Modelling hazardous surface hoar layers across western Canada with a coupled weather and snow cover model

Cold Regions Science and Technology 128 (2016) 22–31

Contents lists available at ScienceDirect

Cold Regions Science and Technology journal homepage: www.elsevier.com/locate/coldregions

Modelling hazardous surface hoar layers across western Canada with a coupled weather and snow cover model Simon Horton* , Bruce Jamieson Department of Civil Engineering, University of Calgary, Calgary, Alberta T2N 1N4, Canada

A R T I C L E

I N F O

Article history: Received 20 September 2015 Received in revised form 20 April 2016 Accepted 11 May 2016 Available online 14 May 2016 Keywords: Surface hoar Avalanche forecasting Persistent weak layer Numerical weather prediction SNOWPACK model

A B S T R A C T Destructive snow avalanches in western Canada are often caused by failure in buried surface hoar layers. Numerical snow cover models can simulate the formation and evolution of these layers, which could help avalanche forecasters assess the location and timing of avalanches. To investigate this application, we compared modelled surface hoar layers with snow and avalanche observations from the Coast, Columbia, and Rocky Mountains of western Canada. Surface hoar formation and evolution was modelled by forcing the snow cover model SNOWPACK with data from a high-resolution numerical weather prediction model. Surface hoar formation was verified with daily snow surface observations at 88 observation sites over two winters, and the evolution of buried layers was verified with avalanche observations and persistent weak layer assessments from 67 avalanche forecast regions. The frequency of surface hoar formation was overpredicted by 40%, although since modelled crystal sizes were moderately correlated with observed sizes, the more hazardous layers were often distinguished. A structural stability index in SNOWPACK often identified surface hoar layers during storm slab avalanche activity, but identifying layers during persistent slab avalanche activity was more difficult. Model limitations included uncertain meteorological inputs, errors in SNOWPACK’s snow surface energy balance, and representing the necessary spatial scales. Despite these limitations, the coupled model resolved differences between major Canadian mountain ranges and could improve avalanche forecasts in data-sparse regions. © 2016 Elsevier B.V. All rights reserved.

1. Introduction Failure in buried surface hoar layers causes many destructive snow slab avalanches in western Canada (Haegeli and McClung, 2007). Surface hoar crystals form when water vapour deposits on the snow surface. While formation is common on clear, calm, and humid nights (Colbeck, 1988), the distribution of surface hoar is affected by complex mountain weather and topography (Feick et al., 2007, Horton et al., 2015). Once buried, surface hoar crystals form thin layers prone to releasing avalanches (Jamieson and Johnston, 1992, Jamieson and Schweizer, 2000). Surface hoar layers release avalanches because of their unique structural properties. Schweizer and Jamieson (2007) found most skier-triggered avalanches released on layers with persistent grain forms, large grain sizes, and low hardness. Buried surface hoar layers often satisfy all three conditions, as crystals can reach several centimetres in length, much larger than the 1.3 mm threshold reported

* Corresponding author. E-mail address: [email protected] (S. Horton).

http://dx.doi.org/10.1016/j.coldregions.2016.05.002 0165-232X/© 2016 Elsevier B.V. All rights reserved.

by Schweizer and Jamieson (2007). Large grains persist in the snowpack for weeks to months and can form weak truss-like structures prone to fracture propagation (Jamieson and Schweizer, 2000, Lutz et al., 2007). Avalanche forecasters have used numerical models to simulate mountain snow covers since the 1990s. The French CROCUS model (Brun et al., 1992) and the Swiss SNOWPACK model (Lehning et al., 1999) use meteorological inputs to simulate the structural, thermal, and mechanical stratigraphy of the snow cover. Models predicting vapour deposition rates for surface hoar formation are implemented in both CROCUS and SNOWPACK (Hachikubo, 2001, Lehning et al., 2002). The SNOWPACK model calculates surface hoar size from the accumulated mass of deposited vapour. Stössel et al. (2010) and Horton et al. (2014) found modelled surface hoar sizes were moderately correlated with sizes observed at study plots. SNOWPACK also identifies potential avalanche failure layers with algorithms based on modelled grain size, layer hardness, and shear strength (Chalmers and Jamieson, 2003, Monti et al., 2014, Schweizer et al., 2006). Schirmer et al. (2010) and Monti et al. (2012) successfully identified critical surface hoar layers in simulated snow profiles with these algorithms.

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

Although snow cover models can simulate critical weak layers, avalanche forecasters are usually already familiar with snow conditions at weather stations where simulations are run. However, forcing snow cover models with data from numerical weather prediction (NWP) models could provide valuable information about snow conditions in remote areas (e.g. Storm, 2012). Bellaire et al. (2011, 2013) tested this method at the Mt. Fidelity snow study plot in western Canada, finding coupled weather and snow models approximated the overall snow stratigraphy, including critical surface hoar and melt-freeze crust layers. Further verifications highlighted the need to address compounding errors with coupled models, as NWP model errors affected modelled surface hoar layers (Bellaire and Jamieson, 2013, Horton et al., 2014). Recently, increasing horizontal resolution of NWP models has improved weather forecasts in complex terrain (Schirmer and Jamieson, 2015, Vionnet et al., 2014). For example, Schirmer and Jamieson (2015) found high-elevation snowfalls in western North America were modelled substantially better with a high-resolution NWP model (2.5 km) than with a regional-scale NWP model (15 km). Previous verifications of coupled weather and snow cover models in Canada used regional-scale NWP models, suggesting errors could be reduced by switching to higher resolution models. Furthermore, past verifications of critical layers focused on the Mt. Fidelity snow study plot, giving limited confidence in extrapolating the results to other mountain ranges. Surface hoar layers vary on a wide range of scales from individual slopes to entire mountain ranges (Schweizer and Kronholm, 2007). Field campaigns by Feick et al. (2007) and Borish (2014) observed spatial correlations on scale of several hundred metres (i.e. basin scale), later supported by spatial patterns modelled on a 30 m grid by Helbig and van Herwijnen (2012). Gridded NWP models do not resolve weather conditions at these scales, suggesting snow cover properties could only be resolved at similar scales to the NWP models. Horton et al. (2015) investigated spatial weather patterns modelled by a 2.5 km NWP model across a small mountainous region. They found air temperature, humidity, and longwave radiation patterns that affected surface hoar distributions were resolved over different elevation bands on a regional scale, but finer scales were not resolved (e.g. basin or slope scales), primarily because local wind patterns had a substantial effect on surface hoar formation. Accordingly, coupled NWP and snow cover models are most suitable

23

for regional-scale applications and should be further verified at those scales. In this study, we modelled surface hoar layers at various locations in western Canada by coupling a high-resolution NWP model with SNOWPACK. The formation and evolution of modelled surface hoar layers over two winters was verified with a large data set of snow observations, avalanche observations, and persistent weak layer assessments. This paper reports the benefits and limitations of modelling hazardous surface hoar layers at scales relevant for avalanche forecasting with coupled weather and snow cover models. 2. Verification data 2.1. Surface grain observations Snow grain observations were collected at weather observation sites operated by Canadian avalanche forecasting programs (e.g. ski resorts, backcountry skiing operations, parks, and transportation corridors). Observations were made at 14 sites in the Coast Mountains, 50 sites in the Columbia Mountains, and 24 sites in the Rocky Mountains (Fig. 1). Observations were collected for operational forecasting purposes and shared on the Canadian Avalanche Association’s Information Exchange (InfoEx) database (Haegeli et al., 2014). Most observation sites were flat sheltered fields chosen to represent conditions in nearby avalanche starting zones (Canadian Avalanche Association, 2014). The form and size of surface snow grains were observed daily following standard observation guidelines (Fierz et al., 2009). A total of 7147 surface grain observations were collected between 1 December and 31 March during the 2013–2014 and 2014–2015 winters. 2.2. Avalanche observations Avalanche observations were compiled from 15 avalanche forecasting programs in the Coast Mountains, 35 programs in the Columbia Mountains, and 17 programs in the Rocky Mountains (Fig. 1). The programs had defined regions that ranged in size from 2.5 to 8540 km2 , including ski resorts, backcountry skiing tenures, and national parks. Avalanches were observed by professionals travelling through their forecast regions and reported on the InfoEx following standard guidelines (Canadian Avalanche Association, 2014).

Fig. 1. Map of southwestern Canada with locations of mountain ranges, weather observation sites, automatic weather stations, and avalanche observation regions (ESRI basemap).

24

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

Avalanches observed in motion and evidence of recent avalanches were reported in terms of the number, size, trigger, elevation, slab depth, and character. Avalanche character describes eight common avalanche scenarios that affect risk-management decisions, such as “loose dry” or “persistent slab” avalanches (Klassen et al., 2013, Lazar et al., 2012). Characteristics of the failure layers were also reported, including the grain forms in the failure layer and bed surface. Over the 2013–2014 and 2014–2015 winters, a total of 2341 avalanches were reported with surface hoar in the failure layer or bed surface. Observations were only recorded when avalanche paths were visited during good visibility, making the data set relatively incomplete (e.g. Haegeli and McClung, 2003). Furthermore, many details were estimates rather than direct measurements from the avalanche starting zones. 2.3. Persistent weak layer assessments The avalanche forecasting programs also tracked persistent weak layers as part of their subjective hazard assessments (Canadian Avalanche Association, 2016). Persistent weak layers in each forecast region were identified by a burial date and grain form (e.g. “Feb 10 surface hoar”). Forecasters updated the depth of each layer daily, and classified it as either developing when a layer was buried but not yet releasing slab avalanches, active when a layer was known or suspected of releasing avalanches, dormant when no recent avalanches had occurred on a layer but reactivation was considered possible, and inactive when avalanches were not expected on a layer. A total of 7976 daily assessments for 225 different surface hoar layers were reported on the InfoEx over the two winters. We suspect the number of layers was under-reported because the layer tracking feature was relatively new to the InfoEx. For example, 47% of the surface hoar avalanches occurred on days when no surface hoar layers were reported in the corresponding region. Therefore, neither avalanche observations nor persistent weak layer assessments completely characterized the true avalanche hazard. 3. Methods Surface hoar layers were modelled by forcing the snow cover model SNOWPACK with weather model data. Following Schirmer and Jamieson (2015) and Horton et al. (2015), weather data was compiled from the Canadian High-Resolution Deterministic Prediction System (HRDPS). The HRDPS, also known as GEM-LAM, is an operational NWP model with 2.5 km horizontal grid spacing (Milbrandt, 2014). Air temperature, relative humidity, wind, incoming shortwave and longwave radiation, and precipitation were taken from the lowest available model levels from the 06:00 and 18:00 model initiations and compiled into a continuous time series of weather data. Surface hoar formation was evaluated by comparing modelled surface grains with surface grain observations, while surface hoar evolution was evaluated by comparing buried modelled layers with avalanche observations and persistent weak layer assessments. Different model configurations were used for each data set, as the surface observations were made at specific locations and the avalanche observations and assessments were made across broad regions. The model configurations were chosen to suit the observation data, and were not necessarily the optimal configurations for spatial forecasting applications. 3.1. Surface hoar formation Since surface grain observations were made at specific locations, the weather model data was downscaled to match the observation sites. Weather data were taken from the NWP model grid point with the closest elevation out of the four nearest grid points. Temperature,

humidity, and precipitation were corrected for elevation differences using lapse rates for the month of year following Liston and Elder (2006). Lapse rate corrections were not applied for wind speed or incoming radiation, as they strongly depend on the surrounding topography, which was unknown for most of the observation sites. The site elevations ranged from 1025 to 2350 m above sea level. Grid point elevations were similar to observation site elevations, with over half of the grid points within 25 vertical metres of the observation sites. The most extreme elevation differences were grid points 310 m below and 334 m above the corresponding observation sites. SNOWPACK simulates surface hoar formation by modelling water vapour deposition on the snow surface over time (Lehning et al., 2002). Deposition can occur when the surface temperature is subzero and no precipitation occurs. The vapour deposition rate is calculated from the air–surface vapour pressure gradient using an aerodynamic bulk method (Föhn, 2001). A bulk transfer coefficient is iteratively calculated from the wind speed, the air–surface temperature difference, and the aerodynamic roughness length (Lehning et al., 2000). Since the NWP model does not forecast snow temperatures, snow surface temperature was modelled with SNOWPACK using Neumann boundary conditions (Bartelt and Lehning, 2002). SNOWPACK calculated surface hoar size by assuming a linear relationship between the accumulated mass of deposited vapour and crystal size (Horton et al., 2014, Stössel et al., 2010). Contingency tables were produced to compare modelled and observed grain types at the observation sites. Modelled and observed surface hoar sizes were also compared, and a contingency table for detecting large surface hoar was produced. Since large surface hoar can be more hazardous (Horton et al., 2014), we classified large surface hoar as crystals greater or equal to 4 mm. The threshold was based on the size reported by Conlan (2015) for deep persistent slab avalanche releases on surface hoar. The threshold was larger than grain size thresholds from other studies (e.g. 1.3 mm by Schweizer and Jamieson, 2007), but surface hoar crystals are often larger when observed on the snow surface than when observed in snow profiles. Surface hoar formation was also modelled at six additional observation sites equipped with automatic weather stations (Fig. 1). Surface grains were intermittently observed at these sites, with a total of 20 surface hoar observations over three winters at Mt. Fidelity and Mt. St. Anne (2012–2015) and 5 observations at the other four sites during 2014–2015. To investigate model errors, surface hoar formation was modelled at these sites with both station measurements and downscaled NWP model data. 3.2. Buried surface hoar layers Avalanche observations were made throughout broad forecast regions and persistent weak layer assessments summarized conditions in those regions. To verify this regional-scale data, a single SNOWPACK simulation was produced for each region to represent typical snow conditions. While surface hoar layers can exist at any elevation, they are most prevalent around treeline elevations. For example, over half of the avalanches observed over two winters occurred between 1900 and 2200 m above sea level. Lower elevations are often below the freezing level and have limited avalanche terrain, while higher elevations are exposed to more wind and solar radiation which limits surface hoar formation (Horton et al., 2015). To simulate treeline conditions, HRDPS grid points at treeline elevations within the boundaries of each region were selected, and the weather data were averaged. Simulated snow profiles were produced with the averaged treeline weather data from each forecast region. While this approach removed spatial details from the NWP model, it was a simple way to summarize regional-scale conditions and verify with regional-scale data. Surface hoar layers were identified in each simulated snow profile and tracked from initial burial until SNOWPACK removed the layers

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

25

from the simulated profile. In SNOWPACK, surface hoar layers eventually merge with other layers when they become negligibly thin (¡1.2 mm), exceed a density of 300 kg m −3 , or melt. To preserve critical structural properties, SNOWPACK reduces the thickness of buried surface hoar layers over time with an empirical equation rather than a purely microstructure-based model. The equation forces thin layers (with small crystals) to disappear faster than thick layers (with large crystals). The daily change in layer thickness DT (in mm per day) is given by

layer interface, and DR is the difference in a modelled hand hardness index at a layer interface. SNOWPACK calculates SSI at every layer interface in the simulated profile and identifies a single critical weak interface where the SSI value is lowest. We classified any surface hoar layer adjacent to a critical weak interface as a critical surface hoar layer. Trends in the timing, age, and depth of critical surface hoar layers were compared to trends in observed avalanches and persistent weak layer assessments.

⎧ ⎪ ⎪ ⎨−0.1107 DT = −0.391 ⎪ ⎪ ⎩−0.0807

4. Results and discussion if

E ≤ 6mm

if

E > 6mm

and

t < 21days

if

E > 6mm

and

t ≥ 21days

(1)

where E is the modelled surface hoar size and t is age of the layer in days since burial. A new equation was implemented for Canadian snow conditions, where surface hoar layers often persist longer than in the European Alps (Schweizer and Jamieson, 2001). The empirical equation was changed to  DT =

−0.1

if E ≤ 4mm

−0.2

if E > 4mm

(2)

to match trends observed in 128 different surface hoar layers at Canadian snow study plots (a similar data set to Horton et al. (2014)). The revised equation allowed modelled surface hoar layers to persist for comparable periods of time to the 128 layers observed at study plots. In addition to tracking the age and depth of modelled surface hoar layers, a structural stability index (SSI) in SNOWPACK was used to identify critical weak layers in each profile. The SSI proposed by Schweizer et al. (2006) combines a skier stability index (SK38 ) based on mechanical properties of snow layers (Jamieson and Johnston, 1998) with a threshold sums index (D) based on the structural properties at each layer interface SSI = SK38 + D

(3)

with SK38 =

t txz + Dtxz

(4)

and ⎧ ⎪ ⎪ ⎨0 D= 1 ⎪ ⎪ ⎩2

if

DE ≥ 0.5mm

if

DE < 0.5mm or

if

DE < 0.5mm and DR < 1.5

and DR ≥ 1.5 (5)

DR < 1.5

where t is the modelled layer shear strength, txz is the modelled shear stress due to the overlying slab, Dtxz is the additional shear stress from a skier, DE is the difference in modelled grain size at a

Modelling hazardous surface hoar layers requires simulating their formation on the snow surface and their evolution in the snow cover. This section begins by evaluating surface hoar formation (Section 4.1), then presents observed and modelled trends for buried surface hoar layers (Sections 4.2 and 4.3), and finishes by discussing the implications for spatial modelling (Section 4.4). 4.1. Layer formation Precipitation particles were the most common surface grains in western Canada. Over two winters, precipitation particles accounted for 47% of all modelled surface grains and 62% of all observed surface grains at weather observation sites (Table 1). Surface hoar was the second most common surface grain, accounting for 31% of all modelled grains and 22% of all observed grains. The model chain over-predicted the frequency of surface hoar formation by 40%. Surface hoar was frequently modelled when precipitation particles were observed, suggesting the HRDPS weather model under-predicted of frequency of precipitation events. Schirmer and Jamieson (2015) found the HRDPS underestimated both the frequency and intensity of winter precipitation events in western Canada. Precipitation errors affected the length of fair weather periods when surface hoar could form (Horton et al., 2014), and with fewer precipitation events, SNOWPACK predicted too much surface hoar. Modelling the formation of large surface hoar is more relevant for avalanche forecasting, as large crystals are more likely to become avalanche failure layers. By assigning a size of 0 mm to grains other than surface hoar, the Spearman rank correlation coefficient between modelled and observed surface hoar sizes was 0.50 (p < 0.001). Although modelled and observed sizes were moderately correlated, modelled sizes were often too small when large (≥4 mm) surface hoar formed (Fig. 2a). When large surface hoar was modelled, the modelled sizes were often too large (Fig. 2b). In either case, the model had difficulty modelling the correct sizes for large surface hoar. Neglecting the exact size and focusing on whether the surface hoar was large enough to become a persistent weak layer, the contingency table for modelling large surface hoar again shows an over-prediction bias (Table 2). The frequency of large modelled surface hoar was higher than the frequency of large observed surface hoar. This resulted in 61% of the observed events to be modelled (hit rate) and 39% of the events to be missed (non-detection). Since large surface hoar crystals were often modelled when small or no surface hoar was observed, the false alarm rate was relatively high (54%). The

Table 1 Contingency table for observed and modelled surface grains at weather observation sites. Observed

Modelled

Surface hoar (SH) Precipitation particles (PP) Decomposing fragments (DF) Rounded grains (RG) Faceted crystals (FC) Melt forms (MF) Total

SH

PP

DF

RG

FC

MF

Total

1045 254 150 39 0 84 1572

893 2758 282 70 0 426 4429

176 251 104 54 0 69 654

24 17 2 1 0 6 50

49 27 22 13 0 8 119

48 55 9 14 0 197 323

2235 3362 569 191 0 790 7147

26

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

40

(a) Observed size (mm)

Modelled size (mm)

40 30 20 10 0

(b)

30 20 10 0

0

5

10

15

20

Observed size (mm)

0

5

10

15

20

Modelled size (mm)

Fig. 2. Comparison of observed and modelled surface hoar sizes at weather observation sites for (a) all cases when surface hoar was observed and (b) all cases surface hoar was modelled. Misclassified grains are assigned a size of 0mm. Sizes are grouped into 1mm wide bins, with the median value for each bin shown with a black line, the interquartile range with boxes, values within 1.5 times the interquartile range with whiskers, and outliers with dots. Box widths are proportional to the square root of the number of observations in each bin. The dashed line shows a one-to-one relationship.

high false alarm rate explains why many of the modelled crystals in Fig. 2b were larger than the observed crystals (i.e. because some of the crystals may not have formed at all). Modelled surface grains depended on the meteorological inputs provided by the NWP model and on SNOWPACK’s surface energy balance. As suggested previously, the NWP model underestimated the number of precipitation events, which allowed more time for surface hoar to form. The size of the surface hoar in SNOWPACK depends on modelled vapour deposition rates, which are sensitive to meteorological inputs such as longwave radiation, air temperature, humidity, and wind speed (Horton et al., 2015, Slaughter, 2010). Errors in any of these inputs could affect surface hoar formation. For example, one of the layers observed at Bow Summit was not modelled with downscaled NWP model data because the relative humidity was substantially under-predicted. Wind speed had complex effects on modelling surface grains at specific sites, as local winds are hard to forecast with NWP models and SNOWPACK has difficulty modelling the effects of wind on vapour deposition rates. Even with station measurements as inputs, 6 of the 25 layers observed at automated weather stations were not modelled. All six misses occurred at Mt. St. Anne and Marmot Basin, which were the two windiest stations. The average wind speed when surface hoar formed at these sites was greater than 1 m s −1 . The SNOWPACK model tends to reduce modelled vapour deposition rates at higher wind speeds (Horton et al., 2015), but strong winds did not inhibit surface hoar formation in these cases. To complicate wind speed effects, these six layers were all modelled with NWP model inputs because the wind speeds were under-predicted by the NWP model. Wind errors likely contributed to many of the errors in surface hoar size at the observation sites (Fig. 2). SNOWPACK’s surface energy balance model can be evaluated by comparing modelled surface temperatures with infrared surface temperature measurements (e.g. Fierz et al., 2003). When using weather model data, SNOWPACK predicts snow surface temperatures by modelling heat exchanges at the surface (Neumann boundary conditions). This approach causes errors that affect modelled Table 2 Contingency table for observed and modelled large surface hoar formation (≥ 4mm) at weather observation sites. Observed Large surface hoar Modelled Large surface hoar 512 Other surface grain 331 Total 843

Other surface grain

Total

606 5698 6304

1118 6029 7147

vapour deposition rates. For example, when modelling the surface temperature at six automated weather stations with actual station measurements as inputs, the standard error for surface temperature was 2.4 ◦ C with a cold bias of −0.7 ◦ C. When switching to NWP model inputs, the standard error increased to 4.3 ◦ C and the bias increased to −1.4 ◦ C. Various atmospheric stability parameters and aerodynamic roughness lengths were tested in SNOWPACK to reduce these errors, but gave similar or results to the default settings. Cold surface temperature biases likely increased the modelled air–surface vapour pressure gradient, which in turn contributed to larger vapour deposition rates and over-prediction of surface hoar size. However, surface temperature errors can affect surface hoar formation in many ways. For example, two of the layers observed at Mt. Fidelity in late March were not modelled with HRDPS inputs because the modelled surface temperature reached 0 ◦ C and melted the surface hoar, while measured surface temperatures were actually sub-zero. 4.2. Observed layers Several widespread avalanche cycles were observed on buried surface hoar layers throughout the 67 avalanche forecast regions (Fig. 3a). Weather site observations suggest widespread layers formed in January 2014, December 2014, and January 2015. Avalanches on these layers were observed throughout western Canada between 11 February and 17 March 2014 (35 days), 21 December 2014 to 9 January 2015 (20 days), and 16 January to 9 February 2015 (25 days), respectively. Avalanches in individual forecast regions were more intermittent, with shorter cycles of activity (Fig. 4). Some surface hoar layers only released avalanches during the storms that initially buried them (i.e. storm slab avalanches), while other layers also released deeper avalanches long after burial (i.e. persistent slab avalanches). Forty nine percent of all surface hoar avalanches in western Canada were classified as storm slab character, while 40% were classified as persistent slab character (Table 3). Storm slab avalanches had a median slab depth of 40 cm, while persistent slab avalanches had an median slab depth of 60 cm. Haegeli and McClung (2003) found most avalanches on buried surface hoar released during one to three distinct avalanche cycles over a period of three to four weeks, consistent with these findings. Active surface hoar layers were reported during the many of the observed avalanche cycles (Fig. 3b). However, active layers were reported for longer periods of time and to greater depths than the observed avalanches. For example, few avalanches were observed in late March 2014 and March 2015, although a substantial number of active layers were reported over this period. Also, the median slab depth of active layers was greater than the median slab depth of

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

27

(a) Surface hoar avalanches > 100 cm 50 - 100 cm 20 - 50 cm < 20 cm

Count

60 40 20 0

(b) Active reported layers Count

30 20 10 0

(c) Critical modelled layers Count

60 40 20 0

Dec

Jan

Feb

Mar

2013-14

Apr Dec

Jan

Feb

Mar

Apr

2014-15

Fig. 3. Number of (a) surface hoar avalanches observed, (b) surface hoar layers reported as active in persistent weak layer assessments, and (c) surface hoar layers modelled as critical by the structural stability index. Totals are pooled from 67 forecast regions in western Canada. Avalanches are colour-coded by slab depth and reported and modelled layers are colour-coded by layer depth. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

surface hoar avalanches (60 versus 45 cm, respectively). Forecasters may classify layers as active following periods of avalanche activity to remain cautious of potentially hazardous releases (e.g. Klassen, 2010), or maybe because buried surface hoar layers can show signs of instability in snowpack tests for at least six weeks (Horton et al., 2014). The status of surface hoar layers reported in persistent weak layer assessments changed with the depth and age of the layers (Table 4). Most layers were initially classified as developing for the first 10 to 20 days after burial until they were roughly 40 cm deep, and then were classified as either active or inactive. Only 68% of the reported surface hoar layers were ever classified as active (152 of 225), as the remaining layers likely formed strong bonds before cohesive slabs developed above them. Haegeli and McClung (2007) also reported a substantial number of surface hoar layers that did not release avalanches during other winters in western Canada. Active layers were reported for 10 to 34 days after burial and to depths between 40 and 90 cm (interquartile ranges), after which they were usually classified as dormant or inactive for the rest of the winter. While the status of surface hoar layers changed with depth and age, the likelihood of triggering specific layers likely depended on the progression of weather events (e.g. storms, temperatures, and wind events).

4.3. Modelled layers Substantially more layers were modelled in the 67 forecast regions than reported in persistent weak layer assessments. A total of 152 different surface hoar layers were reported as active over two winters, while 473 different modelled layers were identified as critical in the same 67 regions. While we suspect the assessments under-reported the number of layers, additional modelled layers likely arose from over-predicting surface hoar formation. Avalanche observations and layer assessments and were too inconsistent to

quantitatively verify buried surface hoar layers in individual regions, so results from the 67 avalanche forecast regions were aggregated to show general trends (Fig. 3). Critical surface hoar layers were typically identified at modelled depths between 18 and 50 cm and between 6 and 19 days after burial (interquartile ranges). Active surface hoar layers were reported to greater depths and longer after burial (Table 4), suggesting modelled layers were not critical for long enough periods. Layers with large initial crystal sizes persisted longer according to Eq. 2, and therefore were more likely to become critical. As a result, the agreement between modelled and observed layers largely depended on whether layer formation was modelled correctly. The time periods when critical layers were modelled often corresponded to the times storm slab avalanches were observed. For example, storm slab avalanches were observed during late February 2014 in all three regions in Fig. 4. Critical surface hoar layers were also modelled over this period. These layers subsequently released persistent slab avalanches several weeks later in March. By March, the critical layer in Fig. 4a was no longer a surface hoar layer, but was at the height where the surface hoar layer had previously existed. Buried surface hoar often merges with adjacent layers of faceted crystals in SNOWPACK, as cold surface temperatures during surface hoar formation can form faceted crystals beneath the surface hoar. Faceting below surface hoar was observed by Hachikubo and Akitaya (1998), and the process is often modelled in SNOWPACK (Monti et al., 2014). As a result, buried surface hoar layers can eventually become faceted crystal layers that are still critically weak, as was the case for the layer in Fig. 4a. Most simulated snow profiles in Fig. 4 had critical surface hoar layers during storm slab avalanche cycles. When persistent slab avalanches released, SNOWPACK usually identified the relevant critical weakness, but often as a layer of critical faceted crystals rather than surface hoar. Fewer critical surface hoar layers were identified over time in the 67 forecast regions as layer depths increased

28

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

(a)

(b)

(c)

Fig. 4. Simulated snow profiles and avalanche observations for three forecast regions. Snow heights, surface hoar layers, and critical weak layers were modelled by SNOWPACK. The daily number of surface hoar avalanches observed in each region is also shown.

Table 3 Summary of surface hoar avalanches. Character

Count

Storm slab Persistent slab Wind slab Loose dry Deep persistent slab Unknown Total

892 739 73 46 34 557 2341

Slab depth (cm) 1st quartile

Median

3rd quartile

25 40 15 5 100 25 30

40 60 25 10 105 40 45

50 75 45 15 125 60 65

(Fig. 3c). Two mechanisms accounted for this; (1) SNOWPACK eventually merged surface hoar with adjacent layers, and (2) shallower layers eventually became more critical because SNOWPACK only selects one critical layer at a time. Agreement between the relative number of modelled and observed layers was fair during the months of January and February, but greater discrepancies arose in the months of December and

March (Fig. 3b and c). Under-reporting is likely in December when forecasters begin operations and have limited information about snow conditions. This could be the case in December 2013, when approximately 20 layers were modelled and less than 5 layers were reported. By March, most of the surface hoar layers that were modelled earlier in the winter were too deep to be critical in SNOWPACK. Warm temperatures and rain-on-snow events later in the winter also melted modelled layers. On 7 February 2015, heavy rainfall with freezing levels above 3000 m resulted in widespread snowmelt across western Canada. Snowmelt destroyed the majority of modelled surface hoar layers, as the number of critical layers dropped from forty to zero (Fig. 3c). Liquid water can destroy too many layers in SNOWPACK because the model does not simulate preferential flow channels (Ikeda et al., 2014, Mitterer et al., 2011). While the number of observed avalanches and reported layers also dropped after 7 February 2015, it appears the model exaggerated the effects of liquid water. Trends in buried surface hoar layers highlight the importance of accurately modelling surface hoar formation. While the formation of some layers was not modelled at all, layers that produced

Table 4 Summary of surface hoar layers reported in persistent weak layer assessments. Status

Developing Active Dormant Inactive

Layer depth (cm)

Layer age (days)

1st quartile

Median

3rd quartile

1st quartile

Median

3rd quartile

10 40 60 45

20 60 95 70

40 90 130 100

4 10 20 12

9 20 35 25

19 34 52 43

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

Fig. 5. Average snow depths observed and modelled at 88 weather observation sites.

widespread avalanche activity were often modelled with large enough crystals to satisfy critical grain size conditions (>0.5 mm in Eq. (5)). Although the modelled crystal sizes likely had errors at the time of burial, most modelled surface hoar was large enough to be critical. Other structural indices such as the threshold sums method proposed by Monti et al. (2014) would also likely detect important surface hoar layers, because modelled layers consistently have large grains and low hardness. However, this also makes the model susceptible to false alarms. In addition to modelling layer formation, accurate modelling of the complete snow cover is needed to properly identify critical layers. Modelled precipitation can dramatically affect properties of the snow above buried layers. Modelled snow depths were typically underestimated, as shown by comparing the average snow depths from 88 weather observation sites with the average snow depth modelled at the same 88 locations (Fig. 5). Schirmer and Jamieson (2015) also found underestimated snowfall amounts in western Canada when coupling SNOWPACK with the HRDPS. Snow height errors were particularly large during 2014–2015 due to frequent warm temperatures and rain-on-snow events. By underestimating snowfall amounts, the slabs above critical surface hoar layers could be thinner and softer, and the layers could be slower to strengthen (Chalmers and Jamieson, 2003). 4.4. Spatial applications The primary advantage of forcing SNOWPACK with weather model data is continuous spatial and temporal coverage. However, forecasting applications must consider relevant spatial scales and error propagation through the models. In this study, the models were configured to suit the verification data. One configuration used downscaled weather data to correct for the elevations of specific observation sites. Another configuration averaged weather data at treeline elevations to simulate conditions at a similar scale to regional avalanche observations and assessments. Other configurations may be more appropriate for avalanche forecasting applications, such as the spatially continuous models of Helbig and van Herwijnen (2012) or Horton et al. (2015). While neither of the configurations in this study were optimal for forecasting, they provide insight into the quality and accuracy of such model chains.

29

Downscaled weather model data was only marginally successful at modelling surface hoar formation at specific observation sites. Modelling site-specific formation was affected by errors in local precipitation, wind, and the surface energy balance. Horton et al. (2015) found weather model data resolved trends relevant to surface hoar formation on a sub-range scale, but not finer scales because local winds and valley clouds were not adequately resolved. Furthermore, they also found SNOWPACK simulations of surface hoar on slopes were less reliable than flat field simulations. More complex downscaling could potentially improve small-scale surface hoar models, for example by accounting for wind exposure with terrainbased exposure parameters (e.g. Winstral et al., 2009) and correcting incoming radiation with sky view factors (e.g. Helbig et al., 2009). Without more accurate weather inputs and advanced downscaling, coupled weather and snow models are currently best suited for modelling flat field processes at regional scales, which could be relevant for forecasting in large or data-sparse regions (Storm, 2012). The verification data in this study spanned regions in different avalanche climate regimes (Haegeli and McClung, 2007). Surface hoar was observed at weather sites in Columbia Mountains more frequently than the Coast and Rocky Mountains (Table 5). Large surface hoar crystals were observed on the surface an average of 16 days per winter in the Columbia Mountains, 12 days per winter in the Coast Mountains, and 8 days per winter in the Rocky Mountains. The average crystal size was also largest in the Columbia Mountains, which have a favourable climate for surface hoar formation (Haegeli and McClung, 2003). The model chain correctly predicted more surface hoar formation in the Columbia Mountains than other ranges. However, the frequency of large surface hoar was over-predicted in the Columbia and Rocky Mountains and under-predicted in the Coast Mountains (Table 5). The discrepancy could be caused by air temperatures biases in the NWP model. Warm biases in the Coast Mountains often caused above freezing temperatures, while cold biases in the Rockies Mountains often exaggerated surface hoar growth. Although surface hoar was frequently observed at weather observation sites in the Coast Mountains, few avalanches or active layers were reported. An average of 0.2 active layers were reported per winter in forecast regions in the Coast Mountain, compared to 1.4 layers per winter in the Columbia and Rocky Mountains (Table 5). The model chain predicted a similar distribution of critical layers between mountain ranges, although more layers were modelled than observed. The Columbia Mountains had the most modelled critical layers with an average of 4.5 per winter, compared to 3.1 in the Rocky Mountains and 1.8 in the Coast Mountains (Fig. 6). The Coast and Rocky Mountains averaged fewer critical layers because surface hoar formation was modelled less often and the crystals were smaller. Furthermore, modelled layers in the Coast Mountains tended to increase in density rapidly after heavy snowfalls and were destroyed by snowmelt. Haegeli and McClung (2007) also reported fewer active surface hoar layers in the Rocky and Coast Mountains than in the Columbia Mountains. They also observed sub-range patterns, such as more active surface hoar layers in the eastern side of the Coast Mountains and the western side of the Rocky Mountains. Nearest-neighbour

Table 5 Comparison of observed and modelled surface hoar by mountain range. Mountain range Days with large surface hoar on surface (per winter) Average crystal size on surface (mm) Observed avalanches (per region per winter) Active layers reported (per region per winter) Critical layers modelled (per region per winter)

Observed Modelled Observed Modelled

Coast

Columbia

12 8 5.0 3.5 4 0 .2 1 .8

16 22 6. 1 6. 2 24 1. 4 4. 5

Rocky 8 15 4. 3 5. 6 17 1. 4 3. 1

30

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31

Fig. 6. The average number of critical surface hoar layers modelled per winter in simulated snow profiles across western Canada. Contours were produced by interpolating between simulations in 67 forecast regions with a nearest-neighbour method (region centres shown with black dots) (ESRI basemap).

interpolation between the 67 forecast regions shows similar subrange patterns were coarsely modelled (Fig. 6), suggesting coupled weather and snow models could possibly resolve relevant sub-range patterns as well.

5. Conclusions Hazardous surface hoar layers were modelled throughout western Canada by coupling a high-resolution NWP model with the snow cover model SNOWPACK. Surface hoar formation was over-predicted at many locations, resulting in a high hit rate for detecting the formation of large crystals, but also a high false alarm rate. Modelled crystal size was crucial to identify buried layers that persisted long enough to become critical weak layers. Although surface hoar formation was subject to errors in the meteorological inputs and errors in the SNOWPACK model, modelled and observed crystal sizes were moderately correlated. Avalanches on buried surface hoar layers initially consisted of storm slab avalanches, which were possibly followed by more destructive persistent slab avalanches for several weeks. The structural stability index in SNOWPACK identified critical surface hoar layers during most of the storm avalanche cycles, but also identified false alarm layers when no avalanches were observed. Modelling critical weakness at older and deeper interfaces when persistent slab avalanches occurred was more complex. For modelled layers to remain critically weak, they either needed large crystals at the time of burial or needed to merge with adjacent layers of weak faceted crystals. Snow cover simulations with high-resolution weather model data could improve avalanche hazard forecasts in data-sparse regions. Weather model data had several advantages over traditional station measurements including continuous temporal and spatial coverage, including in the early winter when field observations are limited. The model correctly predicted more critical surface hoar layers in the Columbia Mountains than the Rocky and Coast Mountains. Resolving patterns between mountain ranges and sub-ranges could be useful in large and data-sparse regions; however, downscaling simulations to finer scales is difficult given the importance of local wind and radiation on surface hoar formation. We anticipate future improvements to NWP models and snow surface energy balance models will contribute to valuable avalanche forecasting products.

Acknowledgements The authors thank the ASARC field team at the University of Calgary, Luke Norman of the Canadian Avalanche Association, and the many subscribers of the InfoEx for collecting and reporting field data. We thank Michael Schirmer and Erik Kulyk for the help in preparing the NWP data and Charles Fierz of the WSL Institute for Snow and Avalanche Research SLF for providing valuable advice on SNOWPACK. We are grateful to the Avalanche Control Section of Glacier National Park, Mike Wiegele Helicopter Skiing, Curtis Pawliuk from the Valemount Area Recreation Development Association, Kerry MacDonald from Marmot Basin Ski Resort, Alexandre Langlois and his team from the University of Sherbrooke, William Golley from Northwest Avalanche Solutions, Bradford White from Banff National Park, and Mike Smith for their support building and maintaining weather stations. For their support of this research we thank TECTERRA, Canadian Pacific, Canadian Avalanche Association (1209-UNI-014), Natural Sciences and Engineering Research Council of Canada, HeliCat Canada Association, Parks Canada, Mike Wiegele Helicopter Skiing, Canada West Ski Areas Association, Backcountry Lodges of BC Association, and Teck Mining Company (IRCPJ 312795 2008). We also thank Michael Conlan and Scott Thumlert for proofreading this paper and the anonymous reviewers for their helpful feedback.

References Bartelt, P., Lehning, M., 2002. A physical SNOWPACK model for the Swiss avalanche warning: part I: numerical model. Cold Reg. Sci. Technol. 35 (3), 123–145. Bellaire, S., Jamieson, B., 2013. Forecasting the formation of critical snow layers using a coupled snow cover and weather model. Cold Reg. Sci. Technol. 94, 37–44. Bellaire, S., Jamieson, B., Fierz, C., 2011. Forcing the snow-cover model SNOWPACK with forecasted weather data. The Cryosphere 5 (4), 1115–1125. Bellaire, S., Jamieson, B., Fierz, C., 2013. Corrigendum to “Forcing the snow-cover model SNOWPACK with forecasted weather data” published in The Cryosphere, 5, 1115–1125, 2011. The Cryosphere 7 (2), 511–513. Borish, M., 2014. Surface Hoar Observations at the Scale of a Helicopter Skiing Operation. Department of Earth Sciences, Montana State University, Bozeman, MT, USA. Master’s thesis Brun, E., David, P., Sudul, M., Brunot, G., 1992. A numerical model to simulate snow– cover stratigraphy for operational avalanche forecasting. J. Glaciol. 38 (1), 13–22. Canadian Avalanche Association, 2014. Observation Guidelines and Recording Standards for Weather, Snowpack and Avalanches. Revelstoke, BC, Canada. Canadian Avalanche Association, 2016. Technical Aspects of Snow Avalanche Risk

S. Horton, B. Jamieson / Cold Regions Science and Technology 128 (2016) 22–31 Management — Resources and Guidelines for Avalanche Practitioners in Canada. Revelstoke, BC, Canada. Chalmers, T., Jamieson, B., 2003. A snow-profile-based forecasting model for skier-triggered avalanches on surface hoar layers in the Columbia Mountains of Canada. Cold Reg. Sci. Technol. 37 (3), 373–383. Colbeck, S., 1988. On the micrometeorology of surface hoar growth on snow in mountainous area. Boundary-Layer Meteor. 44 (1-2), 1–12. Conlan, M., 2015. Forecasting Persistent Deep Slab Avalanches. Department of Civil Engineering, University of Calgary, Calgary, AB, Canada. (Ph.D. thesis) Feick, S., Kronholm, K., Schweizer, J., 2007. Field observations on spatial variability of surface hoar at the basin scale. J. Geophys. Res. 112 (F2), F02002 Fierz, C., Armstrong, R., Durand, Y., Etchevers, P., Greene, E., McClung, D., Nishimura, K., Satyawali, P., Sokratov, S., 2009. The International Classification for Seasonal Snow on the Ground. IHP-VII Tech. Doc. Hyrdol. No. 83, IACS Contrib. No. 1, UNESCO-IHP, Paris Fierz, C., Riber, P., Adams, E., Curran, A., Föhn, P., Lehning, M., Plüss, C., 2003. Evaluation of snow-surface energy balance models in alpine terrain. J. Hydrol. 282 (1), 76–94. Föhn, P., 2001. Simulation of surface-hoar layers for snow-cover models. Ann. Glaciol. 32 (1), 19–26. Hachikubo, A., 2001. Numerical modelling of sublimation on snow and comparison with field measurements. Ann. Glaciol. 32 (1), 27–32. Hachikubo, A., Akitaya, E., 1998. Daytime preservation of surface-hoar crystals. Ann. Glaciol. 26, 22–26. Haegeli, P., McClung, D., 2003. Avalanche characteristics of a transitional snow climate — Columbia Mountains, British Columbia, Canada. Cold Reg. Sci. Technol. 37 (3), 255–276. Haegeli, P., McClung, D., 2007. Expanding the snow-climate classification with avalanche-relevant information: initial description of avalanche winter regimes for southwestern Canada. J. Glaciol. 53 (181), 266–276. Haegeli, P., Obad, J., Harrison, B., Murray, B., Engblom, J., Neufeld, J., 2014. InfoEx 3.0 — advancing the data analysis capabilities of Canada’s diverse avalanche community. Proceedings of the 2014 International Snow Science Workshop, Banff, AB, Canada. pp. 910–917. Helbig, N., Löwe, H., Lehning, M., 2009. Radiosity approach for the shortwave surface radiation balance in complex terrain. J. Atmos. Sci. 66 (9), 2900–2912. Helbig, N., van Herwijnen, A., 2012. Modeling the spatial distribution of surface hoar in complex topography. Cold Reg. Sci. Technol. 82, 68–74. Horton, S., Bellaire, S., Jamieson, B., 2014. Modelling the formation of surface hoar layers and tracking post-burial changes for avalanche forecasting. Cold Reg. Sci. Technol. 97, 81–89. Horton, S., Schirmer, M., Jamieson, B., 2015. Meteorological, elevation, and slope effects on surface hoar formation. The Cryosphere 9 (4), 1523–1533. Ikeda, S., Katsushima, T., Matsushita, H., Ito, Y., Takeuchi, Y., Akiyama, K., 2014. Comparison of snowpack on a slope and on flat land focusing on the effects of water infiltration. Cold Reg. Sci. Technol. 108, 91–97. Jamieson, B., Johnston, C., 1992. Snowpack characteristics associated with avalanche accidents. Can. Geotech. J. 29 (5), 862–866. Jamieson, B., Johnston, C., 1998. Refinements to the stability index for skier-triggered dry-slab avalanches. Ann. Glaciol. 26, 296–302. Jamieson, B., Schweizer, J., 2000. Texture and strength changes of buried surface-hoar layers with implications for dry snow-slab avalanche release. J. Glaciol. 46 (152), 151–160. Klassen, K., 2010. Experience is not enough: persistent weak layers in western Canada 2007–2010. Proceedings of the 2010 International Snow Science Workshop, Squaw Valley, CA, USA. pp. 172–177. Klassen, K., Haegeli, P., Statham, G., 2013. The role of avalanche character in public avalanche safety products. Proceedings of the 2013 International Snow Science Workshop, Grenoble, France. pp. 493–499. Lazar, B., Greene, E., Birkeland, K., 2012. Avalanche problems defined. Avalanche Rev. 31 (2), 14–23.

31

Lehning, M., Bartelt, P., Brown, B., Fierz, C., 2002. A physical SNOWPACK model for the Swiss avalanche warning: part III: meteorological forcing, thin layer formation and evaluation. Cold Reg. Sci. Technol. 35 (3), 169–184. Lehning, M., Bartelt, P., Brown, B., Russi, T., Stöckli, U., Zimmerli, M., 1999. SNOWPACK model calculations for avalanche warning based upon a new network of weather and snow stations. Cold Reg. Sci. Technol. 30 (1), 145–157. Lehning, M., Doorschot, J., Bartelt, P., 2000. A snowdrift index based on SNOWPACK model calculations. Ann. Glaciol. 31 (1), 382–386. Liston, G., Elder, K., 2006. A meteorological distribution system for high-resolution terrestrial modeling (MicroMet). J. Hydrometeor. 7 (2), 217–234. Lutz, E., Birkeland, K., Kronholm, K., Hansen, K., Aspinall, R., 2007. Surface hoar characteristics derived from a snow micropenetrometer using moving window statistical operations. Cold Reg. Sci. Technol. 47 (1), 118–133. Milbrandt, J., 2014. Technical Note for the High Resolution Deterministic Prediction System 4.0.0. Tech. Rep., Environment Canada. Mitterer, C., Hirashima, H., Schweizer, J., 2011. Wet-snow instabilities: comparison of measured and modelled liquid water content and snow stratigraphy. Ann. Glaciol. 52 (58), 201–208. Monti, F., Cagnati, A., Valt, M., Schweizer, J., 2012. A new method for visualizing snow stability profiles. Cold Reg. Sci. Technol. 78, 64–72. Monti, F., Schweizer, J., Fierz, C., 2014. Hardness estimation and weak layer detection in simulated snow stratigraphy. Cold Reg. Sci. Technol. 103, 82–90. Schirmer, M., Jamieson, B., 2015. Verification of analysed and forecasted winter precipitation in complex terrain. The Cryosphere 9 (2), 587–601. Schirmer, M., Schweizer, J., Lehning, M., 2010. Statistical evaluation of local to regional snowpack stability using simulated snow-cover data. Cold Reg. Sci. Technol. 64 (2), 110–118. Schweizer, J., Bellaire, S., Fierz, C., Lehning, M., Pielmeier, C., 2006. Evaluating and improving the stability predictions of the snow cover model SNOWPACK. Cold Reg. Sci. Technol. 46 (1), 52–59. Schweizer, J., Jamieson, B., 2001. Snow cover properties for skier triggering of avalanches. Cold Reg. Sci. Technol. 33 (2), 207–221. Schweizer, J., Jamieson, B., 2007. A threshold sum approach to stability evaluation of manual snow profiles. Cold Reg. Sci. Technol. 47 (1), 50–59. Schweizer, J., Kronholm, K., 2007. Snow cover spatial variability at multiple scales: characteristics of a layer of buried surface hoar. Cold Reg. Sci. Technol. 47 (3), 207–223. Slaughter, A., 2010. Numerical Analysis of Conditions Necessary for Near-surface Snow Metamorphism. College of Engineering, Montana State University, Bozeman, MT, USA. (Ph.D. thesis) Storm, I., 2012. Public avalanche forecast challenges: Canada’s large data-sparse regions. Proceedings of the 2012 International Snow Science Workshop, Anchorage, AK, USA. pp. 908–912. Stössel, F., Guala, M., Fierz, C., Manes, C., Lehning, M., 2010. Micrometeorological and morphological observations of surface hoar dynamics on a mountain snow cover. Water Resour. Res. 46 (4), W04511 Vionnet, V., Bélair, S., Girard, C., Plante, A., 2014. Wintertime sub-kilometer numerical forecasts of near-surface variables in the Canadian Rocky Mountains. Mon. Wea. Rev. 143, 666–686. Winstral, A., Marks, D., Gurney, R., 2009. An efficient method for distributing wind speeds over heterogeneous terrain. Hydrol. Process. 23 (17), 2526–2535.