The influence of wildfire on debris flows in a landscape of persistent disequilibrium: Columbia River Gorge, OR, USA

Abstract

Extensive debris flow fans emanate from steep catchments in the Columbia River Gorge (CRG), Oregon, a landscape dramatically reshaped by Late Pleistocene megafloods. In 2017, the Eagle Creek Fire burned 200 km² of the CRG, prompting concerns of heightened debris flow activity, yet its impact on hazard potential remains unclear. To assess the fire's effect on 10 CRG catchments, we quantify pre- and postfire debris flow erosion using airborne lidar, imagery, and field observations, as well as long-term (10⁴- to 10⁶-year) erosion from debris fans and volcanic edifice reconstruction. Fan-derived 10⁴-year erosion rates (1 to 9 mm year⁻¹) exceed 10⁶-year rates by 10 to 50 times, suggesting sustained, rapid unraveling of these catchments following the megafloods. Pre- and postfire debris flow erosion rates are consistent with 10⁴-year rates, such that fire-derived erosion accounts for a relatively small fraction of post-megaflood denudation (<10%), implying persistent landscape disequilibrium that manifests as ongoing high hazard potential in the CRG, regardless of wildfire conditions.

Debris flow erosion in rocky catchments of northern Oregon (USA) is found to be relatively insensitive to fire conditions.

INTRODUCTION

In mountainous landscapes, hillslope and channel morphology are constantly adjusted through process-landform feedbacks, striving to balance long-term denudation rates and regional rock uplift (1). However, the timescale of these adjustments can be on the order of tens of thousands to millions of years, resulting in many landscapes being in a state of transient disequilibrium (2–4). Erosion rate compilations have shown intermittent processes, such as debris flows or landsliding—rather than gradual processes, such as soil creep, runoff erosion, or chemical weathering—typically govern long-term rates (5). Anderson et al. (6) reported that the equivalent of 300 years’ worth of the regional long-term erosion rate was produced by debris flows initiated in just one storm event in 2013. Furthermore, in these steepland landscapes, debris flows are typically the most efficient sediment transport mechanism, conveying accumulated rockfall and landslide material through kilometers of channel length and down hundreds of meters of relief to low-lying, fluvial valleys (7). This also highlights their capacity for hazard, as humans often inhabit debris flow fans, which lie along the distal reaches of their sediment conveyance pathways (8). The frequency and magnitude of debris flows are intermittent and varied depending on the rate of sediment recharge and the frequency of triggering events (9); thus, quantifying the relationship between recharge and triggering is at the core of predicting debris flow hazards (10, 11). Understanding the drivers of debris flow frequency and magnitude in disequilibrium landscapes, which constitute a major portion of global landscapes, remains to be an outstanding task.

Short-term perturbations (e.g., timber harvest and wildfires) have been shown to exacerbate debris flow magnitude through increasing both sediment recharge rates and triggering frequency. Wildfire can reduce surface cover by incinerating vegetation, thus releasing sediment as dry ravel over sufficiently steep slopes, as well as reducing surface soil cohesion and increasing the availability of unconsolidated material for debris flows to entrain along hillslopes and within channels (12–16). Furthermore, fire may increase the potential for landslide initiation during less intense storms due to loss of strength within burned root systems (12) or the formation of a hydrophobic layer at the surface (17). Orem and Pelletier (18) quantified the contribution of postfire debris flow erosion to long-term erosion rates for two catchments in New Mexico, finding that postfire erosion contributes nearly 90% of the long-term erosion. Roering and Gerber (19) performed a similar calculation for postfire erosion rates in the Pacific Northwest and estimated that postfire conditions constitute 40 to 80% of the long-term erosion for steep soil-mantled catchments. Although these analyses are sensitive to estimates of fire recurrence interval and thus ecological setting, evidence suggests that wildfire-related conditions are likely the primary driver of debris flow cycles in their respective regions (fig. S1). However, the landscapes in these studies were not actively responding to an external perturbation. In disequilibrium landscapes, we hypothesize that the imbalance between regional uplift and long-term erosion rates has an outsized and more sustained control on the rate of sediment recharge compared to fire and is thus a rate-limiting factor for debris flow frequency and magnitude.

The Columbia River Gorge (CRG) runs along the Oregon–Washington state border and is infamous for frequent debris flows and rockfall events causing fatalities and affecting structures and transportation corridors (20–26). The region is uplifting at ~0.2 mm year⁻¹ due to a combination of intrusive magmatic bulging and extrusive volcanic eruption (27) from volcanic edifices lining the CRG (28). Mean annual precipitation in the CRG exceeds 2500 mm year⁻¹, falling primarily in the form of rain (23) with intermittent snow (29). Because of regional bedrock structure, the CRG has distinct morphology on the northern Washington state and southern Oregon side of the Columbia River that generate vastly different hazards. The Oregon side is generally characterized by antidip basalt flows, where steep, rocky slopes and iconic waterfalls are abundant. On the Washington side, intermittent dip-sloping basalt flows lead to extensive and prominent deep-seated landslide complexes (30, 31), which define a particularly narrow corridor of the Columbia River (Fig. 1A). Imprinted along both sides of the CRG are the effects of the Missoula Floods—dozens of cataclysmic glacial lake outburst floods that flowed along the Columbia River at the end of the Last Glacial Maximum [20 to 15 thousand years ago (ka)]—causing erosion along this relatively narrow western section of the CRG (32). The Missoula Floods and ensuing erosion has resulted in widespread disequilibrium across the CRG, characterized by low relief landscapes giving way to steep, high relief catchments or retreating cliffs along the southern edge of the gorge (Fig. 1A).

Fig. 1. CRG overview.

(A) Eastward perspective down the Columbia River, highlighting the river-adjacent Washington state (31) and Oregon (63, 64) deep-seated landslides and the steep escarpment face on the Oregon side, where the ECF burned in September 2017. Hundreds of mass wasting events were mapped across the southern escarpment using differential lidar—50 events were mapped from 2009 to 2018, and 266 events were mapped from 2018 to 2022 (22). The dashed yellow bounding box highlights the Dodson catchments and outlines the extent of Figs. 2 and 3. (B) Timeline of recent geologic, historic, and modern events near the Dodson catchments, highlighting volcanism, megaflooding, fires, landslides, and lidar collections.

Historically, many of the most destructive debris flows in the CRG have initiated from 10 catchments near Dodson, Oregon. Notable historic events include debris flows in February 1996 that amassed to more than 1 × 10⁶ m³ of material due to an exceptional atmospheric river (26), multiple events depositing 1.5 × 10⁵ m³ of material in November 2001 (33), and numerous smaller events reported from 1964 to 1987 [Fig. 2 (20, 26)]. In September 2017, the Eagle Creek Fire (ECF) burned nearly 200 km² of the Oregon side of the CRG, prompting public safety concerns and a multi-institution response, which included government and academic entities supporting a 2018 aerial lidar acquisition, to characterize the risk of debris flows due to postfire conditions. Following the ECF, widespread debris flow events were triggered during January 2021 and 2022 storms across the CRG. Additional lidar collections in 2021 and 2022, along with a 2009 survey, enabled the creation of a high-resolution, prefire and postfire mass wasting event inventory (Fig. 1B). The “prefire” epoch is based on observations that minimal sediment was transported in the form of landsliding between the fire and the 2018 lidar acquisition (22).

Fig. 2. Debris flows occurrence in the Dodson catchments from 1976 to 2022, mapped over soil burn severity.

Soil burn severity data collected and developed by the USFS Burn Area Emergency Response team (39). Note the recurrence of historical and modern postfire debris flow events traveling down the same channels (21, 22).

In the postfire landslide inventory, nearly 30% of events were mapped within the infamous Dodson catchments, which constitute just 10% of the total mapping area. The number of postfire events in the Dodson catchments was 10 times that of the prefire period, suggesting that fire extrinsically affected the debris flow cycle. However, numerous debris flows that occurred during the postfire mapping epoch were initiated on unburned slopes and traveled down reaches with low burn severity during the same initiating storms, alluding to additional controlling factors, such as sediment recharge (34).

In this study, we investigate the influence of wildfire on debris flow erosion in disequilibrium landscapes by quantifying the contribution of fire to sediment transport erosion in the 10 Dodson catchments. Our investigation reveals the steep, high-relief morphology and abundant rock outcroppings, characteristic of disequilibrium landscapes, in the debris flow-prone Dodson catchments. In addition, we show observations of heterogeneous lithology, which promotes the persistence of steep escarpment retreat. We compare these source catchment characteristics an erosion rate compilation that spans a wide range of timescales, including (i) time since the last local volcanic edifice eruption (~10⁶ years), (ii) time since the Missoula Floods (~10⁴ years), and (iii) recent field-, historical imagery–, and differential lidar–based volume estimates of pre- and postfire debris flow events (10¹ to 10⁰ years).

RESULTS

Geomorphic analysis

The Dodson catchments are <2 km² in area, with catchment relief (defined here as elevation spanned by the catchment) ranging from ~420 to 910 m and catchment-averaged slope values ranging from 42° to 50° (Table 1). Notably, 20 to 50% of the catchment relief is young High Cascades basalt [~0.928 to 3.5 million years ago (Ma) (35, 36)] erupted from local volcanic edifices, such as Nesmith Point (Fig. 3). This additional basalt has made the Dodson catchments prominent high-relief features along the CRG escarpment (Figs. 1 and 3), with exposed basalt cliffs making up to 14 to 27% of the respective catchment area [Materials and Methods; (37); Table 1].

Table 1. Dodson catchment characteristics summary.

Percent burned includes only zones that were deemed moderate to high burn severity following the ECF in 2017 within each catchment [labeled in Fig. 3; (39)].

Basin ID	Mean slope (°)	Area (km2)	Relief (m)	Rock exposure index (−)	Percent burned
MP 34.6	50.1	0.13	420	26	72
Ainsworth	43.3	0.21	610	17	64
Rock of Ages	47.4	0.33	620	22	64
Leavens	43.0	1.1	830	14	94
MP 35.5	45.4	0.17	740	20	62
Tumalt	45.5	1.3	910	23	34
Nesmith Point	45.6	0.69	870	22	65
Warrendale West	45.7	0.30	770	23	67
Warrendale East	47.8	0.36	860	27	42
MP 37.5	42.4	0.63	900	14	89

Fig. 3. Geologic overview of the Dodson Catchments.

Fans, geologic units of interest (36), and Nesmith Point highlighted (28). The fans extend into the Columbia River, with bounds mapped using high- resolution 1-m bathymetry (54).

Debris flows and shallow landslides are the primary mass wasting process observed in the CRG, along with intermittent rockfall (22, 26), and each of the 10 Dodson catchments has a large debris flow fan at its outlet, which we hypothesize as having been accumulating since the last of the Missoula Floods [~15 ka; (32)]. The fans are steep (≥10°), mantled by prevalent debris flow levees, and exposures exhibit poorly sorted, matrix-supported deposits lacking stratification or bedding. We propose that the fans must postdate the Missoula Floods because of the high velocities and maximum depths of 200 m achieved by the megafloods, which mobilized sediment orders of magnitude larger than the fan material (38). Furthermore, during these floods, the apex of the largest fan would be inundated and susceptible to erosion (Fig. 3).

We characterize the impact of the ECF on the Dodson catchments using remotely sensed and field-verified soil burn severity data (39) and find that the proportion of each catchment exhibiting moderate to high burn severity widely ranges from 34 to 94% (Table 1).

Long-term (million- and millennial-year) erosion rates

Over 10⁶- and 10⁴-year timescales, the Dodson catchments exhibit similar order-of-magnitude erosion rates to one another within the respective timescale (Fig. 4 and Table 2). Erosion rates are calculated over 10⁶ years by measuring the volume of material eroded using the difference between modern topography and a paleotopography reconstruction of the landscape ~0.928 Ma, the time of the last lava emplacement in the region, and then dividing the volume by the time since eruption and each respective catchment area (Materials and Methods and fig. S2). The 10⁴-year erosion rates are calculated from fan volume estimates, which are constrained using topographic and well log data (Materials and Methods and fig. S6), and we then divide by time since the Missoula Floods (~15 ka) and respective catchment areas. The median 10⁶-year rate is 0.23 mm year⁻¹ [interquartile range (IQR) = 0.15 to 0.28], which is consistent with long-term regional uplift rates (27). In contrast, our fan volume–derived 10⁴-year erosion rates range from 1 to 9 mm year⁻¹ (median = 5 mm year⁻¹), nearly 5 to 50 times faster than the 10⁶-year rate, implying rapid Quaternary landscape adjustment and frequent fan building debris flow events (Table 2). The 10⁴-year erosion rates increase nonlinearly with their respective catchment-averaged slope (40), extending a proposed erosion-slope relationship generated in the San Gabriel Mountains (SGM), a well-studied steep, rocky landscape (41).

Fig. 4. Dodson catchment fan-derived 10⁴-year erosion rates plotted against slope and compared to 10⁶-year erosion rates derived from a paleotopography reconstruction.

The nonlinear slope–erosion rate relationship measured in the SGM (Southern California, USA) via basin-averaged cosmogenic ¹⁰Be erosion rates (41) holds in our study area, where the relatively steep topography reflects the pace of rapid incision. Nonlinear soil transport model fits for soil diffusivity (K) and critical slope (S_c) values were derived visually for this study, and previously published values were used for the SGM.

Table 2. Summary of erosion rates, median P_F values, and volumetric proportion of catchments filled by its respective fan.

Catchments are labeled in Fig. 3.

	Erosion rates (mm year−1)
Catchment ID	106-year edifice reconstruction	104-year fan volume	101 prefire obs. (1996–2018)	100 postfire lidar (2021–2022)	100 Postfire lidar (2018–2022)	Median PF (−)	Percent of catchment filled by fan (%)
MP 34.6	0.11	4.7	-	20	5.2	0.04	70
Ainsworth	0.10	4.5	-	2.2	5.5	0.05	70
Rock of Ages	0.13	4.8	5.6	0.39	0.1	0.0006	60
Leavens	0.19	1.2	2.8	3.0	9.8	0.4	10
MP 35.5	0.22	2.6	-	1.9	1.6	0.03	20
Tumalt	0.29	8.8	5.5	0.12	3.1	0.02	50
Nesmith Point	0.30	1.9	-	0.97	1.7	0.04	10
Warrendale West	0.28	5.0	-	0.84	4.8	0.04	30
Warrendale East	0.27	7.3	3.6	7.9	3.4	0.02	40
MP 37.5	0.24	2.7	-	-	-	0	20

Short-term debris flow erosion rates

We derive modern erosion rates from differential lidar estimates of net-erosional debris flow volumes that occurred from 2009 to 2022. We calculated postfire erosion rates over two epochs, 2021–2022 and 2018–2022, and prefire erosion rates were calculated over 2009–2018, correlating to the bounding lidar survey dates. In addition, we use debris flow deposits mapped via field- and imagery-based methods from 1996 to 2001 to incorporate the last widespread debris flow event. The imagery analysis extends our prefire period from 1996 to 2018. Four of the 10 catchments had large, mappable debris flows that, when averaged over February 1996 to May 2018 (~23 years), resulted in prefire erosion, ranging from 2.7 to 5.6 mm year⁻¹ (Fig. 5). The prefire lidar epoch that spans from 2009 to 2018 revealed negligible net erosion in all catchments (Table 2 and Materials and Methods) yet prevalent sediment loading in burned and unburned catchments (fig. S5). The 2021–2022 postfire erosion rates range from 0 to 20 mm year⁻¹ (median = 1.4 mm year⁻¹) from 16 shallow landslides and 22 debris flows, and the total 2018–2022 postfire epoch rates ranged from 0 to 10 mm year⁻¹ (median = 3.3 mm year⁻¹) based on an additional 15 shallow landslides and 19 debris flows (Figs. 2 and 5 and Table 2). During the postfire epochs, the total eroded volume was ~8.4 × 10⁴ m³, an order of magnitude lower than the total debris flow event erosion in 1996, which was estimated to be nearly 1 × 10⁶ m³ (24, 26).

Fig. 5. Erosion rate results for the Dodson catchments plotted over the respective measurement interval.

We compare our rates to previous studies compiling erosion rates over similar timescales (2, 13) and observe substantial deviations at the 10¹- and 10⁴-year timescales. Notably, we find sustained rapid erosion signal over ~10,000 years, suggesting that the signal of a transient landscape may overshadow the erosional signature of fire.

DISCUSSION

In the Dodson catchments, post–Missoula Flood erosion rates, characterized over time intervals spanning five orders of magnitude, vastly exceed the million-year timescale rates, implying the onset and persistence of transient conditions that led to rapid catchment formation and sediment flux (Fig. 5). Large steep, colluvial fans emanate from the Dodson catchments, and debris flow deposits and the prevalence of levees along the surface (Figs. 2 and 3) imply that debris flows are the primary fan-building mechanism over the past ~15 ka. Subsequently, it stands to reason that debris flow frequency and magnitude appear to set the rapid (>1 mm year⁻¹) 10⁴-year erosion rates we calculated. Using observed debris flow volumes, we estimate that the fans would require debris flow recurrence intervals ranging from 1 to 33 years⁻¹ to account for the observed fan volumes (Materials and Methods). These recurrence interval estimates are consistent with annual observations of small debris flows from local landowners, the Oregon Department of Transportation (ODOT), and the 25-year period between the widespread 1996 and 2021 events.

Classically, fire is implicated in accelerating sediment recharge rates (42) and increasing debris flow susceptibility (14), and previous work has shown that fire-related debris flow events are the primary contributor to long-term erosion rates (18, 43) and thus induce hazardous conditions. To assess whether the theories are valid within the CRG, we quantify the fractional contribution of the 2017 ECF to erosion over millennial (10⁴-year) timescales, P_F, using the following equation (18)

$$P_F=\frac{t_{\mathrm{R}}{E}_{\mathrm{F}}}{{\mathit{RI}}_{F}D}$$ (1)

where RI_F is the fire recurrence interval, t_R is the duration of time with fire-driven erosion, E_F is the postfire erosion rate (2018–2022), and D is the long-term, in this case, 10⁴-year, erosion rate. We calculate P_F values using 10,000 randomly sampled RI_F values from 100 to 300 years (44) and t_R ranging from 3 to 15 years for each catchment, and we find a median P_F of 0.03 across all catchments (IQR = 0.02 to 0.05) (Table 2 and figs. S1 and S3). P_F values in the CRG exhibit an order of magnitude decrease in the contribution of fire to long-term denudation compared to other postfire studies (12, 18, 19, 43), owing to high 10⁴-year erosion rates. The exception is Leavens Creek, where we estimate P_F to be 0.4, which may reflect localized high-severity burning and failure of heavily forested and soil-mantled slopes (45), which is less characteristic of the other steep and rocky catchments in our study area (Table 1 and fig. S1). We observe a positive correlation between increasing 10⁴-year fan-derived erosion rate and rock exposure index (REI) in our catchments, which demonstrates the signature of disequilibrium over long timescales depicted by rock exposure (Fig. 6A). As such, catchments with a low REI, or higher proportion of soil- or colluvium-mantled area, such as Leavens (Table 1), are more responsive to fire, and variability in rockier catchments may be indicative of sediment availability conditions at the time of the ECF (Fig. 6B).

Fig. 6. Erosion rates as a function of rock exposure index (REI).

(A) 10⁴-year fan-derived erosion rates increase with rock exposure in catchments and (B) postfire erosion rates (2018–2022) show a decrease in postfire erosion with increasing REI.

During the prefire differential lidar epoch (2009–2018), few landslides were triggered, and none with sufficient momentum to exit the catchments, traverse the fans, and generate net catchment erosion. Postfire epoch debris flow events were triggered by rainfall events with total combined rainfall and snowmelt amounts nearing the upper bounds of all storm events observed over a 30-year period similar to those measured during the prefire lidar epoch, which did not trigger widespread debris flows (fig. S4). Although the postfire period shows a 10-fold increase in landsliding, nearly half of the landslides (44%) within the Dodson catchments were initiated from unburned or low burn severity zones (Fig. 4), and channels with unburned contributing areas experienced meters of sediment loading in the prefire (2009–2018) interval (fig. S5).

Through the analysis of erosion rates, both pre- and postfire, we find the erosion rates in the Dodson catchments to be insensitive to fire. In the prefire observation window from 1996 to 2018 that incorporates widespread debris flow events in 1996 and 2001 (Fig. 2), we observe erosion rates comparable to the 10⁴-year erosion rates and the postfire epoch that is averaged over 23 years (Table 1 and Fig. 5). Negligible landslide activity from 2009 to 2018, despite the presence of storms with similar magnitudes as 1996 and postfire storms (fig. S3)—and a dampened postfire response—may be the product of extensive sediment excavation by debris flow events in 1996 and 2001. Together, our observations suggest a limited erosional contribution from fire and support our hypothesis of an intrinsic sediment recharge–dominated period that fueled debris flow events with sufficient volume to affect infrastructure at the distal fan extent, rather than an exceptional event, similar to the 1996 event.

The Dodson catchments are eroding at 1 to 9 mm year⁻¹ over ~10⁴ years, similar in magnitude to one of the world’s most hazardous catchments, the Illgraben in Switzerland [~7 mm year⁻¹ (46)]. Basin-averaged hillslope angles exceed 40° because of outcropping bedrock cliffs, and a substantial fraction of the erosion likely occurred post–15 ka, as the fans make up 10 to 70% of the residual (i.e., eroded) catchment volume (Table 1). These observations suggest that a recent transition to rapid erosion has occurred, potentially coincident with or exacerbated by the Missoula Floods. Outburst flooding substantially increased river channel flow velocities and depths, leading to undercutting and incision along the Columbia River (47), potentially affecting base level and erosion on surrounding slopes. It is possible that the megafloods reactivated large landslide complexes along the northern banks, which are persistently active over the past 15 ka (31), and promoted a prolonged southward constriction of the Columbia River. A recent example of this behavior is the Bonneville landslide (Fig. 1A), which dammed the Columbia River ~600 years ago (48). Within decades after that event, the Columbia River overtopped the landslide dam and, for centuries, has sustained a narrower channel, incising and undercutting the southern edge of the CRG, which led to the initiation of Ruckel landslide on the southern banks of the river (31). Similarly, the Skamania landslide complex was active between 1 and 15 ka across from the Dodson catchments and may have promoted the current rapid escarpment retreat by imparting a lateral base-level control.

An additional factor contributing to rapid incision in the Dodson catchments is the young age of the local volcanic units that span >100 m of relief. The High Cascades basalt flows that line the headwaters of the Dodson catchments (Fig. 3) erupted 0.928 Ma (35), which coincides with the hypothesized timescale for basalt’s vertical permeability to break down via the filling in of pore spaces. This hydrologic shift allows for surficial processes, such as debris flows, to begin to rapidly erode the surface (49). Furthermore, the Nesmith Point edifice is one of the closest lava sources to the Columbia River on the southern edge of the gorge, which sets up these extrusive eruptions to have an outsized influence on setting the relief structure along the gorge and generating high-elevation, undissected surfaces.

Rapid debris flow fan construction persists with modern implications on debris flow hazard. The low-gradient (<20°), high-relief plateau that abuts the southern margin (or headwaters) of the Dodson catchments (Fig. 1A) will likely continue to accommodate rapid lateral retreat for millennia due to high relief built by volcanoes within the past million years and sustained undercutting of weak beds within the basalt flows, particularly a hyaloclastite layer, allowing the escarpment to sustain relief (50).

Here, we have shown that, in the CRG, long-term topographic adjustment to a disequilibrium state, not fire, controls the cycle of debris flows, but this may not be exclusive to the CRG. Complex base-level changes may induce disequilibrium and landscape response through rapid erosion and localize debris flow hazard in a range of geologic settings (51, 52). Identifying these regions will provide the opportunity to investigate the relationship between sediment recharge cycles, debris flow frequency and magnitude, and the role of fire, contributing critical knowledge to forecast the evolving hazard in debris flow–dominated steeplands. Steep, rocky, postglacial landscapes are scattered throughout the Pacific Northwest region on rapidly eroding volcanoes along the Cascade Range (53) that are projected to experience more frequent wildfires due to anthropogenic climate change (44). Understanding triggering mechanisms and recharge volumes required for debris flows is vital for hazard mitigation efforts.

MATERIALS AND METHODS

Experimental design

An array of geologic, geomorphic, historic, and remote sensing maps and datasets for 10 debris flow–dominated catchments near Dodson, Oregon, provided us the opportunity to analyze landscape-scale perturbations across multiple timescales to better understand the influence of fire in steep, rocky landscapes (events summarized in Fig. 1B).

Catchment morphometrics

Catchments were mapped by hand along escarpment margins, drainage divides, and apices of fans. Relief was calculated by the difference between the highest and lowest elevation within the catchment. Basin-averaged slope was determined using the Geospatial Data Abstraction Library (GDAL) slope tool on the 1-m digital elevation model (DEM), which implements Horn’s method to calculate slope. We average over a 5 m by 5 m–moving window using bilinear sampling techniques and then average across the entire catchment, following DiBiase et al. (37). There are no alluvial fills that skew the slope values. We calculate the REI value, which is the proportion of catchment that exceeds the threshold slope, S*, which defines the onset of rock outcrops (37). We choose S* to be 55°, which is visually constrained in the 1-m DEM by increasing the coverage of vertical cliffs and reducing the incorporation of colluvium.

Paleotopography reconstruction

To calculate the long-term 10⁶-year erosion rate, we measure the total volume of eroded material by reconstructing the paleotopography at 0.928 Ma, when the most recent High Cascades lava flows from Nesmith Point were deposited [Fig. 3; (35)]. We calculate the volume of material eroded for each catchment and divide this volume by the catchment area and time since the 0.928 Ma eruption. We use a B-spline interpolation between the mapped Nesmith Point location (28) and relict high-elevation ridges determined by high-elevation topography with relatively minimal incision (fig. S2). We implemented the Geographic Resources Analysis Support System (GRASS) r.resamp.bspline tool in QGIS to fit a surface between elevation values sampled at the points mapped in fig. S2. We used the bicubic interpolation method, with 10-m steps in the north-south and east-west direction and a Tikhonov regularization parameter of 0.01. We calculate the volume of material eroded since the last eruption for each catchment by subtracting the 2018 lidar data from the reconstructed paleotopography and then applying the QGIS Zonal Statistics tool to sum the differenced volume per 100-m² pixel.

Debris flow fan volume reconstruction

Fans were mapped from a 1-m DEM that included both bathymetric data of the Columbia River and topographic data of the fans (54). Fan boundaries were determined primarily by surface elevation profiles, as well as fan surface channels and recent debris flow paths. The Dodson debris flow fan material extends into the Columbia River, which is evident in the bathymetry and previous 1996 debris flow events that have been observed to deposit into the Columbia River (Fig. 2).

At the time of the Missoula flooding, the Columbia River surface was >50 m deeper than its current level (30). Using the point measurements at local well logs, we subsample the deepest wells mapped within the Dodson fan areas [fig. S6 (55)] to reconstruct the fan basal geometry along material boundaries reported in the well data (see Supplementary Text). We subtract this lower bound from the 2018 lidar DEM elevation values and then calculate fan volumes using the Zonal Statistics tool in QGIS over each fan area to sum the difference in elevation between the surface and the interpolated planar basal surface per square meter pixel.

To calculate erosion rates, we average the debris flow fan volume estimates over each respective catchment area to obtain an average lowering thickness over 15,000 years, the time since the last recorded Missoula Flood (32). A simple bulk density correction was applied to account for the bulk density of fan material

$$V_{\mathit{\text{Fan}},\mathit{\text{corr}}}=\frac{{\mathrm{\rho }}_{\mathit{\text{Fan}}}}{{\mathrm{\rho }}_B}·V_{\mathit{\text{Fan}}}$$ (2)

where ρ_B is the average bedrock bulk density and ρ_Fan is the bulk density of fan material, following Palucis et al. (56). The lithology of the Dodson catchments is primarily basalt, whose fracture density is highly variable depending on the texture and age of the flows. We use bulk density data reported by Zakharova et al. (57) along a core collected from Columbia River basalts (similar to flows found at the base of the Dodson catchments) where ρ_B ranges from 1.8 to 2.9 kg m⁻³, which we average to 2.4 kg m⁻³. Their core crosses variably aged and fractured basalt flows, similar to the Dodson catchment lithologies. ρ_Fan is estimated from an average of reported dry bulk densities of natural and experimental debris flows, 2.1 kg m⁻³ (58). The bulk density ratio used for the correction is 0.87. Overestimates in volume may arise from the inclusion of deep-seated landslide deposits or post–Missoula Flood fluvial aggradation, of which the magnitude is unknown; however, Columbia River erosion along the fan edge, debris flow material transported by the Columbia River, and uncertainty in the timing of fan initiation also introduce sources of underestimations.

Differential lidar–derived erosion rates

We determine the erosion rates post-ECF by compiling debris flow volume estimates measured from differential lidar within mapped debris flow polygons from the post-ECF inventory (22). We calculate postfire erosion rates over two epochs: a 1-year period from 2021 to 2022 and a 4-year period from 2018 to 2022 (~10⁰), corresponding to an instantaneous and average estimate of the ECF impacts. The volume of postfire erosion was calculated by summing all debris flow–associated depositional volumes (i.e., within mapped debris flow levee polygons) and subtracting it from the sum of all debris flow–associated erosional volumes within each delineated catchment (Fig. 2) and then averaging the net erosional volume over each measurement interval and the catchment area. The error for each differenced dataset is reported as root mean square deviation (RMSD) values, which are 0.91, 0.88, and 0.95 m for 2009–2018, 2018–2021, and 2021–2022, respectively (22). For many of the mapped landslides, we find that the error reported exceeds the differenced lidar values within the landslide, yet the elevation change is still clearly visible and mappable (see fig. S5). The high RMSD values are likely a result of errors upward of 40 m along vertical cliffs.

Prefire erosion was analyzed over a 9-year period, from 2009 to 2018 (~10¹). The quality of the 2009 lidar data was lower than the 2018, 2021, and 2022 lidar surveys, and few initiation events within the catchments were measurable. For the prefire epoch, we reanalyzed the differenced lidar data by removing all differenced values with no ground points and all differenced values on slopes greater than 55°. Using these thresholds, we found no events that exited the catchment and produced net erosion. We, however, did find that many channels were loaded with sediment, including regions with unburned contributing areas (fig. S5). Furthermore, there were minimal observations of postfire ravel between the fire and lidar collection (23), although we cannot rule out the potential contribution of wildfire-associated ravel affecting postfire debris flow volumes.

To quantify average debris flow recurrence intervals over the fan building period, given observed debris flow volumes, we divide the fan volume by the respective total debris flow volume that exited the catchment for the postfire events and divide by the time since the last Missoula Floods (15 ka). Values range from requiring 1 to 33 years⁻¹ to rebuild the fans over 15,000 years for the catchments with cumulative erosion over the 2018–2022 postfire period (i.e., MP 37.5 experienced no net erosion).

Historic inventory–derived erosion rates

We incorporated 1996 and 2001 landslides from an inventory compiled by Burns and Lindsey to estimate a multidecadal prefire interval [Fig. 2 (21)]. The Burns and Lindsey inventory calculates debris flow volumes by multiplying the mapped deposition area by 1 m, which assumes a constant thickness over the entire mapped extent. For 2001 events, the Multnomah County Natural Hazards Mitigation Plan reported the removal of ~150,000 m³ of material from the Dodson catchments (33), and the Burns and Lindsey inventory reports ~100,000 m³ from the Tumalt and Rock of Ages catchments (21). Powell (26) reports that cumulative 1996 debris flow volume from the Dodson catchments was nearly 1,000,000 m³, and the event from Tumalt Creek catchment was reported to make up nearly 500,000 m³, half of the total (26). Other reports of the landslide deposit volume estimates from field surveys of the 1996 Leavens Creek event (Fig. 2) of around 22,000 to 25,000 m³ (24) yet describes them as “preserved” deposits, which likely do not account for material cleared away from the interstate by the ODOT. Debris flow deposition (and therefore inferred erosion) was also underestimated in the Burns and Lindsey inventory due to material removal by ODOT before the imagery used for the mapping was collected (21). Thus, there are likely under- and overestimates in the Burns and Lindsey inventory, so we decided to scale the inventory values down by 50% for 1996 and 2001 events as a conservative estimate.

With these volume estimates, we calculate a cumulative prefire erosion rate, averaging over the time between 1996 and 2018 and including any measured debris flow volumes in the prefire, lidar-based measurements, which we report as the 10¹-year prefire erosion rate.

Rainfall analysis

We combine three records of precipitation data to analyze storm characteristics at the time of landslide initiation. We use (i) National Weather Service Cooperative Observer Network (COOP) weather station 1-hour precipitation data covering 1996–2002 [COOP ID: 350897 (59)], (ii) Bureau of Reclamation AgriMet weather station 15-min to 1-hour precipitation data covering 2002–2023 [AgriMet ID: bndw (60)], and (iii) National Resources Conservation Service Snow Telemetry (SNOTEL) Network weather station daily snow water equivalent (SWE) and daily precipitation values [SNOTEL ID: 666 (29)]. We differentiate storms by (i) 8-hour periods of no recorded precipitation, (ii) a minimum of 5 mm of precipitation, and (iii) a storm duration less than 300 hours [e.g., (61)], and each storm depth included the addition of total integrated SWE loss value during the storm, which we coin the terrestrial water input (TWI).

We identified the four debris flow–inducing storms that initiated events on 8 February 1996 (25, 26), 13 January 2021, 6 January 2022 (22), and 1 December 2023 in the Dodson region (fig. S4). The 8 February 1996 storm is infamous in the Pacific Northwest, as it produced thousands of debris flows across the state of Oregon (62), and the December 2023 storm we characterize vastly exceeds the duration and TWI values of the 1996 storm, yet it produced smaller debris flows with relatively low impacts to infrastructure. Over the entire recorded period, triggering events are distinct according to their duration and total rain and snowmelt inputs (fig. S4).

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S6