School of Science Department of Mathematics 43 A Machine Learning Approach to Study the Relationship between Urban Morphology and Urban Heat Island Supervisor: FUNG Jimmy Chi Hung / MATH Student: ZENG Lingqi / DSCT Course: UROP1100, Spring The Land Surface Temperature (LST) is the radiative skin temperature of the land surface derived from solar radiation. A better estimation of the LST could be useful for predicting the urban climate. A well-trained AI model could be used to predict the future urban heat environment based on different scenarios and provide guidance for mitigation strategies. In this study, four machine learning models: linear regression, multi-layer perceptron, random forest regressor, and epsilon-support vector regression, are trained with urban structure data to predict land surface temperature. Bivariate spline approximation is applied while retrieving the data. The results and performance are also compared with a set of reference training data from a previous study. Retrieving of Urban Morphology from Satellite Images and Machine Learning Supervisor: FUNG Jimmy Chi Hung / MATH Student: KAN Wan On / DSCT Course: UROP1000, Summer Urban morphology, the study of urban forms, is essential to environmental science and city planning because it focuses on the agents and processes responsible for urban transformation over time. This research aims to apply machine learning techniques to satellite images in order to estimate building height information with low cost and high efficiency. As this project progressed in the spring of 2022, we conducted research and developed multiple image segmentation machines and learning models. Currently, our efforts are focused on enhancing pioneer models for more accurate ground truth predictions and image processing techniques for extracting and manipulating more appropriate open-source data sets. These techniques, as well as their outcomes and obstacles, are discussed in detail in this report. Retrieving of Urban Morphology from Satellite Images and Machine Learning Supervisor: FUNG Jimmy Chi Hung / MATH Student: NG Ka Ho / MATH-IRE Course: UROP1100, Summer In this project, we aim to improve the prediction of building dimensions from satellite images by upgrading the UNet approach developed in the previous progress in Spring 2022. By experimenting with modern architecture, Swin-UNet, enhancing data augmentation, altering prediction goals, and running systematically organized trials on a greater scale, we achieved minimum validation categorical cross-entropy losses of 0.322 and 0.048 respectively, on the low-quality target dataset and the high-quality control dataset. This report first revisits the motivations for extracting building dimensions for urban morphology. With the project objective stated, we record the rationale to introduce the above additions to the model together with the experiment results and analysis, during which a small diversion to a literature review of Swin-UNet is included for completeness. The programming side and the optimization of experimental conditions are isolated and discussed.