Blending science and technology in addressing urbanisation challenges

Geomatics, an interdisciplinary field blending geography and informatics, leverages advanced techniques and methods in data analysis, modelling and visualisation to provide wide-ranging geographic information about the environment and the physical world we are living in.

Geospatial data and analyses provide invaluable insights that help guide urban planning and development, from improving urban infrastructure and predicting traffic flow, to optimising land use and many more actions that make our cities more liveable, functional, sustainable and resilient to the impact of climate change.

At PAIR, Prof. WENG Qihao, Chair Professor of Geomatics and Artificial Intelligence, is leading studies that integrate geospatial artificial intelligence (GeoAI) to obtain important insights about urbanisation and mitigate the impact of urbanisation. He is the Associate Director of the Research Institute for Land and Space (RILS), Director of the JC STEM Lab of Earth Observations, Director of the Research Centre for Artificial Intelligence in Geomatics, and Member of the Otto Poon Charitable Foundation Smart Cities Research Institute (SCRI).

Deep learning: An artificial brain that maps the horizontal and vertical perspectives of landscapes

“In geomatics, deep learning serves as a powerful tool for analysing a wide range of spatial data, such as aerial photos, satellite imagery, video and terrain data.”

Deep learning is an AI technique that mimics the way in which human brains function. The human brain includes billions of interconnected neurons. In a similar manner, deep learning, a biologically inspired technology, constitutes numerous multi-layered artificial neural networks, called “deep neural networks”. Deep learning models are capable of recognising and processing vast amounts of data, such as images, texts, sounds and patterns, to generate knowledge and make predictions without human intervention. In geomatics, deep learning serves as a powerful tool for analysing a wide range of spatial data, such as aerial photos, satellite imagery, video and terrain data. Prof. Weng has been utilising deep learning-driven methods to generate maps and insights which provide us with an accurate understanding of the vertical and horizontal landscapes.

In one study, Prof. Weng’s team generated building height maps using a deep learning-based method that they developed. Building height is an important indicator of the level of urban development along the vertical dimension. The accuracy and precision of building height estimation are important. However, existing large-scale building height estimation studies are based on coarse spatial resolution (e.g., 10 m, 500 m, 1,000 m), and cannot show the fine-scale height variation among buildings in urban areas. Furthermore, the high-resolution images (e.g., < 5 m resolution) available for building-scale height estimation studies often have small spatial coverage and are not openly accessible.

To address the issues of resolution and accessibility, the team developed a deep learning-based super-resolution method for predicting building height at a spatial resolution of 2.5 m from 10-m Sentinel-1/2 images and created an open building height dataset. This repository contains 45,000 samples covering 301 cities in the Northern Hemisphere, including China, the conterminous United States, and Europe. These research outcomes have great application potential in high-resolution database updating, urban planning, and natural disaster assessment.

Automatising the generation of maps from images

Land use land cover (LULC) maps gives a clear indication of the ways in which different categories of land are distributed, as well as the various purposes and manners of land use. For example, the blue colour on an LULC map refers to “water” (e.g., lakes, rivers and streams), while the red colour refers to “built-up area” (e.g., residential, industrial, commercial activities). LULC mapping works by categorising a large quantity of remotely sensed images obtained from different sources. In cloud-prone areas, LULC mapping becomes less accurate due to cloud cover. Prof. Weng’s team integrates deep learning models to address this technological bottleneck.

The team developed an integrated time series mapping method to enhance LULC mapping accuracy and frequency in cloud-prone areas. The researchers applied this novel method to the cloud- and rain-prone Pearl River Delta, China, and achieved an overall mapping accuracy of up to 87.01%, which is higher than that of existing major LULC products around the world. They demonstrated the deep learning-based method’s capacity to provide high-quality LULC data sets at shorter time intervals for various land and water dynamics in cloud- and rain-prone regions.

The urbanisation paradox: What the past tells us about climate change

“Geomatics not only looks at spatial objects and interactions, landscapes and terrains, but also other measurements including human and natural activities on Earth.”

Geomatics not only looks at spatial objects and interactions, landscapes and terrains, but also other measurements including human and natural activities on Earth. By pinpointing the connections between human activities and the environment, we can better understand the causes, progression and impact of climate change, make predictions about future climate, and take sustainability actions.

Prof. Weng’s team probes into longitudinal data related to carbon emission and energy use, and attains critical insights about climate change and its links with human activities. In one study, the team developed a method for mapping urban industrial land (IND) areas in ten countries around the world from the year 2000 to 2019. They explored the way in which IND expansion during the period was correlated with the economic growth and carbon emissions observed in the subnational regions of these countries. According to their findings, the impact of IND expansion on economic growth and carbon emissions varies among regions. Industrial land expansion was found to be a leading factor in economic growth and emissions in developing regions (contributing 31% and 55%, respectively), but not in developed regions (contributing only 8% and 3%, respectively). Education emerged as the primary driver of economic growth in developed regions. These results hold profound implications: the rapid expansion of industrial land accelerates climate change, and the engines for economic growth shift and evolve at different development stages.

Cooling without heating the planet

Cooling is increasingly important in a rapidly warming climate. Cooling the world without heating the planet has become a major scientific endeavour. Developers are now looking for passive cooling solutions that enable more effective dissipation of building heat and prevent heat gain, thus reducing the energy and electricity consumption in buildings.

Green and cool roofs, i.e., building roofs coated with light-reflecting coating materials or plants, are popular cooling solutions. To ascertain the effectiveness of green and cool roofs in reducing energy use in buildings under current and future climate conditions, Prof. Weng’s team evaluated the green and cool roof strategies in six global cities located in different climate zones. The team projected that, by the year 2100, the implementation of green roofs and cool roofs at the city level would lead to substantially lower annual energy use, with reductions of up to 65.51% and 71.72%, respectively, in the energy consumed for heating, ventilation and air conditioning (HVAC) purposes. The study also revealed that the energy saving potentials of cool roofs and green roofs are influenced by local climate conditions. These findings provide useful references for choosing and designing roof strategies suitable to specific urban contexts.

Gearing up for the era of global boiling

“The era of global warming has ended; the era of global boiling has arrived,” United Nations Secretary-General António Guterres warned at press conference in July 2023, the hottest month on record at that time. The record was broken a year later, as July 2024 was 0.03 °C warmer than the record set the previous year.

“Heatwave predictions and projections are crucial for enhancing our preparedness for the hazards and reducing adverse impacts.”

Heatwaves can kill. Heatwave predictions and projections are crucial for enhancing our preparedness for the hazards and reducing adverse impacts. Nevertheless, existing standards for identifying “dangerous” heatwaves have been unsuccessful in capturing these heatwaves in certain climate conditions. The same temperature can feel warmer or cooler in different places. In humid conditions, our body temperature rises faster, and heat dissipation becomes more difficult, thus increasing the risk of heat stroke. This hazard is deadlier in Asia than in other parts of the world. A recent paper suggested that 45% of heat-related deaths each year occur in Asia, and 36% in Europe.

Current heatwave indices need to be more precise and context-specific. Prof. Weng’s team suggests the inclusion of two additional factors—humidity and indoor environment—into existing assessments. The group examined six existing heatwave indices (i.e., maximum daily air temperature, humidity index, humidex, wet bulb globe temperature, lethal heat stress index and universal thermal climate index), all of which are based on outdoor environment conditions. Five of the indices studied were not effective in identifying dangerous heatwave conditions in diverse geographical regions and climate conditions. Furthermore, the team points to the occurrence of heat-related deaths indoors, which have been overlooked by existing outdoor data-based indices. These gaps highlight the need for society to take a retrospective look, re-examine the current system’s ability to assess the severity of heatwaves, and demand further enhancements.

Closing the methodological gap to bring science solutions to wider issues

From building maps and geospatial technologies to studying the impacts of urbanisation, Prof. Weng uses geospatial artificial intelligence to derive insights from data and to design holistic solutions to many environmental, ecological, and climatic issues. He is one of the earliest researchers in the area of urban remote sensing. Previously, Prof. Weng developed a method for estimating land surface temperature from satellite-derived measures of vegetation, and this approach has become a core technique in urban climate studies. His research has now expanded to urban studies on human-environment interactions during the urbanisation process in different geographical settings and stages, from a local to a global scope, with the use of geospatial analytics, GeoAI and big data methods.

“The ultimate goal of my research is to obtain better knowledge about urban environments and global urbanisation processes through remote sensing and geospatial methods for sustainable development.”

“I see myself as an environmental geographer who uses GeoAI, remote sensing, geographic information systems, and spatial modelling methods and techniques to study urban environmental issues and ecosystem dynamics,” said Prof. Weng. “The ultimate goal of my research is to obtain better knowledge about urban environments and global urbanisation processes through remote sensing and geospatial methods for sustainable development,” he added.

Prof. Weng considers GeoAI to be an important interdisciplinary tool that closes the methodological gaps among geography, urban science, landscape ecology and environmental science. “GeoAI represents a powerful convergence of geospatial technology, artificial intelligence, and geographical and environmental studies. In my view, this interdisciplinary integration is crucial because no single discipline can adequately address today’s complex environmental challenges. The global challenges we are facing, such as climate change, urbanisation impacts, and biodiversity loss, are inherently interconnected. Understanding these issues requires us to bridge theoretical frameworks with real-world applications. For instance, GeoAI helps us analyse vast amounts of geospatial data while incorporating physical and ecological principles to develop practical environmental management strategies. To work across disciplines, scholars must cultivate intellectual humility and curiosity. We need to be open to learning new methods, embracing unfamiliar perspectives, and stepping outside our comfort zones. This means being willing to learn new technical skills while also appreciating the theoretical foundations and methods of other fields. Interdisciplinary collaborations are crucial. I believe that dramatic breakthroughs in science and technology and practical solutions for society come from synergistic collaborations. Without collaboration with diverse groups of people, time is lost, and shared insights never emerge. Looking ahead, I believe higher education must evolve to better support interdisciplinary research. This includes developing integrated curricula, creating collaborative research spaces, and training students to think across traditional academic boundaries,” he said.

Advancing the field of urban remote sensing

In the past year, Prof. Weng received the 2024 American Association of Geographers (AAG) Wilbanks Prize for Transformational Research in Geography and the 2024 AAG Remote Sensing Specialty Group Lifetime Achievement Honor Award for his ground-breaking contributions in geography. He is the first Chinese scholar to receive both AAG awards at the same time. Over the years, the pioneer and leader in urban remote sensing has not only conducted research that opened a critical new frontier, but has also authored educational resources adopted widely by universities worldwide and has served the scientific communities (e.g., journals, societies, committees) with his leadership. Prof. Weng is an elected member of many prestigious academic societies, including Academia Europaea (The Academy of Europe).

When asked about his upcoming plans in research, Prof. Weng describes several scientific questions and developments that he is eagerly pursuing. “We have successfully established interdisciplinary research centres at The Hong Kong Polytechnic University, namely, the Research Centre for Artificial Intelligence in Geomatics and JC STEM Lab of Earth Observations. These establishments will not only facilitate interdisciplinary collaboration for impactful research within PolyU to address key societal and environmental challenges in areas of GeoAI applications; they will also fully utilise and expand PolyU’s research capital, fostering Hong Kong’s status as a global research and development (R&D) centre. These research facilities allow us to work strategically with various industries, governments and the public and to translate our research to address critical challenges in society, as well as educating and training tomorrow’s talents with state-of-the-art science and technology,” Prof. Weng explains. In the future, he will continue his research in the field of urban remote sensing. He is particularly motivated by the following key issues and questions, which are crucial to furthering our understanding of urban environments and urbanisation processes worldwide:

• In the twenty-first century there have been significant advances in geospatial technology and GeoAI capacity; however, these techniques have not been well integrated with more established fields, such as geography, urban science, and earth and environmental sciences, to serve as the catalyst for research development and applications. How can all these technologies be integrated for better detection, interpretation, characterisation, and modelling of urban structures and environments?
• Urban landscapes are extremely heterogeneous, temporally dynamic, and spectrally diverse. Can a global model be developed for urban landscape space-time analysis that can account for urban morphologies in different geographical settings to support climate modelling and to benefit the Sustainable Development Goals?
• Human economic and social activities, energy use, and demographic characteristics in cities in the nighttime are distinct from those in the daytime. What is the relationship between the daytime and nighttime urban environments and ecosystems?
• Advances in artificial intelligence (AI) and Earth observation (EO) have transformed urban studies. AI will provide a deeper interpretation and autonomous identification of urban issues and the creation of customised urban designs. Open issues remain, especially in integrating diverse geospatial big data, ensuring data security, and developing a general analytical framework. To explore research directions and emerging trends, new and directed research questions should be considered to address the following questions: How will AI transform urban observing, sensing, imaging, and mapping? How can urban landscapes, phenomena, and events be better perceived and recognised with AI models using EO and geospatial big data?