Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

University of Toronto

$51,200.00

Advancing Machine Learning Approaches for Environmental Modeling

FY 2024-25

Summary
With the explosion of environmental data from a variety of sources, environmental scientists have been adopting data-driven solutions to model environmental processes such as pollution and climate change rather than relying solely on more traditional deterministic approaches. To handle large and often multi-source geospatial data, machine learning (ML), including deep learning (DL), has become increasingly popular, as simpler classification or regression methods do not provide the flexibility of allowing for non-linearities and complex interactions. Characterizing and modeling environmental processes are challenging tasks given the complex nature of their spatial and temporal scales, interactions between components, and dynamic nature. Spatial and temporal correlations are important features to capture in data integration and to generate accurate and physically plausible model predictions. Furthermore, the systematic process of identifying, representing, and quantifying uncertainties inherent in data-driven models and plays a crucial role in how predictions are used and interpreted. Through this research program I aim to advance methods for data-driven pollution and climate modeling, bring focus to the importance of uncertainty quantification, and better understand the implications of uncertainty when applying environmental predictions in downstream applications. These advances will significantly impact how environmental processes are modeled and interpreted, which will in turn have relevant public health and policy implications. My long-term objective is to bring together and lead a highly innovative and interdisciplinary team that conducts groundbreaking research with the goal of generating solutions to these deficiencies. In the short-term, this will be achieved through the implementation of three specific research aims. My first research aim is to leverage concepts from geostatistics to characterize spatial covariance functions into features and functions of ML to both inform data integration and improve predictive performance. Focus will be on modeling environmental factors including air quality, greenhouse gases, and climate change indicators (temperature, drought, wildfires). Predictions from ML rarely come with an estimation of error, thus my second research aim is to develop the methods necessary to enable uncertainty estimates from popular ML algorithms, including DL, by leveraging quantile regression. Finally, despite their importance, uncertainty estimates from environmental model predictions have traditionally been underutilized in health effects assessments. Failure to account for uncertainty can lead to biased results in exposure-response analyses. My third research aim is to develop methods to use the prediction intervals generated from ML models, linking the uncertain environmental exposure values to health data so that we can conduct robust health effects assessments.
Program
Discovery Grants Program - Individual
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2024-06444
Name nom
Franklin, Meredith
Department département
Statistical Sciences and School of the Environment
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
51200
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1506
Committee name en
Geosciences
Committee name fr
Géosciences
Area of application code
401
Area of application group en
Environment
Area of application group fr
Environnement
Area of application en
Climate and atmosphere
Area of application fr
Climat et atmosphère
Research subject code
4405
Research subject group en
Atmospheric science
Research subject group fr
Étude de l'atmosphère
Research subject en
Atmospheric chemistry (including air pollution)
Research subject fr
Chimie atmosphérique (y compris pollution atmosphérique)
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Advancing Machine Learning Approaches for Environmental Modeling
Keywords
environmental modeling; exposure estimation; air quality; machine learning; spatial statistics; uncertainty quantification; environmental health; atmospheric science; climate
Application summary
With the explosion of environmental data from a variety of sources, environmental scientists have been adopting data-driven solutions to model environmental processes such as pollution and climate change rather than relying solely on more traditional deterministic approaches. To handle large and often multi-source geospatial data, machine learning (ML), including deep learning (DL), has become increasingly popular, as simpler classification or regression methods do not provide the flexibility of allowing for non-linearities and complex interactions. Characterizing and modeling environmental processes are challenging tasks given the complex nature of their spatial and temporal scales, interactions between components, and dynamic nature. Spatial and temporal correlations are important features to capture in data integration and to generate accurate and physically plausible model predictions. Furthermore, the systematic process of identifying, representing, and quantifying uncertainties inherent in data-driven models and plays a crucial role in how predictions are used and interpreted. Through this research program I aim to advance methods for data-driven pollution and climate modeling, bring focus to the importance of uncertainty quantification, and better understand the implications of uncertainty when applying environmental predictions in downstream applications. These advances will significantly impact how environmental processes are modeled and interpreted, which will in turn have relevant public health and policy implications. My long-term objective is to bring together and lead a highly innovative and interdisciplinary team that conducts groundbreaking research with the goal of generating solutions to these deficiencies. In the short-term, this will be achieved through the implementation of three specific research aims. My first research aim is to leverage concepts from geostatistics to characterize spatial covariance functions into features and functions of ML to both inform data integration and improve predictive performance. Focus will be on modeling environmental factors including air quality, greenhouse gases, and climate change indicators (temperature, drought, wildfires). Predictions from ML rarely come with an estimation of error, thus my second research aim is to develop the methods necessary to enable uncertainty estimates from popular ML algorithms, including DL, by leveraging quantile regression. Finally, despite their importance, uncertainty estimates from environmental model predictions have traditionally been underutilized in health effects assessments. Failure to account for uncertainty can lead to biased results in exposure-response analyses. My third research aim is to develop methods to use the prediction intervals generated from ML models, linking the uncertain environmental exposure values to health data so that we can conduct robust health effects assessments.