Using deep learning and satellite imagery to predict spatial and temporal variations in transport and employment accessibility in data-sparse urban co

This project seeks to develop a neural network model (NNM) to produce estimates of transport and
employment accessibility in data-sparse urban areas by drawing on data from both rich and sparse
urban data contexts and satellite imagery. Specifically, the project seeks to:
1) Measure Transport and Employment Accessibility (TA and EA) at neighbourhood level in richand
sparse-data urban settings;
2) Build a NNM based on satellite imagery to predict longitudinal and geographical changes in
TA and EA;
3) Develop a time-adjusted weighting system for the NNM to create reliable longitudinal TA and EA
predictions in sparse urban data settings.
By addressing these aims, the project will innovate by developing a scalable, transferable
methodology for leveraging existing data and open access satellite imagery to create annual TA and
EA estimates at high spatial resolution in sparse urban data contexts. By doing so, it ultimately seeks
to help progress the United Nations' (UN) Sustainable Development Goals (SDGs) by generating
nonexistent, timely, geographically disaggregated data on TA and EA in these areas (SDG 17).
Specifically it seeks to inform interventions to reduce inequalities (SDG 10) and improve access to
Employment and public transport (SDG 11).