The Lund University Department of Physical Geography and Ecosystem Science, and its Centre for Geographical Information System (GIS Centre), was tasked to develop a course for commissioned education on Geospatial Artificial Intelligence (GeoAI) for Urban Planning. The project was funded by Lund University Commissioned Education (LUCE) through seed money awarded in 2021 to develop new courses.

The suggested course plan is presented below.

Learning Objectives

By studying the course students should be able to:

Knowledge and understanding

• Understand and explain basic GeoAI concepts
• Understand and explain how GeoAI applications, building on GIS and remote sensing (RS) technologies, can be used for urban planning

Skills and abilities

• Use GeoAI methods for satellite image classification
• Use GeoAI methods for spatial modelling and analysis in urban applications

Assessment skills and approaches

• Bring a critical and judicious assessment to GeoAI processing and applications
• Appreciate the advantages and disadvantages of different GeoAi methods

Course Structure

The course is structured in five modules as below:

Module 1: Theory, half a day = 4 hours

In this module, theories of GeoAI and why it is more suitable for spatial modelling and analysis in comparison to traditional AI will be discussed. Optimisation, Machine Learning and simulation techniques will be reviewed.

Module 2: Satellite Image classification, half a day = 4 hours

The overall aim of this module is to learn how to use AI and deep learning methods for satellite image classification to extract urban objects. Questions on how to create a model, what are the parameters to set, etc. will be addressed and discussed. The training is based on a practical exercise using Copernicus data to identify urban objects, e.g., roofs, parks, and vegetation.

Module 3: Optimisation, half a day = 4 hours

Multi-objective optimisation to solve spatial problems in urban areas is the focus of this module. The training will be based on a practical exercise on using a multi-objective optimisation technique, e.g. NSGA-II, for site selection of e.g. solar/wind farms.

Module 4: Simulation, half a day = 4 hours

Developing spatiotemporal simulation models is the aim of this module. The training is based on a practical exercise for flood risk modelling in urban areas.

Module 5: Project, 3 weeks

Students define a project topic and work in groups to use a GeoAI technique for the implementation of the project. The topic might be linked to one of the ongoing projects at SWECO, or an old problem that students want to solve using GeoAI, instead of traditional methods.

Students have 3 weeks to work on the project. The results will be presented and discussed in a half-a- day session.

For each of these modules, interesting data sources for practical work were discussed and identified.