Urban Forecasting

Alexandre Boucher, Michail Fragkias

A pattern-sensitive approach to urban forecasting.


This research proposes a geostatistical approach to urban land cover forecasting. The approach separates in two the task of forecasting urban growth. First, an urban growth model is calibrated to produce the probability that any pixel will be converted from non-urban to urban in a certain period of times. This urban model take into consideration relevant geographical parameters, e.g. roads, topography, existing land covers, as well as socio-economics parameters, e.g. immigration policy, opening to foreign investments. That model yields a map of transition probability not a map of forecasted urban land covers.

The second step is to morph with a geostatistical technique chosen growth patterns on the probability map generated by the urban growth model. This is done by choosing the growth patterns expected to be seen in the future and simulating them conditional to the probability map.

The aim of simulating patterns is not local accuracy but patterns reproduction. Locally accurate maps are smooth and aim for the most likely classification for any pixel. A stochastic simulation map aims at yielding representations with the correct spatial patterns. These two objectives are mutually exclusive; a locally accurate map must be more conservative, hence smoother, than a map aiming at patterns reproduction. Furthermore, contrary to a locally accurate map a simulated map is not unique. There can be several alternate representations that share the same patterns. The potential solutions are explored by generating multiple maps all with the correct patterns and all conditioned with the probability data from the urban growth model. These alternate maps model the inherent uncertainty of the forecasting process.

This research addresses the second step that is the urban growth simulation with geostatistical methods. It is a generic approach in the sense that it can be done independently of the urban growth model chosen. That urban growth model is normally tailored for the particular area being forecasted. An example is shown with the Fragkias-Seto (FS) urban growth model.

Probability of urban growth


Collaborator:Karen Seto (Yale University),Michail Fragkias ( International Project Office IHDP Urbanization and Global Environmental Change project (UGEC))

location: 23°06′32″N 113°15′53″E Guangzhou city in China