Post-fire debris flow modeling analyses: case study of the post-Thomas Fire event in California

Abstract

There is an increased risk in post-fire debris flow (DF) occurrences in the western USA with recent increase in wildfire frequencies. DFs are destructive, causing high loss to lives and infrastructure. A lot of effort is going into possible preventive and/or mitigation measures. Recent research efforts in this niche focus on developing statistical models that assist emergency responders in isolating high-hazard locations after fires. There are two general approaches to this statistical modeling: linear and nonlinear. This study has looked into applying a linear-based logistic regression model and a nonlinear-based C5.0 decision tree model to assess the strength of each in predicting the locations within the Thomas Fire boundary that produced DFs. To do this, DF scars were delineated by running a change detection protocol known as delta normalized difference vegetation index (dNDVI), using high spatial resolution data from Planet Labs. These scars were further validated with data gathered by Santa Barbara County officials on affected areas. Results from the two statistical models were then overlain on the delineated DF scars and compared against each other to determine predictive strengths. The results revealed both models to perform well in predicting high probabilities for locations that were shown to produce DFs. The logistic regression model predicted an overall ~ 44,800 ha (49%) more high-hazard coverage compared to the C5.0 tree and therefore showed greater urgency. However, a closer look at the basin predictions with the delineated DF scars showed that most of these high-hazard basins identified by the logistic regression did not to have any discernible scars. It was projected that most of these locations were likely false positives and further fine-tuning of the model was recommended. Further recommendation was made concerning the development of additional models to predict potential DF inundation paths. Combining origination and inundation models will be most beneficial since the greatest associated danger with DFs is not necessarily where they start, but rather further downstream where most communities and infrastructure are situated.

Incorporating Shear Stiffness into Post-Fire Debris Flow Triggering Models

“…Statistical models also have a plethora of metrics to assess performance to observed outcomes and between models directly. While the metrics used in this study are by no means exhaustive, they are commonly accepted in statistical modeling and in previous work in debris flow [Rupert et al, 2008;Negri, 2016;Staley et al, 2016Staley et al, , 2017Addison and Oommen, 2020]. Works by [Rengers et al, 2020] suggest additional metrics could be of use in certain areas but produce values with comparable utility as the ones provided in this study.…”

“…Such speed could offer emergency modeling for recently burned areas a distinct advantage with potentially little to no loss in model performance depending on acceptable risk acceptance. Such removal of the F factor in the two term model mentioned prior measured by the normalized burn ratio may be moot when considering its measurement is used in identifying burned watersheds prior to a debris flow hazard assessment [Key and Benson, 2005;Addison and Oommen, 2020] suggesting context should be carefully evaluated in such cases.…”

“…Though these categories could be considered relatively broad, the combination of these factors statistically captures behavior of debris flow triggering across watersheds with heterogenous scale and characteristics [Staley et al, 2017]. The goal of the input variables is to quantify the interaction between hydrologic and physical properties during debris flow initiation [Addison and Oommen, 2020;Rengers et al, 2020].…”

“…Utilized by the USGS/NOAA Debris flow hazard program, models from Staley et al, (2017) for the Southern California Region include; slopes ≥ 23° with a high or moderate burn severity, burn severity measured as the proportion of the reflectance of certain wavelengths associated with vegetation before and after a fire per remote sensing [Addison and Oommen, 2020], and surface qualities measuring properties of the sediment such as clay content, grain size, and/or erodibility. Also included are rain measurements in amount and intensity of rainfall for differing storm intervals.…”

Susceptibility Prediction of Post-Fire Debris Flows in Xichang, China, Using a Logistic Regression Model from a Spatiotemporal Perspective

“…Wildfire is defined as a natural disaster accompanied by devastating consequences to the infrastructures, ecosystems, and human lives [1][2][3]. The wildfires considerably alter the landscape characteristics while generating a cascade of hydrogeomorphic hazards [4][5][6]. Post-fire debris flow (hereinafter PFDF) is one of the most frequent post-fire hazards [7][8][9].…”