Debris flood risk assessment for Mosquito Creek, British Columbia, Canada

Nat Hazards (2013) 65:1653–1681 DOI 10.1007/s11069-012-0436-6 ORIGINAL PAPER

Debris flood risk assessment for Mosquito Creek, British Columbia, Canada Matthias Jakob • Kris Holm • Hamish Weatherly • Shielan Liu Neil Ripley

•

Received: 9 May 2012 / Accepted: 29 September 2012 / Published online: 16 October 2012  Springer Science+Business Media Dordrecht 2012

Abstract Mosquito Creek drains a 15.5 km2 watershed on the North Shore Mountains north of Vancouver, British Columbia, Canada, and flows through the densely urbanized District and then City of North Vancouver. Previous studies determined that the creek is subject to debris floods (hyperconcentrated flows). The National Research Council of Canada is applying multi-hazard risk assessment procedures for various regions in B.C. and chose Mosquito Creek as one of its target areas. As part of its natural hazard management plan, the District of North Vancouver (DNV) requested an assessment of debris flood hazards and associated risk to life. Using a combination of empirical methods, dendrochronology and some judgment, BGC Engineering Inc. assessed debris flood hazard extent, velocity and depth for estimated 100-, 200-, 500- and 2,500-year debris flow return periods. Based on the results from the hazard assessment, risk for individuals and groups living within the hazard area, including residential homes and a fire hall, was estimated. Compared to risk tolerance criteria accepted on an interim basis by the DNV, we estimate that societal risk exceeds tolerable standards and that individual risk exceeds tolerable standards for 10 homes. The results from the risk to loss of life study have prompted DNV to implement a series of risk reduction measures including installation of a debris containment net and watershed restoration measures. Keywords Debris floods  Hazard assessment  Debris flood modelling  Runout analysis  Risk assessment

1 Introduction 1.1 General Mosquito Creek, located within a 15.5 km2 watershed in the District of North Vancouver (DNV), British Columbia (B.C.), Canada, is subject to debris floods (hyperconcentrated M. Jakob (&)  K. Holm  H. Weatherly  S. Liu  N. Ripley BGC Engineering Inc., 1045 Howe Street, Vancouver, BC, Canada e-mail: [email protected]

123

1654

Nat Hazards (2013) 65:1653–1681

flows) triggered in the upper watershed by the failure of temporary landslide dams. They typically result in higher sediment concentrations, larger clasts moved and thus higher impact forces than normal rainfall-generated floods. Residential development and a fire hall border the lower channel reaches and are subject to debris flood risk, including an October 31, 1981, event that killed one person and inundated the fire hall. As part of a DNV natural hazard risk reduction initiative and a Natural Resources Canada (NRCan) programme to quantify multiple hazards for key communities, BGC Engineering Inc. (BGC) completed a quantitative debris flood hazard and risk assessment for Mosquito Creek. The objectives of the project were to define the area subject to debris flood hazards and to assess whether the risk to life posed by this hazard should be considered intolerable by the DNV. In the hazard assessment, we quantify the extent and intensity of debris floods and estimate risk to life for the potentially affected residential development and fire hall. This work included field investigation into previous debris floods and generation of debris flood intensity maps based on modelled landslide dam outbreak floods. We then use the debris flood intensity maps, showing estimated debris flood extent, velocity and depth, as the basis to quantify individual and group risk for people living and working within the affected area. Many published studies deal with situations in which comprehensive data sets could be constructed from various sources of physical evidence and historic data. While the same sources were exhausted for this study, the data presented are still discontinuous and are associated with significant error. It is notable that data scarcity and considerable error are typical for urban landslide assessments, which we believe enhances the value of such case. Furthermore, this paper employs quantitative risk assessment techniques that are an emerging practice in geotechnique in B.C. but have yet to gain a strong foothold. The intent of this paper is to encourage a broader adoption of quantitative landslide risk assessments because they provide a more defensible and transparent approach to management of landslide risks than previous methods that rely on a fixed return period for which mitigation measures are to implemented, and fail to consider the often wide range of possible landslide consequences. 1.2 Previous work Kerr Wood Leidal (KWL 2003) completed a hazard assessment on Mosquito Creek that identified debris floods as the most hazardous hydrogeomorphic process in the watershed. This is in accordance with the suggestion by Jakob and Jordan (2001) that many steep creeks in mountainous areas of B.C. are capable of producing debris floods as a consequence of landslide dam outbreaks. KWL’s work included an estimate of debris flood magnitude and frequency. At the time, a design event magnitude approach was chosen based on a prior hazard tolerance criterion of 10 % in 50-year probability recommended by the B.C. Ministry of Transportation (i.e. a 500-year event) (see APEGBC 2010). This design event magnitude was estimated through a combination of dendrochronological investigations and assumed landslide damming of Mosquito Creek, resulting in a landslide dam outbreak flood. KWL (2003) estimated a 500-year return period peak discharge of 250 m3/s with an associated total sediment volume transported of 10,000 m3. These hazard intensities were then entered into a qualitative risk matrix. The consequence of the design event was considered ‘‘Very High’’, defined by KWL (2003) as ‘‘direct or indirect debris impact with extensive structural damage’’. The KWL study did not explicitly include loss of life in the risk assessment, although it is inferred since extensive structural damage has a high potential to result in loss of life if the building is occupied at the time of impact.

123

Nat Hazards (2013) 65:1653–1681

1655

1.3 Study area The headwaters of the 15.5 km2 Mosquito Creek watershed are located between Grouse Mountain and Mount Fromme at an elevation of approximately 1,000 m. The main channel is about 9 km in length and drains southwards into Burrard Inlet (Fig. 1). A majority of the watershed is located within the District of North Vancouver, but lower reaches pass through the City of North Vancouver and Mission Indian Reserve No. 1. We define the upper watershed as a 4.3 km2 area above the upper fan apex at elevation 345 m (Fig. 1). This area is mostly forested with the Grouse Mountain ski area on the west slopes and a service access road and trail network extending across the middle slopes (Fig. 1). Mosquito Creek is confined within a steep and narrow V-shaped valley above the upper fan apex, and channel gradients vary between 13 and 20 % (Fig. 2). Significant deposits of sediment and woody debris occur where the local channel gradient drops significantly, the channel is obstructed by a log jam, or the channel widens. A high percentage of the sedimentation appears to be the result of debris slides associated with past logging or as debris avalanches originating downstream of forest roads. In this paper, we prefer the term debris avalanche as a mostly turbulent landslide process (Hungr et al. 2001). Logging in the upper watershed extends back to the early 1900s. Logging continued in the mid to late 1960s when a forestry road was constructed below the access road on both sides of the valley (shown as deactivated logging road on Fig. 1). Upon the completion of tree removal in 1970, the logging road was abandoned, and over time, a number of debris slides have initiated downslope of the road. This landslide activity resulted in accelerated sediment accumulation within Mosquito Creek, and impact scars and the formation of traumatic resin tissue on trees adjacent to the channel provide evidence of recent debris flood events. A footbridge crosses the upper fan apex, immediately downstream of which is an intermittent deposition area where the channel loses confinement and widens considerably. This upper fan area consists of an up to 125-m-wide deposition zone that extends about 300 m downstream before confining once again (Fig. 1). At the fan apex, a 40-m-long training berm was constructed along the left (east) bank in 1991 to prevent channel avulsions. Downstream of this deposition zone, Mosquito Creek watershed is urbanized with residential housing and a fire hall. Although geomorphic activity is lower than the upper watershed, large boulder trains were identified in the upper fan with little or no organic soil cover, indicating that debris floods with high transport capacity have occurred in the recent past. Several abandoned channels exist on the east side of the deposition zone, with the active channel flowing along the western fringe. In the 400-m-long section downstream from the fan apex to Montroyal Boulevard (see Fig. 1), the creek is well confined, has a width of approximately 8 m and an average channel gradient of 15 %. Lower sections of this reach have been stabilized by placing and grouting large angular boulders in-channel and along banks. A trail is located on the western bank, and an overgrown side channel is located on the east side of the ravine bottom (Fig. 3). This channel can convey flows during extreme runoff events, as experienced during the October flood of 1981. During this flood, the fire hall, which is located at the downstream end of this side channel, witnessed water and mud damage. Between Montroyal Boulevard and Evergreen Basin (125 m elevation) (Fig. 4), the channel substrate and banks are riprapped within a ravine extending 10–15 m above the channel bed. Although residential homes are located on both sides of the creek, they are located mostly outside the bounds of the ravine and thus less susceptible to floods or debris floods. This situation differs fundamentally to the situation downstream of Evergreen Basin

123

1656

Nat Hazards (2013) 65:1653–1681

Fig. 1 Study area overview

where buildings have been placed directly into the former creek channel. On the east side of Mosquito Creek, just below Montroyal Boulevard ravine slopes are gentler and the valley top is poorly defined. Downstream of Montroyal Boulevard, the channel is wider (between 10 and 15 m) and the creek corridor is up to 150 m wide. On the right bank, an old channel runs adjacent to the ravine sidewalls. A 4-m-wide trail runs parallel to the right bank of the creek, approximately 2 m above the channel bottom.

123

Nat Hazards (2013) 65:1653–1681

1657

Fig. 2 Channel profile and key landmarks

Fig. 3 Overflow channel near Montroyal Boulevard

Evergreen Basin (Fig. 4), with a capacity of approximately 6,000 m3 forms the inlet to an 850-m-long concrete box culvert (approximately 3 m 9 3 m), constructed in the early 1960s to reduce the potential for flood damage. The culvert has a cross-section area of 9 m2 and presumed capacity of 25 m3/s. The basin is intended to capture sediment and large woody debris transported by the creek during extreme flood events and provides a surcharge pond for culvert inflows. The box culvert slopes at 5 % has two entry points and extends to immediately downstream of West Queens Road where it daylights and discharges back into the natural channel (Fig. 4). Based on recent LiDAR data collected by the DNV, Evergreen Basin has an approximate storage capacity of 6,000 m3.

123

1658

Nat Hazards (2013) 65:1653–1681

Fig. 4 Mosquito Creek between Montroyal Boulevard and West Queens Road

In the event that the debris cage is blocked an overflow spillway direct water flow downstream. A steel grillage is located at the entrance to the overflow spillway, to reduce the potential of large woody debris blocking the culvert entrance. In case the amount of woody debris overwhelms the steel grillage and blocks the culvert, inlet floodwaters and debris could overflow the basin and inundate the historic channel of Mosquito Creek, eroding the channel. Downstream of Evergreen Basin for about 450 m, the former channel of Mosquito Creek has been retained as park land and a trail meanders along the alignment of the old channel. Development adjacent to this reach has not encroached into the old creek corridor, and the old channel remains as a pronounced drainage that could convey significant discharge. However, a number of houses have been constructed on the right bank (east side of Fairmont Road, Fig. 4), within the creek corridor. Reportedly during the 1981 flood, the basin was within a few feet of overtopping due to large woody debris plugging the culvert inlet.

123

Nat Hazards (2013) 65:1653–1681

1659

The south end of the trail eventually emerges at the north end of Del Rio Drive (Fig. 4). Here, the land above the culverted section of Mosquito Creek has not been retained as a park and the original channel is no longer discernible due to extensive landscaping. The former channel of Mosquito Creek runs through the backyards of the residential properties on the west side of Del Rio Drive. The Evergreen Basin culvert outfall is located to the immediate south of West Queens Road, where Mosquito Creek once again enters a city park (William Griffin Park) (Fig. 4). The large William Griffin recreational centre is located to the immediate east of the creek at this location. As with upstream reaches, the park encompasses the bounds of the ravine that confines Mosquito Creek. 1.4 Previous events Between Montroyal Boulevard and Evergreen Basin, a number of coniferous trees along the right bank have visible impact scars and are likely indicative of boulders hitting the trees during one or more debris flood events. Based on a dendrochronological analysis of these scars and supplemented by historical documentation of flood events, KWL (2003) concluded that significant (defined by being intensive enough to scar trees adjacent to the channel) creek events occurred in 1896, 1924, 1949, 1961, 1968, 1971, 1975 and 1981. These events were probably debris floods as they exceeded the likely wetted cross-section from peak water flows, suggesting a return period of approximately 18 years for debris floods. The cluster of events from 1961 to 1981 may be attributable to poor logging practices in the 1960s and 1970s that led to watershed destabilization, particularly from road fill-related failures that were identified on historic air photographs. The most recent storm to cause significant flooding and erosion problems on Mosquito Creek occurred on October 31, 1981. Damage sustained included erosion above Montroyal Boulevard and significant deposition and blockage at Evergreen Basin. One person drowned while attempting to cross the creek. In addition, the Montroyal fire hall was flooded as ‘‘a wall of mud and raging water burst into the hall, cascading up to the ceilings in some parts’’ (Vancouver Sun, November 1, 1981). This description is consistent with the characteristics of a debris flood. In response to the 1981 debris flood, numerous projects were undertaken throughout the DNV to restore creek channels over a 4-year period. These works included significant bank and channel protection along Mosquito Creek from above Montroyal Boulevard to Evergreen Basin. These channel works have proven to be successful in reducing the damage of floods that have occurred over the last 30 years. Moreover, logging has ceased in the watershed since the 1970s and tree regeneration with the corresponding increase in root strength, and road deactivation likely further reduced the frequency of debris floods triggered by temporary landslide blockages. Despite this apparent decline in high frequency events, there is no guarantee that large, rare events can re-occur and pose a significant risk to downstream residents and infrastructure.

2 Hazard assessment 2.1 Methods The most credible future debris flood scenario at Mosquito Creek is for a debris avalanche to initiate in the upper watershed and descend to the main channel where, at impact, a

123

1660

Nat Hazards (2013) 65:1653–1681

landslide dam is formed. Continuation of flow in the form of a debris flow is less likely due to the right angle impact that is unlikely to transfer momentum through impulsive loading (Bovis and Dagg 1992). Furthermore, Mosquito Creek’s channel gradient of 15–28 % in the upper watershed limits debris flow mobilization. The resulting landslide dam would impound a small lake that would subsequently breach, most likely by overtopping. Depending on the breach mechanism and time to evolve, an outbreak flood (i.e. debris flood) could result. This process has been well documented throughout B.C. and elsewhere (Jakob and Jordan 2001; Jakob and Hungr 2005). For the purposes of this study, a digital elevation model (DEM) of the watershed was created. Inputs to the DEM included 1- and 2-m resolution LiDAR data. Because these two LiDAR images do not cover the entire Mosquito Creek watershed, gaps in the coverage were supplemented with a 20 m DEM from the NRCan Geobase. The resulting DEM was then re-sampled to a 10-m resolution grid. 2.1.1 Landslide frequency and volume We examined air photographs and Google Earth imagery and estimated the potential volume of rare (200-year return period and 500-year return period) debris avalanches that could potentially initiate from the valley sideslopes. It is difficult to assign return periods to debris avalanches in the absence of accurate dating control, and making reasonable estimates requires some geomorphic judgment and experience. We assumed two scenarios. One is a slightly larger debris avalanche than those that can be observed on historic air photographs in the Mosquito Creek watershed. Interpretation of 70 years of historic air photographs showed that debris avalanches of [1,000 m3 volume likely occur with a frequency of approximately 5 years in the Mosquito Creek watershed and over the past 100 years have largely been associated with poorly managed road drainage (KWL 2003). Debris avalanches of [10,000 m3 likely occur with a return period between 50 and 100 years, while debris avalanches [*20,000 m3 are estimated to occur at return periods of exceeding 200 years, although an exact determination of the associated return period is difficult (Fig. 5). There is little botanical evidence for debris of avalanches in excess of 20,000 m3 as the majority of the watershed was logged in the early 1900s. However, road construction and associated drainage problems have created the potential for such large debris avalanches to be triggered. Debris avalanches in Mosquito Creek can only achieve a finite volume, limited by the shallow depth (\1 m) of soil available to fail and slope length. The estimated 500-year return period debris avalanche magnitude is larger than events observed on historic air photographs. The failure scenario would be very intense rainfall on bare soils or a thin and wet snowpack occurring after an extended wet period, conditions that most likely to occur in the study area in late November to early January. Misdirected runoff from the Grouse Mountain ski area was observed during a field visit. Such drainage concentration could potentially trigger a debris slide, entraining material along its downstream path and evolve into a debris avalanche (Hungr et al. 2001). Debris avalanches increase in width along their downslope path through margin shear and lateral sediment entrainment but are limited in volume by the underlying bedrock or basal till that resists erosion. Based on Google Earth imagery, interpretation of historic air photographs and fieldwork, two locations were identified where large debris avalanches could initiate (Fig. 1). Landslide A is located on the west side of Mosquito Creek, approximately 1.1 km

123

Nat Hazards (2013) 65:1653–1681

1661

Fig. 5 Upstream view of Evergreen Basin from overflow spillway

upstream from the fan apex. The location of Landslide A was chosen because of the connectivity with a logging road below which landslides have been triggered in the past, the redirected runoff from the ski runs above and the fact that the area encompassed by the presumed landslide is an area of flow convergence. Landslide A could be approximately 80 m wide, 570 m long, and has an average slope gradient of approximately 47 %. Landslide B (Fig. 1) is located on the east side of the creek, approximately 400 m downstream from the toe of Landslide A. Possible elevation differences from the crest to the toe of the landslide could vary from approximately 620–480 m, resulting in a total landslide slope length of 170 m with an average slope gradient of 55. Debris avalanches have a tendency to widen on open slopes by entrainment of colluvial materials along the flow margin. The angle of landslide widening was determined by measuring the widening angle of a random sample of 50 recent (occurrence in the past 10 years or so) debris avalanches on relatively planar slopes on Vancouver Island and the North Shore Mountains on satellite images. The angle was measured as the intersecting angle of the two widest landslide margins to the central point of initiation. The 50 debris avalanches measured yielded a mean and median widening angle of 17 and 15, respectively, and a maximum widening angle of 44. Lacking a scientific basis to ‘‘calculate’’ the angle of widening, judgment was applied to determine the most likely widening angle. Since the lower portion of Landslide A would be confined, we assumed that widening beyond the natural flow convergence is unlikely. Therefore, a widening angle of 9 was chosen for Landslide A. The areas of these two hypothetical landslides are estimated at 18,000 and 36,000 m2, estimated as having return periods of roughly 200 and 500 years, respectively. Both landslides are well covered by a mix of hemlock, fir and cedar trees. Using a hand auger, soil depth was measured at seven locations at Landslide A and at three sites at Landslide B. Average soil depth was approximately 0.7–1 m. Slope gradients were also measured during the slope traverse. At the potential impact sites of the landslides with Mosquito Creek, channel widths were measured as well as upstream and downstream channel gradients. Channel

123

1662

Nat Hazards (2013) 65:1653–1681

Fig. 6 Cross-section of impact zone of landslide dams resulting from hypothetical Landslide B

Fig. 7 Cross-section of impact zone of landslide dams resulting from hypothetical Landslide A Table 1 Approximate thicknesses and volumes of hypothetical landslides A and B Hypothetical landslide

Landslide area (m2)

Landslide thickness (m)

Landslide volume (m3)

Landslide dam height (m)

A

36,000

1.0

36,000

17

B

18,000

0.7

12,000

11

cross-sections were sketched at the potential impact zones along with the potential configuration of landslide dams (Fig. 6 and 7). Based on the field work, areas, thicknesses, volumes and landslide, dam heights were estimated for Landslides A and B (Table 1). 2.1.2 Landslide dam outbreak peak flow estimate KWL (2003) dated tree impact scars near the Baden Powell Trail footbridge (Fig. 1) and measured the cross-section between the impact points on either creek bank. The most likely date of the debris flood leading to the tree damage was determined to January 19, 1968, which (at the time) was the highest 24-h rainfall recorded at the DNV District Hall some 2 km east of Mosquito Creek. Adjusting for channel scour subsequent to the debris flood, the peak discharge of the 1968 event was estimated by KWL (2003) to lie between 120 and 180 m3/s with an approximate corresponding debris flood return period of 200 years. A more precise method to estimate peak discharge from the failure of a landslide dam is the use of a numerical model. For this study, the physically based mathematical model BREACH (Fread 1991), which is distributed and maintained by NOAA’s National

123

Nat Hazards (2013) 65:1653–1681

1663

Weather Service, was used. BREACH simulates the physical processes of an overtopping or piping failure using the principles of hydraulics, sediment transport and slope stability. Generally, given the same initial conditions, an overtopping failure will produce a higher peak flow and shorter duration raising limb of hydrograph than a piping failure. The former failure scenario was therefore used as a conservative assumption in this study. An assumption of the BREACH model is that a naturally formed landslide dam has no crest width (Fread 1991). Therefore, erosion initially commences with the breach channel on the downstream slope. Inputs to BREACH include the geometry and physical soil properties of the dam and impounded lake, inflow hydrograph (i.e. baseflow), tail water cross-sections, roughness co-efficient and numerical simulation control parameters. Table 2 summarizes the design flood events, landslides and landslide dam characteristics for the 100-, 200-, 500- and 2,500-year return period debris floods. The number of model runs is dictated by the objectives of the debris flood risk assessment. In this instance, it was desirable to determine existing risk over a large spectrum of return periods, and secondly, to identify the return period range that is successfully mitigated by a range of engineered mitigation structures. Debris floods were simulated for the 100-, 200-, 500-, and 2,500-year return period debris floods, with the 200-year event represented by failure of Landslide B and the 500-year event represented by failure of Landslide A. In Table 2, n/a indicates that a landslide dam was not considered at this return period. The initial water surface elevations were set equal to the elevations of the dam crests (i.e. the landslide location elevation plus the assumed landslide dam height presented in Table 1). Baseflow in Mosquito Creek, for both scenarios, was conservatively assumed to be equal to the 200-year return period peak instantaneous flow of approximately 22 m3/s as estimated by KWL (2003) for both the 200- and 500-year return period events. The primary output from BREACH is the breach peak discharge and outflow hydrograph. Both landslide dams are assumed to have properties similar to earthen dams. BREACH requires that a number of soil property parameters be input to the model, as shown in Table 3. While some of these parameters could be estimated with detailed field investigations, in general, there is a restricted range over which they are sensitive. Parameter values shown in Table 3 are therefore either based on field estimations (e.g. D50 and porosity) or generally accepted values in the scientific literature. Estimated reservoir volumes and surface area are summarized in Table 4 for the 200- and 500-year return period events. The 100- and 2,500-year event peak flows have been estimated through curve extrapolation as detailed in the results section.

Table 2 Design flood events and associated landslide and associated landslide and landslide dam characteristics Return period (years)

Landslide volume (m3)

Landslide dam failure mode

Initial elevation of water surface at t = 0 (m)

Base flowa (m3/s)

100

n/a

overtopping

n/a

22

200

12,000 (landslide A)

Overtopping

491

22

500

36,000 (landslide B)

Overtopping

576

22

2,500

Multiple landslides

Sequential overtopping

Not determined

22

Baseflow is assumed as equivalent to approximately the 200-year flood (KWL 2003) a

Refer to text for notes and assumptions

123

1664

Nat Hazards (2013) 65:1653–1681

Table 3 Assumed input parameters for the outbreak flood modelling using BREACH Input parameters

Lower value

Upper value

Best estimate

D50 grain size (mm)

1

10

7

Porosity ratio

0.32

0.40

0.36

Unit weight (kg/m3)

1,600

2,000

1,800

Internal friction angle (degrees)

15

25

20

Cohesive strength (kN)

10

20

12

Ratio of D90 to D30 grain sizes

100

400

300

Table 4 Estimated impoundment water volumes Scenario (return period)

Upstream face slope (H:V)

Downstream face slope (H:V)

Reservoir surface area (m2)

Maximum water depth (m)

Impounded water volume (m3)

Large dam (500-year) landslide A

2.7:1

2.7:1

5,200

10.5

21,000

Small dam (200-year) landslide B

2.7:1

2.7:1

1,700

6.0

4,300

2.1.3 Debris flood modelling Debris flood hazards at Mosquito Creek were defined based on a combination of onedimensional dam breach modelling with BREACH (which defines the hydrograph shape for a natural dam breach in the upper watershed) and two-dimensional hydraulic modelling (which routes the dam breach downstream). The two-dimensional hydraulic model FLO-2D (2004) was chosen to route the outbreak flood downstream. FLO-2D can model unconfined flows across fan surfaces and simulate flows of varying sediment concentrations. FLO-2D is a depth-averaged volume conservation-based flood routing model that was developed specifically for the analysis of mud flows. Flow progression is controlled by topography and flow resistance. The governing equations include the continuity equation and the dynamic wave momentum equation. The debris flow/debris flood runout area is discretized using a rectangular grid, and the equations of motion are solved using a central finite difference scheme, in which the average flow velocity across a grid element boundary is computed one direction at a time, with eight potential flow directions. For modelling hyperconcentrated flows, FLO-2D requires input parameters for viscosity and yield stress. For this assessment, we used empirical coefficient and exponent values associated with the low viscosity scenario of Table 5. Table 5 Empirical coefficients used for FLO-2D debris flow modelling Scenario

Viscosity coefficient (a1)

Viscosity exponent (b1)

Yield stress coefficient (a2)

Yield stress exponent (b2)

High viscosity

2.7

11.0

0.05

14.5

Intermediate viscosity

1.0

11.0

0.1

15.0

Low viscosity

0.13

12.0

2.7

10.4

123

Nat Hazards (2013) 65:1653–1681

1665

1000

Debris Floods (multiple landslide dams)

500

300 200

Debris Floods

Peak Discharge (m3/s)

(single landslide dam)

100

50

Rainfall-Generated Floods via Regiona lAnalysis

30 20

Rainfall-generated floods via rainfall-runoff method 10

5

1

2

5

10

20

50

100

200

500

1,000

2000

5000

Return Period (years) Fig. 8 Debris flood return period–peak discharge graph

2.2 Hazard assessment results 2.2.1 Debris flood frequency–magnitude relationships The basis for numerical modelling of debris floods with variable return periods requires the establishment of a reliable frequency–magnitude relationship. Figure 8 shows the frequency– magnitude (return period peak discharge) for floods and debris floods on Mosquito Creek. This figure has been generated using a variety of datasets and methods. The data for rainfallgenerated floods have been obtained from KWL (2003). The debris flood frequency and magnitude relationship for the 200- and 500-year return period events assume debris avalanches (Landslides A and B) impacting the main channel of Mosquito Creek, creating landslide dams that subsequently breach. The resulting peak flows are compared with dendrochronologic methods. The 100-year return period debris flood peak flow has been determined by linear extrapolation of the existing 200- to 500-year frequency–magnitude curve. The 2,500-year return period event is associated with greater uncertainty. A linear extrapolation of the existing frequency–magnitude curve would yield a peak flow for the 2,500-year event of 900 m3/s. Work by Jakob and Friele (2010) and Jakob (2012) has shown that such simple extrapolations may not provide reliable results because the underlying hazard processes change. In the case of Jakob and Friele’s (2010) work at Cheekye River, some 80 km north of Mosquito Creek, linear extrapolation would have yielded a 10,000-year return period debris flow 2.7 times higher than the one estimated from application of detailed field-based methods.

123

1666

Nat Hazards (2013) 65:1653–1681

At Mosquito Creek, slope length and overburden colluvium thickness cannot yield a single debris avalanche large enough to produce a lake with sufficient water volume and dam height to produce a breach with a peak discharge of 900 m3/s. The only process by which even higher peak flows as determined for the 500-year event can be conceived is multiple quasi-simultaneous slope failures. There are several cases in the literature where this has occurred leading to extreme discharge events (i.e. Guadagno and Revellino 2005; Garcia-Martinez and Lopez 2005). However, in the case of Mosquito Creek, flow convergence to form a ‘‘super debris flow’’ is not considered to be conceivable because the channel is of too low a gradient to allow bulk debris flow transport. Instead, multiple debris avalanches would create multiple landslide dams. Once one dam is overtopped and breached, downstream dams would likely fail in sequence. However, this process does not yield additive discharges due to flow attenuation within the intermittent reaches. Following the arguments above, the 2,500-year return period debris flood is likely to lie between the 310 m3/s estimated for the 500-year debris flood and the 900 m3/s that would be obtained from linear extrapolation of the frequency–magnitude curve shown in Fig. 6. Conditions that would lead to a 2,500-year event could include an extreme rainstorm or rain-on-snow event ([500-year return period) following a period of exceptionally high antecedent moisture conditions, or a significant earthquake during very wet conditions that would lead to widespread shallow landsliding. In either scenario, it is plausible that several simultaneous failures could occur in the Mosquito Creek watershed. In the absence of detailed fan trenching and radiocarbon dating, it is currently estimated that the peak discharge of the 2,500-year return period event corresponds to a doubling of the QDF500, which would yield approximately 600 m3/s. To account for the inherent uncertainty of such an estimate, a tentative return period range from 1,000 to 4,000 years has been assigned for this event. The only rational approach to estimate sediment volumes associated with such an event would be to drill into the Mosquito Creek delta at several locations or trench the Mosquito Creek fan which cannot be completed due to the existing dense development on the fan and the excessive costs associated with such investigations. 2.2.2 Breach modelling The following results can be summarized: • Peak discharges for the 200- and 500-year return period events are approximately 120 and 310 m3/s, respectively; • the hydrograph shape is insensitive to changes in soil property; • for both cases, the peak outflow occurs rapidly after the breach: approximately 2 min for the 500-year event and 1 min for the 200-year event for the upper value range (Figs. 7 and 8); and • BREACH assumes that the final breach has a trapezoidal cross-section. For the 500-year event and the 200-year event, the top width of the breach at peak breach flow is 17.5 m and 10 m, and the bottom width is approximately 6.7 m and 6 m, respectively. The best estimate and the upper range of the peak discharge for the smaller hypothetical dam (200-year return period) are nearly identical because small landslide dams that collapse mainly due to water pressure are insensitive to soil properties as was tested. The empirically derived results show a wide range of possible results. However, they are generally consistent with the BREACH modelling results with Froehlich’s equations coming closest to the results obtained in this study (Table 6).

123

Nat Hazards (2013) 65:1653–1681

1667

Discharge(m3/s)

300

Breach Outflow 200 Year 500 Year

200

100

0 0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

6.5

7

7.5

8

Time (Minutes) Fig. 9 Hypothetical landslide dams breach outflow hydrographs on Mosquito Creek

2.2.3 Evergreen basin A major modelling assumption is the capacity of the Evergreen Basin culvert during a debris flood event. The culvert capacity is a critical element of this analysis as overflows from the culvert have the potential to impact a number of houses along Del Rio Drive (Fig. 4). These houses were constructed following the construction the culvert on Mosquito Creek between Evergreen Basin and West Queens Road in the 1960s. At a water depth (H) to culvert diameter (D) ratio of 1 the culvert has a capacity of 25 m3/s. Given the layout of the basin and culvert, H/D ratios of up to 3 are possible, theoretically tripling the potential unobstructed culvert capacity. For the 100-year event, the attenuated peak flow estimate is 25 m3/s, which matches the culvert capacity at H/ D = 1. This event is likely be associated with less bedload and organic debris transport than higher return period events, implying that Evergreen Basin is more likely to function as designed and intercept the debris before it can reach the downstream culvert intake. For these reasons, BGC assumed that the culvert would operate at or near capacity and allow the entire attenuated peak flow to pass. This assumption is still considered to be reasonably conservative because it is likely that the H/D ratio will exceed 1 and therefore the culvert would convey more than the 25 m3/s. Table 6 Empirical estimates of landslide dam breach peak discharge at the breach location Authors

Formula

Q500 (m3/s)

Q200 (m3/s)

Blown and Church (1985)

Q = 2.7(HV)0.42

470

190

Costa and Schuster (1988)

Q = 0.763(VH)0.42

130

50

Froehlich (1995)

Q = 0.607 V0.295H1.24

210

70

H is the lake level drop and V is the volume of water impounded

123

1668

Nat Hazards (2013) 65:1653–1681

Table 7 Modelled peak flows and attenuated flows for the four scenarios Reference return period (years)

Return period range (years)

Modelled peak flow (m3/s)

Attenuated flow at evergreen basin (m3/s)

100

50–150

70

25

200

100–300

120

38

500

300–700

310

84

2500

1,000–4,000

600

193

However, the 200-, 500- and 1,000-year return period debris floods will likely carry a large amount of bedload and organic debris with the potential to overwhelm Evergreen Basin and block the culvert intake. The attenuated flow of the 200-year flood (QDF200) and the 500-year debris flood (QDF500) both exceed the culvert capacity at a H/D ratio of 1 (Table 7). H/D ratios in excess of one would then promote accumulation of debris at the intake and result in a high likelihood of culvert clogging. By inference, this would also occur for the 2,500-year debris flood (QDF2500) as it has a higher peak flow and will transport higher volumes of bedload and organic debris than either the 500- or the 200-year return period events. These assumptions are believed to be reasonably conservative. Ideally, one would assign probabilities for culvert exceedance for different attenuated flows. Such probabilities, however, would be subjective because the effects of bedload and organic debris in clogging the culvert are unknown and cannot be modelled. 2.2.4 Debris flood modelling Debris flood hazard can be defined by the intensity of such events: that is the maximum flow depth and the maximum flow velocity. These two variables are estimated based on FLO-2D hydraulic modelling and includes an estimation of the peak flow in the upper watershed and the downstream routing of the peak flow, which attenuates in a downstream direction due to frictional losses. Table 7 summarizes the modelled peak flows and attenuated flows at Evergreen Basin for the four scenarios considered herein: 100-, 200-, 500- and 2,500-year events. Figure 9 shows the outflow hydrograph for the 200- and 500-year debris floods.Only the 100-, 200-, 500- and 2,500-year return period debris floods were modelled. Higher return periods were not considered because (a) the error associated with professional judgment becomes too large for the estimate to be credible and (b) a return period of, say 10,000 years, would need to consider extremely rare scenarios such as strong earthquakes occurring during periods of high antecedent moisture and wet weather, or an extremely rare rainstorm following a very wet months. Such combination of events could result in many simultaneous debris avalanches and landslide dams. Predicting the water and sediment discharge of multiple sequential dam breaches cannot be accomplished with any degree of confidence at this time. The main features observed from these model results can be summarized as follows: • The 100-year event will likely not overwhelm the intake of the culvert at Evergreen basin. • The 200-, 500- and 2,500-year return period events would destroy the Baden Powell Footbridge. • The 200-, 500- and 2,500-year return period events are likely to stay within the confined floodplain of Mosquito Creek until reaching Forest Hills Drive. Therefore, the only damage in this area could be caused through bank erosion that cannot be modelled

123

Nat Hazards (2013) 65:1653–1681

1669

reliably. The highest potential for bank erosion is given along confined sections with high flow velocities (between 5 and 9 m/s) such as the upper Skyline Drive between Chelsea Crescent and Handsworth Road on the west side of Mosquito Creek and Glencanyon Drive in the area delineated by the two intersections with Dolores Drive. In this area, as much as 30 houses could be affected. • The 200-, 500- and 2,500-year return period events are likely to avulse below Evergreen Basin and flood the area west of Del Rio Drive and Fairmont Road north of Queens Road West. This area is located within the Mosquito Creek valley that has been culverted. Flow depth in this area is estimated as between a few centimetres and 1 metre, and flow velocity between 1 and 4 m/s. • Water is expected to flow the fastest in the lowest and most confined sections and down Del Rio Drive. • Damage to the homes will depend to a large degree on the amount of organic and mineral debris that will be carried with the flow at this stage. This will depend on how much material is being contained in Evergreen Basin or stored in log jams that are likely to form in the channel of Mosquito Creek. At a minimum, one can expect significant nuisance damage due to flooding of basements and gardens. Hydrostatic pressures are likely to break basement windows facing the flow, which would result in water entering basements and cellars. 2.2.5 Discussion While the defined return periods of Table 7 suggest precision, it is not possible to assign precise return periods because the methods used do not allow such precision. The 2,500year return period event is associated with significant uncertainty. The only approach possible to estimate sediment volumes associated with such an event would be to drill into the Mosquito Creek delta at several locations or trench the Mosquito Creek fan. The latter cannot be completed due to the existing dense development on the fan. In the former case, the sediment cores obtained would likely show coarse-grained layers indicative of debris floods. Organic sediment within, below or above these debris flood layers could be used to date the events. Measurement of the deposit thickness and its areal extent would yield approximate bedload volumes, but would still not allow an estimate of peak discharge. Because of the inherent uncertainties in determining peak flows, and to avoid the allusion of precision, it is more appropriate to be viewed as ranges that likely correspond to the reported peak discharge value (second column of Table 7). For simplicity, however, in the following risk assessment, we refer to the single return periods reported in the first column of Tables 2 and 7. 3 Risk assessment 3.1 Methods Risk to individuals and groups were quantified based on the methodology outlined below. To improve the defensibility of the quantitative risk assessment (QRA), the results were then compared to an empirical method of Jonkman et al. (2008) relating flood depth and mortality, and to case studies at other locations. We report the results as Probability of Death of an Individual (PDI) (individual risk), on F–N curves showing the expected frequency of N or more fatalities (societal risk), and as

123

1670

Nat Hazards (2013) 65:1653–1681

mortality rates based on probability of life loss (PLL) estimates for comparison to other case studies. We compare the results to risk tolerance thresholds established on an interim basis by the DNV. 3.1.1 Elements at risk The elements at risk considered in this study include the people residing in homes that are potentially impacted by debris floods. People outside of their homes or recreational users of the trails paralleling Mosquito Creek are excluded because such numbers are unknown, and it is assumed that no evacuation occurs prior to the hazard event. We used data provided by the DNV for building locations and occupancy, including residential buildings and the fire hall. Buildings considered include those within the footprint area inundated by modelled debris floods, where modelled flow depth exceeded 0.1 m and modelled flow velocity exceeded 0.1 m/s. It was assumed that flow depths and flow velocity lower than those values do not constitute a credible risk to loss-of-life. Residential building occupancy figures provided by the DNV are based on 2006 Canadian Census data and average per-unit occupancy data for multi-family dwellings (e.g. condominiums) from Metro Vancouver. Average detached home occupancy is calculated by first subtracting estimated populations for any multi-family dwellings, and then dividing the remainder by the number of detached homes within the census block. BGC understands that these figures are approximate but are considered reasonable for areas that have not undergone significant re-development since the 2006 Census as is the case for the Mosquito Creek study area. For the fire hall, we assume that 4 firemen are present, working 12 h shifts, 4 days per week including holidays. 3.2 Risk analysis For an example, event tree showing the steps involved in the debris flood risk assessment for the 500-year debris flood is shown in Fig. 10. Risk was determined for individuals and groups based on the following equation (AGS 2007): R ¼ PA  PS:H  PT:H  E  V

ð1Þ

where PA, is the annual hazard probability of a debris flow or debris flood; PS:H, is the spatial probability of direct building impact (S stands for spatial, H for hazard); PT:H, is the temporal probability of impact (T stands for temporal); E, is the number of lives potentially at risk in the building; V is the likelihood of loss-of-life, should the building be directly impacted. For individuals, risk is reported as the annual Probability of Death of an Individual (PDI), with E = 1 and values for PT:S assigned for the individual most at risk within each building. For groups, risk is reported as the expected frequency (F) of N or more fatalities, as shown on an F–N curve. The y axis on the F–N curve, F, is calculated as: F ¼ Rf

ð2Þ

where f is the product of probabilities along each branch of the event tree shown in Fig. 10. The x axis, N, is calculated as: N ¼V E

ð3Þ

where V is the probability of death in case of impact and E is the number of elements at risk.

123

Nat Hazards (2013) 65:1653–1681

1671

Fig. 10 Example of event tree for risk analysis (500-year return period shown completely)

Probability of life loss (PLL) estimates was used to estimate mortality for an event of a particular likelihood, where mortality is defined as the proportion of life loss for the exposed population. PLL is calculated as: PLL ¼ Rfi Ni

ð4Þ

where f and N are defined as per Eqs. 2 and 3, and i is the hazard scenario considered (e.g. QDF200, QDF500, QDF2500). It is important to note that PLL is not equivalent to the Probability of Death of an Individual (PDI). The PLL considers the overall likelihood of life loss, whereas PDI estimates the likelihood that a particular life will be lost.Assuming no evacuation occurs, mortality and PLL are related as: Mortality ¼ PLL=PA

ð5Þ

where PA is the event probability, as per Eq. 1. Given the occurrence of an event, PLL can be used according to Eq. 4 to estimate mortality. This was done for each building within the study area, and results summed to obtain estimates of mortality for the various event scenarios. 3.2.1 Estimation of PA Values of PA used in the risk analysis correspond to the 100-, 200-, 500- and 2,500-year return period categories modelled in the hazard analysis. 3.2.2 Estimation of PS:H PS:H provides an estimate of the probability that the element at risk (homes) are impacted by the specific return period event, given that it occurs. It ranges from 0 (no chance of building impact) to 1 (100 % chance of building impact). The value of PS:H was assumed

123

1672

Nat Hazards (2013) 65:1653–1681

to be 0.9 for all buildings intersecting modelled flows, corresponding to a high likelihood of impact during an event. Absolute certainty of impact (PS:H = 1) cannot be guaranteed due to the inherent uncertainties associated with numerical modelling, including the interaction of buildings with the flow and the potential for unpredictable avulsions, log jams, bank erosion or slumps. 3.2.3 Estimation of PT:S PT:S provides an estimate of the temporal probability that the element at is present when the hazard occurs, corresponding to the proportion of time spent by persons in a building. For residential buildings, an average value of 0.5 was assigned for groups implying that half of the population in the consultation zone is present at any given time. This value also implies that there is no warning available for a debris flood. A more conservative value of 0.9 was used for the estimation of the individual most at risk, used for estimates of individual risk. This corresponds to the person spending the most time (21–22 h per day) at home, such as a young child, stay-at-home parent, or an elderly person. For the fire hall, PT:S was assigned an average value of 0.7 for groups, corresponding to 17 h/day when workers are present. For risk to individuals, PT:S was assigned an average value of 0.3, corresponding to 12-h work shifts, 4 days per week (not including holidays). 3.2.4 Estimation of vulnerability Vulnerability is defined as the likelihood that a particular person will perish given impact by the hazard. For example, a vulnerability of 1 implies absolute certainty that the person will die, while a vulnerability of 0.1 implies a 10 % chance of death given the hazard impact for the respective annual probability event. Mortality is defined as the rate of death for persons exposed to the hazard when it occurs, where ‘‘exposed’’ is defined as the net population after evacuation (if any). If the entire population is present and certain to be impacted, mortality and vulnerability metrics are numerically equivalent. Estimates of vulnerability strongly affect QRA results and are also subject to a large degree of uncertainty because factors controlling vulnerability are poorly known. These factors include the presence of large organic debris increasing destructive potential of the flow, the level of building damage, the nature of debris entering the building, the location of individuals inside the building and human factors such as age, infirmity and decisions to enter hazardous locations to rescue items of value. The criteria used in this study are shown in Table 8 and were assigned to persons within individual buildings. To address some of the uncertainties noted above, we compared the vulnerability criteria to relations between debris flood intensity and recorded building damage, and to specific case studies. For building damage, we compared the vulnerability criteria to empirical results of Jakob et al. (2011), who quantified levels of building damage for 68 published global debris flow events in terms of a debris flow intensity index (IDF), calculated as maximum flow depth multiplied by the square of the maximum flow velocity. This comparison is shown in Table 8 for four building damage categories (Table 9). We then estimated potential life loss at Mosquito Creek independently from the QRA, based on mortality rates for other cases studies. These were compared for consistency with the QRA results, considering similarities and differences between the case studies.

123

Nat Hazards (2013) 65:1653–1681

1673

Table 8 Matrix to compute vulnerability of loss-of-life for debris floods Variable

Maximum flow depth (m) \0.3

0.3–1

1–1.5

1.5–2

2–3

3–4

\0.001

0.001–0.01

0.01–0.03

0.03–0.05

0.05–0.08

0.08–0.15

0.001

0.006

0.02

0.04

0.065

0.12

I

II

II

II

II

II

Maximum flow velocity (m/s) \2

2–5

0.001–0.01

0.01–0.03

0.03–0.05

0.05–0.08

0.08–0.15

0.15–0.30

0.006

0.02

0.04

0.065

0.12

0.23

II

III

III

III

III–IV

IV

The top value is the estimated range, values in bold are the mean, and Roman numerals correspond to the building damage classes shown in Table 9

Table 9 Building damage classes of Jakob et al. (2011)

Building damage I

Some sedimentation

II

Some structural damage

III

Major structural damage

IV

Complete destruction

3.2.5 Mortality and flood depth An empirical relationship was developed by Jonkman et al. (2008) that correlates mortality rates with flood depth at previous major flood events. Data plotted on this figure are classified whether warning was given prior to the event. The most conservative cases, corresponding to events with no warning, were used to plot an approximate relationship for debris floods at Mosquito Creek (shown as black diamonds on Fig. 11) for comparison to mortality estimates based on PLL. The number of deaths at each water depth is estimated by multiplying the mortality rate by the number of persons occupying buildings impacted at a particular modelled water depth, for each event scenario. 3.3 Results 3.3.1 Risk estimation for individuals (PDI) Risk to individuals (PDI) was quantified according to Eq. 1 and for people living in each home. Risk is reported as the sum of risk for the 200-, 500-, and 2,500-year debris flood scenarios. The 100-year return period scenario was not considered, because no buildings were impacted by flow depths exceeding 0.1 m for this scenario. Results are shown in Table 10. Minimum PDI values are based on the lower end of vulnerability ranges in Table 8, and the lower bounds of the probability ranges. Maximum PDI values are based on the higher end of vulnerability ranges in Table 8 and the higher bounds of the probability ranges. Mean values are the average of the minimum and maximum PDI estimates and are considered a ‘‘best estimate’’.

123

1674

Nat Hazards (2013) 65:1653–1681 1.E-01

Frequency of N or more fatalities (F)

1.E-02

UNACCEPTABLE

1.E-03

1.E-04

ALARP 1.E-05

ACCEPTABLE 1.E-06

1.E-07 0.001

0.01

0.1

1

10

100

1000

N (Number of Fatalities) Fig. 11 Relationship between mortality and water depth for different warning times (after Jonkman et al. 2008). The dashed line was hand-drawn as a reasonable conservative approximation to account for that fact that there would likely be no warning

3.3.2 Group risk Figure 12 shows group risk summarized on an F–N graph, based on Eq. 1. The bold line represents the mean (‘‘best estimate’’), and the upper and lower curves (grey lines) are based on the upper and lower estimates for event probability and vulnerability (Table 8). According to Fig. 12, debris flood risk for loss-of-life of groups is unacceptable as measured by DNV’s risk tolerance criteria; the highest risk return period corresponds to the 200-year debris flood because its minimum and maximum value range plots the furthest from the line that separates the ‘‘unacceptable’’ region from ‘‘ALARP’’; the 100-year event or more frequent events (not shown) will plot either in the ‘‘ALARP’’ or in the ‘‘acceptable’’ region; and higher return period (less frequent) debris floods are likely to plot in the ‘‘ALARP’’ region. Table 10 Summary of minimum and maximum PDI values

No. of buildings with PDI [ 10-4 Minimum PDI

1

Maximum PDI

10

30

5

30

Best estimate PDI

123

No. of buildings with PDI [ 10-5 10

Nat Hazards (2013) 65:1653–1681

1675

Fig. 12 Group risk (F–N curve) for the affected population along Mosquito Creek within the District of North Vancouver. The three plotted points present the risks calculated for the 200-, 500- and 2500-year return period debris floods. The black line indicates the mean (‘‘best’’) estimate, bracketed by the minimum and maximum estimates in grey. ALARP stands for as low as reasonably practicable. Notes: Warning levels are based on Tsuchiya and Yasuda (1980). The data on the graph are from Jonkman et al. 2008. The black diamond indicated no warning given. The dashed line is drawn by BGC and symbolizes a debris flood scenario with no warning.

Table 11 Mortality estimates based on PLL estimates

Event (year)

Mortality estimate

100

0

200

0.2–0.5

500

0.3–1

2,500

1–2

3.3.3 Probability of life loss (PLL) Mortality estimates based on PLL estimates for each return period are shown in Table 11. According to Table 11, no fatalities are estimated for a 100-year event, and 1–2 fatalities are estimated for a 2,500-year event. Fractional results shown for 200- and 500-year can be interpreted as 0–1 fatalities estimated for these return periods. 3.3.4 Mortality case studies 3.3.4.1 October 1921 debris flood, Britannia Beach, BC On October 28, 1921, after a day of heavy rain, a massive flood destroyed much of that portion of the community and mine operations that existed on the lower beach area. 50 of 110 homes were destroyed and thirty-seven people lost their lives. Railway construction activities had dammed up a portion of the creek. When this dam collapsed the town below was flooded. This event has some similarities with a debris flood triggered by a potential landslide dam break at Mosquito Creek. In both cases, the channel gradient does not lend itself to debris flows but rather debris floods and significant amounts of organic and mineral debris would be entrained. Finally, there was little to no warning on October 28, 1921, and little or no warning can be expected on Mosquito Creek.

123

1676

Nat Hazards (2013) 65:1653–1681

BGC reviewed historical documents to estimate the flow velocities and flow depths associated with the Britannia Creek debris flood. Eye witness accounts talking about a ‘‘20 m high wave of water’’ are likely misinterpreted from ‘‘20 feet of water’’, since the imperial system prevailed in those days. Even 20 feet (*7 m) appears unlikely given the photographic evidence from the flood. The photographs suggest that an area alongside and south of the current creek was overwhelmed by debris and water with flow depth to perhaps 3 m near the fan apex and 1 m near the fan fringe. Because the loss of confinement on the fan decreased flow velocities, it is expected that velocities ranged between 4 m/s just downstream of the fan apex and perhaps 2 m/s at the fan margins. Of the 300 people living in the community on the Britannia Creek fan, 37 were killed, resulting in a mortality of 0.12 (12 %). For a single person, the chance of death was 37/300 = 0.124. Of the 300 people living on the fan, 15 suffered severe injuries (5 % injury rate). Per home destroyed, there was on average one (0.74) fatality, and 45 % of all buildings on the fan were destroyed. Using the estimated flow velocities (2–4 m/s) and flow depth (1–3 m), Table 8 contains vulnerability ratings that are reasonably similar to these data, from 0.03 to 0.15 with an average value of 0.1. At Mosquito Creek, 133 people are living in the area potentially subjected to debris floods. Applying the same mortality rate for the affected community (12 %), a debris flood on Mosquito Creek would result in about 16 fatalities for the areas affected by either the 200- or 500-year model events. However, this estimate is conservative given the minimal protection offered by small miners’ shacks without concrete foundations, compared to modern homes. This exemplifies that focus on only one case study may be misleading. The magnitude and intensity of the debris flood at Britannia Beach have also never been established, but photographic evidence suggests that the total volume of sediment deposited on the fan may have been in the order of 100,000 m3. 3.3.4.2 December 1981 debris flow, Charles Creek, BC On December 4, 1981, a 30,000–40,000 m3 debris flow travelled down Charles Creek, approximately 4 km north of Horseshoe Bay, following a period of heavy rain and snowmelt. Initial surges blocked a bridge under a residential road, resulting in further deposition upstream, blockage of the highway bridge and deposits of up to 6 m high on the surface of the highway. Two houses were inundated by water and gravel, although no structural damage occurred. Of the 40 residents who attempted to evacuate from the houses along Charles Creek, 1 woman was swept away by flood water. This corresponds to a 0.025 (2.5 %) mortality rate for this event. Because the flow depth and velocity at the location of death is unknown, it cannot be used to validate Table 8. Applying a similar mortality rate to Mosquito Creek would result in about 3 fatalities. This may be over-conservative when compared to Mosquito Creek for two reasons. First, in the case of Charles Creek, death occurred not due to initial debris flow impact but rather during evacuation following the event. Second, Charles Creek was a debris flow, not a debris flood, which implies higher flow velocities and accordingly higher impact forces. 3.3.4.3 July 1997 Hummingbird Creek debris flow, Salmon Arm, British Columbia On July 11, 1997, a large debris flow occurred at Hummingbird Creek on Mara Lake. A 25,000 m3 debris avalanche was initiated downstream of a forest road culvert that drained a small catchment. The debris avalanche evolved into a debris flow that reached between 600 and 1,000 m3/s and deposited 92,000 m3 of sediment on the fan (Jakob et al. 2000). There were no impact-related fatalities recorded, but one heart attack was apparently related to the trauma of seeing the debris flow.

123

Nat Hazards (2013) 65:1653–1681

1677

Deposition depths ranged between 3.5 and 1 m upstream of highway 97A and between 0.1 and 0.5 m downstream of the highway. Flow velocities upstream of the highway ranged downstream between 6 m/s and perhaps 12 m/s. Downstream of highway 97A flow velocities ranged between an estimated 1 and 3 m/s. Of the five cabins upstream of the highway, two were destroyed. There were no people present in these cabins at the time of impact. Lower Hummingbird Creek fan is largely settled with private residences, mostly for weekend use. The total number of cabins on the fan that were affected by the event is approximately 20. According to Table 8, the vulnerability range upstream of the highway ranged between 0.05 and 0.3. For the fan below the highway, vulnerabilities range between 0.001 and 0.03. Assuming a potential occupancy of two people per cabin, mortality for the upper fan would then range between 0.1 and 3. For the lower fan, mortality would range between 0.04 and 1.2. The fact that no one died through impact is clearly associated with the absence of many property owners at the time of impact, which underlines the necessity to include temporal probabilities in risk calculations. 3.3.4.4 June 2010 Testalinden Creek debris flow, Oliver, British Columbia On June 13, 2010, a debris flow was triggered by the overtopping and subsequent incision of an earthfill dam at Testalinden Lake. The debris flow destroyed five houses, severely damaged two and obliterated several orchards and vineyards as well as shed debris on a major highway. This event was highly publicized and photographed which allowed an estimation of flow depths which appeared to have ranged between 1 and 2 m at impact with homes. Eye witness accounts of 20-feet-high1 (7 m) flowing debris are believed to be exaggerated given the photographic evidence. Associated flow velocities are difficult to estimate. On film footage, mud lines were identified on the uphill-facing portions of buildings that were used to calculate flow velocities. Run-up against poles on the fan was estimated to be between 0.5 and 1.0 m (Jordan pers. comm. June 2011). Using an empirical formula [v = (2gh)-2] with g being the acceleration of gravity (9.8 m/s2) and h being the run-up, the range in velocities in mid and lower fan may range between 1 and 3 m/s. These values would result in mortalities between 0.03 and 0.08 (3–8 %) according to Table 8. Given that 7 homes were severely damaged and assuming an occupation of 4 persons per home, this would result in a statistical number of deaths of 0.8–2.2. However, given that the event occurred in the afternoon on a Sunday during summer, it is likely that only a portion of the residents were in their home. Estimating that perhaps only half of the residents were in their homes at the time of impact, the expected number of deaths drops to a range of 0.4–1.1. The fact that no one died is a fortunate circumstance, but given the approximate number of deaths expected to occur in such event (average of 1 person), the outcome is not unexpected. 3.3.4.5 August 2005 flooding, New Orleans, USA During landfall on August 29, 2005, hurricane Katrina caused massive flooding and devastation along a 270 km stretch of the US Gulf Coast. The storm surge caused overtopping and breaching of levees around New Orleans. An area of 260 km2 of the city flooded at some locations up to 4 m deep. It took over 40 days to dewater the city. Flow depths reached up to 3 m. Water rise rate over the first 1.5 m reached up to 50 m/hr or roughly one cm/min. The total death toll associated 1

Estimated by Roop Smagh (resident) to CBC news (Monday, June 15).

123

1678

Nat Hazards (2013) 65:1653–1681

with hurricane Katrina amounted to 1464 persons. Of the 746 fatalities that were recovered in their location of death, 54 % died in their residence, 20 % in medical facilities, 10 % in nursing homes and 7 % in the open. The typical causes of death were drowning or physical trauma due to debris impacts and collapsing buildings. Mortalities were calculated for various neighbourhoods in New Orleans that could reasonably be homogenized. Mortalities range between 0 and 0.15 (15 %). For the whole of New Orleans (including Orleans, St. Bernard and New Orleans East), a mortality of 1.2 % was calculated. For the Lower 9th Ward, which was one of the worst affected areas and suffered the direct impact of a wave due to dike breach, mortalities ranged between 0.03 (3 %) and 0.07 (7 %). Simple application of the mortality rate of 0.03–0.07 to Mosquito Creek would yield a total of 1–2 fatalities for either the 200- or 500-year debris flood scenarios since flow velocities and flow depth for the two events lie within this uncertainty range. The above case studies have yielded mortalities ranging over one order of magnitude from 0.012 (1.2 %) to 0.12 (12 %). Applied to Mosquito Creek, these mortalities would results in a number of expected deaths between 1.6 and 16. 3.3.5 Mortality estimates based on depth–mortality relations Table 12 summarizes mortality rates estimated for Mosquito Creek. Given the distribution of properties within areas of variable flow depths, Table 12 suggests that the number of statistical deaths from debris floods for all modelled return periods at Mosquito would range from 0.004 to 1.5 (rounded zero to two) and that the 100-year magnitude event is extremely unlikely to result in fatalities. That the 100-year event is unlikely to result in fatalities is logical given that such flows are unlikely to overtop the box culvert downstream of Evergreen Basin. 3.3.6 Discussion Table 13 summarizes mortality rates for the case studies examined. Applying a similar average mortality rate (3–4 %) to Mosquito Creek would yield approximately 7–15 statistical deaths for the QDF100 to the QDF2500. This estimate is higher than the estimated 0.004–1.5 (0–2) fatalities based on modelled flow depths and the empirical method of Jonkman et al. (2008). The difference may be due to comparatively low maximum flow depths (\1 m) at most buildings impacted within the study area. Damage to buildings will be minor and, therefore, the number of lives lost will likely be lower than that estimated from case studies. According to Table 10, approximately 5–10 properties exceed the risk tolerance threshold of 10-4 for PDI established by the DNV. According to Fig. 12, group risk for the study area overall exceeds tolerable thresholds. The QDF200 results in the highest absolute risk to loss-of-life, while the QDF100 results in the lowest absolute risk. Both the PDI and group risk demonstrate that risk will need to be reduced to render it at least tolerable. The above results are considered best estimates given the methods and data available, and allow an approximation of the possible range in fatalities using transparent and replicable methods. However, several uncertainties limit the accuracy of such estimates. These include (a) accuracy of the frequency–magnitude relationships; (b) adequacy of the hydraulic model and the associated input errors; (c) the exact nature of debris flood impact (water vs. boulders vs. organic debris); and (d) multiple factors affecting the vulnerability of persons following debris flood impact.

123

Nat Hazards (2013) 65:1653–1681

1679

Table 12 Mortality for different water depths at Mosquito Creek (based on Jonkman et al. 2008) Water depth (m)

Mortality rate

QDF100 No. of death

QDF200 No. of death

QDF500 No. of death

QDF2500 No. of death

0.1–0.3

0.001

0.004

0.05

0.04

0.04

0.3–0.5

0.003

0

0.1

0.1

0.8

0.5–1.0

0.008

0

0.1

0.3

0.3

1.0–1.5

0.02

0

0.1

0.1

0.4

0.004

0.3

0.5

1.5

Total

Water depths below 0.1 m were not considered, as such flow is considered exceedingly unlikely to result in fatalities

Table 13 Case study mortalities and theoretical number of deaths Location and year

Mortality (%)

N  (QDF100)

N  (QDF200)

N  (QDF500)

N  (QDF2500)

New Orleans (2005)

0.7–5.5

0–0.2

1–5

1–6

3–20

Britannia Beach (1921)

12

0–0.2

10

13

44

Hummingbird Creek (1997)

0

0

0

0

0

Testalinden Creek (2010)

0

0

0

0

0

Charles Creek (1981)

2.5

0.1

2

3

9

Mean (rounded)

3–4

0.1

2–3

2–4

7–15

Range

0–12

0–0.2

0–3

0–13

0–44

 

N stands for the theoretical number of deaths at Mosquito Creek

The estimates used in this paper could be improved by performing more detailed analysis on the ratio of people inside and outside of buildings to vehicles in the area, including analysis of risk to trail users along Mosquito Creek. Various methods exist to estimate the vulnerability of unprotected people on streets (e.g. McClelland and Bowles 2002). Such analyses would allow a more complete assessment of risk to life for all persons within the hazard area. However, if one were to include people on streets, in cars and on trails, the existing risk will necessarily increase since (a) the total number of people affected would increase and (b) people outside of houses and in the area affected will have higher vulnerabilities than those considered inside buildings. Because the risk to individuals and groups is already considered unacceptable, such study would only reemphasize this conclusion.

4 Conclusion In this paper, we quantify debris flood hazards and associated risks to people residing along Mosquito Creek. Landslide dam outbreak flood extent, depth and velocity are modelled at magnitudes corresponding to 100-, 200-, 500- and 2,500-year return period categories; although we prefer that these be interpreted as approximate ranges (100–300, 300–1,000 and 1,000–4,000 years). Risk to loss of life is then determined for individuals and groups exposed to debris flood hazard, and compared to risk tolerance standards accepted on an interim basis by DNV. Both individual and group risk estimates suggest that current risk levels are intolerable.

123

1680

Nat Hazards (2013) 65:1653–1681

This study outlines a systematic approach to identify requirements for risk reduction measures and to assess whether such measures can achieve a desired level of residual risk. The DNV has assumed a leadership role with respect to risk-based decision making for geohazard management, and the authors hope that this example will help promote the adoption of similar methods by other jurisdictions as well. The results from the risk to loss of life study have prompted DNV to implement a series of risk reduction measures including installation of a flexible debris containment net in 2012 and watershed restoration measures. Acknowledgments This contribution was made possible with the support of Jozsef Dioszeghy and Fiona Dercole of the District of North Vancouver who also encouraged publication of this work. A draft has been reviewed by Mike Porter.

References Association of Professional Engineers and Geoscientists of British Columbia (APEGBC) (2010) Guidelines for Legislated Landslide Assessments for Proposed Residential Developments in BC. Revised May 2010 Australian Geomechanics Society (AGS) (2007) A national landslide risk management framework for Australia. Aust Geomech 42(1):69 Blown I, Church M (1985) Catastrophic lake drainage within the Homathko River basin, British Columbia. Can Geotech J 22:551–563 Bovis MJ, Dagg BR (1992) Debris flow triggering by impulsive loading: mechanical modeling and case studies. Can Geotech J 29:345–352 Costa JE, Schuster RL (1988) The formation and failure of natural dams. Geol Soc Am Bull 100:1054–1068 FLO-2D Software Inc (2004) FLO-2D users manual version 2004.10, October 2004 Fread DL (1991) BREACH: an erosion model for earthen dam failures. NWS Report, National oceanic and atmospheric administration, Silver Spring, Maryland Froehlich DC (1995) Peak outflow from breached embankment dam. Journal of Water Resources Planning and Management. J Water Resour Plann Manag 121(1):90–97 Garcia-Martinez R, Lopez JL (2005) Debris flows of December 1999 in Venezuela. In: Jakob M, Hungr O (eds) Debris-flow hazards and related phenomena. Praxis and Springer, Heidelberg Guadagno FM, Revellino P (2005) Debris avalanches and debris flows of the campania region (southern Italy). In: Jakob M, Hungr O (eds) Debris-flow hazards and related phenomena. Praxis and Springer, Heidelberg Hungr O, Evans SG, Bovis MJ, Hutchinson MJ (2001) A review of the classification of landslides of the flow type. Environ Eng Geosci VII(3):221–238 Jakob M (2012) The fallacy of frequency: statistical techniques for debris-flow frequency—magnitude analyses. In: Eberhardt E, Froese C, Turner AK, Leroueil S (eds) Landslides and engineered slopes: protecting society through improved understanding. Taylor & Francis Group, London, pp 741–750 Jakob M, Friele P (2010) Frequency and magnitude of debris flows on Cheekye River, British Columbia. Geomorphology 114:382–395 Jakob M, Hungr O (eds) (2004) Debris flows and related phenomena. Praxis and Springer, Heidelberg Jakob M, Hungr O (2005) Debris-flow hazards and related phenomena. Praxis and Springer, Heidelberg Jakob M, Jordan P (2001) Design flood estimates in mountain stream—the need for a geomorphic approach. Can J Civ Eng 28:425–439 Jakob MJ, Hungr O, Anderson D, Fuller T, Ayotte D (2000) The July 7, 1997 debris flow at Hummingbird Creek. Can Geotech J 37(5):1109–1125 Jakob M, Stein D, Ulmi M (2011) Vulnerability of buildings to debris flows. Nat Hazards. doi: 10.1007/sl1069-011-0007-2 Jonkman S, Vrijling J, Vrouwenvelder ACWM (2008) Methods for the estimation of loss of life due to floods: a literature review and proposal for a new method. J Nat Hazards 46:353–389 Kerr Wood Leidal Associates Ltd (2003) Debris flow study and risk mitigation alternatives for Mosquito Creek. Report for the District of North Vancouver, December 2003 McClelland DM, Bowles DS (2002) Estimating life loss for dam safety risk assessment—a review and new approach. Report prepared for the Institute for Water Resources, U.S Army Corps of Engineers, U.S.

123

Nat Hazards (2013) 65:1653–1681

1681

Bureau of Reclamation, Australian National Committee on Large Dams, Department of Natural Resources and Environment, Victoria, Australia and Department of Natural Resources, Queensland, Australia by the Institute for Dam Safety Risk Management, Utah State University, IWR Report 02-R3, 420 pp Tsuchiya Y, Yasuda T (1980) High tide and life risk, refuge warning in relation to the Ise Bay Typhoon. J Nat Disaster Sci 2(2):27–60

123