Date of Award

5-28-2025

Publication Type

Thesis

Degree Name

M.Sc.

Department

Mathematics and Statistics

Keywords

Forecasting; Teleconnections; Trend analysis

Supervisor

Kevin Granville

Rights

info:eu-repo/semantics/openAccess

Abstract

Climate change can impact various facets of a region’s fire regime, such as the frequency and timing of fire ignitions. This study investigates the temporal trends of monthly fire counts in the northwest region of Ontario, Canada, between 1960 and 2023. Fires ignited by human activities or lightning are analysed separately. The significance of trends are determined using the trend-free pre-whitened Mann-Kendall test with a Thiel-Sen slope estimate, and are contrasted with those obtained using the Cochrane-Orcutt method. Both of these approaches consider and adjust for autocorrelations in the time series data. We also consider the forecasting of future monthly fire counts using a Negative Binomial Auto-Regressive (NB-AR) model suitable for count time series data with the presence of overdispersion while investigating the use of climate teleconnections such as ENSO, NAO, AO, PDO, and AMO as predictors at differing temporal lags. Several candidate models having varying time lags for each predictor are identified and their predictive skills are quantified through cross-validation estimates of Mean Absolute Error. These models omit months when there are historically no or few ignitions and use monthly indicator functions to capture seasonal trends alongside an AR(1) term of the previous month's count. We find that a model with 6-month-lagged AMO is the most suitable for forecasting counts of human-caused fires, while no teleconnection predictor was found to be significant when forecasting lightning counts.

Download

Share

COinS