Geodata and AI
August 31st, 2025 | by Marius Dupuis
(6 min read)
Artificial Intelligence has made it into most areas of our life. It’s no surprise that the geodata domain is widely affected by it, too. In this article, we try to give our first impression on the current situation and activities.
Definition of terms
Artificial intelligence (AI) is an umbrella term for computer software which mimics human intelligence in order to perform tasks and learn from them.
Machine learning (ML) is a subset of AI. Whereas computer programs were implemented using detailed code and instructions, ML enables such programs to take in a (specialized or vast) set of data, derive principles aka patterns from that, and subsequently apply those principles. It is important to understand that most of the ML applications are using pattern recognition (e.g., by using multi-layer neural networks aka “deep learning”) to create similar results. A very common mistake is to think that general purpose tools are search engines just because they also have access to the internet but at the end this is again just set of data as input and pattern recognition is doing the rest. But to get good results the input data (training data) should be well-selected and labeled and supervised reinforcement learning is applied. Labeling explicitly tells which piece of data is correct and which is wrong. Everything else (which is used for general purpose tools) is unsupervised learning because the management and labeling of this vast amount of data can’t be handled (and it is no solution to use AI for generating training data). A good overview about the general terms are given in the blog of IBM.
Coming back to the definition of terms, the confusion arise because, as usual, by trying to give new things a name which is not precisely representing the purpose or the capability of the tool. In geodata domain we always have to attach “geo” to it to make it “special purpose”. The term GeoAI has emerged, and we should look at its definition. A good starting point is the explanation by ESRI:
Geospatial artificial intelligence (GeoAI) is the application of artificial intelligence (AI) fused with geospatial data, science, and technology to accelerate real-world understanding of business opportunities, environmental impacts, and operational risks.
This definition draws a bigger picture around the understanding of the use cases. A more hands-on approach is given at a blog called Geodata Insights:
GeoAI is the fusion of geospatial data (maps, satellite images, GPS coordinates) with AI techniques like machine learning, deep learning, and computer vision. This combination allows machines to recognize patterns, detect anomalies, and make predictions in ways that were once impossible.
Both definitions and a few others can also be found on a website of the International Cartographic Association’s Commission on GeoAI.
And now – what can we do with it?
Coming from the broad range of definitions, the potential set of applications seems unlimited – as is the case with most emerging technologies before the market starts some consolidation. So, let’s look at what we currently see in the pipeline:
Areas of applications that might benefit from a wider use of AI are
- Smart cities & Urban planning (AEC – architecture, civil and traffic engineering as well as construction)
- Disaster management & Humanitarian aid
- Agriculture & Precision farming
- Environmental monitoring & Climate change analysis
- Defense & National Security
- Management & Exploitation of natural resources
For them, geodata aspects that will be influenced by AI are
- Making geodata understandable by providing improved visualization
- Developing algorithms to analyze, interpret, transform and adapt information
- Scanning vast amounts of data to identify patterns and draw conclusions
This may, for example, lead to faster processing of raw data (e.g., from aerial imagery to map data), conditioning data for specific use cases (i.e., reducing information to relevant information), and performing quality checks on results.
Last but not least, it is expected that human access to geodata can be improved by having new AI-based means. Just think of prompt engineering that is already being used on a large scale in other areas, e.g. programming and authoring of vast amounts of training data. If this hope materializes, geodata could be made available quicker to a wider range of users.
Major beneficiaries of the expected speed and accuracy improvements from data acquisition to interpretation will be governments (e.g. for infrastructure maintenance), national agencies (e.g., for updating statistics data), insurance (e.g. for understanding the impact of a disaster), public health and safety, and business in general (did you ever wonder why some shops ask for your ZIP code at the cashier?).
A good, very specific example and quite some background information including market data are given in a report by PwC (Leveraging GeoAI: a strategic approach for utility companies).
Where there’s AI, there’s regulation
Especially in the EU, regulation is part of the AI business. We haven’t been able to identify a specific regulation on geodata in connection with AI, but existing regulations on both topics should apply. A good overview of AI regulation is given by Gcore (based in Luxembourg), another one is provided by the not-for-profit organization IAPP on their website. Especially if we are talking about location data (e.g., GPS traces) geodata can become personal data, because you can identify living, leisure and working places if you have enough traces available. In this case GeoAI at least falls under the General Data Protection Regulation (GDPR).
For geodata, regulation is highly dependent on the country whose data is to be processed, transferred etc. As we know, map data is considered a national resource, for example, in China and, therefore, highly restricted in terms of export and/or acquisition and processing. Only few licensed companies are handling this kind of data.
But, going back to Europe, there is, indeed, a good example of geodata regulation that might be correlated with AI. The EU regulation on deforestation-free supply chains (EUDR) will come into effect at the end of 2025 (for large enterprises) and mid 2026 (for small/medium enterprises), respectively. It obliges companies that import raw materials such as soy, coffee, palm oil or wood into the EU to provide precise geodata on the areas under cultivation. This brings us back to the previously mentioned aspects that AI may largely improve acquisition, accessibility and transparency of underlying data — or, in the negative case, that GeoAI might help to generate fake geodata to receive approvals that are hard to be revealed as fraud. Especially in a case like EUDR where not everyone obliged to provide proof of compliance will be an expert on geodata.
Other resources
For every problem, there’s a GitHub repository. This is also the case with GeoAI. Just have a look at the OpenGeoAI website and their link to the respective repository. We haven’t analyzed it further yet but it scanning the market for GeoAI tools and repositories might indeed be an interesting task for an upcoming blog post.
If you want to know a bit more about how LLMs can be taught the taxonomy of geodata and how LLMs can be used for geodata, we recommend the following videos from FOSS4G NA 2024 on Project Geospatial:
- Bringing Geospatial Awareness to LLMs Using Open-Source Software
- Applying Large Language Models to Geospatial Search and Analysis
And, finally, not topic would be to be taken serious without an accompanying journal. We found Geodata and AI an interesting read.
Conclusion
Sure, we were not surprised to see AI technology make it into the geodata domain on a large scale. Its potential to make data processing, interpretation and understanding much quicker and provide insights to a broad range of users beyond today’s experts is huge. As for all emerging technologies, we will have to see how much will be left after the typical peaks of the first hype, but GeoAI with its handling of large amounts of different kinds of data is a good candidate to see the benefits of AI-based improvements for the time to come.
About The Author
