Post-fire debris flows (PFDFs) threaten life, property, and infrastructure in steep wildfire-prone terrain worldwide (e.g., Nyman et al. 2011; Kean et al. 2011; Parise and Cannon 2012; Raymond et al. 2020; Wall et al. 2020; Esposito et al. 2023). PFDFs are mixtures of water and sediment, typically with a sediment concentration exceeding 50% by volume (Iverson 1997). In the first few years following a wildfire, PFDFs often initiate when short-duration (typically <1 h), high-intensity rainfall produces runoff that rapidly entrains sediment on steep slopes (e.g., Meyer and Wells 1997; Gabet and Bookter 2008; Cannon et al. 2001; Kean et al. 2011; McGuire et al. 2017). Note that PFDFs are distinct from shallow landslides; they are runoff-driven, not infiltration-driven, and do not require antecedent rainfall nor high storm-total or multihour rainfall (Fig. 1a; Kean et al. 2011). With increasing wildfire activity (e.g., Westerling et al. 2006; Abatzoglou and Williams 2016) and projected intensification of short-duration, high-intensity rainfall in a warming climate (e.g., Prein et al. 2017; Fowler et al. 2021), the frequency of PFDFs is likely to increase. For example, Kean and Staley (2021) show that the likelihood of major PFDFs could more than double in Southern California with only modest warming. As PFDFs become a more frequent hazard affecting larger areas, more requests emerge for decision-support tools to effectively convey uncertainty around rainfall intensities, PFDF likelihood, volume, and potential impacts (e.g., Gourley et al. 2020). PFDFs are rainfall-driven, thus there is a particular need for tools that merge rainfall forecasts from numerical weather prediction models with models designed to assess PFDF hazards. Herein, we provide an overview of PFDFs, current methods for hazard assessment, and the challenges in operational forecasting and communicating information about these hazards. Through two case study events, we propose a framework for integrating ensemble forecasts from mesoscale models with PFDF likelihood and volume models to create decision-support tools. Several research and development activities needed to transition this framework to operational status are reviewed in the section “Benefits, challenges, and next steps.” This manuscript provides a compact reference on the science of PFDF hazards for the atmospheric science community, states atmospheric science-based needs to address PFDF hazards, and aims to inspire continued collaboration between atmospheric scientists and researchers in other scientific communities.
(a) Example of PFDF source material regions: steep channels (upper arrow) and rilling, which appears as rake-like patterns on the hillslope (lower arrow), from the Santa Ynez Mountains above Montecito, California, following the 9 Jan 2018 PFDF event. (b) Structure impacted by an extreme PFDF in the Montecito event. The arrow highlights ∼3–4-m splash marks on the building, an indicator of the depth and velocity of the PFDF. (c) Impacts of a moderate-size debris flow (10 Mar 2021) below the 2020 Bond Fire in Orange County, California; arrow highlights splash height (∼1 m). (d) Impacts of a small-volume PFDF in June 2022 on the Dixie Fire in Northern California. Photo credit for (a) and (b) U.S. Geological Survey; (c) David Erickson, Orange County Fire Authority; and (d) Don Lindsay, California Geological Survey.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
(a) Example of PFDF source material regions: steep channels (upper arrow) and rilling, which appears as rake-like patterns on the hillslope (lower arrow), from the Santa Ynez Mountains above Montecito, California, following the 9 Jan 2018 PFDF event. (b) Structure impacted by an extreme PFDF in the Montecito event. The arrow highlights ∼3–4-m splash marks on the building, an indicator of the depth and velocity of the PFDF. (c) Impacts of a moderate-size debris flow (10 Mar 2021) below the 2020 Bond Fire in Orange County, California; arrow highlights splash height (∼1 m). (d) Impacts of a small-volume PFDF in June 2022 on the Dixie Fire in Northern California. Photo credit for (a) and (b) U.S. Geological Survey; (c) David Erickson, Orange County Fire Authority; and (d) Don Lindsay, California Geological Survey.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
(a) Example of PFDF source material regions: steep channels (upper arrow) and rilling, which appears as rake-like patterns on the hillslope (lower arrow), from the Santa Ynez Mountains above Montecito, California, following the 9 Jan 2018 PFDF event. (b) Structure impacted by an extreme PFDF in the Montecito event. The arrow highlights ∼3–4-m splash marks on the building, an indicator of the depth and velocity of the PFDF. (c) Impacts of a moderate-size debris flow (10 Mar 2021) below the 2020 Bond Fire in Orange County, California; arrow highlights splash height (∼1 m). (d) Impacts of a small-volume PFDF in June 2022 on the Dixie Fire in Northern California. Photo credit for (a) and (b) U.S. Geological Survey; (c) David Erickson, Orange County Fire Authority; and (d) Don Lindsay, California Geological Survey.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Assessing post-fire debris-flow hazards
The threat to life and property downstream from burned areas increases as the likelihood and volume of PFDFs increases; larger volume PFDFs can inundate greater areas (Scheidl and Rickenmann 2010; Barnhart et al. 2021). PFDFs vary in magnitude and impacts, from those with negligible or minimal impacts to catastrophic events (Fig. 1). Rainfall intensity–duration thresholds are commonly used following wildfire to assess the potential for PFDFs in response to design or forecast rainstorms (NOAA-USGS Debris Flow Task Force 2005; Cannon et al. 2008). The M1 likelihood model proposed by Staley et al. (2017) provides estimates of PFDF likelihood at the small watershed (∼a few square kilometers) scale; referred to herein as “basins.” PFDF likelihood is based on the fraction of the basin area with slopes ≥ 23° with moderate or high soil burn severity (MH23) (Parson et al. 2010; Fig. A1 in the appendix), the basin-average difference normalized burn ratio (dNBR), a soil erodibility factor (KF), and the peak 15-min basin-average rainfall accumulated depth (R15) in millimeters. Basins are more likely to produce debris flows as MH23, dNBR, KF, or R15 increase. Therefore, even in the absence of a spatially variable rainfall field, PFDF likelihood varies throughout a burned area due to spatial variations in topography, burn severity, and soil erodibility. Accounting for anticipated spatial variations in rainfall intensity, however, would be particularly important because the model is more sensitive to changes in R15 relative to the other three input variables (Staley et al. 2017). Using the M1 likelihood model, a rainfall threshold can be defined for each basin in a burn area based on the R15 required for PFDF likelihood to exceed 50% (Staley et al. 2017).
The emergency assessment volume model proposed by Gartner et al. (2014) estimates PFDF volume based on basin relief, area burned at moderate-to-high severity, and the peak 15-min rainfall intensity (I15) in millimeters per hour. The I15 variable is simply the 15-min rainfall depth in millimeters (R15, described above) multiplied by 4 to provide a 15-min rainfall rate in millimeters per hour. There is a positive relationship between PFDF volume and each of the three predictor variables. We expect a positive relationship between debris-flow volume and area burned at moderate-to-high severity because these areas generally experience the most extreme reductions in vegetation cover and infiltration capacity, which increases the potential for erosion (Moody et al. 2013). Greater rainfall intensities are linked with enhanced runoff in small, recently burned basins, potentially mobilizing more sediment (Kean et al. 2011). Increases in PFDF volume with basin relief could be related to more energy being available for erosion processes (Gartner et al. 2014).
The M1 likelihood model and the emergency assessment volume model are most applicable in the first year following fire before the landscape recovers and becomes less susceptible to runoff-generated debris flows (Cannon et al. 2008; Hoch et al. 2021; Thomas et al. 2021). Both models rely on terrain, soil, and burn severity variables that capture basin-scale properties relevant to debris-flow processes. For a given watershed, these variables are computed once following the fire and then are assumed to be constant, at least during the first rainy season. Erosion and deposition of sediment during post-fire rainstorms will modify topography at small scales, but basin-scale terrain and soil properties used as model inputs will not vary substantially on this time scale. Therefore, within the context of the M1 likelihood model and emergency assessment volume model, it is the spatial distribution of peak 15-min rainfall accumulation (or intensity) that will vary from storm-to-storm and affect debris-flow hazard assessments.
The U.S. Geological Survey (USGS) Emergency Assessment of Post-Fire Debris-Flow Hazards operationalizes the M1 likelihood model and the emergency volume assessment model and is currently used throughout the western United States (Fig. 2; USGS 2022). For the first year following a wildfire, it provides estimates of PFDF likelihood and volume within each basin or stream segment of the burn area and a combined likelihood and volume hazard classification at various spatially constant rainfall intensities. This product does not provide predictions on PFDF runout or area inundated. These assessments are available at https://landslides.usgs.gov/hazards/postfire_debrisflow/.
Likelihood of a PFDF in response to the design rainstorm with a peak 15-min rainfall intensity (I15) of 24 mm h−1 by basin across the 2017 Thomas Fire burn area in Santa Barbara and Ventura Counties in California. Image is from the USGS Emergency Assessment of Post-Fire Debris-Flow Hazards page, https://landslides.usgs.gov/hazards/postfire_debrisflow/.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Likelihood of a PFDF in response to the design rainstorm with a peak 15-min rainfall intensity (I15) of 24 mm h−1 by basin across the 2017 Thomas Fire burn area in Santa Barbara and Ventura Counties in California. Image is from the USGS Emergency Assessment of Post-Fire Debris-Flow Hazards page, https://landslides.usgs.gov/hazards/postfire_debrisflow/.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Likelihood of a PFDF in response to the design rainstorm with a peak 15-min rainfall intensity (I15) of 24 mm h−1 by basin across the 2017 Thomas Fire burn area in Santa Barbara and Ventura Counties in California. Image is from the USGS Emergency Assessment of Post-Fire Debris-Flow Hazards page, https://landslides.usgs.gov/hazards/postfire_debrisflow/.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
These emergency PFDF assessments provide valuable information to stakeholders tasked with decision-making about actions such as using PFDF mitigation strategies (e.g., installation of debris fences or cleaning out debris basins), positioning resources, or initiating evacuations. As these efforts are costly, the goal is to protect life and property while avoiding unnecessary preparations or evacuations. The emergency PFDF assessments apply “design storms” that assume spatially constant rainfall intensities across the burn area (Fig. 2). Therefore, they do not represent the actual spatial variations in rainfall intensity that would be observed in any particular storm. Thus, the emergency PFDF assessments are valuable for identifying hazard areas ahead of a storm, but do not indicate specific areas most likely to observe rainfall intensities capable of initiating PFDFs on a storm-by-storm basis. Advances in computing resources and high-resolution meteorological modeling provide the opportunity to generate spatially varying estimates of PFDF likelihood and volume at the basin scale by applying spatially variable rainfall from high-resolution forecasts. Providing storm-specific information about PFDF hazards may enable decision-makers to better assess risk, which is a function of PFDF likelihood and impact on a downstream value-at-risk.
Forecasting and messaging post-fire debris-flow hazards
The M1 likelihood model provides basin-specific thresholds across the burn area, although it is often more practical for operational forecasters and emergency managers to consider a single intensity–duration threshold that can be applied across the entire burned area. These fire-wide thresholds are developed in part using the USGS emergency PFDF hazard assessment tool. A starting point is often the I15 at which 50% of the basins within a burn area have a PFDF likelihood > 50%. Subsequently, collaborations among various agencies [e.g., USGS, National Weather Service (NWS), U.S. Forest Service, state geological surveys] may refine the fire-wide threshold considering factors such as values-at-risk or previous impactful post-fire hydrologic events in a particular area. Thresholds vary for burn areas within and across regions and may be adjusted based on observed response following storm events. Thus, the operationally used thresholds may not be posted publicly.
NWS forecasters monitor forecast models to assess whether conditions are favorable for rainfall exceeding the established threshold in the vicinity of a burn area and issue flash flood watches, flash flood warnings, and flood advisories as needed. PFDFs fall under the flash flood hazard classification. Forecasters also communicate forecast rainfall rates and confidence in these rates to emergency managers, ideally at an 18–24-h lead time (Fig. 3). Although this approach has proven successful in avoiding PFDFs that occur without issuance of a watch, warning, or advisory, the uncertainty around forecast short-duration, high-intensity rainfall in mesoscale models (e.g., Cannon et al. 2020) at the lead time needed for decision support creates a situation conducive to “messaging fatigue” (Mackie 2013; Kim and So 2018; Koh et al. 2020). While it is necessary for forecasters to highlight that there is a potential for rainfall exceeding PFDF thresholds at lead times sufficient for decision-making, non-occurrence of events over time can lead to apathy about the hazard.
Example of NWS San Diego forecast office communication to stakeholders ahead of a potential PFDF event on the Bond Fire. This template provides (right) established rainfall depth thresholds at the 15-, 30-, and 60-min duration and (left) maximum 1-h rainfall for 6-h blocks for each burn scar over a 42-h period. Forecaster confidence in 1-h rainfall (duration or rate) and storm total rainfall are also provided. This template is not consistent across NWS offices; it was developed specifically to support stakeholder needs in the NWS San Diego County Warning Area.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Example of NWS San Diego forecast office communication to stakeholders ahead of a potential PFDF event on the Bond Fire. This template provides (right) established rainfall depth thresholds at the 15-, 30-, and 60-min duration and (left) maximum 1-h rainfall for 6-h blocks for each burn scar over a 42-h period. Forecaster confidence in 1-h rainfall (duration or rate) and storm total rainfall are also provided. This template is not consistent across NWS offices; it was developed specifically to support stakeholder needs in the NWS San Diego County Warning Area.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Example of NWS San Diego forecast office communication to stakeholders ahead of a potential PFDF event on the Bond Fire. This template provides (right) established rainfall depth thresholds at the 15-, 30-, and 60-min duration and (left) maximum 1-h rainfall for 6-h blocks for each burn scar over a 42-h period. Forecaster confidence in 1-h rainfall (duration or rate) and storm total rainfall are also provided. This template is not consistent across NWS offices; it was developed specifically to support stakeholder needs in the NWS San Diego County Warning Area.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Short-duration, high-intensity rainfall is typically driven by mesoscale (∼2–200 km, subdaily) atmospheric features (Lin 2007), although the synoptic environment can indicate the potential for mesoscale features to develop (e.g., Houze et al. 1976; Cannon et al. 2020; de Orla‐Barile et al. 2022). Forecasters primarily rely on deterministic mesoscale models (e.g., High-Resolution Rapid Refresh), small mesoscale ensembles (e.g., the High-Resolution Ensemble Forecast system, HREF) or global-scale ensembles (e.g., Global Ensemble Forecast System) to provide information on rainfall rates, potential for rainfall to intersect a burn area, and their confidence in these conditions. Deterministic mesoscale models provide information at a spatial and temporal scale relevant to PFDF hazards (≤a few kilometers, ≤1 h), but do not provide information about forecast uncertainty. The HREF provides a 10-member ensemble forecast at 3-km spatial resolution and 1-h temporal resolution. While a valuable tool, the spatial and temporal resolution are too coarse to ideally represent PFDF hazards. Additionally, the size of the ensemble may be too small to capture the full spread of possibilities and extremes. Global ensemble forecasts provide uncertainty information but are too spatially and temporally coarse to provide actionable information for PFDF hazards.
Operational meteorology and hydrology increasingly utilize ensemble-based forecasts that provide probabilistic hazard information as part of impact-based decision support to better account for the uncertainty present in weather forecasts (AMS 2007; Uccellini and Ten Hoeve 2019; Carr et al. 2021). For probabilistic information on PFDF hazards, precipitation forecasts from ensembles of mesoscale models are necessary inputs to PFDF hazard models. Additionally, as wildfires grow larger (e.g., Abatzoglou and Williams 2016; Kitzberger et al. 2017; Abatzoglou et al. 2021), PFDF hazard assessments cover greater areas relative to mesoscale features capable of producing high-intensity rainfall. Thus, there is a growing need to highlight specific portions of burned areas that are most likely to experience PFDFs with an incoming storm. To address these needs, we experiment with the integration of high-resolution large ensemble mesoscale atmospheric modeling to operational PFDF likelihood and volume models. Our aim is to assess the benefits and challenges of this integration to enhance impact-based decision support.
Selection of study area and storm events
Although PFDFs are observed in steep terrain across western North America, we focus on Southern California. This densely populated and wildfire-prone area experiences frequent damaging PFDFs (e.g., Cannon et al. 2008; Staley et al. 2013; Oakley et al. 2017; Kean et al. 2019; Schwartz et al. 2021). We focus on the 2017 Thomas Fire burn area (Figs. 2 and 4a) spanning eastern Santa Barbara and western Ventura Counties. The Thomas Fire ignited on 8 December 2017 during a hot Santa Ana downslope wind event (Gershunov et al. 2021), reaching 114,078 ha by the time containment occurred on 12 January 2018. The fire burned in steep terrain in an area with a history of post-fire debris flows (Kean et al. 2019; Lancaster et al. 2021). Numerous basins in the burn area were modeled to have a >60% likelihood of PFDFs in response to an I15 of 24 mm h−1 during the first year following the fire (Fig. 2), an average recurrence interval of <1 year (NOAA 2022).
(a) Southern California study area with topography (filled contours) encompassing the innermost 1-km Weather Research and Forecasting (WRF) Model domain. The Thomas Fire perimeter is shown in black. The inset map shows outermost (d01; dx = 9 km) and middle (d02; dx = 3 km) WRF domains. (b) Gold fill denotes areas inundated by the 9 Jan 2018 PFDFs (Lukashov et al. 2019; Kean et al. 2019).
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
(a) Southern California study area with topography (filled contours) encompassing the innermost 1-km Weather Research and Forecasting (WRF) Model domain. The Thomas Fire perimeter is shown in black. The inset map shows outermost (d01; dx = 9 km) and middle (d02; dx = 3 km) WRF domains. (b) Gold fill denotes areas inundated by the 9 Jan 2018 PFDFs (Lukashov et al. 2019; Kean et al. 2019).
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
(a) Southern California study area with topography (filled contours) encompassing the innermost 1-km Weather Research and Forecasting (WRF) Model domain. The Thomas Fire perimeter is shown in black. The inset map shows outermost (d01; dx = 9 km) and middle (d02; dx = 3 km) WRF domains. (b) Gold fill denotes areas inundated by the 9 Jan 2018 PFDFs (Lukashov et al. 2019; Kean et al. 2019).
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
For atmospheric modeling, we selected two events featuring narrow cold-frontal rainbands (NCFRs), 9 January 2018 (Fig. 5) and 2 February 2019. These mesoscale features are narrow (a few kilometers wide and tens to ∼100 km long) bands of high-intensity rainfall coincident with a storm’s cold front (e.g., Houze et al. 1976; Jorgensen et al. 2003). NCFRs have previously triggered damaging PFDFs in Southern California as well as produced other hydrologic impacts (Oakley et al. 2017, 2018; Cannon et al. 2018). NCFRs are common in Southern California, with an average of three events per year across the Southern California Bight (de Orla‐Barile et al. 2022). They vary in intensity and not all NCFRs produce impacts. Both selected NCFR events generated post-fire hydrologic impacts on the Thomas Fire burn area, although for brevity we focus this demonstration on the far more impactful 9 January 2018 event. Modeling results for the 2 February 2019 event are included in the supplemental material.
Radar reflectivity (dBZ) showing an NCFR impacting the Thomas Fire burn area for (a) 9 Jan 2018 at 1155 UTC and (b) 30 min later at 1225 UTC. The light red line shows the burn perimeter. Image source: California Nevada River Forecast Center.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Radar reflectivity (dBZ) showing an NCFR impacting the Thomas Fire burn area for (a) 9 Jan 2018 at 1155 UTC and (b) 30 min later at 1225 UTC. The light red line shows the burn perimeter. Image source: California Nevada River Forecast Center.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Radar reflectivity (dBZ) showing an NCFR impacting the Thomas Fire burn area for (a) 9 Jan 2018 at 1155 UTC and (b) 30 min later at 1225 UTC. The light red line shows the burn perimeter. Image source: California Nevada River Forecast Center.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
The 9 January 2018 NCFR event occurred before the Thomas Fire was 100% contained and led to PFDFs from burned basins above the community of Montecito, California (Fig. 4b). The PFDFs from this localized portion of the burn area mobilized approximately 680,000 m3 of debris (Kean et al. 2019), overwhelming debris basins and resulting in 23 deaths, 408 damaged homes, and nearly $1 billion in damages (Lukashov et al. 2019; Lancaster et al. 2021). Observed I15 ranged from 78 to 105 mm h−1 over the basins above Montecito (Kean et al. 2019), more than 3 times the I15 threshold of 25.5 mm h−1 defined for this burn area (USGS 2017). Numerous PFDFs were also observed in unpopulated, remote areas in the northwestern portion of the burn area (Swanson et al. 2022). The intense precipitation associated with the NCFR dissipated as it moved eastward into Ventura County (Oakley et al. 2018).
High-resolution, large ensemble simulation design
We used the Weather Research and Forecast (WRF) atmospheric model version 4.3 (Skamarock et al. 2021) to generate high-resolution precipitation forecasts over the study area. We applied a double-nested WRF Model domain configuration with 9-, 3-, and 1-km horizontal grid spacing (Fig. 4a). The 1-km domain extends from west of Point Conception to allow us to observe the evolution of the NCFR features as they propagate eastward toward the Thomas Fire study site. This domain features other areas of steep terrain prone to wildfire and post-fire debris flows such that the forecast rainfall output for these two storms could be applied to any burned watershed area, past or future, in subsequent studies. A total of 100 vertical terrain-following levels are used for the WRF simulations, with a domain model top of 10 hPa. The simulations were run at a 24-h forecast lead time with respect to the time of NCFR impacts on the burn area (e.g., 24-h lead time ahead of 1200 UTC 9 January 2018).
A 100-member WRF ensemble was developed to account for uncertainty in the precipitation forecasts. Initial and lateral boundary conditions from both the Global Forecast System (GFS) and European Centre for Medium-Range Weather Forecasts (ECMWF) weather models were used to account for model initial condition uncertainty. A stochastic kinetic energy perturbation (Berner et al. 2009, 2011) was applied to the potential temperature and horizontal wind at each numerical time step across all WRF Model domains. Last, multiple cumulus, microphysics, and boundary layer schemes were utilized to account for model physics uncertainty (a complete list is provided in Table A1). For each 1-km model grid cell, we generated 5-min accumulated rainfall for the 100 unique 24-h forecast realizations for each of the two storms. This provides a physically plausible range of rain rates and a probability distribution of forecast precipitation that are used as an input to debris-flow likelihood and volume models.
The goal of this work is not to assess weather forecast model performance, but rather to explore the feasibility and value of providing probabilistic information on PFDF likelihood and volume driven by rainfall forecast uncertainty. However, it is valuable to compare the range of model precipitation forecast values to precipitation observations to build confidence in the model representation of the weather event and inform interpretation of output from PFDF likelihood and volume models. When assessing the performance of model forecasts, especially to a point observation such as a rain gauge, a “neighborhood” method can be used in which adjacent grid cells (e.g., English et al. 2021) and timesteps are evaluated to avoid overpenalizing the model for slight errors in time and location. We utilize rain gauge data and evaluate the grid cell in which the station lies as well as the eight adjacent 1-km grid cells. We consider observed timing and ±1 h around the observed 15-min window of maximum interval precipitation at the gauge.
A goal of ensemble forecasting is to encapsulate the observed value within the ensemble spread. Although the observed value is an outlier within the ensemble, it is represented as a potential outcome. Using the neighborhood method to assess WRF ensemble performance at the Doulton Tunnel gauge above the debris-flow area in Montecito (Fig. 4), only one ensemble member exceeds the observed 15-min interval value of 97 mm h−1 (Fig. 6a). At the KTYD gauge, 11 ensemble members exceed the observed 15-min interval value of 76 mm h−1 (Fig. 6c). For other rain gauges examined in the Montecito area (not shown), the observed event was also an outlier within the ensemble forecast distribution. However, for assessed gauges farther west that were not directly impacted by the most intense rainfall in the NCFR, ensemble-mean 15-min rainfall intensities exceeded observed. Although observed values for the 9 January 2018 event lie at the very high end of the ensemble spread, the observations are much closer to the ensemble mean for the 2 February NCFR event assessed (Fig. S2). This suggests there may be specific challenges with modeling the 9 January 2018 event.
(a) Histogram of 15-min precipitation forecasts (in mm h−1) from the 100 WRF ensemble members at the location of the Doulton Tunnel rain gauge, using a 1-km neighborhood and temporal window of ±1 h about the time of maximum observed rainfall. Vertical lines indicate ensemble mean and gauge observation. (b) Time series of the observed (dots) and ensemble forecast 15-min rainfall rates at the WRF grid cell closest to the location of the Doulton Tunnel rain gauge; no neighborhood method is used. The black star indicates the maximum 15-min observed rainfall. The ensemble median, minimum, and maximum forecast values are denoted by the black, red, and blue lines, respectively, and the gray shading represents the ensemble interquartile range. (c) As in (a) for the KTYD gauge. (d) As in (b) for the KTYD gauge. See Fig. 4b for gauge locations. Station elevations provided under station names in (a) and (c). Gauge data acquired from County of Santa Barbara Department of Public Works via https://rain.cosbpw.net/.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
(a) Histogram of 15-min precipitation forecasts (in mm h−1) from the 100 WRF ensemble members at the location of the Doulton Tunnel rain gauge, using a 1-km neighborhood and temporal window of ±1 h about the time of maximum observed rainfall. Vertical lines indicate ensemble mean and gauge observation. (b) Time series of the observed (dots) and ensemble forecast 15-min rainfall rates at the WRF grid cell closest to the location of the Doulton Tunnel rain gauge; no neighborhood method is used. The black star indicates the maximum 15-min observed rainfall. The ensemble median, minimum, and maximum forecast values are denoted by the black, red, and blue lines, respectively, and the gray shading represents the ensemble interquartile range. (c) As in (a) for the KTYD gauge. (d) As in (b) for the KTYD gauge. See Fig. 4b for gauge locations. Station elevations provided under station names in (a) and (c). Gauge data acquired from County of Santa Barbara Department of Public Works via https://rain.cosbpw.net/.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
(a) Histogram of 15-min precipitation forecasts (in mm h−1) from the 100 WRF ensemble members at the location of the Doulton Tunnel rain gauge, using a 1-km neighborhood and temporal window of ±1 h about the time of maximum observed rainfall. Vertical lines indicate ensemble mean and gauge observation. (b) Time series of the observed (dots) and ensemble forecast 15-min rainfall rates at the WRF grid cell closest to the location of the Doulton Tunnel rain gauge; no neighborhood method is used. The black star indicates the maximum 15-min observed rainfall. The ensemble median, minimum, and maximum forecast values are denoted by the black, red, and blue lines, respectively, and the gray shading represents the ensemble interquartile range. (c) As in (a) for the KTYD gauge. (d) As in (b) for the KTYD gauge. See Fig. 4b for gauge locations. Station elevations provided under station names in (a) and (c). Gauge data acquired from County of Santa Barbara Department of Public Works via https://rain.cosbpw.net/.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
We also examine a time series of ensemble spread for the event at the location of the gauge (Figs. 6b,d). Here, we do not incorporate the neighborhood method, but rather focus on the grid cell closest to the station, and step through each 15-min interval. Modeled precipitation at Doulton Tunnel correctly displays a peak I15 at the time it was observed, but the maximum modeled peak intensity is roughly 15 mm h−1 lower than observed (Fig. 6b). At KTYD, the observed peak I15 is within the ensemble spread (Fig. 6d).
Applying mesoscale model output to debris-flow likelihood and volume models
The debris-flow likelihood and volume models accept a single value of R15 or I15, respectively, for each basin. For each of the 100 ensemble members, the storm peak R15 and I15 are calculated from the 5-min model precipitation at each 1-km grid cell within the burn area. Then, peak R15 and I15 are calculated for each basin within the burn area based on a weighted average among grid cells within the basin. With R15 and I15 calculated for each ensemble member in each basin, we compute the debris-flow likelihood and volume associated with each ensemble member. We summarize this information by calculating the likelihood and volume associated with the ensemble mean (i.e., as a measure of a likely outcome) or the ensemble 90th percentile (i.e., as a measure of a more extreme outcome).
Threshold exceedance by basin
Although forecasters typically consider rainfall exceeding fire-wide thresholds, development of forecast tools using basin-specific thresholds highlight the areas of greatest concern for incoming storms. PFDFs may occur if a threshold is exceeded but have no direct impacts on people or infrastructure if the PFDF is small or occurs in a remote location. PFDF hazard (likelihood and volume) generally increases with rainfall intensity; most major, impactful PFDFs in Southern California were associated with rainfall exceeding three times the established threshold (Kean and Staley 2021). The basin-specific rainfall threshold is defined as the I15 at which there is a >50% likelihood of PFDF occurrence in the basin according to the Staley et al. (2017) M1 likelihood model.
Most of the basins across the northern tier of the burn area have >75% of ensemble members exceeding the threshold (Fig. 7a). The northwestern portion of the burn area has numerous basins with >50% of ensemble members exceeding the threshold by a factor of 2, although the basins above Montecito were only in the 26%–50% of ensemble member range (Fig. 7b). Very few basins in the northern portion of the burn area feature >15% ensemble members exceeding the threshold by a factor of 3. The area above Montecito has 1%–15% of ensemble members exceeding the 3-times threshold (Fig. 7c), consistent with observed rain rates that were at the high end of the ensemble spread (Fig. 6). The spatial distribution of basins with larger numbers of ensemble members at or above the 3-times-threshold rainfall (Fig. 7c) may indicate either lower thresholds in these areas, or the model favoring the highest intensities in that area.
Number of WRF ensemble members exceeding basin-specific thresholds, defined as the I15 at which there is a >50% likelihood of PFDF occurrence for the basin, for (a) 1-times threshold, (b) 2-times threshold, and (c) 3-times threshold. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Number of WRF ensemble members exceeding basin-specific thresholds, defined as the I15 at which there is a >50% likelihood of PFDF occurrence for the basin, for (a) 1-times threshold, (b) 2-times threshold, and (c) 3-times threshold. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Number of WRF ensemble members exceeding basin-specific thresholds, defined as the I15 at which there is a >50% likelihood of PFDF occurrence for the basin, for (a) 1-times threshold, (b) 2-times threshold, and (c) 3-times threshold. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
PFDF likelihood and volume forecasts from ensemble-mean I15 and R15
The ensemble mean is a commonly used summary statistic. We utilize ensemble-mean R15 and I15 for each basin as an input to likelihood and volume models, and examine the standard deviation of I15, likelihood, and volume to represent uncertainty. For the 9 January 2018 event, the ensemble-mean forecast values of I15 are greater than 35 mm h−1 across the northern and western portions of the burn area. The highest I15 values (>40 mm h−1) were confined to several basins within this area (Fig. 8a). The western part of the burn area is also the region with the highest standard deviation (Fig. 8b), indicating the greatest uncertainty in rainfall intensities.
For the 9 Jan 2018 storm event, (left) the ensemble mean (average) and (right) the standard deviation for (a),(b) basin for peak I15 rainfall; (c),(d) probability of debris-flow occurrence; and (e),(f) debris-flow volume estimate. Basins outlined in red in (a) experienced impacts (Fig. 3b; Kean et al. 2019). The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
For the 9 Jan 2018 storm event, (left) the ensemble mean (average) and (right) the standard deviation for (a),(b) basin for peak I15 rainfall; (c),(d) probability of debris-flow occurrence; and (e),(f) debris-flow volume estimate. Basins outlined in red in (a) experienced impacts (Fig. 3b; Kean et al. 2019). The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
For the 9 Jan 2018 storm event, (left) the ensemble mean (average) and (right) the standard deviation for (a),(b) basin for peak I15 rainfall; (c),(d) probability of debris-flow occurrence; and (e),(f) debris-flow volume estimate. Basins outlined in red in (a) experienced impacts (Fig. 3b; Kean et al. 2019). The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
A majority of basins across the northern half of the burn area fell in the 80%–99% PFDF likelihood category, although several in the area above Montecito were in the 60%–80% range (Fig. 8c). High standard deviations (20%–31%; Fig. 8d) in PFDF likelihood are apparent in the western and southwestern portions of the burn area where there were also high standard deviations in I15 (>15%; Fig. 8b). In the southeastern portion of the burn area, likelihoods based on ensemble mean were lower, generally in the <60% category, although standard deviations were greater than in the northern portion of the burn area. Two factors likely contribute to this pattern. First, soil burn severity was generally lower in the southeastern portion of the burn scar (Fig. A1), which lowers PFDF likelihood, all else being equal. Second, the ensemble mean of I15 across the southeast portion of the burn scar was generally lower (25–35 mm versus 35–48 mm; Fig. 8a). Since PFDF likelihood is estimated with a logistic regression model, likelihood sharply increases for a given basin as I15 crosses a threshold value. The ensemble-mean I15 placed many basins in the southeastern portion of the burn scar in a position where likelihood was sensitive to I15. In contrast, many basins in the northern portion of the burn scar were less sensitive to similar variations in I15 given their burn severity and the ensemble-mean I15.
The largest volumes (>50,000 m3) predicted with ensemble-mean I15 are found across the northern tier of the burn area as well as far western and far eastern areas (Fig. 8e). The San Ysidro Creek basin (Fig. 4b) produced the largest observed PFDF during the actual event (297,000 m3). The predicted ensemble-mean volume for San Ysidro Creek was 76,564 m3. Volume observations were made for five basins following the 9 January 2018 PFDF event (Kean et al. 2019). Using the ensemble I15 as an input to the volume model, the observed volume was captured within the ensemble spread for all basins excepting San Ysidro, where the observed volume was higher than any in the ensemble spread (Fig. S7). As with precipitation observations versus ensemble forecast precipitation (Fig. 6), the observed volumes were at the high end of the ensemble distribution. Kean et al. (2019) found that the volume model overestimated volume in four out of five basins assessed using the best available estimates of I15; San Ysidro was the only one where volume was correctly predicted. This indicates that observed volume being at the higher end of the ensemble distribution in this study (Fig. S7) is driven by the observed I15 also being at the higher end of the forecast ensemble. Note that basins producing the greatest volumes do not always correspond to those with the highest rainfall intensities (Fig. 8a) or highest likelihoods (Fig. 8b). The volume model utilizes basin relief whereas the likelihood model incorporates information about topographic slope. Thus, it is possible that basins with low slope but large drainage area and relief (i.e., gentle gradient) may produce large volume PFDFs with a low likelihood.
Ensemble 90th-percentile, median, and 10th-percentile I15 and PFDF likelihood and volume
Users of weather forecast information tend to make less biased interpretations and better understand future weather conditions when they are provided with both upper and lower bounds of a forecast (Joslyn et al. 2011). Thus, in addition to the mean and standard deviation presented in Fig. 8, we also assess the ensemble 90th and 10th percentiles and the median (Fig. 9). For the 90th percentile of I15, nearly all basins in the western and northern portions of the burn area exceed 45 mm h−1, and a few basins across the northern tier exceed 55 mm h−1. Using the 90th-percentile I15 as an input to the likelihood and volume models, nearly all basins have an 80%–99% chance of debris flows (Fig. 9b). Several basins across the northern half of the burn area display predicted volumes > 100,000 m3 (Fig. 9c). Note that the 90th-percentile I15 for any two basins may not come from the same ensemble member; extreme basin I15 values may result from different ensemble members affecting different portions of the burn area with high-intensity rainfall. The 10th-percentile PFDF likelihood map (Fig. 9h) reveals several basins in the northern portion of the burn area have a high likelihood (80%–99%) of a PFDF even when rainfall intensities are at the lower end of the ensemble spread, indicating their basin characteristics and burn severity are very conducive to PFDF activity. The 10th-percentile predicted volumes (Fig. 9i) highlight some basins in the northeastern and north-central portion of the burn area that are projected to produce volumes in the 50,000–100,000 m3 range, despite relatively low I15 at the 10th percentile, also indicating that these basins may be conducive to high-volume debris flows based on their terrain and burn severity characteristics.
For the 9 Jan 2018 event, (top) ensemble 90th-percentile (a) peak I15, (b) probability of debris flow occurrence, and (c) predicted volume; (middle) ensemble median (d) peak I15, (e) probability of debris flow occurrence, and (f) predicted volume; and (bottom) ensemble 10th-percentile (g) peak I15, (h) probability of debris flow occurrence, and (i) predicted volume. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
For the 9 Jan 2018 event, (top) ensemble 90th-percentile (a) peak I15, (b) probability of debris flow occurrence, and (c) predicted volume; (middle) ensemble median (d) peak I15, (e) probability of debris flow occurrence, and (f) predicted volume; and (bottom) ensemble 10th-percentile (g) peak I15, (h) probability of debris flow occurrence, and (i) predicted volume. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
For the 9 Jan 2018 event, (top) ensemble 90th-percentile (a) peak I15, (b) probability of debris flow occurrence, and (c) predicted volume; (middle) ensemble median (d) peak I15, (e) probability of debris flow occurrence, and (f) predicted volume; and (bottom) ensemble 10th-percentile (g) peak I15, (h) probability of debris flow occurrence, and (i) predicted volume. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Number of ensemble members exceeding a likelihood, volume, or combined hazard threshold
Another way to visualize PFDF hazard is to show the number of ensemble members exceeding a likelihood or volume threshold of interest. We evaluated the number of basins with likelihood > 80%, considering this threshold as “high confidence” in PFDF occurrence. The basins with 81–100 ensemble members exceeding this threshold are confined to the northern tier of the burn area (Fig. 10a). Upslope from Montecito, three basins had 61–80 members predicting > 80% likelihood. Sixteen basins across the northern half of the burn scar have 81–100 ensemble members predicting volumes > 50,000 m3; none of these basins are upslope from Montecito. One basin that drains to Montecito has 61–80 members exceeding this threshold (Fig. 10b). We chose a threshold volume of 50,000 m3 as an order-of-magnitude cutoff for PFDFs that could have a substantial impact since the five most impactful PFDFs from the Montecito event had volumes ranging from 10,000 to 297,000 m3 (Kean et al. 2019).
For 9 Jan 2018, (a) number of WRF ensemble members > 80% post-fire debris-flow likelihood, (b) number of WRF ensemble members that produce volumes > 50,000 m3, and (c) number of WRF ensemble members with a “high” Combined Hazard (CH) class. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
For 9 Jan 2018, (a) number of WRF ensemble members > 80% post-fire debris-flow likelihood, (b) number of WRF ensemble members that produce volumes > 50,000 m3, and (c) number of WRF ensemble members with a “high” Combined Hazard (CH) class. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
For 9 Jan 2018, (a) number of WRF ensemble members > 80% post-fire debris-flow likelihood, (b) number of WRF ensemble members that produce volumes > 50,000 m3, and (c) number of WRF ensemble members with a “high” Combined Hazard (CH) class. The coordinates of the top-left and bottom-right corners of each map are 34°39′N, 119°41′W and 34°15′N, 118°56′W, respectively.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
The PFDF likelihood model presents the likelihood of triggering any PFDF, even if small and non-impactful. Thus, we also calculate Combined Hazard (CH) class, a metric that accounts for both likelihood and volume, to highlight greatest-hazard areas. To compute CH, we rank each basin into a likelihood class of 1–5 based on PFDF likelihood and a volume class of 1–4 based on PFDF volume. The two ranked values are summed to determine the CH class. A CH of 2–3 is classified as “low,” 4–6 is classified as “moderate,” and 7–9 is classified as “high” (USGS 2022). We display the basins that have a high CH and find 81–100 ensemble members are in agreement in high CH across much of the northern portion of the burn area (Fig. 10c). In the most-impacted area above Montecito, only two basins had 81–100 members with high CH.
Benefits, challenges, and next steps
At the 24-h forecast lead time, the precipitation forecast and PFDF models focus the greatest rainfall intensities, highest likelihoods, and highest combined hazard on the northern tier of the burn area. PFDFs occurred in this area (Swanson et al. 2022), although the greatest impacts were observed in the far western portion of the burn area associated with PFDFs originating in basins upslope from Montecito (Kean et al. 2019) (Fig. 4b). This pattern is due to both the precipitation forecast placing the highest intensities across the northern tier (Figs. 7 and 8a) as well as the basin conditions in this area being conducive to high PFDF likelihood, even for rainfall intensities at the lower end of the ensemble spread (Fig. 2). Although observed rainfall in the basins above Montecito was at the upper end of the ensemble spread (Fig. 6), the analyses presented still indicate potential for extreme I15 and PFDF occurrence in that area. Between 1 and 15 ensemble members in that area exceed 3 times the threshold (Fig. 7c), indicating a possibility of major PFDFs. The Montecito area has the highest standard deviation of I15 across the burn area (Fig. 8b), indicating high uncertainty in this variable. The 90th-percentile analyses (Fig. 9) also indicate the Montecito area had a potential for high I15, high PFDF likelihood, and high volume PFDFs.
The spatial pattern of I15 and PFDF hazard would likely draw the attention of persons interpreting the map to areas north of the impacted area at the 24-h lead time. However, much of the northern portion of the burn area is sparsely populated or wilderness areas. Filtering high-hazard basins by values-at-risk below the basin outlets may be one way to modify these maps to draw focus to areas where risk may still be high despite a lower probability of an impactful PFDF based on metrics such as likelihood, I15 > 3 times the threshold, volume, and/or high CH.
At a 24-h lead time, we anticipate uncertainty in mesoscale precipitation forecasts from numerical weather prediction, as uncertainty is present even at shorter lead times (e.g., Cannon et al. 2020; English et al. 2021). As lead time decreases, the ensemble will likely begin to converge on a solution more representative of observations, thus it is necessary to rerun the simulation at some frequency. While the PFDF likelihood and volume models run in a matter of minutes, the 100-member, 1-km resolution WRF simulation presented here would not be feasible to run operationally without access to a supercomputer due to computational requirements and processing time (each storm event 100-member forecast ensemble simulation required 125,000 CPU hours). One approach to make the ensemble forecast capability less computationally expensive is to run the higher-resolution domains only when specific thresholds of precipitation rates are predicted by the coarse-resolution domain. The 1-km grid spacing is critical to providing model input at a spatial scale relevant to the small, steep watersheds where debris flows initiate (e.g., Kean et al. 2011; Tang et al. 2019). Further evaluation and optimization would be needed to determine the ideal domain coverage, horizontal grid spacing, ensemble size and configuration, and frequency of simulation for operational evaluation of PFDF hazards. In addition, we focus on one storm type here, but storms with varying characteristics may result in different model performance at the 24-h lead time and would need to be evaluated in progress toward operationalization. Forecast performance at varying lead times should also be evaluated, with a goal of striking a balance among forecast accuracy, lead times needed for decision-making, and available computing resources.
In addition to PFDF likelihood and volume, the interaction of a flow with downstream topography helps determine its impacts. PFDFs with greater volumes are more likely to inundate large areas, although inundation extent and other debris-flow behavior relevant to assessing impacts (i.e., peak flow depth) also depends on flow mobility and the degree to which topography and infrastructure confine the flow. Visualizations from an operational PFDF inundation model, which could show the projected path and/or depth of PFDFs, are an important next step to provide impact-based decision support. Since volume is a necessary input to most debris-flow inundation models (e.g., Barnhart et al. 2021; Gorr et al. 2022), our experiment motivates creating a PFDF volume ensemble that can serve as an input to an inundation model. From ensemble volume, an ensemble of area inundated and peak flow depth can be generated that could highlight areas most likely to be impacted and provide insight to the severity of impacts.
The framework presented here can be applied anywhere, given two key considerations. First, the PFDF likelihood and volume models perform best where abundant PFDF data have been collected and utilized to train these models, which to date primarily includes Southern California (Gartner et al. 2014; Staley et al. 2017). Continued and enhanced monitoring of PFDF activity in recently burned areas throughout the western United States is necessary to improve model performance. Second, with additional configuration to optimize performance, the WRF ensemble approach used here for a cool-season frontal storm in California can be applied elsewhere for different seasons (e.g., for warm-season deep convection over the southwestern United States).
Probabilistic forecasts present valuable information for decision-making across time scales leading up to events (Dale et al. 2014), though have proven challenging for stakeholders to interpret or trust (Mass et al. 2009). With appropriate presentation, probabilistic information generally improves decision quality (Ripberger et al. 2022). Thus, how forecast data are displayed and communicated warrants consideration (Mass et al. 2009; Lambrecht et al. 2019; Carr et al. 2021). An important next step in this work is to evaluate stakeholder ability to interpret storm-specific, spatially variable information about PFDF hazards and establish best practices for teaching them how to apply this information as well as how to reduce messaging fatigue. Verification of coupled precipitation forecast–PFDF models may also help build stakeholder confidence in these tools (Carr et al. 2021). Ultimately, as users make decisions that increasingly account for the uncertainty in weather forecasts (AMS 2007), the actions taken based on probabilistic information may support reduced impacts to life and property associated with PFDFs and other high-impact weather events.
Conclusions
This study presents coupling of a high-resolution, 100-member ensemble precipitation forecast with PFDF likelihood and volume models to provide probabilistic information on PFDF hazards. This approach combines precipitation forecast and PFDF information, including estimates of uncertainty, which is valuable for decision-making. However, at the 24-h lead time, errors in forecasts of rainfall intensity models may lead to misrepresentation of PFDF hazard at a specific location in the burn area. Thus, examining the full model ensemble spread in areas with values-at-risk is necessary. The proposed approach demonstrates the feasibility of moving toward an operational tool for PFDFs that provides impact-based decision support at relevant time scales for the protection of life and property and a resource to support the reduction of messaging fatigue resulting from false alarms. To continue forward progress on this topic, it is critical that the meteorology community collaborate with the geomorphology community (e.g., Moody et al. 2013). This article highlights the benefits of such interactions.
Acknowledgments.
This work was supported by the California Department of Water Resources Atmospheric River Program (4600013361) and the NOAA Collaborative Science, Technology, and Applied Research (CSTAR) program (NA19NWS4680004). We thank Don Lindsay at the California Geological Survey and three anonymous reviewers for feedback that improved this manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
Data availability statement.
The precipitation output from the WRF simulations used in the analyses presented in this manuscript are available from the University of California, San Diego Library Digital Collections at https://doi.org/10.6075/J0251JCQ
Appendix: Additional materials
Figure A1 shows a soil burn severity map of the Thomas Fire, and Table A1 shows the Weather Research and Forecasting (WRF) Model setup.
Thomas Fire soil burn severity map from the U.S. Forest Service Burned Area Emergency Response (BAER) team report. Reds indicate areas of high burn severity, yellow indicates moderate severity, green indicates low severity, and gray indicates unburned areas. The Thomas Fire BAER report can be accessed online at https://www.fs.usda.gov/detail/lpnf/home/?cid=fseprd570093.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Thomas Fire soil burn severity map from the U.S. Forest Service Burned Area Emergency Response (BAER) team report. Reds indicate areas of high burn severity, yellow indicates moderate severity, green indicates low severity, and gray indicates unburned areas. The Thomas Fire BAER report can be accessed online at https://www.fs.usda.gov/detail/lpnf/home/?cid=fseprd570093.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Thomas Fire soil burn severity map from the U.S. Forest Service Burned Area Emergency Response (BAER) team report. Reds indicate areas of high burn severity, yellow indicates moderate severity, green indicates low severity, and gray indicates unburned areas. The Thomas Fire BAER report can be accessed online at https://www.fs.usda.gov/detail/lpnf/home/?cid=fseprd570093.
Citation: Bulletin of the American Meteorological Society 104, 9; 10.1175/BAMS-D-22-0188.1
Physics models included in the WRF 100-member ensemble.
References
Abatzoglou, J. T., and A. P. Williams, 2016: Impact of anthropogenic climate change on wildfire across western US forests. Proc. Natl. Acad. Sci. USA, 113, 11 770–11 775, https://doi.org/10.1073/pnas.1607171113.
Abatzoglou, J. T., D. S. Battisti, A. P. Williams, W. D. Hansen, B. J. Harvey, and C. A. Kolden, 2021: Projected increases in western US forest fire despite growing fuel constraints. Commun. Earth Environ., 2, 227, https://doi.org/10.1038/s43247-021-00299-0.
AMS, 2007: American Meteorological Society information statement on weather analysis and forecasting. Accessed 18 September 2022, https://www.ametsoc.org/ams/index.cfm/about-ams/ams-statements/statements-of-the-ams-in-force/enhancing-weather-information-with-probability-forecasts/.
Barnhart, K. R., R. P. Jones, D. L. George, B.W. McArdell, F. K. Rengers, D. M. Staley, and J. W. Kean, 2021: Multi‐model comparison of computed debris flow runout for the 9 January 2018 Montecito, California post‐wildfire event. J. Geophys. Res. Earth Surf., 126, e2021JF006245, https://doi.org/10.1029/2021JF006245.
Berner, J., G. J. Shutts, M. Leutbecher, and T. N. Palmer, 2009: A spectral stochastic kinetic energy backscatter scheme and its impact on flow-dependent predictability in the ECMWF ensemble prediction system. J. Atmos. Sci., 66, 603–626, https://doi.org/10.1175/2008JAS2677.1.
Berner, J., S. Y. Ha, J. P. Hacker, A. Fournier, and C. Snyder, 2011: Model uncertainty in a mesoscale ensemble prediction system: Stochastic versus multiphysics representations. Mon. Wea. Rev., 139, 1972–1995, https://doi.org/10.1175/2010MWR3595.1.
Cannon, F., C. W. Hecht, J. M. Cordeira, and F. M. Ralph, 2018: Synoptic and mesoscale forcing of Southern California extreme precipitation. J. Geophys. Res. Atmos., 123, 13714–13730, https://doi.org/10.1029/2018JD029045.
Cannon, F., and Coauthors, 2020: Observations and predictability of a high-impact narrow cold-frontal rainband over Southern California on 2 February 2019. Wea. Forecasting, 35, 2083–2097, https://doi.org/10.1175/WAF-D-20-0012.1.
Cannon, S. H., E. R. Bigio, and E. Mine, 2001: A process for fire‐related debris flow initiation, Cerro Grande fire, New Mexico. Hydrol. Processes, 15, 3011–3023, https://doi.org/10.1002/hyp.388.
Cannon, S. H., J. E. Gartner, R. C. Wilson, J. C. Bowers, and J. L. Laber, 2008: Storm rainfall conditions for floods and debris flows from recently burned areas in southwestern Colorado and Southern California. Geomorphology, 96, 250–269, https://doi.org/10.1016/j.geomorph.2007.03.019.
Carr, R. H., K. Semmens, B. Montz, and K. Maxfield, 2021: Improving the use of hydrologic probabilistic and deterministic information in decision-making. Bull. Amer. Meteor. Soc., 102, E1878–E1896, https://doi.org/10.1175/BAMS-D-21-0019.1.
Dale, M., J. Wicks, K. Mylne, F. Pappenberger, S. Laeger, and S. Taylor, 2014: Probabilistic flood forecasting and decision-making: An innovative risk-based approach. Nat. Hazards, 70, 159–172, https://doi.org/10.1007/s11069-012-0483-z.
de Orla‐Barile, M., F. Cannon, N. S. Oakley, and F.M. Ralph, 2022: A climatology of narrow cold‐frontal rainbands in Southern California. Geophys. Res. Lett., 49, e2021GL095362, https://doi.org/10.1029/2021GL095362.
English, J. M., D. D. Turner, T. I. Alcott, W. R. Moninger, J. L. Bytheway, R. Cifelli, and M. Marquis, 2021: Evaluating operational and experimental HRRR model forecasts of atmospheric river events in California. Wea. Forecasting, 36, 1925–1944, https://doi.org/10.1175/WAF-D-21-0081.1.
Esposito, G., S. L. Gariano, R. Masi, S. Alfano, and G. Giannatiempo, 2023: Rainfall conditions leading to runoff-initiated post-fire debris flows in Campania, southern Italy. Geomorphology, 423, 108557, https://doi.org/10.1016/j.geomorph.2022.108557.
Fowler, H. J., and Coauthors, 2021: Anthropogenic intensification of short-duration rainfall extremes. Nat. Rev. Earth Environ., 2, 107–122, https://doi.org/10.1038/s43017-020-00128-6.
Gabet, E. J., and A. Bookter, 2008: A morphometric analysis of gullies scoured by post-fire progressively bulked debris flows in southwest Montana, USA. Geomorphology, 96, 298–309, https://doi.org/10.1016/j.geomorph.2007.03.016.
Gartner, J. E., S. H. Cannon, and P. M. Santi, 2014: Empirical models for predicting volumes of sediment deposited by debris flows and sediment-laden floods in the Transverse Ranges of Southern California. Eng. Geol., 176, 45–56, https://doi.org/10.1016/j.enggeo.2014.04.008.
Gershunov, A., and Coauthors, 2021: Hot and cold flavors of Southern California’s Santa Ana winds: Their causes, trends, and links with wildfire. Climate Dyn., 57, 2233–2248, https://doi.org/10.1007/s00382-021-05802-z.
Gorr, A. N., L. A. McGuire, A. M. Youberg, and F. K. Rengers, 2022: A progressive flow-routing model for rapid assessment of debris-flow inundation. Landslides, 19, 2055–2073, https://doi.org/10.1007/s10346-022-01890-y.
Gourley, J. J., and Coauthors, 2020: Predicting the floods that follow the flames. Bull. Amer. Meteor. Soc., 101, E1101–E1106, https://doi.org/10.1175/BAMS-D-20-0040.1.
Hoch, O. J., L. A. McGuire, A. M. Youberg, and F. K. Rengers, 2021: Hydrogeomorphic recovery and temporal changes in rainfall thresholds for debris flows following wildfire. J. Geophys. Res. Earth Surf., 126, e2021JF006374, https://doi.org/10.1029/2021JF006374.
Houze, R. A., Jr., P. V. Hobbs, K. R. Biswas, and W. M. Davis, 1976: Mesoscale rainbands in extratropical cyclones. Mon. Wea. Rev., 104, 868–878, https://doi.org/10.1175/1520-0493(1976)104<0868:MRIEC>2.0.CO;2.
Iverson, R. M., 1997: The physics of debris flows. Rev. Geophys., 35, 245–296, https://doi.org/10.1029/97RG00426.
Jorgensen, D. P., Z. Pu, P. O. G. Persson, and W. K. Tao, 2003: Variations associated with cores and gaps of a Pacific narrow cold frontal rainband. Mon. Wea. Rev., 131, 2705–2729, https://doi.org/10.1175/1520-0493(2003)131<2705:VAWCAG>2.0.CO;2.
Joslyn, S., S. Savelli, and L. Nadav-Greenberg, 2011: Reducing probabilistic weather forecasts to the worst-case scenario: Anchoring effects. J. Exp. Psychol. Appl., 17, 342–353, https://doi.org/10.1037/a0025901.
Kean, J. W., and D. M. Staley, 2021: Forecasting the frequency and magnitude of postfire debris flows across Southern California. Earth’s Future, 9, e2020EF001735, https://doi.org/10.1029/2020EF001735.
Kean, J. W., D. M. Staley, and S. H. Cannon, 2011: In situ measurements of post‐fire debris flows in Southern California: Comparisons of the timing and magnitude of 24 debris‐flow events with rainfall and soil moisture conditions. J. Geophys. Res., 116, F04019, https://doi.org/10.1029/2011JF002005.
Kean, J. W., and Coauthors, 2019: Inundation, flow dynamics, and damage in the 9 January 2018 Montecito debris-flow event, California, USA: Opportunities and challenges for post-wildfire risk assessment. Geosphere, 15, 1140–1163, https://doi.org/10.1130/GES02048.1.
Kim, S., and J. So, 2018: How message fatigue toward health messages leads to ineffective persuasive outcomes: Examining the mediating roles of reactance and inattention. J. Health Commun., 23, 109–116, https://doi.org/10.1080/10810730.2017.1414900.
Kitzberger, T., D. A. Falk, A. L. Westerling, and T. W. Swetnam, 2017: Direct and indirect climate controls predict heterogeneous early-mid 21st century wildfire burned area across western and boreal North America. PLOS ONE, 12, e0188486, https://doi.org/10.1371/journal.pone.0188486.
Koh, P. K., L. L. Chan, and E. K. Tan, 2020: Messaging fatigue and desensitisation to information during pandemic. Arch. Med. Res., 51, 716–717, https://doi.org/10.1016/j.arcmed.2020.06.014.
Lambrecht, K. M., B. J. Hatchett, L. C. Walsh, M. Collins, and Z. Tolby, 2019: Improving visual communication of weather forecasts with rhetoric. Bull. Amer. Meteor. Soc., 100, 557–563, https://doi.org/10.1175/BAMS-D-18-0186.1.
Lancaster, J. T., and Coauthors, 2021: Observations and analyses of the 9 January 2018 debris-flow disaster, Santa Barbara County, California. Environ. Eng. Geosci., 27, 3–27, https://doi.org/10.2113/EEG-D-20-00015.
Lin, Y. L., 2007: Mesoscale Dynamics. Cambridge University Press, 630 pp., https://doi.org/10.1017/CBO9780511619649.
Lukashov, S. G., J. T. Lancaster, N. S. Oakley, and B. J. Swanson, 2019: Post-fire debris flows of 9 January 2018, Thomas Fire, Southern California: Initiation areas, precipitation and impacts. Debris-Flow Hazards Mitigation: Mechanics, Monitoring, Modeling, and Assessment, J. W. Kean et al., Eds., Association of Environmental and Engineering Geologists, 774–781, https://doi.org/10.25676/11124/173144.
Mackie, B., 2013: Warning fatigue: Insights from the Australian bushfire context. Ph.D. dissertation, Media and Communication, University of Canterbury, 216 pp., https://doi.org/10.26021/4033.
Mass, C., and Coauthors, 2009: PROBCAST: A web-based portal to mesoscale probabilistic forecasts. Bull. Amer. Meteor. Soc., 90, 1009–1014, https://doi.org/10.1175/2009BAMS2775.1.
McGuire, L. A., F. K. Rengers, J. W. Kean, and D. M. Staley, 2017: Debris flow initiation by runoff in a recently burned basin: Is grain‐by‐grain sediment bulking or en masse failure to blame? Geophys. Res. Lett., 44, 7310–7319, https://doi.org/10.1002/2017GL074243.
Meyer, G. A., and S. G. Wells, 1997: Fire-related sedimentation events on alluvial fans, Yellowstone National Park, USA. J. Sediment. Res., 67, 776–791, https://doi.org/10.1306/D426863A-2B26-11D7-8648000102C1865D.
Moody, J. A., R. A. Shakesby, P. R. Robichaud, S. H. Cannon, and D. A. Martin, 2013: Current research issues related to post-wildfire runoff and erosion processes. Earth-Sci. Rev., 122, 10–37, https://doi.org/10.1016/j.earscirev.2013.03.004.
NOAA, 2022: Atlas 14 point precipitation frequency estimates. Accessed 18 September 2022, https://hdsc.nws.noaa.gov/hdsc/pfds/pfds_map_cont.html.
NOAA-USGS Debris Flow Task Force, 2005: NOAA-USGS debris-flow warning system: Final report. U.S. Geological Survey Circular 1283, accessed 20 August 2022, http://pubs.usgs.gov/circ/2005/1283/pdf/Circular1283.pdf.
Nyman, P., G. J. Sheridan, H. G. Smith, and P. N. Lane, 2011: Evidence of debris flow occurrence after wildfire in upland catchments of south-east Australia. Geomorphology, 125, 383–401, https://doi.org/10.1016/j.geomorph.2010.10.016.
Oakley, N. S., J. T. Lancaster, M. L. Kaplan, and F. M. Ralph, 2017: Synoptic conditions associated with cool season post-fire debris flows in the Transverse Ranges of Southern California. Nat. Hazards, 88, 327–354, https://doi.org/10.1007/s11069-017-2867-6.
Oakley, N. S., F. Cannon, R. Munroe, J. T. Lancaster, D. Gomberg, and F. M. Ralph, 2018: Brief communication: Meteorological and climatological conditions associated with the 9 January 2018 post-fire debris flows in Montecito and Carpinteria, California, USA. Nat. Hazards Earth Syst. Sci., 18, 3037–3043, https://doi.org/10.5194/nhess-18-3037-2018.
Parise, M., and S. H. Cannon, 2012: Wildfire impacts on the processes that generate debris flows in burned watersheds. Nat. Hazards, 61, 217–227, https://doi.org/10.1007/s11069-011-9769-9.
Parson, A., P. R. Robichaud, S. A. Lewis, C. Napper, and J. T. Clark, 2010: Field guide for mapping post-fire soil burn severity. General Tech. Rep. RMRS-GTR-243, U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, 49 pp., https://doi.org/10.2737/RMRS-GTR-243.
Prein, A. F., R. M. Rasmussen, K. Ikeda, C. Liu, M. P. Clark, and G. J. Holland, 2017: The future intensification of hourly precipitation extremes. Nat. Climate Change, 7, 48–52, https://doi.org/10.1038/nclimate3168.
Raymond, C. A., L. A. McGuire, A. M. Youberg, D. M. Staley, and J. W. Kean, 2020: Thresholds for post‐wildfire debris flows: Insights from the Pinal Fire, Arizona, USA. Earth Surf. Processes Landforms, 45, 1349–1360, https://doi.org/10.1002/esp.4805.
Ripberger, J., A. Bell, A. Fox, A. Forney, W. Livingston, C. Gaddie, C. Silva, and H. Jenkins-Smith, 2022: Communicating probability information in weather forecasts: Findings and recommendations from a living systematic review of the research literature. Wea. Climate Soc., 14, 481–498, https://doi.org/10.1175/WCAS-D-21-0034.1.
Scheidl, C., and D. Rickenmann, 2010: Empirical prediction of debris‐flow mobility and deposition on fans. Earth Surf. Processes Landforms, 35, 157–173, https://doi.org/10.1002/esp.1897.
Schwartz, J. Y., N. S. Oakley, and P. Alessio, 2021: Assessment of a post-fire debris flow impacting El Capitan watershed, Santa Barbara County, California, USA. Environ. Eng. Geosci., 27, 423–437, https://doi.org/10.2113/EEG-D-21-00008.
Skamarock, W. C., and Coauthors, 2021: A description of the Advanced Research WRF Model version 4.3. NCAR Tech. Note NCAR/TN-556+STR, 145 pp., https://doi.org/10.5065/1dfh-6p97.
Staley, D. M., J. W. Kean, S. H. Cannon, K. M. Schmidt, and J. L. Laber, 2013: Objective definition of rainfall intensity–duration thresholds for the initiation of post-fire debris flows in Southern California. Landslides, 10, 547–562, https://doi.org/10.1007/s10346-012-0341-9.
Staley, D. M., J. A. Negri, J. W. Kean, J. L. Laber, A. C. Tillery, and A. M. Youberg, 2017: Prediction of spatially explicit rainfall intensity–duration thresholds for post-fire debris-flow generation in the western United States. Geomorphology, 278, 149–162, https://doi.org/10.1016/j.geomorph.2016.10.019.
Swanson, B. J., S. G. Lukashov, J. Y. Schwartz, D. N. Lindsay, and J. T. Lancaster, 2022: Assessment of logistic regression model performance and physical controls on January 9, 2018, debris flows, Thomas Fire, California. Environ. Eng. Geosci., 28, 113–131, https://doi.org/10.2113/EEG-D-21-00026.
Tang, H., L. A. McGuire, F. K. Rengers, J. W. Kean, D. M. Staley, and J. B. Smith, 2019: Evolution of debris‐flow initiation mechanisms and sediment sources during a sequence of postwildfire rainstorms. J. Geophys. Res. Earth Surf., 124, 1572–1595, https://doi.org/10.1029/2018JF004837.
Thomas, M. A., F. K. Rengers, J. W. Kean, L. A. McGuire, D. M. Staley, K. R. Barnhart, and B. A. Ebel, 2021: Postwildfire soil‐hydraulic recovery and the persistence of debris flow hazards. J. Geophys. Res. Earth Surf., 126, e2021JF006091, https://doi.org/10.1029/2021JF006091.
Uccellini, L. W., and J. E. Ten Hoeve, 2019: Evolving the National Weather Service to build a weather-ready nation: Connecting observations, forecasts, and warnings to decision-makers through impact-based decision support services. Bull. Amer. Meteor. Soc., 100, 1923–1942, https://doi.org/10.1175/BAMS-D-18-0159.1.
USGS, 2017: Emergency assessment of post-fire debris flow hazards: Thomas Fire. Accessed 19 September 2022, https://landslides.usgs.gov/hazards/postfire_debrisflow/detail.php?objectid=177.
USGS, 2022: Emergency assessment of post-fire debris flow hazards. Accessed 19 September 2022, https://landslides.usgs.gov/hazards/postfire_debrisflow/.
Wall, S. A., J. J. Roering, and F. K. Rengers, 2020: Runoff-initiated post-fire debris flow Western Cascades, Oregon. Landslides, 17, 1649–1661, https://doi.org/10.1007/s10346-020-01376-9.
Westerling, A. L., H. G. Hidalgo, D. R. Cayan, and T. W. Swetnam, 2006: Warming and earlier spring increase western US forest wildfire activity. Science, 313, 940–943, https://doi.org/10.1126/science.1128834.