Population and urbanisation

National population projections say how many people a country will have. They do not say where those people will live: and for infrastructure, housing and climate adaptation that is the question that matters. We downscale projections to grids and simulate how built-up area follows.

The work went global first at PBL, with 2UP, and has grown considerably since, particularly for the Joint Research Centre of the European Commission.

2UP, the first global model

2UP, Towards an Urban Preview, is PBL's global urbanisation model, and the first time the allocation logic of the RuimteScanner was applied to the entire world. It downscales national population and GDP projections to a grid of 30 arc-seconds, roughly one kilometre at the equator, running to 2100. First it simulates where urban area expands, then it distributes urban and rural population within the result, with each time step feeding the next. Like the Land Use Scanner, it is built in GeoDMS.

The scenarios it downscales are the Shared Socioeconomic Pathways, the same set of futures the IPCC assessments are built on. That is what made its output useful to other people's models: the World Resources Institute uses it as exposure data in the Aqueduct Flood Risk Analyzer, developed with Deltares, the VU, Utrecht University and PBL, and a 2026 paper in Earth's Future uses it to project exposure to floods and landslides. That paper carries an uncomfortable finding for planners: steering urban expansion away from flood-prone land can push it onto unstable slopes, and the other way around.

The model is described in Towards an urban preview (PBL, 2018) and in An integrated global model of local urban development and population change (Koomen et al., 2023).

CRISP, and the World Urbanization Prospects

CRISP (Cities and Rural Integrated Spatial Projections) is the successor. The Joint Research Centre describes it as building on 2UP: built-up development and population change are still modelled in consecutive iterations, but the allocation is new. Where 2UP treated a cell as urban or not, CRISP represents built-up surface as a fraction of the cell, which suits a world where most growth is densification rather than a frontier moving outward.

It turns national population projections into a global one-kilometre grid, from 2020 to 2100, and it works in three steps. First it estimates population and built-up area change for roughly a thousand functional areas, constrained by the national projection. Then it allocates new built-up area to individual grid cells, weighing distance to existing settlement, roads and water against the share already built up. Finally it moves population into the newly built-up and more suitable cells and out of the less suitable ones, which is how internal migration and decline enter the picture.

Modelling the built-up surface first, and only then using it to downscale population, is what makes the result behave like a settlement pattern rather than a smoothed statistical surface.

CRISP underpins the spatial projections in the United Nations' World Urbanization Prospects 2025, published by the Population Division of UN DESA, which covers 237 countries from 1950 to 2050. The UN methodology report sets out how the projections were produced, and the Vrije Universiteit described the contribution here. The gridded results are published by the Joint Research Centre as part of the GHSL data package.

One finding from that report says a good deal about why this work exists: between 1975 and 2025, built-up land grew almost twice as fast as the world's population. Urbanisation is not only a question of how many people there are, but of how much space each of them takes.

The model itself is described in Introducing the CRISP model to downscale future population projections, by Jacobs-Crisioni, Schiavina, Krasnodębska, Dijkstra, Claassens, Hilferink, Van der Wielen and Koomen. Documentation: geodms.nl/crisp.

IGOR

Where CRISP projects totals forward, IGOR breaks them down. The name stands for Individuals Grouped Over Ratios: an iterative proportional fitting method that splits gridded population totals into narrow five-year sex and age groups, currently 36 classes.

It reconciles two sources at once, cell-level sex and broad-age shares from the European 1 km rasters, and region-level population pyramids from census microdata, and produces one population grid per sex and age combination. That level of detail is what makes it possible to ask who lives somewhere, not just how many.

Documentation: the IGOR wiki.