Cooling the public realm : data analytics, simulation, and optimization in high-density cities

Abstract

Urban cooling design is of growing importance due to the sustained rise in urban heat island effect and heatwaves globally. The impact has been felt disproportionally in the public realm, in which unfavourable thermal conditions disrupt outdoor activities or enjoyment of places. The design of thermally acceptable public realms, despite recent progress in research literature and professional practices, is confronted with three obstacles. First, the scientific community disagreed upon the quantitative impact of heat on outdoor activities; nor is the range of outdoor thermal conditions conducive to activities identified, since research evidence drawn across climate zones often contradicted each other. Second, design interventions were often governed by quantitative indicators, i.e. the green coverage ratio; less attention was paid to configurational parameters, such as where to plant trees or how to arrange building masses for maximizing cooling potentials; Lastly, simulation models based on the first principles are computationally expensive and do not automatically provide design suggestion, limited their application in optimization. For the obstacles above, advanced digital tools such as analytics, simulation and optimization have not been fully integrated into the decision-making process of urban cooling design.

This thesis aims at methodological innovations for urban cooling design in humid subtropical high-density cities. The objectives are to 1) explore a novel data-driven approach to quantify the heat-activity relationship in open spaces; 2) develop a simulation-based optimization method in search of optimal locations for trees; 3) develop a machine learning algorithm to enhance the traditional simulation-based optimization workflow and its application in urban cooling design.

The thesis is structured in response to the objectives above. Chapter 1 introduces the research background and Chapter 2 outlines the methodologies. In Chapter 3, geo-coded Twitter data have been collected and analyzed in Hong Kong’s major public open spaces; statistical models were used to identify the heat-tweet relationships, and results were compared with field studies in an urban park. In Chapter 4, a simulation-based optimization method was developed to automatically search for the optimal locations for urban trees, and the method was tested using the same urban park given contextual constraints. Chapter 5 presents a machine-learning enhanced optimization approach to identify the optimal building layout for cooling purposes. Specifically, an Artificial Neural Network model was trained as a surrogate in replacement of traditional physics-based models. The new approach has significantly outperformed traditional ones in terms of efficiency and the cooling performances of optimal design. Chapter 6 describes the potential implementations of developed methodologies in urban design.

The original contributions of this research lie in three-fold: First, the study is among the first to use geo-coded social media data to study outdoor activities under thermal variations. Second, the study presents a realistic simulation-based optimization method developed for urban greenery, which is capable to consider the full range of cooling effects of plants such as shading, evapotranspiration, and localized wind at the street scale. Third, the study successfully integrated machine learning algorithms in traditional simulation-based optimization, which marks a significant contribution to modelling and optimization methodology.