Conference Agenda

Cities grow by extending at their edges and by densifying their existing urban fabric. The latter process has advantages in terms of saving open space, allowing more efficient use of space, sustaining higher amenity levels and limiting travel, but also threatens urban green spaces and results in (over)crowding. The balance between these impacts depends to a large extent on how urban areas are redeveloped. This paper looks at the dominant form of urban redevelopment in the Netherlands: the addition of new residences within existing urban areas.

We investigate the complexity of residential (re)development by analysing 30 million housing-stock mutations recorded in the national building administration for 2012-2025. By combining several spatial data layers, we identify four processes: (1) replacement (demolition followed by new construction on the same parcel), (2) new build on unbuild land, (3) within-building additions or splits, and (4) transformation of non-residential buildings to residential. The results show that three-quarters of the net addition of around one million dwellings took place within the existing urban fabric. Replacement accounts for a growing share of net additions over time, while new build development slightly declines. High-density neighbourhoods—just 7 % of Dutch land—host five times more replacement additions per square kilometre than low-density areas. Regression models reveal that housing growth per neighbourhood correlates positively with higher social rent shares, high densities of urban amenities, and a higher share of protected heritage areas, but negatively with potential available land, indicating that scarcity incentivises densification. Effect sizes differ markedly across density categories and development processes; for example, the positive impact of urban attractiveness on transformation and within-building additions strengthens as neighbourhood density falls.

The study contributes by: (1) providing a reproducible micro-scale method to detect residential development processes; (2) demonstrating that densification is a substantial—and still growing—component of Dutch housing supply; and (3) offering policy-relevant evidence that densification succeeds mainly in neighbourhoods with historic building stock, high amenity levels and limited expansion space. These insights support spatial strategies that blend targeted infill with selective outward growth to meet future housing demand.

Over the last decade, significant advances have been made in the systematic and quantitative characterisation of urban form, for which tools and indicators were developed. At the same time, urban planning authorities have remained poorly equipped both theoretically and practically for facing contemporary urban challenges, such as densification. Built density is still often apprehended through simple surface-based indicators such as floor area ratio (FAR) and building coverage ratio (BCR). Although these indicators are easily intelligible, many papers have argued that they cannot totally encompass the multifaceted nature of density, especially in complex urban landscapes. Yet, too little attempt was made to link tools developed in the academic literature and urban planning practices.

Our communication proposes a simple methodological framework to characterise built density using a typological approach. The method is intended to be fully reproducible and easily adaptable to many cities (for the resulting ‘density types’ are defined on local data) in order to be adopted in planning decision-making. Our work is conducted in the context of the new Brussels Land Use Plan drafting, in collaboration with the Brussels Planning Agency.

First, we identify a series of indicators able to account for the many dimensions of built density and compute them. Second, we perform a Principal Component Analysis (PCA) on these indicators to create uncorrelated multidimensional variables that we map, describe and define as components of density. Third, we apply a clustering algorithm on the most informative components and map the resulting clusters, i.e. density types.

Our preliminary results show that considering more indicators of built density than FAR and BCR helps at differentiating more complex situations and nuancing our view on the urban landscape. In the context of regulating future densification and planning de-densification of certain areas in Brussels, we initiate a reflexion on how a quantitative multidimensional understanding of built density can translate into tangible urban planning policies.

Cities are increasingly under pressure to accommodate growing populations, leading to a rising demand for new housing. While urban expansion and large-scale brownfield conversions continue to play a role, much of this demand is met by small-scale, incremental densification, particularly in suburban areas. These subtle yet frequent changes are considered to alter urban form and increase population densities in locations often characterized by limited infrastructure capacity. For urban planning this poses significant challenges as it contradicts long-term assumptions about where development is likely to occur. However, two key issues complicate insights into incremental densification: first, small-scale processes often take place within existing building right and outside the scope of formal planning and are poorly captured by existing statistics—especially in contexts like Germany; second, estimating future processes is inherently difficult, as decisions to densify depend heavily on individual preferences of landowners, which frequently diverge from market-driven logic. Addressing these challenges is essential for creating more resilient and responsive planning frameworks for incremental densification.

In this contribution, we thus present an approach that allows us to measure and predict incremental densification. As a case, we look at densification in suburban areas of the city of Munich in the period of 2013-2023 to then predict the probability for future development at the level of individual plots. We measure incremental changes based on building footprints and train a random forest model. As predictor variables we use high-resolution socio-demographic data, accessibility and neighbourhood indicators as well as plot and building characteristics, e.g. land prices, age of residents, type of owner, the built-up ratio or the year of construction of existing buildings. The trained model is then used to predict probabilities for future development from 2023 onwards. We develop two models: Whether incremental densification takes place or not (classification problem) and how much new floor area can be expected on a parcel (regression problem).

The results show a 10% overall increase of floor area in suburban areas of Munich. The predominant process is replacement of single-family homes by apartment blocks. For the classification if densification takes place or not, we achieve an overall accuracy of 0.77 and a ROC-AUC of 0.86. The initial floor area ratio, the size and form of the parcel as well as the year of construction of the existing building showed to be important predictors in the model. The prediction of new floor area yields weaker results, with an R² value of 0.27. For the prediction, we applied a robust estimation strategy and opted to first predict the probability for future development per parcel and use a factor derived from the empirical data to derive an average increase of floor area. Overall, the results are highly promising for estimating future densification in suburban areas, providing an evidence-based support for strategic planning.