions and reduce potential health risks.
Abstract: What is the current state-of-the-art in integrating results from artificial intelligence research into geographic information science and the earth sciences more broadly? Does GeoAI research contribute to the broader field of AI, or does it merely apply existing results? What are the historical roots of GeoAI? Are there core topics and maybe even moonshots that jointly drive this emerging community forward? In this editorial, we answer these questions by providing an overview of past and present work, explain how a change in data culture is fueling the rapid growth of GeoAI work, and point to future research directions that may serve as common measures of success.
Moonshot (Editorial): Can we develop an artificial GIS analyst that passes a domain-specific Turing Test by 2030?
Keywords: Spatial Data Science, GeoAI, Machine Learning, Knowledge Graphs, Geo-Semantics, Data Infrastructure
Acknowledgement: we sincerely thank all the reviewers who contribute their time to the peer-review process and ensure the quality of the accepted papers.
Recently, Dr. Song Gao (Co-PI) received a NSF grant together with Dr. Qunying Huang (PI), Dr. Daniel Wright (Co-PI), Dr. Nick Fang (Co-PI), and Dr. Yi Qiang (Co-PI).
Title: A GeoAI Data-Fusion Framework for Real-Time Assessment of Flood Damage and Transportation Resilience by Integrating Complex Sensor Datasets
Abstract: Traditional modeling approaches for flood damage assessment are often labor-intensive and time-consuming due to requirements for domain expertise, training data, and field surveys. Additionally, the lack of data and standard methodologies makes it more challenging to assess transportation network resilience in real-time during flood disasters. To address these challenges, this project aims to integrate novel data streams from both physical sensor networks (e.g., remotely-sensed data using unmanned aerial vehicles [UAVs]), and citizen sensor networks (e.g., crowdsourced traffic data, social media and community responsive teams connected through a developed mobile app). The goal is to develop a framework for real-time assessment of damage and the resilience of urban transportation infrastructures after coastal floods via the state-of-the-art computer vision, deep learning and data fusion technologies. The project will also advance Data Science through multi-disciplinary and multi-institutional collaborations. The project is expected to improve the sustainability, resilience, livability, and general well-being of coastal communities by having a direct impact on the effectiveness, capability, and potential of using both physical and social sensor data. This will in turn enable and transform damage assessments, and identify critical and vulnerable components in transportation networks in a more effective and efficient manner. The interdisciplinary research team, along with students and collaborators from different coastal regions, will facilitate the sharing of knowledge and technologies from different socio-environmental contexts and testing the transferability of the research outcomes.
The project will harmonize physical and citizen sensors within a geospatial artificial intelligence (GeoAI) data-fusion framework with a focus on three research thrusts: (1) unsupervised flood extent detection by integrating UAV images collected throughout this project with existing geospatial data (e.g., road networks and building footprints); (2) flood depth estimation using deep learning and computer vision techniques combined with crowdsourced photos and UAV imagery; and (3) assessment of the impact on and resilience of transportation networks based on near real-time flood and damage information. The innovative methodology will be demonstrated and deployed through collaborative efforts in response to future flood events as well as several historical storms. The project will produce open-source algorithms for future educational use, raw and processed datasets and associated processing software, a mobile app to engage community responsive science teams, and three research publications.
The problem of discovering regions that support particular functionalities in an urban setting has been approached in literature using two general methodologies: top-down, encoding expert knowledge on urban planning and design and discovering regions that conform to that knowledge; and bottom-up, using data to train machine learning models, which can discover similar regions. Both methodologies face limitations, with knowledge-based approaches being criticized for scalability and transferability issues and data-driven approaches for lacking interpretability and depending heavily on data quality.
To mitigate these disadvantages, we propose a novel framework that fuses a knowledge-based approach using design patterns and a data-driven approach using latent Dirichlet allocation (LDA) topic modeling in three different ways: Functional regions discovered using either approach are evaluated against each other to identify cases of significant agreement or disagreement; knowledge from patterns is used to adjust topic probabilities in the learning model; and topic probabilities are used to adjust pattern-based results. The proposed methodologies are demonstrated through the use case of identifying shopping-related regions in the Los Angeles metropolitan area. Results show that the combination of pattern-based discovery and topic modeling extraction helps uncover discrepancies between the two approaches and smooth inaccuracies caused by the limitations of each approach.
Figure. The proposed framework of fusing knowledge-based and data-driven approachesFigure. Extracted shopping regions by combining data-to-knowledge and knowledge-to-data approaches.
[Madison, WI/USA] – [August 8, 2019] – Professor Song Gao as the Principal Investigator (PI) has been awarded an AI for Earth research grant from Microsoft to help further the efforts in the area of Geospatial Artificial Intelligence (GeoAI).
This new grant will provide Dr. Song Gao and his research assistants Yuhao Kang and Jake Kruse at the GeoDS@UW-Madison lab, and Dr. Fan Zhang (Postdoc Researcher at the MIT Senseable city Lab and Peking University) with the Azure cloud computing resources and AI data labelling services to accelerate their work on understanding the playability of cities and metropolitan areas from the human-environment interaction perspective using multi-source geospatial big data (e.g., images, texts, and videos).
The Microsoft AI for Earth is a $50 million, 5-year program that brings the full advantage of Microsoft technology to those working to solve global environmental challenges in the key focus areas of climate, agriculture, water and biodiversity. Through grants that provide access to cloud and AI tools, opportunities for education and training on AI, and investments in innovative, scalable solutions, AI for Earth works to advance sustainability across the globe.
Liu, Y., Yuan, Y., & Gao, S. (2019). Modeling the Vagueness of Areal Geographic Objects: A Categorization System. ISPRS International Journal of Geo-Information, 8(7), 306. DOI: https://doi.org/10.3390/ijgi8070306
Abstract: Modeling vague objects with indeterminate boundaries has drawn much attention in geographic information science (GIScience). Because fields and objects are two perspectives in modeling geographic phenomena, this paper investigates the characteristics of vague regions from the perspective of the field/object dichotomy. Based on the assumption that a vague object can be viewed as the conceptualization of a field, we defined five categories of vague objects: (1) direct field-cutting objects, (2) focal operation-based field-cutting objects, (3) element-clustering objects, (4) object-referenced objects, and (5) dynamic boundary objects. We then established a categorization system to formalize the semantic differences between vague objects using the fuzzy set theory. The proposed framework provides valuable input for the conceptualization, interpretation, and modeling of vague geographical objects.
Figure. The categorization system of the five categories of fuzzy regions and their relations.
Understanding human mobility is important in many fields, such as geography, urban planning, transportation, and sociology. Due to the wide spatiotemporal coverage and low operational cost, mobile phone data have been recognized as a major resource for human mobility research. However, due to conflicts between the data sparsity problem of mobile phone data and the requirement of fine-scale solutions, trajectory reconstruction is of considerable importance. Although there have been initial studies on this problem, existing methods rarely consider the effect of similarities among individuals and the spatiotemporal patterns of missing data. To address this issue, we propose a multi-criteria data partitioning trajectory reconstruction (MDP-TR) method for large-scale mobile phone data. In the proposed method, a multi-criteria data partitioning (MDP) technique is used to measure the similarity among individuals in near real-time and investigate the spatiotemporal patterns of missing data. With this technique, the trajectory reconstruction from mobile phone data is then conducted using classic machine learning models. We verified the method using a real mobile phone dataset including 1 million individuals with over 15 million trajectories in a large city. Results indicate that the MDP-TR method outperforms competing methods in both accuracy and robustness. We argue that the MDP-TR method can be effectively utilized for grasping highly dynamic human movement status and improving the spatiotemporal resolution of human mobility research.