A 21st Century Domesday: thoughts on a data-driven Land Use Framework

How do we ensure land is used efficiently, productively, sustainably, and responsibly?

How do we balance the competing priorities of food security, clean energy production, adequate housing and infrastructure, space for leisure and recreation, and environmental stewardship?

These questions lie at the heart of the UK Government's plans for a new Land Use Framework, which has kick-started with the Department for Environment, Food and Rural Affairs announcement of a 12-week public consultation.

A highly charged topic

The very concept of ‘land use’ is highly emotive. Even atavistic.

Civilisation was enabled because ancient humans learnt how to shape landscapes, rather than be shaped by them. The origin of the ‘land economy’ can be traced back to at least the age of feudalism, and arguably much earlier. It has been a common thread of our history and culture, intertwined with and driving concepts of class structure and governance, as well as national identity.

Perhaps the most important and controversial concept that has emerged over centuries of land use is the idea of land ‘value’ – be it economic, cultural or intrinsic:

• Economic value arises if land directly or indirectly factors into a financial transaction, such as the output of a piece of farmland or the mortgage associated with some real estate.
• Cultural value exists if land has symbolic meaning or offers practical, but not necessarily financial, benefits to a group of people, such as a green belt or parkland.
• Intrinsic value exists if land is associated with natural capital that transcends concepts of ‘use’ and must be protected, such as Sites of Special Scientific Interest.?

These values are not mutually exclusive, and therein lies a problem. Conflict arises when there is disagreement between different stakeholders as to the perceived priority of the value of a particular piece of land. One group’s precious green space is another’s new housing estate. Is it better to grow food or install a solar farm? What strategy is correct for ensuring the conservation of a threatened habitat?

The evolving needs and demands of a growing population, emerging technology, volatile geopolitics and even the changing climate add complexity to the mix.

Decisions, decisions?

Decisions made now have ramifications for future generations.

Those decisions must take into account the conditions that our children and grandchildren will have to contend with, whilst recognising that today’s decisions will also shape those future conditions. What will energy demand look like? Where will our food be grown? Where will people live? If tomorrow’s climate will increase the risk of crop failure here then where should we start planning to grow today? The scenarios to consider are myriad.

How do we make the right decisions around land use that maximise positive outcomes and minimise negative ones, both in the near- and long-term of our society?

Considering the development of a Land Use policy, the The Royal Society makes several key recommendations in its Multifunctional Landscapes report, with specific reference to "increasing, and enhancing access to, science and innovation relevant to land use.". A key recommendation of the report is that

A novel data science-driven approach is needed to develop a high-quality common evidence?base to underpin land?use decisions.

?At Aspia we couldn’t agree more.

An evidential approach, accelerated by AI

We believe that a fundamental ingredient of any modern Land Use policy is evidence. Evidence of past land use, evidence of current activity, evidence of change. Without consistent, comprehensive, continuous evidence it is difficult not only to make informed decisions, but also to measure the outcomes where a decision has led to action.

And we believe that the most powerful asset in any evidence-based approach is observational data.

That is what we are building at Aspia. We are creating a new lingua franca for land use intelligence, rooted in observational science and accelerated by AI.

Our Large Observation Model (LOM) – EarthPT – was the world’s first and remains the largest GPT-based foundation model for multi-modal spatio-temporal observational data. Drawing inspiration from the field of natural language processing, EarthPT has been trained on the criterion of next-token-prediction. Although simple, this has been proven to result in highly capable Large Language Models (LLMs), and – in-the-limit – appears to approximate human reasoning, with the trained weights representing relatively compact ‘world models’ that comprise consolidated knowledge about the datasets they are trained on.

EarthPT is learning about land use from vast quantities of observational data.

Unlike language models, the number of unique observational tokens will only increase with time.

Our model allows us to extract meaning from data: learning the relationships between noisy variables, inferring properties and conditions where gaps in knowledge exist, and even accurately forecasting future conditions. It draws on the learning that can be gleaned from data as a whole to deliver insights at a hyper-local level.

In the context of Land Use, these abilities correspond to a more accurate and complete picture of the key questions underpinning decision-making: What, Where, and When?

A 21st Century Domesday

Imagine having access to a consistent, comprehensive, and regularly updated ledger of intelligence – those what’s, where’s and when’s – for any given parcel of land, at a granular spatial and temporal level. The ledger would represent a dynamic record of the properties of interest, qualitative and quantitative, for any given location. It could record change for that location, compared to the ledger for any other location and time, and aggregated with other ledgers across a larger area of interest. The database containing these ledgers could become an unequivocal reference point for decision-making pertinent to Land Use, continually updated as new observations become available. What habitat class does this land parcel represent? Where are the sites at highest risk from flood? When was this field harvested?

At Aspia our vision is to create the world’s most advanced foundation model for planetary intelligence, capable not only of inference and prediction, but also reasoning about the world. What if we could learn the causal link between event or intervention (both natural and anthropogenic) and observed outcomes? We could use such a tool to aid decision making by simulating potential scenarios within the safety of a computer model. In cases where decisions involve an action or intervention, we could pick the one with the highest likelihood of the most positive outcome. In cases where events are beyond our control – extreme weather for example – we could be better prepared by placing firmer confidence intervals around potential deleterious consequences.

AI is not a panacea in any domain, and we should be suspicious of anyone who claims otherwise. It is, however, an extraordinarily powerful tool. And it is the data that make it possible. We don’t want to replace human decision-makers, nor do we want to trivialise the broader issues that must factor into the development of any effective Land Use policy: historical, cultural and sociological. But by providing a data-driven, self-consistent, and trustworthy framework for land use intelligence, and by serving that information in a highly consumable and interoperable manner, we can not only inform decision-making and monitor compliance, we can use this resource to continuously learn how policy is working in practice. Where there is evidence of failure, we can curtail its impact. Where success is evident, we can recommend that good practice is replicated elsewhere.

The answers will be in the data, and all we need to do is observe.