Remote sensing, ecosystem analysis and population statistics are useful tools for modelling urban society environment interactions, including risks related to urban heat and water. Considering urbanization and climate change, the evolution of these risks should be assessed for the coming decades to support sustainable urban development. On the one hand we need knowledge on likely future distributions of vulnerable population segments. We also need to have an idea how the urban environment will look like, particularly in terms of sealed surface density given how this affects the urban heat island and flooding. We propose scenario analysis coupled to a framework of complementary modelling and simulation approaches to produce this information.
Develop scenarios for alternative visions on sustainable urban development.
Produce a residential Microsimulation framework to estimate future spatial distributions of the urban population for each scenario.
Estimate local increases in sealed surface cover given simulated increases in household and job density.
Use Cellular Automata to allocate estimated aggregate increases in imperviousness to individual pixels.
We design 2 main scenarios: a Business As Usual (BAU) and a Sustainable (SUS) vision on urban development in our study area, the Brussels Capital Region and its surrounding districts. In the Business As Usual scenario, urban development is not steered toward more sustainability, meaning that traditional planning praxis, that has lead to extremely scattered urban development today, is simply continued. In the Sustainable Development scenario on the other hand, we implement guidelines from contemporary sustainable urban planning research and policy. The core principle of the SUS scenario demands more and denser urban development in or near employment, service and transport hubs, and less or no further development in absence thereof. Sustainable urban development decreases our ecologic footprint considerably. Reuse of historic brownfield development and preventing of new development in flood risk or ecologically important areas are also import features of this scenario.
We use Microsimulation to model residential dynamics (locations of households) on the spatial level of statistical sectors. Using detailed census data, a household segmentation is first developed to cluster households with similar location behavior. Separate discrete choice models are developed for each household segment using a range of relevant geographic features. Simulation is first performed between 2001 and 2011, the years of the last 2 censuses, to assess accuracy. Then, simulations are run between 2012 and 2040 using constraints derived from our 2 scenarios BAU and SUS.
Sealed Surface Estimation
Using a decision tree algorithm called model tree, we estimate sealed surface density in function of observed and simulated increases in population and job density. The model tree serves 2 purposes. On the one hand, it approximates complex non-linear relations using a small set of simple linear regression models. It also identifies clusters of sectors that share underlying linear relations between residential-economic development and sealed surface density. The model tree produces accurate results and is useful for spatial interpretation and prediction of urban dynamics.
A Cellular Automata (CA) model is developed to allocate simulated increases in sealed surface density on statistical sector level to individual pixels. This allows us to produce land cover maps sharing the same format as those derived from Remote Sensing. To train the CA model, neighborhood sealed surface density and distance to roads are considered for observed changes between 2001 and 2013. Historic sealed surface density maps are derived from Landsat satellite imagery. Probability densities are used to model non-linear transition potentials.