{"id":{"repo_id":"cambridge","oai_identifier":"oai:www.repository.cam.ac.uk:1810/390929"},"canonical_url":"https://search.dev.ndltd.org/etd/cambridge/oai:www.repository.cam.ac.uk:1810/390929","repository":{"repo_id":"cambridge","name":"Cambridge University","base_url":"https://api.repository.cam.ac.uk/server/oai/request"},"display":{"title":"Understanding ozone, climate and their interactions with causal machine learning","abstract":"Understanding and projecting environmental risks is essential to inform society's response to climate change. This thesis explores two closely related risks, surface ozone and temperature. Simulating both currently requires the use of complex numerical models, which are limited by incomplete process knowledge, inadequate resolution and subsequent necessary parametrisations. This thesis develops and applies data-driven approaches to model ozone and temperature, aiming to alleviate some of the limitations of numerical models and leverage growing volumes of data characterising the Earth system. Chapters 1 and 2 introduce the main themes of the thesis, including the strengths and weaknesses of numerical Earth system modelling, outline why data-driven modelling, particularly machine learning, may address some weaknesses and describe relevant background on tropospheric ozone, climate projections and causal machine learning. In Chapter 3, we explore the trade-offs when building large deep learning models for ozone air pollution forecasting, an important task to predict risks to human health. We develop a state-of-the-art transformer to forecast ground-level ozone, finding that the model forecasts with high accuracy across Europe, including generalising to unseen regions. However, we also identify current limitations in interpretability and performance in forecasting extremes, limiting the use of deep learning models to improve our process understanding of the factors controlling ozone. Chapter 4, addressing some of the limitations regarding interpretability and process understanding in the previous chapter, derives a novel observationally-based estimate of the sensitivity of ozone to temperature using data-driven causal methods, quantifying how this sensitivity varies under different meteorological conditions and providing insights into how ozone risks may change in future climates. However, the findings in this chapter also reinforced the need for robust physically-based models for out-of-distribution generalisation and climate projections. Given the limiting computational expense of numerical climate models, Chapter 5 develops interpretable causal models capable for the first time of autoregressive climate model emulation at monthly timesteps, facilitating rapid generation of large ensembles of temperature projections, and working towards an assessment of how ozone risks may change given future temperatures. We also show that our interpretable, causal model can be used for counterfactual experiments and attribution studies, paving the way to actionable insights from climate model emulators. Finally, Chapter 6 describes how causal machine learning approaches may complement traditional numerical modelling, and outlines future directions for causal machine learning to provide insights and actionable tools to project and understand the evolution of ozone and temperature under climate change.","abstract_html":"Understanding and projecting environmental risks is essential to inform society's response to climate change. This thesis explores two closely related risks, surface ozone and temperature. Simulating both currently requires the use of complex numerical models, which are limited by incomplete process knowledge, inadequate resolution and subsequent necessary parametrisations. This thesis develops and applies data-driven approaches to model ozone and temperature, aiming to alleviate some of the limitations of numerical models and leverage growing volumes of data characterising the Earth system. Chapters 1 and 2 introduce the main themes of the thesis, including the strengths and weaknesses of numerical Earth system modelling, outline why data-driven modelling, particularly machine learning, may address some weaknesses and describe relevant background on tropospheric ozone, climate projections and causal machine learning. In Chapter 3, we explore the trade-offs when building large deep learning models for ozone air pollution forecasting, an important task to predict risks to human health. We develop a state-of-the-art transformer to forecast ground-level ozone, finding that the model forecasts with high accuracy across Europe, including generalising to unseen regions. However, we also identify current limitations in interpretability and performance in forecasting extremes, limiting the use of deep learning models to improve our process understanding of the factors controlling ozone. Chapter 4, addressing some of the limitations regarding interpretability and process understanding in the previous chapter, derives a novel observationally-based estimate of the sensitivity of ozone to temperature using data-driven causal methods, quantifying how this sensitivity varies under different meteorological conditions and providing insights into how ozone risks may change in future climates. However, the findings in this chapter also reinforced the need for robust physically-based models for out-of-distribution generalisation and climate projections. Given the limiting computational expense of numerical climate models, Chapter 5 develops interpretable causal models capable for the first time of autoregressive climate model emulation at monthly timesteps, facilitating rapid generation of large ensembles of temperature projections, and working towards an assessment of how ozone risks may change given future temperatures. We also show that our interpretable, causal model can be used for counterfactual experiments and attribution studies, paving the way to actionable insights from climate model emulators. Finally, Chapter 6 describes how causal machine learning approaches may complement traditional numerical modelling, and outlines future directions for causal machine learning to provide insights and actionable tools to project and understand the evolution of ozone and temperature under climate change.","abstract_has_math":false,"creators":["Hickman, Sebastian"],"institution":"University of Cambridge","degree_name":"Doctor of Philosophy (PhD)","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Archibald, Alexander"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-05-03","date_published":"2025-05-03","updated_at":"2026-07-22T22:24:13Z","subjects":["Climate","Machine learning"],"languages":["eng"],"rights":[],"rights_urls":["https://www.repository.cam.ac.uk/bitstreams/02987e18-8360-4f7f-88cc-781b64c9c74a/download","https://creativecommons.org/licenses/by/4.0/"],"identifier_entries":[]},"links":{"outbound_url":"https://doi.org/10.17863/CAM.122273","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Archibald, Alexander"]},{"key":"dc:creator","label":"Author","values":["Hickman, Sebastian"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2025-05-03"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Cambridge"]},{"key":"dc:relation.isreferencedby.uri","label":"Dc Relation Isreferencedby URI","values":["https://www.repository.cam.ac.uk/handle/1810/390929"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["Doctoral"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Climate","Machine learning"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["https://www.repository.cam.ac.uk/bitstreams/02987e18-8360-4f7f-88cc-781b64c9c74a/download","https://creativecommons.org/licenses/by/4.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.17863/CAM.122273"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://www.repository.cam.ac.uk/bitstreams/20e93246-0232-4c42-87fd-b9a05a2a16f7/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Understanding and projecting environmental risks is essential to inform society's response to climate change. This thesis explores two closely related risks, surface ozone and temperature. Simulating both currently requires the use of complex numerical models, which are limited by incomplete process knowledge, inadequate resolution and subsequent necessary parametrisations. This thesis develops and applies data-driven approaches to model ozone and temperature, aiming to alleviate some of the limitations of numerical models and leverage growing volumes of data characterising the Earth system. Chapters 1 and 2 introduce the main themes of the thesis, including the strengths and weaknesses of numerical Earth system modelling, outline why data-driven modelling, particularly machine learning, may address some weaknesses and describe relevant background on tropospheric ozone, climate projections and causal machine learning. In Chapter 3, we explore the trade-offs when building large deep learning models for ozone air pollution forecasting, an important task to predict risks to human health. We develop a state-of-the-art transformer to forecast ground-level ozone, finding that the model forecasts with high accuracy across Europe, including generalising to unseen regions. However, we also identify current limitations in interpretability and performance in forecasting extremes, limiting the use of deep learning models to improve our process understanding of the factors controlling ozone. Chapter 4, addressing some of the limitations regarding interpretability and process understanding in the previous chapter, derives a novel observationally-based estimate of the sensitivity of ozone to temperature using data-driven causal methods, quantifying how this sensitivity varies under different meteorological conditions and providing insights into how ozone risks may change in future climates. However, the findings in this chapter also reinforced the need for robust physically-based models for out-of-distribution generalisation and climate projections. Given the limiting computational expense of numerical climate models, Chapter 5 develops interpretable causal models capable for the first time of autoregressive climate model emulation at monthly timesteps, facilitating rapid generation of large ensembles of temperature projections, and working towards an assessment of how ozone risks may change given future temperatures. We also show that our interpretable, causal model can be used for counterfactual experiments and attribution studies, paving the way to actionable insights from climate model emulators. Finally, Chapter 6 describes how causal machine learning approaches may complement traditional numerical modelling, and outlines future directions for causal machine learning to provide insights and actionable tools to project and understand the evolution of ozone and temperature under climate change."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["000229ed258ae3a784591ad4f2b7d38e","87eda9de84448d1f82354d60eee3eb5f"]},{"key":"dc:title","label":"Title","values":["Understanding ozone, climate and their interactions with causal machine learning"]}]}],"canonical_facts":{"dc:contributor.advisor":["Archibald, Alexander"],"dc:creator":["Hickman, Sebastian"],"dc:date.issued":["2025-05-03"],"dc:description.abstract":["Understanding and projecting environmental risks is essential to inform society's response to climate change. This thesis explores two closely related risks, surface ozone and temperature. Simulating both currently requires the use of complex numerical models, which are limited by incomplete process knowledge, inadequate resolution and subsequent necessary parametrisations. This thesis develops and applies data-driven approaches to model ozone and temperature, aiming to alleviate some of the limitations of numerical models and leverage growing volumes of data characterising the Earth system. Chapters 1 and 2 introduce the main themes of the thesis, including the strengths and weaknesses of numerical Earth system modelling, outline why data-driven modelling, particularly machine learning, may address some weaknesses and describe relevant background on tropospheric ozone, climate projections and causal machine learning. In Chapter 3, we explore the trade-offs when building large deep learning models for ozone air pollution forecasting, an important task to predict risks to human health. We develop a state-of-the-art transformer to forecast ground-level ozone, finding that the model forecasts with high accuracy across Europe, including generalising to unseen regions. However, we also identify current limitations in interpretability and performance in forecasting extremes, limiting the use of deep learning models to improve our process understanding of the factors controlling ozone. Chapter 4, addressing some of the limitations regarding interpretability and process understanding in the previous chapter, derives a novel observationally-based estimate of the sensitivity of ozone to temperature using data-driven causal methods, quantifying how this sensitivity varies under different meteorological conditions and providing insights into how ozone risks may change in future climates. However, the findings in this chapter also reinforced the need for robust physically-based models for out-of-distribution generalisation and climate projections. Given the limiting computational expense of numerical climate models, Chapter 5 develops interpretable causal models capable for the first time of autoregressive climate model emulation at monthly timesteps, facilitating rapid generation of large ensembles of temperature projections, and working towards an assessment of how ozone risks may change given future temperatures. We also show that our interpretable, causal model can be used for counterfactual experiments and attribution studies, paving the way to actionable insights from climate model emulators. Finally, Chapter 6 describes how causal machine learning approaches may complement traditional numerical modelling, and outlines future directions for causal machine learning to provide insights and actionable tools to project and understand the evolution of ozone and temperature under climate change."],"dc:format.checksum.md5":["000229ed258ae3a784591ad4f2b7d38e","87eda9de84448d1f82354d60eee3eb5f"],"dc:identifier.doi":["https://doi.org/10.17863/CAM.122273"],"dc:identifier.uri":["https://www.repository.cam.ac.uk/bitstreams/20e93246-0232-4c42-87fd-b9a05a2a16f7/download"],"dc:language":["eng"],"dc:publisher.institution":["University of Cambridge"],"dc:relation.isreferencedby.uri":["https://www.repository.cam.ac.uk/handle/1810/390929"],"dc:rights":["https://www.repository.cam.ac.uk/bitstreams/02987e18-8360-4f7f-88cc-781b64c9c74a/download","https://creativecommons.org/licenses/by/4.0/"],"dc:subject":["Climate","Machine learning"],"dc:title":["Understanding ozone, climate and their interactions with causal machine learning"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["Doctoral"],"dc:type.qualificationname":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-07-22T22:24:13Z"}