Launch of the GeoAI Research Center

Ghent University in Belgium has launched an international research centre with the aim to address geospatial problems by improving spatial reasoning and analysing capabilities of artificial intelligence (AI) models. The GeoAI Research Center – part of the Faculty of Sciences, Department of Geography – is an interdisciplinary community with a sharp eye on balancing the benefits with the need for ethical considerations and security measures.

In a nutshell, the GeoAI Research Center in Ghent is about enabling machines to process and reason about geospatial data with capabilities that surpass human limitations. Nico Van de Weghe, a professor in GIScience & geoAI and director of the centre, explains: “AI still lacks the intuitive and contextual understanding inherent to human reasoning. For example, although computers currently perform many tasks faster than humans, they do not always do so as efficiently or intelligently as the human brain, which uses minimal energy for tasks like object recognition. The task ahead involves leveraging knowledge-based and data-driven AI approaches to tackle geospatial problems, allowing humans to focus on informed decision-making.”

Hybrid methodological approach

The GeoAI Research Center differentiates itself through several aspects. First and foremost, unlike many existing geoAI initiatives that focus solely on either knowledge-based or data-driven approaches, the centre emphasizes the development and application of hybrid models. Data-driven AI relies on analysing large volumes of data to identify patterns and make predictions using machine learning and statistical models. For example, it uses deep learning to classify satellite images based on pixel data. In contrast, knowledge-based AI uses structured information, rules and logical reasoning derived from domain knowledge, employing ontologies, knowledge graphs and expert systems. An example of this is applying geographical rules to infer land use changes based on predefined criteria. “We chose a hybrid approach because combining the strengths of both achieves more robust, accurate and contextually aware insights,” says Tim Van de Voorde, co-director and a professor in remote sensing. For example, a geoAI system that uses machine learning to analyse spatial patterns from satellite images (data-driven) and applies predefined rules to classify land use based on those patterns (knowledge-based), would be considered a hybrid AI system. ”The hybrid approach not only enhances the system’s accuracy, but also its explainability and reliability in decision-making. The integration improves model performance and transferability, and also training efficiency by reducing the required training data amount,” he continues.

Scope for cooperation

The centre’s scope for international cooperation is very broad. “A diverse collaboration is essential to advance geoAI technologies and their applications effectively,” emphasizes Haosheng Huang, who is also a co-director of the GeoAI Research Center, in addition to being a professor in GIScience & cartography as well as vice-president of the International Cartographic Association (ICA). “At the international level, we plan to bring together researchers, industries, NGOs and governmental agencies. We have also garnered interest from several companies. By working with industry leaders, we aim to address real-world challenges, develop practical solutions, and promote the widespread adoption of geoAI.”

The collaboration scope for the private sector extends from drone and satellite builders, cloud processing specialists and GIS developers, to big AI companies. Prof Haosheng Huang adds: “We’re exploring collaborations with other sectors, such as environmental organizations, urban development agencies and the agricultural sector. These partnerships will help tailor geoAI research and applications to meet the diverse needs of stakeholders, ensuring broader societal benefits.” Recognizing the skills gap in geoAI, the GeoAI Research Center will offer specialized training programmes and workshops for people with different skill levels and professional backgrounds, including geospatial professionals, data scientists, policymakers and students.

Making tasks easier

By 2027, the group leaders expect the integration of artificial intelligence into the geospatial sector to have significantly enhanced efficiency and effectiveness across various functionalities and tasks. This will allow humans to focus on defining problems, interpreting results and making actionable decisions. The projection for automated land surveying is that AI-driven drones and autonomous vehicles equipped with advanced sensors will automate data collection, while AI algorithms will process and analyse this data in real time. ‘Natural Language GIS (NLG) assistants’ are another promising development. These assistants will allow users to query GIS data and perform complex geospatial analysis using natural language. This advancement is expected to spark competition among both existing and new kinds of GIS players – with the new ones focusing exclusively on natural language geospatial interfaces – making GIS more accessible and user-friendly.

In the field of environmental monitoring and automated feature extraction, the 2027 projection is that AI algorithms will do all the work and will not only speed up the mapping process, but will also improve the accuracy and detail of geospatial databases. Consequently, quicker responses to environmental threats will be possible. The same goes for tasks such as predictive urban planning, where AI models use vast amounts of data to forecast urban growth, infrastructure needs and environmental impacts. Planners will be able to simulate various scenarios, making urban planning more proactive rather than reactive.

The integration of AI with location-based technology will include the use of computer vision for object and location identification, leading to commercial competitive advantages. In fact, real-time geospatial intelligence for business strategy will become the new normal; the proliferation of Internet of Things (IoT) devices and the use of generative AI enables organizations to analyse vast quantities of geospatial data generated from smart devices in real time. The adoption of cloud technology will facilitate the delivery of GIS services on a pay-as-you-go basis. This will lower costs for companies who want to differentiate their products with location services. Transformation due to hybrid geoAI will occur in the sports world too. AI-supported analysis of video data will provide easy insights into team dynamics, improving both individual skills and team coordination. And of course, augmented reality (AR) technology will grow in combining real-world geographic data with computer-generated overlays, enhancing spatial understanding and decision-making and allowing users to interact with their surroundings in innovative ways.

Potential risks

While offering significant advancements, open geoAI also presents potential risks and avenues for misuse. That underscores the need for careful consideration of ethical guidelines, robust security measures and regulatory frameworks. A specific threat is the speed and accuracy with which geospatial data can be analysed to identify vulnerabilities facilitating sabotage or terrorist attacks in critical infrastructure such as power grids or transportation networks. Cyber attackers could also use geospatial analysis to coordinate distributed denial-of-service (DDoS) attacks effectively. Additionally, there are concerns about misuse for aggressive military actions or unauthorized surveillance. Prof Haosheng Huang warns: “Even during periods without active conflicts, geoAI’s ability to generate realistic geospatial data like images or maps raises concerns about potential misuse for spreading disinformation. This could involve creating fake satellite images to support false claims about properties, land use, disasters, migrants’ movements or other significant events, potentially leading to public panic or influencing political outcomes. Incorrect or biased analysis of geospatial data could support inappropriate land development, mismanagement of natural resources and discriminatory practices.”

Privacy violations

Besides all the above, another significant concern is the risk of privacy violations and surveillance overkill with the widespread adoption of geoAI in commercial and public applications. Open GeoAI can be used to track individuals’ movements and activities without their consent, leading to serious privacy breaches. This could involve analysing location data from smartphones or social media to infer sensitive information about individuals, such as their routines and relations. Selling detailed spatial profiles to third parties to be combined with other person-related data increases the dangers. And of course, excessive government use of geoAI for surveillance and detection fuels concerns about violating civil liberties.

Prof Van de Weghe comments: “There should be clear distinctions between acceptable uses of geoAI for enhancing services, and invasive practices that infringe on privacy and civil liberties. At our research centre, we prioritize addressing these ethical concerns. We are committed to modelling the trade-offs between AI performance and geoprivacy, and developing best practices and case studies to ensure privacy law-compliant geoAI development. Regulations should mandate transparency in data collection, usage and sharing, with strict consent mechanisms. Developing AI techniques, such as anonymization and differential privacy, can help mitigate these risks. Public awareness and engagement are also essential to build trust and ensure the ethical use of geoAI. International cooperation might be necessary to manage cross-border challenges posed by geoAI, ensuring consistent privacy protections across jurisdictions.”

Challenges

One of the other key challenges for the research centre, besides ensuring funds and so on, is to make the vast amount of existing geospatial data ready for AI training and applications. Many of these of high-quality datasets are proprietary or subject to restrictive copyrights, which raises significant legal challenges – and also ethical ones, since determining the ownership and copyright of AI-generated works is complex. Potential solutions might be found in Linked Open Data, according to Prof Haosheng Huang: “Contributing to the development of Linked Open Data in the geospatial domain is essential, i.e. publishing and interlinking of structured data on the web using open standards and formats to foster a more collaborative ecosystem. Of course, robust data anonymization and encryption methods will be essential to protect user privacy.”

The GeoAI Research Center’s management realizes there are still many unknowns. One identified gap is the potential impact of climate change on geospatial data reliability, says Prof Van de Voorde. “As climate patterns shift, the historical data used to train geoAI models may become less accurate, necessitating new approaches to model training and validation that can adapt to rapidly changing environmental conditions.” Another challenge is the integration of real-time data streams with existing geospatial datasets. As an example, he mentions dynamic urban traffic management. While the collection of real-time traffic data, pedestrian movements and public transport usage from IoT devices is already happening, the effective integration of this data with historical traffic patterns to optimize traffic flow in real time is a field for further research. 

Nico Van de Weghe highlights an additional challenge: the importance of public engagement and policy influence in geoAI development. “Public engagement can make geoAI more accessible, foster informed discourse on its benefits and risks, and guide ethical research practices. We think that engaging with policymakers is essential to develop regulations that support innovation while protecting public interests and securing funding for research. The GeoAI Research Center will actively explore this path.”