Hybrid AI and multiscale physical modelling for optimal urban decarbonisation combating climate change
Lead Research Organisation:
Imperial College London
Department Name: Earth Science and Engineering
Abstract
The challenges articulated in this proposal are how to (1) accurately assess carbon emissions in urban areas; (2) help design and manage cities so that the carbon footprint is reduced; and (3) quantify the impact of urban carbon emissions on global climate change-towards the 1.5 degree climate goal.
Greenhouse gas emission reduction is key to tackling global warming. In 2020, 24% of net greenhouse gas emissions in the UK were estimated to be from the transport sector, 21% from energy supply, 18% from business, 16% from the residential sector and 11% from agriculture [1]. Accurate assessment of urban carbon emissions will help policy makers in their decision-making processes and managers of public and private spaces to optimise energy use, carbon reduction and economic benefit. Models are powerful tools in understanding carbon life cycle and atmospheric processes, making predictions, uncertainty quantification and optimal control/design for decarbonisation. However, integrated assessment of the environment and human development is arguably the most difficult and important "systems" problem faced [2]. The complex carbon cycle and atmospheric physical processes act over a wide range of spatial (from meters to degrees) and temporal (from hours, days to decades) scales. Currently, there is no integrated modelling across neighbourhood, city and global scales which can be used for exploring the complex relationship between carbon emissions associated with human activities and global climate change.
Here I aim to develop a hybrid AI (Artificial Intelligence)-multiscale physics-informed optimal management framework for accurate assessment and mitigation of CO2 in urban areas. Effective carbon assessment and management necessitate the implementation of multiscale carbon models that can capture adequate spatial and temporal variability of urban carbon emissions & dispersion patterns. Current models are either excessively computationally expensive, or fail to capture the detailed variability of such problems. The proposed work will advance the status of science by developing an advanced multiscale carbon model (based on our recently developed Fluidity-Urban model) where, the use of dynamically adapted meshes enables us to resolve complex urban turbulent flows and carbon dispersion processes. The effect of city infrastructures on carbon dispersion processes is considered at different scales. AI-based modelling will then be used for the optimal design of urban infrastructures and layout for mitigation of carbon emissions. Energy efficiency and carbon-based energy usage in cities are measured based on detailed datasets of existing infrastructures in the selected city-London. The modelling framework will include new carbon parameterisation schemes for urban infrastructures/layout, enabling more accurate assessment of urban carbon emissions, and their impact on climate change. Potential improvements to existing urban infrastructures, and optimal designs for new urban developments will be provided through the AI-based optimal control tool proposed here for carbon reduction and energy efficiency. Finally, an AI-based GHG parameterisation module will be developed for coupling the calculated CO2 fluxes at high resolution grids with existing Earth System modelling. The impact of carbon emissions in cities on global climate can then be evaluated accurately based on existing and improved city infrastructure and layouts.
This innovative framework will allow the critical assessment of existing and new policy options on decarbonisation to be carried out, thus improving local and global climate. The tool could potentially change the way in which city infrastructure design, GI and BI for decarbonisation are used in our future cities and pave the way for accurate quantification of the impact of urban carbon emissions on global warming.
[1] BEIS, N.. 2020 UK Greenhouse Gas Emissions, Final Figures.
[2] Navarro et al. 2018. Earth Syst. Dynam., 9, 1045
Greenhouse gas emission reduction is key to tackling global warming. In 2020, 24% of net greenhouse gas emissions in the UK were estimated to be from the transport sector, 21% from energy supply, 18% from business, 16% from the residential sector and 11% from agriculture [1]. Accurate assessment of urban carbon emissions will help policy makers in their decision-making processes and managers of public and private spaces to optimise energy use, carbon reduction and economic benefit. Models are powerful tools in understanding carbon life cycle and atmospheric processes, making predictions, uncertainty quantification and optimal control/design for decarbonisation. However, integrated assessment of the environment and human development is arguably the most difficult and important "systems" problem faced [2]. The complex carbon cycle and atmospheric physical processes act over a wide range of spatial (from meters to degrees) and temporal (from hours, days to decades) scales. Currently, there is no integrated modelling across neighbourhood, city and global scales which can be used for exploring the complex relationship between carbon emissions associated with human activities and global climate change.
Here I aim to develop a hybrid AI (Artificial Intelligence)-multiscale physics-informed optimal management framework for accurate assessment and mitigation of CO2 in urban areas. Effective carbon assessment and management necessitate the implementation of multiscale carbon models that can capture adequate spatial and temporal variability of urban carbon emissions & dispersion patterns. Current models are either excessively computationally expensive, or fail to capture the detailed variability of such problems. The proposed work will advance the status of science by developing an advanced multiscale carbon model (based on our recently developed Fluidity-Urban model) where, the use of dynamically adapted meshes enables us to resolve complex urban turbulent flows and carbon dispersion processes. The effect of city infrastructures on carbon dispersion processes is considered at different scales. AI-based modelling will then be used for the optimal design of urban infrastructures and layout for mitigation of carbon emissions. Energy efficiency and carbon-based energy usage in cities are measured based on detailed datasets of existing infrastructures in the selected city-London. The modelling framework will include new carbon parameterisation schemes for urban infrastructures/layout, enabling more accurate assessment of urban carbon emissions, and their impact on climate change. Potential improvements to existing urban infrastructures, and optimal designs for new urban developments will be provided through the AI-based optimal control tool proposed here for carbon reduction and energy efficiency. Finally, an AI-based GHG parameterisation module will be developed for coupling the calculated CO2 fluxes at high resolution grids with existing Earth System modelling. The impact of carbon emissions in cities on global climate can then be evaluated accurately based on existing and improved city infrastructure and layouts.
This innovative framework will allow the critical assessment of existing and new policy options on decarbonisation to be carried out, thus improving local and global climate. The tool could potentially change the way in which city infrastructure design, GI and BI for decarbonisation are used in our future cities and pave the way for accurate quantification of the impact of urban carbon emissions on global warming.
[1] BEIS, N.. 2020 UK Greenhouse Gas Emissions, Final Figures.
[2] Navarro et al. 2018. Earth Syst. Dynam., 9, 1045
Organisations
- Imperial College London (Lead Research Organisation)
- University of Surrey (Collaboration, Project Partner)
- Chinese Academy of Sciences (Collaboration)
- Chongqing University (Collaboration)
- National Institute for Space Research Brazil (Collaboration)
- China Meteorological Administration (Project Partner)
- King Abdullah University of Sci and Tech (Project Partner)
- DIREK LTD (Project Partner)
- Institute of Urban Environment (Project Partner)
- Suzhou Dahuan Technology Co. Ltd (Project Partner)
- Institute of Atmospheric Physics CAS (Project Partner)
- Arup Group (Project Partner)
- Florida State University (Project Partner)
Publications

Cai S
(2024)
A Hybrid Data-Driven and Data Assimilation Method for Spatiotemporal Forecasting: PM2.5 Forecasting in China
in Journal of Advances in Modeling Earth Systems

Cai S
(2024)
Advancing neural network-based data assimilation for large-scale spatiotemporal systems with sparse observations
in Physics of Fluids

Description | Online lectures: AI Modelling in sustainable urban environmental design and control |
Geographic Reach | Asia |
Policy Influence Type | Influenced training of practitioners or researchers |
Title | A Python tool for generating high-resolution urban CO2 emission gridmap in natural gas consumption sector |
Description | The tool disaggregates natural gas consumption in urban area at designated resolution. The tool uses energy performance certificate (EPC) and display energy certificate (DEC) records for domestic, public, and commercial properties to estimate natural gas consumption per property. This gas consumption can then be aggregated into a high-resolution CO2 emission gridmap. The tool is based on Python, Pandas, and GIS packages, rasterio and geopandas. |
Type Of Material | Data analysis technique |
Year Produced | 2024 |
Provided To Others? | No |
Impact | Natural gas consumption in domestic and non-domestic sectors are significant in urban carbon emissions (55% London). The tool fills in the gap that current emission inventories lack gridmap with less than 1km resolution. Such high-resolution emission sources is essential to physical models in local urban environment, especially the interaction between buildings and green infrastractures. |
Title | Case study: mutiscale physical modelling for urban environment near the BT tower in London |
Description | A case study is setup for atmospheric and environmental research near the BT tower in London. It is used for investigate the impact of green infrastructures on local climate and environment. The outcomes of datasets: velocity, temperature, humidity. |
Type Of Material | Computer model/algorithm |
Year Produced | 2025 |
Provided To Others? | No |
Impact | Providing the details of local flows, temeperature and humidity for further investigating the impact of green infrastructures on local atmospheric and environment. |
Title | Land-use regression model for urban carbon dioxide concentration |
Description | Urbanization and land use change are strongly correlated with increased urban CO2 emissions, highlighting the need to study spatiotemporal trends in intraurban CO2 to inform sustainable spatial planning of cities. This model uses land use regression (LUR) to predict intraurban CO2 concentrations. The LUR model is developed as a case study in the San Francisco Bay Area using data from the BEACO2N monitoring network. Furthermore, LUR is compared to machine learning (ML) algorithms that explore non-linear relationships, representing a two-fold novel contribution. Model performance is evaluated using reserved data from unseen sensor locations. The highest predictive accuracy is achieved using extreme gradient boosting (XGBoost) and a convolutional neural network (CNN), both with R² values of 0.58, outperforming traditional LUR, which achieved an R² of 0.34. XGBoost and CNN also outperformed traditional LUR for unseen sensor locations, accounting for up to 42% of the variability in observed CO2 concentrations. These models can help understand city land use and carbon budgets, thereby aiding urban planning and management. |
Type Of Material | Computer model/algorithm |
Year Produced | 2024 |
Provided To Others? | No |
Impact | This model was the first of its kind to evaluate the viability of using LUR to predict intraurban ambient CO2 concentrations. It has been demonstrated that ML-based algorithms can outperform traditional LUR in terms of predictive accuracy, illustrating that the relationship between environmental features and CO2 should not be presumed to be linear. This conclusion corroborates the findings of a previous study, underscoring the potential and value of expanding upon traditional LUR using novel modeling approaches. Evaluating models using data from unseen node locations further illustrated the models' potential site-to-site transferability at the locations situated centrally within the training node network, but poor performance on fringe test nodes. The availability and spatiotemporal variability of data from intraurban CO2 monitoring networks is scarce and remains a major limitation for research in this field. In light of rising urbanization and the increasingly drastic effects of climate change, efforts to establish intraurban CO2 monitoring networks with high spatiotemporal resolution should be expanded. More research into the modeling of ambient intraurban CO2 is necessary to better understand and predict the effects of anthropogenic urban activities and land use change on climate change. |
Description | AI and data assimilation |
Organisation | National Institute for Space Research Brazil |
Country | Brazil |
Sector | Public |
PI Contribution | Joint supervise PhD student on AI and data assimilation study - weekly meeting |
Collaborator Contribution | Joint supervise PhD student on AI and data assimilation study - weekly meeting |
Impact | Presentation on MONAN Meeting: AI for atmosphere/ocean dynamics: https://www.youtube.com/watch?v=P6g65RMzJ_k |
Start Year | 2024 |
Description | Bulit environment |
Organisation | Chongqing University |
Country | China |
Sector | Academic/University |
PI Contribution | Proving multiscale physical urban models and AI techniques, online lectures |
Collaborator Contribution | Providing data |
Impact | Joint papers: 10.1155/2024/5531325 Joint book: Resilient Urban Environments, eBook ISBN, 978-3-031-55482-7, Print ISBN 978-3-031-55481-0 |
Start Year | 2024 |
Description | Green-blue infrastructures in urban areas |
Organisation | University of Surrey |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | Modelling of impact of green-blue infrastructures on urban environment |
Collaborator Contribution | Monitoring the impact of G-B infrastructures |
Impact | Presentation on RELAIM25 |
Start Year | 2024 |
Description | Mutiscale physical-based atmospheric models |
Organisation | Chinese Academy of Sciences |
Department | Institute of Atmospheric Physics (IAP) |
Country | China |
Sector | Academic/University |
PI Contribution | Working on development next generation adaptive mesh atmopsheric model |
Collaborator Contribution | Working on applications of next generation adaptive mesh atmopsheric model |
Impact | Presentations on MOW2024 (Mathematics of the Weather) (https://www.emetsoc.org/events/event/workshop-mow2024-mathematics-of-the-weather) LI, Jinxi et al. The construction of a three?dimensional dynamically adaptive finite-element atmospheric model Fluidity-Atmos GAN, Pu et al. A Preliminary 3D AI-Driven Adaptive Mesh Technique in Adaptive Atmospheric Model Fluidity-Atmosphere |
Start Year | 2024 |
Description | Mutiscale physical-based urban environmental models |
Organisation | Chinese Academy of Sciences |
Department | Institute of Urban Environment |
Country | China |
Sector | Academic/University |
PI Contribution | Working on development of mutiscale urban environmental models |
Collaborator Contribution | Working on applications of mutiscale urban environmental models |
Impact | https://doi.org/10.3390/atmos15091037 |
Start Year | 2024 |
Description | Invited talk at the workshp of Mathematics Of the Weather 2024 in Germany |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | Talk title: Hybrid AI and multiscale physical modelling for optimal urban decarbonisation combating climate change |
Year(s) Of Engagement Activity | 2024 |
URL | https://tpchange.de/meetings/4310-2/ |
Description | LI, Linfeng gave a talk on MOW24 |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | The talk title: Neural Network Implementation of High-order Discontinuous Galerkin Methods |
Year(s) Of Engagement Activity | 2024 |
URL | https://tpchange.de/meetings/4310-2/ |
Description | Organisation of the first Healthy People and Healthy Planet workshop: Leveraging AI for Decarbonised, Healthy, Inspiring and Energy-Positive Cities |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Other audiences |
Results and Impact | The theme of this workshop is: "Healthy People and Healthy Planet: AI for Decarbonized, Healthy, Inspiring, and Energy Positive Cities." It aims to bring together leading experts, researchers, and practitioners to explore the transformative potential of artificial intelligence in urban sustainability. The meeting will focus on how AI can be leveraged to reduce carbon emissions, enhance public health, inspire innovative urban design, and create energy-positive environments. Key objectives include sharing cutting-edge research, fostering interdisciplinary collaborations and funding opportunities, listening to voices from industry and communities, and developing actionable strategies to integrate AI technologies into urban planning and policy. By the end of the workshop, participants will have identified existing problems in the urban environment and practical AI-driven solutions to advance the development of smart, sustainable, and resilient cities. We discussed the Challenges in NetZero and AI in urban sustainability and management. The main points are listed below: 1. Leveraging technology to promote healthy living 2. Challenges in existing technologies toward Net-Zero 3. Policy changes and development control using digital tools 4. Energy use, economics and carbon reduction 5. Trustworthy and generative AI 6. Healthy built environment 7. Future/Potential Funding |
Year(s) Of Engagement Activity | 2024 |
URL | https://ai4urban.github.io/blog/2024/kickoff-meeting/ |
Description | Organisation of the second Healthy People and Healthy Planet workshop: Delivering Change Managing Growth with Digital Twins |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Policymakers/politicians |
Results and Impact | Aims: ===== Delivering more equitable, resource efficient, healthier homes, whilst achieving COP 29 climate change obligations to ensure that '1.5 is kept alive', demands a major shift in Planning, Housing, Transport and Environmental Policies. Healthy People, Healthy Planet (HP2), founded by Imperial College and the Universities of New York, Surrey, Southampton plus climate tech experts - OpenWeather, aims to navigate the Planning Minefield by ensuring future decision-making is driven by data, science, and collaboration. HP2 is developing integrated AI/ Digital Twin technologies to empower communities to analyse needs, synthesise and prioritise options and test proposed solutions in a virtual world, thereby minimising risk in the real world. Vision: ===== Shape the cities of tomorrow by integrating innovative technologies, sustainable practices, and inclusive planning. This workshop aims to inspire collaborative solutions that balance economic growth, environmental resilience, and social equity, creating thriving urban environments for future generations. |
Year(s) Of Engagement Activity | 2025 |
URL | https://ai4urban.github.io/blog/2025/hp2-2nd-workshop/ |
Description | Presentation (Poster) at RECLAIM Network Plus Conference January 2025 |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Policymakers/politicians |
Results and Impact | We presented our initial outcome of our research project, which received attention from audient. |
Year(s) Of Engagement Activity | 2025 |
Description | Visiting the Environmental Research Group (ERG) at Imperial College London |
Form Of Engagement Activity | A formal working group, expert panel or dialogue |
Part Of Official Scheme? | No |
Geographic Reach | Local |
Primary Audience | Study participants or study members |
Results and Impact | Discussed CO2 monitoring datasets available at ERG whch can be used for validation of CO2 models and AI training. |
Year(s) Of Engagement Activity | 2024 |