How data science and AI is supporting land use decision making

Planners need detailed data about land available for the many competing uses required of it: transport networks, civic buildings and amenities, housing, agricultural land, energy, floodplains, waste disposal and green and blue spaces being some of the most obvious. Geospatial AI can help to supply actionable insights from a wide range of data sources

The Government has set ambitious plans to reform the National Planning Policy Framework (NPPF) and to introduce a vision-led approach to transport planning, challenging outdated assumptions of automatic traffic growth and a reactive tendency to support growth and development with road-building and highway improvements. This reset presents a rare opportunity to reshape the future of transport and spatial planning.

Join us at Transport AI 2025 for a deep dive into data innovation in the transport sector. Call for papers closes 15 November

It also requires that planners have access to accurate detailed data about land available for the many competing uses required of it: transport networks, civic buildings and amenities, housing, agricultural land, energy, floodplains, waste disposal and green and blues spaces being some of the most obvious. Geospatial AI can help to supply actionable insights from a wide range of data sources.

In November 2022, the Alan Turing Institute began a partnership with the Geospatial Commission to develop a prototype decision support tool focused on spatial modelling for land use.

This initial project explored how data science and AI could be used to support land use decision making, in collaboration with Newcastle City Council planners which provided the local context and challenges. The result was a prototype tool, DemoLand, which could suggest land use scenarios that would lead to desired outcomes. In December 2023, the Geospatial Commission and The Turing extended its partnership to further develop DemoLand by introducing geospatial AI.

New pressures on land use

New calls to increase the delivery of transport networks and housing place new and existing pressures on our finite land. We need to meet the needs of our growing population whilst maintaining security, satisfying our net zero commitments and protecting our land. These competing priorities mean that a holistic, evidence-based view of our land is required to understand acceptable trade-offs and the impact of different use cases. 

There has been exponential growth in data about our world. Key attributes of land are being recorded through sensors and social processes in ways that are increasing in detail and machine readable. This accumulation of information is creating the necessary conditions for a revolution in how we understand and manage our land. 

While data is necessary, they are not sufficient. Geospatial AI can process the expanding data about land to unlock its information and turn it into actionable insights. The Alan Turing Institute define ‘geospatial AI’ as AI applied in a geospatial context. This paper refers to AI as a set of computational techniques based on machine learning that allow machines to learn about datasets and use that learning to perform statistical tasks such as prediction, or human-presenting tasks like holding a conversation about a topic. 

The Recommendations

This report makes five key recommendations for the UK geospatial ecosystem to improve the use of geospatial AI for land use decision making. 

1. Identify additional areas of opportunity for satellite data to build the value case for geospatial AI: Satellite data, even of medium resolution, can have a profound impact in land use modelling. 

2. Develop a Geospatial AI Toolkit for LLMs: The tools developed in this project represent an exploration of a much larger body of work that could give LLMs access to the UK’s rich datastores, including census and Office for National Statistics (ONS) data. Developing an open-source library of such tools with a wide community of partners has huge potential to accelerate the deployment of LLMs in land use planning and other geospatial applications. This toolkit would address the limitations of existing LLMs in geospatial tasks by utilising rich data sources while allowing users to select the underlying LLM that aligns with their use case and aiming for model agnosticism to enable easy adaptation to newer models as the landscape evolves. 

3. Expand the conversation on national foundation models to land use and geospatial: While a Geospatial AI Toolkit that can provide capabilities to existing LLMs has the potential to address many of the challenges in this space, a model that has a strong built-in understanding of geography and space is desirable. 

4. Improve access to key computational and data resources: Training foundational models, including ChatGPT and SatlasNet, requires vast computational resources. Data is also required to power the tools that will enhance the capabilities of LLMs and to fine-tune language models and foundational satellite models. Much of this data is hard to licence and fund for individual research groups and organisations. We recommend that government identifies ways to increase access to these resources. 

5. Promote knowledge sharing and cross-discipline collaboration: Building geospatial AI for land use requires hard-to-find technical skills and experience. Expertise in AI, computational infrastructure, geographic theory and policy will all play key roles in growing this new field. We recommend that government organises a series of workshops, conferences and training opportunities to accelerate the growth of this field by facilitating cross-discipline knowledge and skills transfer, capacity building and collaboration.