Degree Name
MS (Master of Science)
Program
Mathematical Sciences
Date of Award
5-2026
Committee Chair or Co-Chairs
Dr. Mostafa Zahed
Committee Members
Dr. Jeff Randall Knisley, Dr. Michele Lynn Joyner
Abstract
Wildfire activity in the western United States has intensified, creating challenges for resource allocation and emergency planners. Accurate long-range forecasting is essential for anticipating multi-state spillover risks, and communicating uncertainty to exposed communities. Yet despite this operational need, current forecasting tools rarely provide rigorous, probabilistic uncertainty quantification. This thesis develops probabilistic forecasts for Colorado, Montana, Utah, and Wyoming using county level fire counts (1992–2020, 172 counties, 10,368 observations). Several statistical frameworks exist for modeling spatially correlated time series data. Space-Time AutoRegressive (STAR) and Space-Time AutoRegressive Moving Average (STARMA) models were evaluated. Stationarity tests revealed mixed results (60% rejection), motivating year fixed effects. STAR(3) achieved optimal performance (RMSE = 0.595, R2 = 66–76%). STARMA models failed due to boundary violations (θ ≈ 1.0, MSPE =339 vs 0.46). Monte Carlo simulations (8,000 datasets) showed mean squared prediction error identified STAR(3) correctly 76% of the time versus 1–3% for AIC/BIC. Seven-year forecasts project declining activity in Colorado, Montana, Utah (59–85%) but increasing in Wyoming (51%), with 90% prediction intervals spanning 300–730%. Findings demonstrate prediction error-based selection outperforms information criteria, with spatial spillover effects (ρ ≈ 0.10–0.14) supporting regional coordination.
Document Type
Thesis - embargo
Recommended Citation
Tagoe, Princess, "Investigating Historical Wildfire Trends and Forecasting Future Events in Colorado, Montana, Utah and Wyoming through Advanced Spatio-Temporal Analysis" (2026). Electronic Theses and Dissertations. Paper 4672. https://dc.etsu.edu/etd/4672
Copyright
Copyright by the authors.
Included in
Longitudinal Data Analysis and Time Series Commons, Natural Resources and Conservation Commons, Statistical Methodology Commons, Statistical Models Commons
