Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

The University of British Columbia

$34,000.00

Dynamic Gaussian Processes to Identify Environmental Mechanisms

EF 2024-25

Résumé
Collaboration with a research lab of hydrologists has identified scientific objectives calling for novel statistical methodology. The lab has rich data sets for over 1,000 watershed catchments, i.e., the area draining into a single streamflow, across Canada and the United States. For each catchment there are multivariate daily time series of streamflow, rainfall, snowfall, and temperature going back nearly four decades, as well as topographical information. The overarching goal is to use the several meteorological time series to understand and quantify the mechanisms generating flow. There are some special attributes of the application. Foremost, the streamflow response of interest today does not depend just on rainfall today: the rainfall time series going back in time over many days will impact the streamflow, i.e., lagged effects. Moreover, this historical rainfall turns into streamflow by several distinct physical mechanisms: overland flow, shallow subsurface flow, and deep groundwater flow. Rainfall that travels overland will increase streamflow relatively quickly, whereas transport via one of the other two mechanisms is more delayed and spread. Thus, we have multiple mechanisms for rainfall turning into flow, each with its own lag profile. Identifying the different lag profiles and their relative contributions from data is important, e.g., a reduction in the delay of fixed rainfall as overland flow to streamflow and/or an increase in its magnitude heightens the risk of flooding and may be attributable to a change of land use. The overarching goal of the research program is to build a flexible framework for multivariate time series that allows modelling of a time series response like streamflow as a function of several lagged effects of one or more time-series covariates such as rainfall. This framework will be based on Gaussian processes (GPs), which can data-adaptively model nonlinear lagged effects. They can also incorporate classical time series concerns such as seasonality and dependence of errors, hence enabling valid uncertainty quantification of the lag profiles. Longer-term trend for patterns of variation spanning say a few years can also be modelled. This set up is quite general and covers other multivariate time series with lagged covariate effects, measurements at a frequency other than days, and seasonality other that annual. There are substantial computational challenges, due to rich daily time series of about 14,000 daily observations for each of 1,000+ catchments. Standard GP computations are prohibitive with that many observations, and efficient algorithms taking advantage of special structure will be important.
Programme
Discovery Grants Program - Individual
Emplacement
British Columbia, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2024-06474
Name nom
Welch, William
Department département
Statistics - Statistics
Organization id
2
Institution établissement
The University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
34000
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
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1203
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Earth sciences
Area of application fr
Sciences de la terre
Research subject code
3000
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Statistics and probability
Research subject fr
Statistique et probabilité
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
Dynamic Gaussian Processes to Identify Environmental Mechanisms
Keywords
gaussian process; time series; multivariate; lag; seasonality; long-term trend; streamflow; meteorology
Application summary
Collaboration with a research lab of hydrologists has identified scientific objectives calling for novel statistical methodology. The lab has rich data sets for over 1,000 watershed catchments, i.e., the area draining into a single streamflow, across Canada and the United States. For each catchment there are multivariate daily time series of streamflow, rainfall, snowfall, and temperature going back nearly four decades, as well as topographical information. The overarching goal is to use the several meteorological time series to understand and quantify the mechanisms generating flow. There are some special attributes of the application. Foremost, the streamflow response of interest today does not depend just on rainfall today: the rainfall time series going back in time over many days will impact the streamflow, i.e., lagged effects. Moreover, this historical rainfall turns into streamflow by several distinct physical mechanisms: overland flow, shallow subsurface flow, and deep groundwater flow. Rainfall that travels overland will increase streamflow relatively quickly, whereas transport via one of the other two mechanisms is more delayed and spread. Thus, we have multiple mechanisms for rainfall turning into flow, each with its own lag profile. Identifying the different lag profiles and their relative contributions from data is important, e.g., a reduction in the delay of fixed rainfall as overland flow to streamflow and/or an increase in its magnitude heightens the risk of flooding and may be attributable to a change of land use. The overarching goal of the research program is to build a flexible framework for multivariate time series that allows modelling of a time series response like streamflow as a function of several lagged effects of one or more time-series covariates such as rainfall. This framework will be based on Gaussian processes (GPs), which can data-adaptively model nonlinear lagged effects. They can also incorporate classical time series concerns such as seasonality and dependence of errors, hence enabling valid uncertainty quantification of the lag profiles. Longer-term trend for patterns of variation spanning say a few years can also be modelled. This set up is quite general and covers other multivariate time series with lagged covariate effects, measurements at a frequency other than days, and seasonality other that annual. There are substantial computational challenges, due to rich daily time series of about 14,000 daily observations for each of 1,000+ catchments. Standard GP computations are prohibitive with that many observations, and efficient algorithms taking advantage of special structure will be important.