Analyzing Debris Flow Risk and Demographics in Post Wildfire Southern California Using ArcGIS Pro

Summary

The primary goal of this project was to determine the debris flow risk for major watersheds in the Eaton and Palisades Burn scars in Southern California. Using ArcGIS pro, a multcriteria raster model was created for the Eaton and Palisades burn scars and surrounding areas. Using watershed data from within the burn scars, the highest risk watersheds could be identified. Then, utilizing demographic census data, statistics were summarized for block groups that intersect streams exiting the burn scars. This project was created in collaboration with partner Serena Morgan, a CU boulder student.

Research Questions

Which watersheds in the burn areas are at highest risk for debris flows?

What are the demographics of the people most likely to be affected by debris flows?

Background

2024 Fires

The palisades and Eaton Fires burned in Los Angeles County in January of 2024. The Eaton fire burned 14,021 Acres and the palisades fire burned 23,448 Acres. Each fire was contained on January 31, 2024. A total of 16,251 structured were destroyed and 30 people were killed. (Link) (Link). In the image above, the Palisades fire is on the left, and Eaton on the right.

Debris Flows

Debris flows are fast-moving landslides that typically occur in streambeds during intense periods of rain. Typically, they begin on the steep sides of mountain valleys, and debris such as boulders and trees becomes funneled into streambeds. If they occur in steep mountain canyons, such as those in Southern California, debris can travel for miles and up to speeds of 35 mph (U.S. Geological Survey). Debris flows are highly related to watersheds, as debris will funnel into tributaries, then into the main riverbeds, exiting at the mouth of the drainage basin (De Haas). Freshly burned areas are particularly susceptible to debris flows due to a lack of vegetation holding soil in place, and ash-covered hillsides create a water repellent effect, leading to little absorption and high amounts of runoff (County of San Mateo).

Wildfires that occur in California in the early winter months are especially problematic due to the debris flow risk. The most precipitation in Southern California occurs in January and February, so areas that are freshly burned in December and January are likely to experience heavy rainfall (NOAA).

Study Area

The study area for this project encompasses the burn scars of the Palisades and Eaton fires as well as their surroundings areas. Debris Flows flow downstream from burn areas, so it was important to include areas outside the burn scars themselves.

Data Sources

Fire Polygons

We downloaded fire polygon data from the California State Geoportal and used “Select by Attribute” to obtain only the Eaton and Palisades fire polygons. These polygons were important to distinguish between burned and unburned areas in our study area.

DEM

We downloaded 1-meter resolution DEM tiles from USGS for each of our two study areas: the general areas of the Eaton and Palisades burn scars. We used the “Mosaic to New Raster” tool to combine the DEM tiles into one DEM for each study area.

Soil Erodability

Soil erodibility data (K-means) polygons were downloaded from the United States of America Soil Survey Geographic Database. This database shows how easily erodable soil is, with higher values indicating more erosion potential (Institute of Water Research). Values ranged from 0 to 0.47, with higher values meaning the soil is more prone to erosion. This data was converted to raster, and a new field was created that took the “erodability” field, which was in “text” format, and made it into a “float” format. This step was essential to use this data in the Reclassify tool in the model, which is described later.

Normalized Burn Ratio

To incorporate burn severity into our model, we downloaded Sentinel-2 bands 8 and 12 from the Copernicus Browser. With the Raster Calculator, we used the equation (B8 - B11)/(B8 + B11) to get the normalized burn ratio raster. This raster was clipped to the burn areas themselves, not the study area rectangles. This was because NBR is only relevant to burned areas. This raster represents how severely burned areas are, with high values indicating high fire severity and low values indicating low severity.

NDVI (Normalized Difference Vegetation Index)

An NDVI raster was created by downloading Bands 4 and 8 from the Copernicus Browser. Using the raster calculator, we used the equation (B8 - B4)/B8 + B4). This raster shows areas of healthy vegetation, with high values near 1 indicating healthy vegetation, and low values near -1 indicating bare land, dead vegetation, or water bodies.

Slope

Using the DEM as input, a slope raster was created using the Slope tool, which is one of the variables used in the model. Values range from 0 to 90 degrees, with 0 degrees indicating a flat surface and 90 degrees indicating a vertical slope.

Tiger/Line Census Blocks Groups

Census block groups were downloaded from the United States Census Bureau. These provided the polygons for the demographic analysis.

Census Block Group Demographic Information

We downloaded the demographic data from the US Census Bureau, which included the most recent available data for median income, median home value, race, and percentage of households with somebody aged 60+. We deleted the margin of error columns, as they were irrelevant for our model. The “Add Join” tool was used to join the Census block group polygons to the demographic information CSV file, using the GeoID as the join field. Then, since the demographic information for each category (eg, white population) was in the number of people per block group, new fields were created to create the percentage per block group. This was the equation used in the calculate field tool: Category population/total population per block group. This was done for these categories: race (White, Black, Asian, Hawaiian/Pacific Islander, Indigenous, and Other), and households with somebody aged 60 or older. The median income and median home value columns had to be copied and changed to “long” format, rather than “text”, so summary statistics could later be run on the block groups.

Some preprocessing steps were taken for all data files. Each data layer was projected to California UTM Zone 11. Each raster variable in the multi-criteria model was resampled to 10-meter resolution. All variables were clipped down to a rectangle surrounding each fire, which was created by making a new polygon feature class in the geodatabase.

Model Variables

The Debris flow risk model utilized 4 different variables. NBR (burn severity) was created by using bands 8 and 11 from sentinel 2 in the the raster calculator: (B8 - B11)/(B8 + B11). NDVI data was create by using bands 8 and 4 in the raster calulator: (B8 - B4)/B8 + B4) Each model variable was resampled to 10m spatial resolution. K-factor, NBR, and NDVI were clipped to a rectangle polygon in the general area of the burn scar. NBR data was clipped to the fire boundaries themselves since burn severity is irrelevant outside of the burn scars. All 4 variables were resampled to 10 meter resolution.

Model

Using modelbuilder in ArcGIS pro, models were created for both the Palisades and Eaton Fires. Above is the model used for the Palisades fire. The model begins with reclassification of each variable into a 1 to 5 scale. 5 being highest likelihood of debris flow and 1 being lowest. After the rasters have been reclassified, the weighted sum tool was used. This tool overlays the relcassified rasters and adds the pixel values. Then, using watersheds in the burn area (described next) zonal statistics was run using the debris flow model.

An identical workflow was used for the Eaton fire.

Variable Reclassification Breakdown

Reclassification Justification

After preprocessing, our debris flow risk map was created in Modelbuilder using a weighted sum multi-criteria model. The variables for this model were normalized burn ratio (NBR), slope, NDVI, and soil erodability (K-factor). These are important factors in determining debris flow risk. More severely burned slopes are at greater risk for mudslides, as vegetation helps prevent soil erosion by keeping soil in place. Steep slopes are at higher risk for debris flows, as water rushing downhill can more easily carry debris with it (Institute of Water Research; Mitsopoulos and D. Mironidis). A high K-factor means the soil is more easily eroded and carried into riverbeds (United States Forest Service). A low NDVI value indicates dead vegetation or bare ground, which decreases soil cohesiveness and shear resistance, increasing debris flow risk (R. Elkadiri et al.)

Watershed Delineation

With our debris flow risk raster, zonal statistics were calculated using watershed areas. Watersheds are an integral part of debris flow risk analysis, as debris flows often flow down creek beds. Using the watershed tool, streams at a high risk for debris flows could be identified. To identify watersheds, several intermediate tools are necessary. Using the DEM, sinks needed to be filled to smooth out pits in the DEM. Next, a flow direction raster was created using the “Flow Direction” tool. A flow accumulation raster was then created from the DEM using the “Flow Accumulation” tool. The output of this tool was a raster that designated the stream beds. Next, a pour point was selected. This was where a potential debris flow could flow out of a canyon into a populated area. Next, the “Snap Pour Point” tool was used on the pour point data, which snapped the pour point to be located on the nearest area of high flow accumulation (streambed). Finally, the “Watershed” tool was used, with inputs of the flow direction raster, flow accumulation raster, and the snapped pour point. The result was a vector polygon of the entire catchment basin where water will flow to the pour point. This process was repeated for a majority of the major watersheds in the Palisades and Eaton fire perimeters.

Watershed Zonal Statistics: Results

Using the watersheds as inputs for zonal statistics, the watersheds with the highest risk were identified. The watersheds were clipped to the fire boundary before running zonal statistics because some of the watersheds extended far outside the burn area, causing many low values to be introduced into the mean calculations for the watersheds. This is because outside the burn area, NBR values were assigned as 0, as the area had not been burned at all. This resulted in some of the watersheds, which extended outside the fire boundary to have low values, even though they included large areas of severely burned, steep terrain. Interestingly, a majority of the watersheds already fit perfectly within the fire boundary, as ridgelines, which are already natural fire breaks, are utilized as fire barriers (Avitt). Clipping the watersheds to the burn area gave a more accurate presentation of the debris flow risk for watersheds that extended outside of the fire boundary. A majority of the watersheds have a high risk of debris flows, as both the Palisades and Eaton fires burned in steep terrain, with low NDVI values.

Pacific Coast Highway Debris Flow

On February 13, 2025, a California fire department firefighter was swept off the road by a debris flow at the location shown while driving down the Pacific Coast highway. The watershed above was determined to have a high debris flow risk from the model.

Eaton Wash Debris Flows

On February 13th, 2025, the same torrential rainstorm triggered a debris flow. Rain rates of over 2 inches per hour were recorded at 5:02pm. At 5:22pm, a 12 ft tall, 100ft wide debris flow cascaded down Eaton Canyon, eventually spreading to 250 ft wide and 5 ft deep in the wash highlighted in the image above.

Part 2: Demographic Analysis

Stream Delineation

From here, the demographic analysis could begin. The first step was to delineate the major streams that exit the watersheds. This involved trial and error using the “Set Null” tool with the input as the flow accumulation raster. The expression used was “Value <= 39,000” for Eaton streams and “Value <= 18,000” for Palisades streams. The output of this was the major streams exiting the watersheds, and it removed the small tributaries. This output was then converted to polygons. Still, some unnecessary streams were removed using the erase tool to further clean the dataset. These major streams had to be clipped to a reasonable distance from the burn area, since the study area was quite large and encompassed areas far downstream that are unlikely to be affected.

Block Group Selection

Using select by location, block groups that intersected major streams were considered to be at risk for debris flows. Large block groups that had their populations upstream far from burn areas were removed, as no people from the block group were at risk.

Results

Block Group Risk Assessment

Once the at-risk block groups were identified, a risk value was calculated based on the watershed values. This was completed manually, by looking at the watersheds that drained into each block group. If only one watershed drained into a block group, that watershed value was assigned to the block group. If multiple watersheds drained into the block group, the values of the watersheds were averaged, and that value was assigned to the block group.

Demographic Statistics: Eaton

The demographics of people living downstream of the Palisades and Eaton burn areas reveal some trends. Demographic statistics were summarized for the block groups that intersect major streams that exit the burn areas. Median income and median home value are high, but lower than in the Palisades areas. 52.6% of the population is White, 8.9% is Black, 17.75% is Asian, and less than 1% is Indigenous. Similar to the Palisades Fire, a little over 50% of the households have somebody over the age of 60.

Demographic Statistics: Palisades

For the Palisades fire, people at risk for debris flows are mostly White, with a very high mean home value of almost $2,000,000. About 1% of the population is Black, 8% is Asian, and less than 1% is Indigenous. A majority of the at-risk block groups had a value of $2,000,000+, so the actual median home value is likely much higher. Over 55% of the households in the block groups have somebody 60 or older. The average median income is almost $200,000.

Uncertainties

It is important to state the limitations of the model. One limitation is that it does not answer how far water and debris from the watersheds will go from the mouth of each watershed. This was outside the scope of this project, which aimed to determine debris flow risk, not exactly where debrius flow is certain.When determining the cutoff point of the rivers in this model, a conservative estimate was used, so the rivers were clipped at a short distance away from the mouth of each canyon. In the event of a debris flow, mud and debris may reach farther than our estimate, or not as far. This uncertainty means that more or fewer block groups may be affected than the ones we selected to run our demographic analysis on.

Another limitation of the model is the determination of risk for each block group. Initially, distance to streams was going to be a variable included in the multi-criteria raster model. This made sense initially because we knew that debris flows are concentrated in creek beds. Zonal statistics would then have been run using the multi-criteria risk raster and the block groups intersecting the creek beds. But this created a problem: from our general knowledge of debris flows, we knew that unburned areas with healthy vegetation on relatively mild slopes can be at very high risk for debris flows if downstream of a steep, severely burned canyon. Our initial model idea would not represent this risk accurately. In addition, the same cutoff decision would have to be made on how far from the burn scar debris flow risk is still present.

We pivoted to using watersheds to determine the risk for each block group. Because our model cannot accurately show where mud and debris may go, we decided to simply select any block group that intersects a main creek. Risk was determined by taking the average values of each basin that drains into a block group. This may not be the best method of determining risk. For example, a long, skinny block group on the Malibu coast in the Palisades burn scar was given a relatively low debris flow risk value. This was because several watersheds drained into it, some with very high risk and others with lower risk. Based on a case study where a car was swept off the road by a debris flow, we now know this block group was at a high risk for debris flows, based on how many watersheds flowed into it (NBC News). Potentially, a better method would be to sum the values of the watersheds that drain into a block group, rather than the mean.

A final uncertainty of the model relates to the resolution of the demographic data. Many of the block groups, especially in the Palisades burn area, were quite large. This introduced uncertainty in the demographic analysis, as maybe only a fraction of the homes in a block group may be affected by debris flows. This likely skewed the results of the demographic analysis, which had the goal of answering: What are the demographics of people most at risk for debris flows?

Conclusion

Using ArcGIS pro, the debris flow risk in the Palisades and Eaton fires were assessed using a multicriteria raster model. 4 variables, soil erodibility, NDVI, burn severity, and slope, were used in the model. Each variable was reclassified to a common 1 to 5 scale, with higher values dignifying higher debris flow risk. The weighted sum tool was used to overlay the rasters and add their values. Watersheds that drain into areas of population were delineated and it was determined which watersheds are at greatest risk.

A demographics analysis was also conducted. Using block group level data, populations living adjacent to streams exiting the burn zone were considered generally at risk for debris flows. Demographic statistics were summarized and was found tha these areas are affluent, with high home values and high average income. Over 50% of people were white with a majority of households having someone over 60 years old.

References

Arcgis.com, 2025, www.arcgis.com/home/item.html?id=f46e037cf3d340379f82c2daff6de405. Accessed 13 Mar. 2025.

Avitt, Andrew. “Clearing the Ridge: Fire for Forest Health and Resilient Communities | US Forest Service.” US Forest Service, 2023, www.fs.usda.gov/about-agency/features/clearing-the-ridge.

Burns, Melinda. “The 1/9 Debris Flow Was Not so Rare.” The Santa Barbara Independent, 19 May 2022, www.independent.com/2022/05/19/the-1-9-debris-flow-was-not-so-rare/.

County of San Mateo. Understanding Debris Flows and Flooding Debris Flows and Flooding. https://www.smcgov.org/sites/default/files/2021-11/Debris%20Flows%20and%20Flooding%20Final.pdf

De Haas, Tjalling, et al. “Debris-Flow Watersheds and Fans: Morphology, Sedimentology and Dynamics.” Springer EBooks, 1 Jan. 2024, pp. 9–73, https://doi.org/10.1007/978-3-031-48691-3_2. Accessed 4 May 2025.

Hayden, Tyler. “County Releases New Debris Flow Map for Montecito.” The Santa Barbara Independent, 5 Dec. 2019, www.independent.com/2019/12/05/county-releases-new-debris-flow-map-for-montecito/. Accessed 14 Apr. 2025.

Institute of Water Research. “RUSLE - an Online Soil Erosion Assessment Tool.” Msu.edu, 2025, iwr.msu.edu/rusle/kfactor.htm.

McGregor, Edgar. “Historic Debris Flow Wipes out Eaton Wash, Altering Creek Route and Uprooting Hundreds of Trees.” MyEatonCanyon.com, 15 Feb. 2025, myeatoncanyon.com/historic-debris-flow-wipes-out-eaton-wash-altering-creek-route-and-uprooting-hundreds-of-trees/. Accessed 14 Apr. 2025.

Mitsopoulos, I.D., and D. Mironidis. “Assessment of Post Fire Debris Flow Potential in a Mediterranean Type Ecosystem.” WIT Transactions on Ecology and the Environment, vol. 1, 16 May 2006, pp. 221–229, https://doi.org/10.2495/deb060211. Accessed 7 Apr. 2025.

NBC News. “Severe Storms Drive Rain, Mudslides, and a Rare Tornado into Southern California.” YouTube, 14 Feb. 2025, www.youtube.com/watch?v=rNO92v9kTvg. Accessed 14 Apr. 2025.

NOAA. “Climate.” Www.weather.gov, www.weather.gov/wrh/climate?wfo=lox.

R. Elkadiri, et al. “A Remote Sensing-Based Approach for Debris-Flow Susceptibility Assessment Using Artificial Neural Networks and Logistic Regression Modeling.” Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 7, no. 12, 11 Aug. 2014, pp. 4818–4835, https://doi.org/10.1109/jstars.2014.2337273. Accessed 29 May 2023.

“TIGER/Line® Shapefiles.” Www.census.gov, www.census.gov/cgi-bin/geo/shapefiles/index.php.

United States Census Bureau. “Explore Census Data.” United States Census Bureau, 2025, data.census.gov/.

United States Forest Service. Clearwater-Nez Perce National Forest Summary of Landslide Prone Delineation for Forest Plan Revision Clearwater National Forest.U.S. Geological Survey. “What Is a Debris Flow?” Usgs.gov, 25 Oct. 2018, www.usgs.gov/faqs/what-a-debris-flow.

U.S. Geological Survey. “TNM Download V2.” Apps.nationalmap.gov, apps.nationalmap.gov/downloader/.

“WFIGS 2025 Wildfire Perimeters.” Ca.gov, 2025, gis.data.ca.gov/datasets/CALFIRE-Forestry::wfigs-2025-wildfire-perimeters/explore?location=34.190495%2C-118.543384%2C9.94. Accessed 16 Feb. 2025.

For more information about this project, view the completed project report here.

Page updated

Google Sites

Report abuse