Land use and spatial planning
Where should the next hundreds of thousands of homes go? What happens to agriculture if water safety takes priority? Which locations are still developable once nature, noise and flood restrictions are respected? These are allocation questions, and they are what our land-use models exist to answer.
Our approach is explicit about its assumptions. A model run states what demand it allocates, which locations it considers available, how suitable each is, and why. That makes results reproducible and, importantly, arguable: a debate about assumptions is a good debate, a debate about a black box is not.
RuimteScanner
The RuimteScanner, known internationally as the Land Use Scanner, has supported Dutch national spatial planning since 1997 and is one of the longest-running policy models in the country. Its place is between two other things: regional projections of what a country will need, and the assessment of what the resulting pattern would mean.
How it works
The model needs three kinds of input, and separating them cleanly is most of the design.
Availability. Before a location can receive anything it is put through a sieve: nature protection, noise contours around airports, water safety, plan capacity, whether the site is already evidently in use. The result is a plain yes or no per cell, per sector, per target year, because some restrictions only apply from a certain date onwards.
Suitability. Available locations are then ranked, and here the model deliberately keeps two very different things apart. The empirical component says what is likely or valuable, derived from property prices or from statistical analysis of developments that have actually occurred. The design component says what is wanted, expressed as a ladder of steps in which hard plan capacity outranks soft plans, and short-term outranks long-term. The first is fixed input that stays constant across scenarios; the second is what a policy variant actually varies.
Density. Finally, how much fits where it lands. Density decides how much space is left over for everything else, so it is a policy lever in its own right, and for housing it is expressed as development packages that describe dwelling types and densities together.
Regional demand enters at the top and effects come out at the bottom. The dashed line is what makes the chain a loop: computed effects feed back into suitability, so a variant can be tuned on its own consequences.
Splitting these steps is not just tidiness. Availability maps are expensive to compute and rarely change, so they are calculated once and reused. That is what makes it possible to change a variant and see the consequence during a workshop or a design session, rather than a week later.
The three-layer model
RuimteScanner 2.0, the version released as open source, replaced the single land-use layer with three: land use, objects (dwellings, business premises, wind turbines, solar parks) and actors (residents, and workers characterised by the jobs they hold). Allocation happens on 25 by 25 metre cells.
That split is what allows genuinely useful questions to be asked. How many dwellings and how much property value sit in an area that floods, computed with the standard damage method. Which heating options a neighbourhood would have, by synchronising with the Hestia energy model. Whether jobs and green space stay within reach. And because those effects can be fed back into the suitability maps, a policy variant can be tuned on its own consequences.
To be precise about what it is not: distinguishing actors does not make this an agent-based model. It describes location choice, not the decision-making of individuals, and the feedbacks it contains are limited and fairly static.
In the wider chain
Regional demand for housing and jobs comes from the transport and land-use interaction model Tigris XL, and that relationship runs both ways. The space available for urbanisation according to RuimteScanner goes into Tigris XL first, so that the two models agree about how much room there actually is. Demand for the other sectors comes from trends, expert judgement or policy targets.
Where it has been used
Recent applications include PBL's Ruimtelijke Verkenning 2023, the Planmonitor NOVI, the Deltascenario land-use update and Grote opgaven in een beperkte ruimte. Research applications have ranged from the reconstruction of Roman-era land use to urban heat islands and the future of the peat meadow areas.
The open-source version, RSopen, is documented in full at geodms.nl/rsopen.
Beyond the Netherlands
- LUISA: the EU-ClueScanner we developed for the Joint Research Centre became the land-use component of the JRC's LUISA platform for EU-wide territorial policy assessment.
- Land Use Scanner Germany: developed with the Federal Office for Building and Regional Planning.
- Bangladesh Delta Scanner: in use by Deltares and CEGIS in Dhaka since 2014 for land-use and water management scenarios.
Over the years the family has grown to include RSlight, RuimteScanner L and XL, Transformation Potential, LUISA-BEES, LUISA Africa and VALUISA. The modelling concepts behind them, from continuous and discrete allocation to land units and suitability, are documented in the land use modelling wiki.
The data underneath
Good allocation needs good base data, and some of that we build ourselves. The delineation of built-up area decides what counts as inside a settlement, which sounds trivial until you have to defend a particular boundary. Pand Hoogte Nederland derives a height for every building in the country from the national elevation rasters, and our BAG tools turn the address and building register into something a model can actually use.