EN Suitability

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

The suitability of a location determines how attractive that location is for a particular sector or subsector. Suitability is, together with availability and density, one of the three filters that determine where and how much is allocated. In the allocation procedure, available locations are sorted by suitability: the location with the highest suitability is developed first, until the regional claim has been met.

The suitability assessment consists of two components that play a different role in each sector.

The two components

Empirical component

The empirical component describes the “intrinsic” attractiveness of a location, based on statistical analyses of observed patterns. This component is independent of scenario and variant: the base map is the same for all variants and is calculated and stored once. For housing and employment, the allocation updates that base map in each projection year with what the model has placed in earlier projection years; see the sections Dynamic green proximity and Role in the allocation below.

The empirical component is determined in a different way for each sector:

Sector	Method	Unit	Source file
Housing	Development balance	Euro	Hedonic price analysis (NVM), land production costs (regression), construction costs (key figures)
Employment	Probability of occurrence	Probability (0-1)	Multinomial logit model per Jobs6 subsector
Holiday accommodation	Probability of occurrence	Probability (0-1)	Logit model
Solar	Half the normalised capacity factor, half the landscape perception map (inverted)	Score	Capacity factor Utrecht University, CBS/WUR perception map
Wind	Half the normalised capacity factor, half the landscape perception map (inverted) + placement rules	Score	Capacity factor Utrecht University, CBS/WUR perception map

Design component (rungs)

The design component describes the relative priority of location types on the basis of policy choices and design principles. It is variant-specific and can be adjusted per application. The design component is operationalised through the rung concept, by analogy with the Sustainable Urbanisation Ladder.

The rung concept works as follows: locations are classified in a hierarchy of rungs, in which lower rungs (higher priority) come up before higher rungs. Within each rung, the empirical suitability determines the order. The allocation first fills all available locations in the lowest rung, and only moves on to the next rung if the claims have not yet been met or the locations in the previous rung have run out.

The design component therefore takes precedence over the empirical component: a location in a low rung with low empirical suitability is developed before a location in a high rung with high empirical suitability.

Housing suitability: the development balance

The suitability for housing is based on a development balance: the difference between the expected revenues and the expected costs of housing development at a location, per development package. What a development package is, which packages exist per variant and per subsector, and which sieve determines whether a package may land on a cell, is described on Housing elaboration (in Dutch).

Revenues

The expected revenue is derived from a hedonic price analysis of NVM transaction data (2000 up to and including 2023). This analysis estimates the dwelling value per dwelling type (WP4) as a function of dwelling characteristics (floor area, rooms, maintenance, high-rise) and location characteristics: accessibility (travel time to 500k inhabitants and to public transport hubs), urban attractiveness (UAI, network-based) and the nature and water fractions in the surroundings (fr_natuur_tot2500m: share of nature within 2.5 km; fr_water_500m: share of water within 500 m, both from LGN2024). The revenues per development package are calculated with the nature and water fractions of the base year; the allocation corrects those revenues in each projection year for the nature and water that are present at the start of that projection year, see the section Dynamic green proximity below. The change relative to the base year is valued afterwards in Dwelling value due to changes in green (in Dutch). The coefficients are estimated externally (PriceIndices, spec rs_groen) and read in via NVM_filedate and NVM_coeff_Year. The full price function, the choice of variables and the assumptions per application are described on Hedonic house price model (in Dutch).

For each development package and each cell, the expected revenue is calculated as the hedonic price of a dwelling of that package, multiplied by the number of dwellings per hectare of the package and the area of the cell in hectares.

Dynamic green proximity

The suitability for housing is not static. In each projection year, the nature and water fraction is determined again for the state at the start of that projection year, with the same template that the indicator uses (Templates/Suitabilities/Groenfracties_T). The difference with the base year gives a relative price mark-up per dwelling type, by which the revenues in the allocation are multiplied (#637). Only the revenues are corrected, not the development balance as a whole.

The lag of one projection year is intentional: what is built in a projection year was built in the period before it, and so cannot have been priced with nature that was created in that same period. As a result there is also no endogeneity within a projection year, because the map is fixed before the first iteration starts. In the first projection year the mark-up is zero, because nothing has yet been realised that could count. Anyone who runs only the final year has just one projection year and therefore notices nothing of this mechanism.

Costs

The cost side consists of eight items:

Construction costs: the construction sum of the real estate, as a key figure in euros per square metre of gross floor area times the gross floor area of the development package. The level comes from CBS table 83673NED; a ratio per dwelling type is applied on top of that.
Additional costs: fees, permit charges, interest during construction and contingencies, as a mark-up on the construction sum.
Profit and risk: the developer’s return on the real estate investment.
Land production costs: making the land ready for building and habitation, the off-site infrastructure and the planning process. Estimated with three semi-logarithmic regressions on land development budgets from 2019, with plot density, travel time to 5,000 inhabitants, the size of the plan area and a soil dummy as location variables.
Subsidence costs: the extra costs of building on subsoil that is prone to subsidence, compared with the sandy-soil reference, per cell and per development package. Only the initial construction costs, because only those end up with the developer.
Cloudburst storage costs: the construction of the storage measures that belong to the package, with the same measure fractions per variant as the storage indicator; in a variant without measures the item is zero.
Acquisition costs: buying up the existing real estate and the land. For dwellings from the hedonic value estimate, with a discount for housing association stock and the replacement factor; for non-residential buildings from the WOZ source data; for undeveloped land from a price per land-use class.
Demolition costs: demolishing the existing buildings, with a basic key figure per building type plus an asbestos surcharge for buildings from before 1994.

The acquisition and demolition costs of the buildings are spread over the cadastral parcel, so that a garden costs what the house costs. Which parcels with dwellings are due to make way in a projection year is determined not by the balance but by the Redevelopment probability (in Dutch).

The development balance per location per development package is:

Residual value      = Revenues
                    - Construction costs - Additional costs - Profit and risk
                    - Land production costs - Subsidence costs - Cloudburst storage costs

Development balance = Residual value - Acquisition costs - Demolition costs

Locations with a higher development balance are more attractive. Locations with a balance below MinimumExploitatieSaldo can be declared restrictive; that threshold is set per variant in VariantK.dms, and a higher value there is stricter, not more lenient.

The full description of each item, with the sources, the coefficients, the layers that make the maps and the assumptions, is on Housing development balance (in Dutch).

Employment suitability: multinomial logit model

The suitability for employment is determined by a multinomial logit model that estimates, per location, the probability of development of each of the six employment subsectors (Jobs6). What those six subsectors are and how the sector is further worked out is described on Employment elaboration (in Dutch). The model was estimated in the report Ruimte voor werken (Spinlab Research Memorandum SL-20, VU/PBL, 2022, chapter 4): per 100 m cell it explains the growth of the building footprint per BAG function of at least 100 m2 per hectare between 2012 and 2022, relative to no growth as the reference category (pseudo-R2 0.186). The coefficients are read from a dated CSV (logitcoeff_mlogit_devType_<datum>.csv, with the date from Werken_coeff_date) that corresponds one to one with table 5 of the report.

The explanatory variables are:

Distance to cargo airport, motorway on-ramp (2018), seaport (2019), railway station (2019), urban contour (2000), all log-transformed and censored at a minimum of 1 km
Travel time to 100,000 inhabitants (log-transformed, censored at 1 to 120 minutes)
Hedonic land price 2007 (log-transformed, plus 1 euro per m2)
Urban Attractivity Index (UAI)
Land-use shares in the cell (not available, building site, nature, housing) and the share of other built-up land within a radius of 1000 m

Per location, the six probabilities are normalised with the multinomial logit formula:

P(sector_j) = exp(V_j) / (1 + sum(exp(V_k)))

What is not applied from the estimation

The formula uses 14 of the 31 estimated terms. The provincial fixed effects have deliberately been left out, because they would produce hard border effects between provinces (#599). The five policy dummies from the estimation (VINEX, Natura2000, bundling areas, buffer zones 2005, EHS 1990) are not in the formula either: RSopen imposes spatial restrictions through Availability (sieve and restriction maps), and the policy from the estimation period is not future policy. The downside is that the map misses the estimated negative pressure where the sieve does not exclude those areas.

Vintage of the input data

The coefficients were estimated on the state at the start of the growth period (2012); the application feeds them with the current state, on the reasoning that the current state predicts the growth that follows. Specifically: the land-use shares come from the BBG year BBG_Year (in the classification of the newer BBG editions, the solar park class counts as other built-up), the travel time to 100,000 inhabitants from a map with reference year 2024, and the UAI is calculated dynamically from BAG counts of BAG_RecentYear with renormalisation on the current maximum, against the coefficient estimated on 2018. Cells with missing data get a utility of -1e9 and thus probability 0 (#599); the land price map is the only input without its own safety net, and an IntegrityCheck ensures that this map remains filled across the whole area.

Role in the allocation

In the allocation, the employment suitability differs from the static map on three points:

Dynamic per projection year (analogous to the dynamic green proximity for housing): the shares of housing, nature and water in the cell and the neighbourhood share of other built-up land (1000 m) are updated in each projection year with what the model allocated or exogenously imposed in earlier projection years, from the state at the start of the projection year. Only additions count: the share that a cell already had in the static map is not reduced, because new use almost always lands on cells that did not contribute to any of these shares. New employment thus attracts follow-up allocations. Building sites and the UAI stay static; in the first projection year the map is equal to the static map.
Raw utility as allocation input: within a subsector only the rank order counts, and the MNL denominator (which differs per cell) wrongly pushes down cells that are attractive to several sectors; that is why the allocation ranks on the raw utility. Cells with missing data are null and never enter the candidate set.
Priority in the subsector choice: when several employment subsectors have the same cell above their cut-off threshold, the subsector for which the cell is an existing dominant office or institutional cell wins first (see point 4), then the subsector with hard plans on that cell, then soft plans, then stimuli (KanOp-filtered), and only within that the highest probability. The rung scores themselves cannot be compared between subsectors, because the number of rungs differs per subsector.

Within a subsector, the map otherwise counts purely ordinally within a rung; see Allocation procedure in formulas (in Dutch).

In addition, two priority rungs determine where new employment lands, both as the highest axis in their own rung table:

Intensification in the existing stock (#668): on cells where business services dominate the employment state in the base year (share above KantoorDominantieDrempel, with a lower bound on the employment state so that a one-person business at home does not count), a potential floor applies of the cell’s own employment state times the intensification factor, the rung ranks above greenfield, and the demolition-age and IBIS-full tests in the projection-year sieve are dropped. In this way desk sharing lands in the buildings where the desks are; see Density for the density side. The same intervention exists for government and quaternary services, with the same dominance threshold, its own lower bound and its own intensification factor.
Local-serving functions near the neighbourhoods (#710): for retail, other consumer services and government and quaternary services, cells within existing built-up area or within a disc of VerzorgendWoongebiedStraal around existing residential area rank above everything outside it; for government and quaternary services, the existing institution comes before that. Bare rural area is only overflow. The disc is anchored on the base year, so neighbourhoods from later projection years do not yet attract facilities; that is listed as an open point in #710.

The underlying estimation has limitations: the estimation period lies before COVID and before the policy on large distribution warehouses, densification and urban extension are mixed in the estimation, and the input data have diverging vintages. A re-estimation has been proposed in #647.

Holiday accommodation suitability: logit model

The suitability for holiday accommodation follows an approach similar to employment, but with a binary logit model (holiday accommodation or not). The explanatory variables are:

Distance to airport, motorway on-ramp, railway station, urban contour, Natura 2000
Travel time to 100,000 inhabitants (log-transformed)
Hedonic land price
Landscape quality (perception map)
Land-use shares in the surroundings (1000m radius): agriculture, infrastructure, forest, other nature, housing, water, diversity

The coefficients are estimated in the model itself, on exactly the terms that the formula applies: the dependent variable is the transition to holiday accommodation between BBG 1996 and BBG 2015, the covariates are the same maps and transformations as in the formula, and the estimation runs with the logit template Templates/Statistiek/Logit_T. The re-estimated set is stored as a decoupled file with a fingerprint; the original csv (chosen via Verblijfsrec_coeff_date) carries the same term names and remains as a reference. The re-estimation is needed because the formula uses only part of the terms from that original estimation: the policy dummies and the EHS distance are left out, because in RSopen restrictions run through the sieve and not through the suitability, and the constant of the full model does not fit a formula with fewer terms. Cells with missing data get probability 0, the same nodata handling as for employment (#599).

Solar and wind suitability

Solar

The suitability for solar energy is based on the solar ladder, a hierarchy of location types ordered by societal preference. The solar ladder is design-based and is defined in VariantParameters/Tredes/Zon.dms. Within each rung of the solar ladder, locations are sorted on the empirical suitability: half the capacity factor normalised on the maximum, half the (inverted) landscape perception map (BaseData/Suitabilities/Zon), so that locations with a higher yield and a lower landscape perception are used first.

The rungs of the solar ladder are in Classifications/Zonneladder, with the map per rung in BaseData/Zonneladder, and include categories such as verges, car parks, noise barriers, and agricultural land.

Wind

The suitability for wind energy combines the capacity factor normalised on the maximum and the (inverted) landscape perception map, each for one half (BaseData/Suitabilities/Wind), with a set of placement rules for wind turbines. These rules determine whether a location is physically suitable for placing a turbine, taking into account the distances between turbines, orientation and minimum group size. The placement rules are evaluated with a stamp pattern that is applied in the allocation phase.

The rung concept in detail

The full description of the ladder is on Rungs (in Dutch): how a ladder is built from axes, which axes there are, which ladders exist per sector and subsector, which axes move with time and what to watch out for when you change one. Below is only what you need to read a suitability map.

A ladder is a Cartesian product of axes, defined per sector, per subsector and per rung variant in VariantParameters/Tredes/. For housing in the BAU ladder, the axes are redevelopment (2: the parcels that the redevelopment probability released, and the rest), planning capacity with stimuli (6 classes), existing built-up area (3: inside, adjoining a residential core, further away), public transport accessibility (3) and land use (2), together 216 classes. The order of the axes is the order of priority: the first axis weighs most, so a released parcel comes before any other cell, and within that, hard plans occupy the top rungs. A location within hard planning capacity, outside built-up area and without public transport therefore still gets priority over a location outside planning capacity, inside built-up area and near public transport.

The planning capacity axis is not a cross product of plans and stimuli but a sequence of six classes in order (hard plan, soft plan, hard, medium and soft stimuli, outside these): a cell falls into the first class that applies to it.

The ladder decides on cells of 25 metres, the same cells on which the allocation chooses, and is built up per axis: each axis gives, per cell, the first class that is true, and the rung follows from the combination of those axis classes in the order of the axes.

The rung variant per sector is controlled through Trede_Variant in VariantK.dms. The names of the rung sets are not the same as the names of the variants of an application: Intensiveren and Transformeren come from the Ruimtelijke Verkenningen 2023, and a later application may refer to them with a different variant name. See Part III for the applications. The building regime axis is a property of the place and is not to be confused with the construction method, a property of the development package; see #614 for that distinction.

Three of the five housing axes move with time: the redevelopment axis is set in each projection year from the Redevelopment probability (in Dutch), the planning capacity lapses after the projection year in PlancapaciteitGeldigTotEnMet (in ModelParameters/Advanced), and the built-up area grows in each projection year by a disc of 100 metres around what the model built in earlier projection years. The public transport accessibility and the building site stay at the base year. See Rungs (in Dutch) for the reasons and the details.

Interplay between the empirical and design components

The final suitability value used in the allocation is a combination of the rung (design component) and the empirical score. The rung determines the coarse order; the empirical score determines the fine order within each rung.

In the implementation, the empirical map per subsector is first converted into a rank number, which is added to an offset that differs per rung. Locations in lower rungs get a higher offset, so that they always end up above locations in higher rungs, regardless of the empirical score.

Suitability_total = Rung_offset x range + rank_number(Empirical_suitability)

This mechanism ensures that the design preference leads, while within the design preference the empirical basis is decisive. Within a rung, therefore, only the rank order of the empirical score counts, not its level. The competition between subsectors of the same sector for the same cell is decided on the level of the empirical score; for employment, the priority of existing dominant cell, planning capacity and stimuli comes first (see Role in the allocation under the employment section).