Demoland

Explaining the science

The project defines four indicators related to the quality of life capturing selected dimensions of the environment, society and economy: air pollution, house price, jobs accessibility, green space accessibility.

For air pollution and house prices, the project develops machine learning-based, predictive models based on land use variables derived from the Urban Grammar project to allow assessment of land use change.

The metric of accessibility to two different opportunities (jobs and green space) is calculated for four modes of transport (walking, bicycle, vehicles, public transit) between a relevant set of origins and destinations.

This project further investigates generating land use proposals for use in urban planning. Based on the generated target outcomes and training data for key indicators, the project develops a Machine Learning algorithm whose results could be used to inform changes in land use to achieve desired outcomes. The project implements a Deep Q Network within a Reinforcement Learning framework. Using this it trains a neural network consisting of current and target indicators in the input layer. The difference between the indicators enters feed-forward mechanisms used to calculate the reward at the output layer. The algorithm initiates a series of actions in the variables that produce indicators to match the targets across the defined geographic areas.

The project is done in collaboration with the Geospatial Commission and Newcastle City Council (NCC). It defines development scenarios of the Tyne and Wear county, for which it reports predicted changes of selected indicators, allowing assessment of a proposed land use change based on machine learning.

The outputs are presented in an interactive visual web-based environment allowing quick comparison and presentation for policy makers and professionals.