Satellite imagery and AI to enhance dwelling yield forecasting

Accurate forecasting of land supply and housing development are central to effective urban and transport planning, and the formulation of policies to guide the provision of infrastructure and services in Western Australia.

Existing forecasting methods are based on various vector-based spatial datasets. An opportunity exists to explore the integrated use of spatially and temporally consistent remotely sensed imagery.

Through leveraging recent advancements in AI-based deep learning analytics and high-resolution time-series satellite and aerial imagery, this project will seek to enhance current dwelling yield forecasting approaches using a data-driven, evidence-based approach.

Participants

• Department of Planning, Lands and Heritage (Western Australia)
• University of Western Australia

Project background

Policy need

Accurate forecasting of development can help to inform effective urban planning and the formulation of policies that guide infrastructure and service provision. Existing local forecasting methods incorporate a number of vector-based spatial and temporal datasets from government agencies including from the Australian Bureau of Statistics (ABS), Valuer General’s Office (VGO), local planning strategies, structure plans and surveys such as the Land Use and Employment survey (LUES).

With increasing availability of high-resolution time-series satellite and aerial imagery and recent advancements in AI-based deep learning analytics where multi-layered neural networks are used to learn patterns found in complex data sets, an opportunity exists to explore the use of spatially and temporally consistent remotely sensed imagery to improve dwelling yield forecasting.

The research will draw on a range of public and quasi-public satellite and aerial imagery datasets (such as Sentinel and Urban Monitor aerial images) and ancillary GIS layers to establish a series of experiments designed to minimise acquisition and computation costs. This will allow for the tailored and flexible extraction of predictive features from remote sensing datasets while ensuring the costs of scaling are minimised and sustainable.

Significance

The significance of this project lies in its ability to deliver a robust AI model that integrates time-series data from high-resolution satellite/aerial imagery, positively contributing to land use forecasting and subsequent planning.

By combining deep learning techniques with spatial analyses, the project will deliver a new level of precision in monitoring and predicting housing developments in both greenfield and infill areas. This comprehensive forecasting tool can provide a detailed understanding of spatial trends, create a scalable infrastructure that can grow to assimilate new and evolving datasets, and enable ongoing assessment of urban growth patterns. Such an approach ensures that the methodology remains adaptable to new datasets, evolving economic conditions, and shifting population dynamics.

Project objectives

The project objective is to develop a robust AI model that integrates time-series data from high-resolution satellite/aerial imagery, combining deep learning techniques with spatial analyses, that will enhance existing land use forecast methodologies for Greater Perth.

Please note …

This page will be a living record of this project. As it matures, hits milestones, etc., we’ll continue to add information, links, images, interviews and more. Watch this space!