Date of Award


Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy in Geological Engineering (PhD)

Administrative Home Department

Department of Geological and Mining Engineering and Sciences

Advisor 1

Thomas Oommen

Committee Member 1

Ann Maclean

Committee Member 2

Stanley Vitton

Committee Member 3

Qiuying Sha


Historically, post-fire debris flows (DFs) have been mostly more deadly than the fires that preceded them. Fires can transform a location that had no history of DFs to one that is primed for it. Studies have found that the higher the severity of the fire, the higher the probability of DF occurrence. Due to high fatalities associated with these events, several statistical models have been developed for use as emergency decision support tools. These previous models used linear modeling approaches that produced subpar results. Our study therefore investigated the application of nonlinear machine learning modeling as an alternative. Existing models identified the burn severity of wildfires as an important input in their development. Currently, the most widespread approach to obtaining this input is the use of the differenced normalized burn ratio (dNBR) index, which is determined using data from optical sensors on satellites. However, progress of this existing protocol is mostly hampered by the presence of cloud coverage during data acquisition since optical sensors cannot penetrate clouds. Radar sensors on the other hand can penetrate clouds and smoke. This study therefore developed a radar based algorithm to be used as an alternative to the dNBR metric. The results showed the SAR metric to perform even better than the dNBR, with an overall accuracy (OA) of ~60% and Kappa of 0.35 in comparison to an OA of ~35% and a kappa of 0.1 from the dNBR approach. Next we developed a nonlinear machine learning model to predict the likelihood of post-wildfire debris flow occurrences. This produced improved results over the linear modeling approach with an average sensitivity of 77%, depicting increased ability to predict ~8 out of 10 DF producing basins. Finally, we performed a case study to validate our DF model that showed the model’s robustness in isolating especially high hazard locations. Having these improved models will furnish emergency responders with an increased ability to better assess the associated risks of potential debris flow producing basins and make informed decisions on mitigation and/ or prevention measures.