Robert Payne

Topic

Dynamical and Deep Learning-Based Climate Downscaling for Fire Weather Applications in British Columbia

School of Earth and Ocean Sciences

Date & location

• Tuesday, September 15, 2026
• 10:00 A.M.
• Clearihue Building, Room B007

Examining Committee

Supervisory Committee

• Dr. Adam Monahan, School of Earth and Ocean Sciences, University of Victoria (Supervisor)
• Dr. Alex Cannon, School of Earth and Ocean Sciences, UVic (Member)
• Dr. Colin Mahony, Research Climatologist, BC Ministry of Forests (Outside Member)
• Dr. Derek van der Kamp, Pacific Forestry Centre, Canadian Forest Service (Outside Member)

External Examiner

• Dr. Piyush Jain, Northern Forestry Centre, Canadian Forest Service

Chair of Oral Examination

• Dr. Christina Chakanyuka, School of Nursing, UVic

Abstract

Multiple unprecedented weather extremes have affected British Columbia (BC) over the past decade, and the frequency and intensity of extreme events are projected to increase over the remainder of the century. Reliable fine-scale weather information has become critical for informing weather forecasts, environmental and climate policy, climate mitigation, and the development of sustainable infrastructure. However, limited computational power caps the resolution at which climate models can be run, and alternatives are required. One such approach is climate downscaling, wherein high-resolution weather and climate information is inferred from corresponding low-resolution information. Countless methods for achieving this have been proposed, which can broadly be categorized into statistical methods and dynamical methods. This thesis concerns both paradigms. First, we propose the use of a novel deep learning architecture known as a stochastic Generative Adversarial Network (GAN) for statistically downscaling fire weather indices from the Canadian Forest Fire Weather Index System over BC. We show that our models trained in a perfect prognosis framework are capable of generating realistic-looking fields that reproduce the correct marginal distribution of values and amount of spatial variability with a reasonable degree of accuracy. Crucially, model predictions were shown to be more accurate on average than simply using the low-resolution information on which the GAN was conditioned, demonstrating the added value of downscaled. Training models within an imperfect prognosis framework was shown to drastically reduce the quality of the downscaled indices, on account of the model having to compensate for intermodel variability and biases in addition to accounting for scale differences. Second, we assess surface wind speed and specific humidity outputs from a new dynamical regional climate model produced at the University of British Columbia dubbed ClimatEx-WRF. We show that ClimatEx-WRF produces realistic seasonal cycles in climatological wind speeds and specific humidity, and produces reasonably accurate temporal and spatial variability relative to station observations from Environment Canada. Overall, we demonstrate the value of climate downscaling as a tool to derive high-resolution weather information through both statistical and dynamical methods. We conclude by proposing several avenues through which our methods can be expanded upon, ultimately to provide valuable insights into the potential of climate downscaling to inform climate policy and decision-making.