EN Model structure overview

This is an English translation of the Dutch page Modelstructuur op hoofdlijnen, which is leading. Translated on 8 November 2026 from the Dutch text committed together with this page, on top of wiki commit 2eddf6a.

A three-layer model

RuimteScanner 2.0 is part of a model chain in which it forms the link between expected or desired regional developments in various sectors on the one hand and (societal) effects on the other (Figure 2). As in reality, the interplay of the parties that demand space on the land and property market is the pivot here. A characteristic of the main structure is that the model does not only allocate dominant land use, but at the same time removes actors and objects from locations and allocates them to locations. For practical reasons (available data and methods, added value in the model) the model uses a limited number of layers: dwellings by dwelling type and market segment; jobs by economic subsector, with the building footprint as the associated property; holiday homes; storage volume for water storage; hectares per agricultural class; and megawatts of wind and solar energy, with the turbines as objects in the case of wind. Which of these layers take part follows from the sectors that an application switches on; see below. An extension with other layers, for example household types, is conceivable but has not been built.

image

Figure 2 RuimteScanner 2.0 as part of a larger model chain.

The main advantage of explicitly including actors and objects is that concrete policy choices regarding these layers can be explored (for example a minimum percentage of green space per residential environment, or a maximum local dwelling density), and that specific effects can be linked to them. Examples of such effects are the dwellings, commercial buildings and their value that can be hit by floods (rebuilt in the model from the damage method of HIS-SSM, see Flood damage (in Dutch)); heating options for buildings and neighbourhoods in the built environment (to be calculated externally with the Hestia model); and local changes in the accessibility of jobs and green space (Accessibility of jobs (in Dutch), Accessibility of green space (in Dutch)). By feeding such effects back into the RuimteScanner, through the local suitability maps (internalising external effects) or through the regional objectives (adjusting the demand for space), policy variants can be tuned partly to these effects. The application pages in Part III show examples of this.

Incidentally, the fact that actors are distinguished and allocated to locations in aggregate does not yet make this an agent-based model as the scientific literature distinguishes it (for example Groeneveld et al., 2017). Such models integrate two components: a grid-cell-based model of the landscape and an agent-based model that describes the decision-making of the main stakeholders in the land-use system being described (Parker et al., 2003). Both components are integrated in that they describe the dependencies and feedbacks between actors and their environment. In RuimteScanner 2.0 such mutual relations are not included extensively, and only location choice is described. Dependencies and feedbacks are included in a limited and relatively static way, for example where the travel time to 500,000 inhabitants and to a public transport hub counts in the dwelling price and thereby in the suitability for housing, or where the nature, water and buildings that the model placed in an earlier projection year adjust the suitability in the next projection year (see Suitability).

The regional development of dwellings and jobs is taken from the Land-Use Transport Interaction (LUTI) model Tigris XL (Zondag et al., 2015; Zondag & Geurs, 2011). The relationship with Tigris XL works both ways: before the regional demand for space for housing and employment is determined in Tigris XL, the amount of space available for urbanisation in the base year is added from RuimteScanner to Tigris XL. As a result, the two models use the same available space for urbanisation. The regional development of the other sectors comes from (combinations of) trends, expert judgement or policy targets; Table 1 at the bottom gives the source per sector.

From regional demand to local possibilities

The regional demand, in numbers of dwellings, jobs, hectares, cubic metres of storage or megawatts, is allocated to individual grid cells of 25 by 25 metres, the AdminDomain in the model. This is done on the basis of the land use that these cells received in the base year or in earlier projection years and, partly based on that, the availability and suitability of the cells for alternative uses. Before the spatial allocation, the model goes through the following phases per sector:

  1. Filling in the relevant layers of the sector in the base year (the existing locations of jobs, dwellings, wind turbines and so on); see Initial state (in Dutch).
  2. Determining the availability of locations for the sector concerned (for example locations without restrictions under environmental and planning law, such as Natura 2000 areas and noise contours around airports); see Availability.
  3. Determining the suitability of the available locations for the sector and its subsectors; see Suitability.
  4. Specifying the density at which the relevant layers are realised in the case of allocation (dwellings per hectare, for example); see Density.

The first phase is generic: the outcome is the same for every scenario and every variant, is calculated once and stored as a decoupled file, and is only run again when a model update requires it, for example when data for a new base year become available (see Decoupled files (in Dutch)). In most applications the other phases are run at least once, and this applies especially to phases two and three. Availability and suitability are separate parts of the model. They differ in the way they are defined and in the units used, and separating them allows them to be specified, adjusted, communicated and stored separately. The last of these saves computing time: a variant that only sets the sieve differently does not need to recalculate the suitability maps.

For similar reasons phase three itself is split further. The suitability map is built from two components. The first component is empirically based. It describes, for example, the value of a location (based on an analysis of property prices) or the relative probability of a particular spatial development (based on statistical analyses of developments already observed). The second component is created by design and describes the relative priority of specific location types at certain locations. It is determined on the basis of design principles or (intended) policy, and is operationalised as a ladder of rungs, such as the Sustainable Urbanisation Ladder: reasoning from the bottom up, the rungs are given a decreasing suitability, for example hard planning capacity (in Dutch) above soft, and both above the rest. In practice a ladder has more axes than planning capacity alone, up to 324 classes for housing; how a ladder is built up is described on Rungs (in Dutch).

The empirical component forms the basis for local suitability in the third phase. It is fixed model input that is always the same across the scenarios and policy variants of an application; the maps are calculated once, stored and reused, and only the adjustment per projection year for what the model has placed itself, such as the nature and water around residential locations, takes place in the allocation itself. The design-oriented component, by contrast, is intended for working out (policy) variants: the ladders are quick to compose, evaluate and adjust, which makes them particularly suited to exploratory, iterative research. The design component takes precedence: a cell in a higher rung always comes before a cell in a lower rung, and only within a rung are the available locations sorted by the empirically determined suitability. An application that wants a sector to allocate without a design preference gives that sector a ladder with a single rung; the empirical suitability then decides on its own.

The last input that RuimteScanner 2.0 needs to allocate the regional projections to available and suitable locations is their density. The higher this density, the greater, in the case of housing, the possibilities for financially feasible housing construction in the existing city, and, more generally, the smaller the additional space needed for these functions. This leaves more space for other functions such as agriculture and nature, and flood risk, accessibility and infrastructure costs generally benefit from it as well. However, high densities are not always possible, for technical reasons (the production of wind and solar energy), economic reasons or reasons of environmental and planning law (planning damage claims, residential amenity). The densities can be set per variant and are bounded per location, for housing through development packages with a location-specific ceiling, so that the user can explore the implications of density choices in conjunction.

Hybrid allocation with competition within sectors

Allocation procedure in formulas (in Dutch)

The allocation of the regional developments (for example numbers of dwellings or jobs per sector) to locations initially takes place successively (sequentially) instead of integrally. The different sectors take their turn in the order of their observed or expected dominance on the land market. First come, first served. Which sectors take part, in which order and at which regional scale level is set out as a table in ModelParameters/SectorAllocRegio. That is the place where an application switches its sectors on and off: the model knows exactly the sectors that have an active row there. A sector can have several rows, first the finest allocation region and then coarser ones to absorb residual claims, and the order in the table is the order of allocation. Housing and employment are usually at the front, because they bid most strongly on the land market. An application can depart from this order, for example in policy variants in which other ambitions come first; which sectors and which order a project chooses is described on the application pages in Part III.

The sequential allocation serves two purposes. It makes the model suitable for exploratory use, to arrive at outcomes iteratively, by drawing and calculating. And it matches Dutch practice, with large price differences on the land and property market between market segments (housing, employment, agriculture) that are related to regulation and to differences in supply and demand (Dekkers & Rietveld, 2011). Because of the high pressure on the housing market, attractive locations within existing settlements develop into residential locations sooner than into work locations. By placing housing before employment in the table, the model expresses this: work locations are only allocated once the demand for residential locations in the region and period concerned has been met, or once the space for it has run out.

Nature does not go through this table. New nature is imposed as a map per variant, and where a delivery gives no map but a requirement or a distribution, a separate allocation or ranking puts it on the map; see Nature elaboration (in Dutch).

Within a sector, however, the model does simulate spatial competition, between so-called subsectors: for housing between dwelling types and market segments with their development packages, for employment between economic subsectors, for agriculture between agricultural classes. The allocation of subsectors within a sector uses an iterative allocation procedure in which the regional developments of the subsectors are allocated to the available locations on the basis of suitability. The model develops as many locations as needed to meet the regional demand of the subsectors, and stops when that demand has been met or when all available locations have been filled. Agriculture uses a different mechanism: that sector is distributed with the discrete allocation, with a lower bound and an upper bound per agricultural class and per region; see Agriculture elaboration (in Dutch). The section below describes the simulation of competition between the subsectors of the other sectors in more detail.

The simulation of competition between subsectors

In the allocation of subsectors to available locations, local suitability is decisive and the law of the strongest applies: the highest bidder (the subsector with the highest suitability at a location) wins. The suitability values, which can be regarded as bid prices, are a given. There is no exchange of suitable locations between subsectors and no equilibrium prices are calculated: there is no price mechanism that balances supply with demand and thereby maximises total suitability (for a region, summed over the locations and subsectors), as an iterative bidding process with shadow prices does (see Dekkers, 2005; Koomen & Buurman, 2002). A location that is eminently suitable for a specific use but has a lower starting value is therefore not chosen after all in this procedure, and total suitability in a region is not automatically maximised.

The procedure works in successive intermediate steps (iterations), in each of which part of the regional demand (claim) for a sector is allocated to the most suitable places. After each intermediate step, residual locations remain (the locations that remain available for other subsectors) as well as residual claims (the regional claims of subsectors that were not the highest bidder anywhere), and these are allocated to each other in the next iteration. The procedure stops when the available locations or the claims have run out; the number of iterations per sector and allocation region is in the same table. The application page Ruimtelijke Verkenningen 2023 (in Dutch) goes deeper into this allocation process on the basis of a concrete application.

In each iteration the model first determines a cut-off threshold per subsector with the GeoDMS function weighted nth element: the lowest suitability needed to realise the entire residual claim of that subsector, given the local densities; locations below that threshold drop out, so that the claim is not exceeded. Then, per location, the subsector with the highest suitability above its cut-off threshold wins, using arg_max. These locations are built on (objects) and occupied (actors) according to the densities of the winning subsector (phase 4). Existing actors and objects from the base year (phase 1) are removed or demolished there: the allocation replaces the existing function, with the local-serving job subsectors as the exception, because they stay in place when housing wins the cell. What has been displaced returns as a residual claim and is still allocated: in a following iteration or sector allocation region if the displaced sector still has its turn to come, in a following sequence if it has already had its turn.

The allocation chooses directly at the 25-metre cell (AdminDomain); the coarser 100-metre grid (AllocDomain) still serves, among other things, for the density ceilings and the delimitation of the computational domain. To counter fragmentation, the raw result of each iteration is post-processed with clustered allocation: narrow ribbons and clusters below a minimum size are rejected, after which a following iteration searches elsewhere, so that coherent units (residential environments, business parks, water storage areas) arise instead of loose cells. The full elaboration, including the cut-off threshold, the shape tests and the displacement, is in Allocation procedure in formulas (in Dutch).

Once allocated, always allocated: cells that the model has allocated do not become available again, for the rest of the series of projection years, for (sub)sectors that come later. Agriculture is the exception, because a crop does not fix the cell and is chosen again in every projection year. Actors and objects from the base year, on the other hand, can be replaced, for example through relocation or demolition. The model keeps track of this, offsets it against the remaining regional developments and, in the case of a residual demand, includes it in the allocation again. Residual requirements that remain when all available space in a region has been used up are either accepted or transferred to a higher regional scale level (overflow (in Dutch)), where the allocation procedure continues. Whether this happens and, if so, at which spatial scale level, is determined by the user with the rows in the allocation table.

The user also judges whether the simulation outcomes are consistent with the objective of the application. What is appropriate can differ per study: when reproducing trend developments, following observed changes matters; in scenario studies, the fit with the storylines matters most. The extent to which the results match the expected outcomes can be judged on the basis of the generated suitability maps and the (effects of the) simulated spatial patterns. Based on this evaluation, new or adjusted (policy) variants can be constructed. With all this, RuimteScanner 2.0 follows a hybrid allocation procedure (both sequential and based on trade-offs), with much room for design.

An extensive set of outcomes

The allocation result consists of a set of maps (grids of 25 by 25 metres) that give per cell which subsector has been newly added, for housing the development package, and the local characteristics belonging to the subsector: the number of allocated objects (dwellings per type, holiday homes), actors (jobs per subsector), land take (building footprint in square metres for housing and employment), storage volume and hectares per agricultural class. In addition, the model keeps a land-use type and the year of change per cell, the land-use state register (in Dutch). At the locations where new development takes place, the existing objects are removed and the cell is filled homogeneously according to the local characteristics of the allocated subsector; only local-serving employment remains in place under new dwellings.

The sector-specific land-use layers each describe part of the expected developments, but in isolation they do not provide an integrated, nationwide picture of future land use. Such an overview is useful for viewing changes in their spatial context and, for example, comparing them with current land use as Statistics Netherlands (CBS) maps it. That is what the Land-use map (in Dutch) is for: it translates the layers with a spatial aggregation method that matches as closely as possible the method that CBS uses for the Bestand Bodemgebruik (Claassens et al., 2023), with threshold values for the number of dwellings, the building footprint and the number of holiday homes to assign a grid cell to a land-use type; nature and agriculture come onto that map from the land-use state register. In addition to the land-use representation, the model calculates various indicators from the results, including the accessibility of jobs, the proximity of green space, flood damage, carbon storage and the change in dwelling value; see Impact modules and indicators. This forms the last layer of the three-layer model in Figure 2.

Elaboration per sector

The exact implementation of the allocation process and of the phases that precede it differs per sector. The following pages give a further elaboration per sector. The focus is on the more or less generic components that go beyond specific applications. The possibilities for more policy-based and design-based elaborations, especially of the availability and suitability of locations, are discussed on the pages concerned.

Including the most recent developments

The start year of the simulations rarely coincides with the most up-to-date input data available from the data holders. For the state of housing, employment and holiday accommodation in the base year, the model uses the BAG, which is updated daily: the developments between the start year and the date of the most recent BAG release in the model are therefore already known. In addition, the BAG contains status information about developments in the near future: buildings for which a building permit has been granted or whose construction has already started, which can be assumed with great certainty to be realised, and buildings for which a demolition permit has been granted, which are very likely to be demolished. The model includes all these recent construction developments (in Dutch) in the first projection year as an adjustment of the initial state, before the allocation starts, so that they no longer need to be allocated.

Synthesis

In summary, RuimteScanner 2.0 can be characterised as a hybrid, semi-integrated allocation model that allocates regional projections of actors, objects and land use to available and suitable locations. In line with practice, land-use functions are allocated sequentially (one after the other). Functions that are allocated sequentially are referred to as sectors. Which functions these are and in which order is a setting of the application, and not something fixed in the model; the model knows housing, employment, water storage, holiday accommodation, solar energy, wind energy and agriculture as sectors that an application can switch on. Competition can be simulated within the sectors. This then takes place between so-called subsectors, in an iterative process in which the highest bidder wins and which stops when the available space has run out or when all regional requirements have been realised. This competition exists within the sectors housing (between dwelling types and market segments, through the development packages), employment (between economic subsectors) and agriculture (between agricultural classes).

The model has a flexible set-up. This concerns not only the order in which functions are allocated and whether or not this happens in competition. It also concerns the question of which functions are included at all. For the plan monitor of the NOVI (Kuiper, Rijken, et al., 2023), for example, only urbanisation was considered: housing, employment and holiday accommodation. In the Ruimtelijke Verkenning 2023 (Kuiper, Hamers, et al., 2023) the solar energy and wind energy sectors also took part, and in NL2120 application (in Dutch) water storage and agriculture. The model explicitly gives room to design here. This mainly concerns filling in the availability and suitability of places for the various functions and the density at which they are (or may be) allocated to places.

Table 1 gives an overview of how the model is set up per sector. The application pages in Part III illustrate how it works and how it can be applied; the Ruimtelijke Verkenning 2023 (in Dutch) is the most extensive of these, with various model aspects side by side. That description is explicitly an illustration of the model aspects and not an account of the substantive background and results of that study; for those, see the available reports (Hamers et al., 2023; Kuiper, Hamers, et al., 2023). The specific model settings of an application are recorded in the model configuration and are given on the application page.

Table 1 Source of the demand for space, the empirical suitability and the density per sector.

Sector Source of the demand for space Empirical suitability Density Competition within the sector
Housing regional model Tigris XL development balance per development package, from a hedonic price analysis, a regression on land production costs and key figures dwellings per ha, from the development package yes, four subsectors and their packages
Employment regional model Tigris XL, anchored per subsector on the base year: at the level of Tigris XL or on the observed LISA trend, adjustable probability of occurrence, multinomial logit model per subsector m2 building footprint per job, from observed densities per region yes, six economic subsectors
Water storage supplied requirement in m3 per subarea weighted sum of criteria, including water depth, proximity to peat and seepage storage volume from the water layer of the cell no
Holiday accommodation trends probability of occurrence, binary logit model holiday homes per ha, from the marginal trend density no
Solar energy policy target, calculated back from the Klimaat- en Energieverkenning half the normalised capacity factor, half the inverse of the landscape perception map, within the rungs of the solar ladder megawatts per ha, key figure no
Wind energy policy target half the normalised capacity factor from the wind speed, half the inverse of the landscape perception map, with placement rules for turbines from the spacing between turbines no
Agriculture a band around the base-year state per region, reduced by what is imposed exogenously and scaled to the space that remains net margin per crop, after transition costs follows from the crop yes, between agricultural classes, via the discrete allocation
Nature no claim from a regional model; per variant an imposed map or requirement not applicable not applicable no

Nature is listed separately in the table because the sector does not receive a claim from a regional model and does not go through the allocation table.