• Barstad, I., , Sorteberg A. , , Flatøy F. , , and Déqué M. , 2009: Precipitation, temperature and wind in Norway: Dynamical downscaling of ERA40. Climate Dyn., 33, 769776, doi:10.1007/s00382-008-0476-5.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., , Crevier L.-P. , , Mailhot J. , , Bilodeau B. , , and Delage Y. , 2003a: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part I: Warm season results. J. Hydrometeor., 4, 352370.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., , Brown R. , , Mailhot J. , , and Bilodeau B. , 2003b: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part II: Cold season results. J. Hydrometeor., 4, 371386.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., , Roch M. , , Leduc A.-M. , , Vaillancourt P. , , Laroche S. , , and Mailhot J. , 2009: Medium-range quantitative precipitation forecasts from Canada’s new 33-km deterministic global operational system. Wea. Forecasting, 24, 690708.

    • Search Google Scholar
    • Export Citation
  • Bhumralkar, C. M., 1975: Numerical experiments on the computation of ground surface temperature in an atmospheric general circulation model. J. Appl. Meteor., 14, 12461258.

    • Search Google Scholar
    • Export Citation
  • de Goncalves, L. G. G., , Shuttleworth W. J. , , Burke E. J. , , Houser P. , , Toll D. L. , , Rodell M. , , and Arsenault K. , 2006: Toward a South America land data assimilation system: Aspects of land surface model spin-up using the simplified simple biosphere. J. Geophys. Res., 111, D17110, doi:10.1029/2005JD006297.

    • Search Google Scholar
    • Export Citation
  • Dodson, R., , and Marks D. , 1997: Daily air temperature interpolated at high spatial resolution over a large mountainous region. Climate Res., 8, 120.

    • Search Google Scholar
    • Export Citation
  • Douville, H., , Royer J.-F. , , and Mahfouf J.-F. , 1995: A new snow parameterization for the Météo-France climate model. Part I: Validation in stand-alone experiments. Climate Dyn., 12, 2135.

    • Search Google Scholar
    • Export Citation
  • Fridley, J. D., 2009: Downscaling climate over complex terrain: High finescale (<1000 m) spatial variation of near-ground temperatures in a montane forested landscape (Great Smokey Mountains). J. Appl. Meteor. Climatol., 48, 10331049.

    • Search Google Scholar
    • Export Citation
  • Grant, A., , and Mason P. , 1990: Observations of boundary-layer structure over complex terrain. Quart. J. Roy. Meteor. Soc., 116, 159186.

    • Search Google Scholar
    • Export Citation
  • Hartman, M. D., , Baron J. S. , , Lammers R. B. , , Cline D. W. , , Band L. E. , , Liston G. E. , , and Tague C. , 1999: Simulations of snow distribution and hydrology in a mountain basin. Water Resour. Res., 35, 15871603.

    • Search Google Scholar
    • Export Citation
  • Jackson, S. I., , and Prowse T. D. , 2009: Spatial variation of snowmelt and sublimation in a high-elevation semi-desert basin of western Canada. Hydrol. Processes, 23, 26112627.

    • Search Google Scholar
    • Export Citation
  • Liston, G. E., 2004: Representing subgrid snow cover heterogeneities in regional and global models. J. Climate, 17, 13811397.

  • Loth, B., , Graf H.-F. , , and Oberhuber J. M. , 1993: Snow cover model for global climate simulations. J. Geophys. Res., 98 (D6), 10 45110 464.

    • Search Google Scholar
    • Export Citation
  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44, 133149.

  • Mailhot, J., and Coauthors, 2010: Environment Canada’s experimental numerical weather prediction systems for the Vancouver 2010 Winter Olympic and Paralympic Games. Bull. Amer. Meteor. Soc., 91, 10731085.

    • Search Google Scholar
    • Export Citation
  • Noilhan, J., , and Planton S. , 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117, 536549.

    • Search Google Scholar
    • Export Citation
  • Sheridan, P., , Smith S. , , Brown A. , , and Vosper S. , 2010: A simple height-based correction for temperature downscaling in complex terrain. Meteor. Appl., 17, 329339, doi:10.1002/met.177.

    • Search Google Scholar
    • Export Citation
  • Slater, A. G., and Coauthors, 2001: The representation of snow in land surface schemes: Results from PILPS 2(d). J. Hydrometeor., 2, 725.

    • Search Google Scholar
    • Export Citation
  • Stahl, K., , Moore R. , , Floyer J. , , Asplin M. , , and McKendry I. , 2006: Comparison of approaches for spatial interpolation of daily air temperature in a large region with complex topography and highly variable station density. Agric. For. Meteor., 139, 224236.

    • Search Google Scholar
    • Export Citation
  • Stewart, I., , Cayan D. R. , , and Dettinger M. , 2004: Changes in snowmelt runoff timing in western North America under a ‘business as usual’ climate change scenario. Climatic Change, 62, 217232.

    • Search Google Scholar
    • Export Citation
  • Strack, J. E., , Liston G. E. , , and Pielke R. A. Sr., 2004: Modeling snow depth for improved simulation of snow–vegetation–atmosphere interactions. J. Hydrometeor., 5, 723734.

    • Search Google Scholar
    • Export Citation
  • Tribbeck, M. J., , Gurney R. J. , , and Morris E. M. , 2006: The radiative effect of a fir canopy on a snowpack. J. Hydrometeor., 7, 880895.

    • Search Google Scholar
    • Export Citation
  • Trivedi, M. R., , Berry P. M. , , Morecroft M. D. , , and Dawson T. P. , 2008: Spatial scale affects bioclimate model projections of climate change impacts on mountain plants. Global Change Biol., 14, 10891103.

    • Search Google Scholar
    • Export Citation
  • Wilson, L. J., , and Vallée M. , 2003: The Canadian Updateable Model Output Statistics (UMOS) system: Validation against perfect prog. Wea. Forecasting, 18, 288302.

    • Search Google Scholar
    • Export Citation
  • Zadra, A., , Roch M. , , Laroche S. , , and Charron M. , 2003: The subgrid-scale orographic blocking parametrization of the GEM model. Atmos.–Ocean, 41, 155170.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Domain of the near-surface and land surface forecast system GEM-SURF. Observation stations are marked with colored dots. The Olympic network is shown in red, standard climate stations are shown in green, and snow courses and snow pillow stations are shown in yellow (section 2 and Tables 13). The color bar indicates elevation in meters. The top and bottom right panels are close-ups of the two principal alpine VO2010 venue sites delimited by boxes in the left panel.

  • View in gallery

    Schematics of the Vancouver Olympics high-resolution forecast system. GEM-SURF is driven by hourly downscaled forecast fields issued by MSC operational forecast models. The first 48 h of forcings are obtained by downscaling GEM-15km forecasts; hours 49–96 are driven using downscaled GEM-33km forecasts. Forcing fields are temperature (T), humidity (q), and winds (U, V) at the atmospheric model’s first level (~40 m above the surface) as well as surface pressure, radiation, and rate of precipitation.

  • View in gallery

    Correlation R between pairs of snow depth observation records (Nov 2007 to Jan 2009). (top) Correlation as a function of elevation separation (m), and (bottom) horizontal separation (km). Red dots are correlations calculated between pairs of stations located above 500 m. Blue dots are for pairs of stations at elevations between 100 and 500 m. The green dots are stations with elevation less than 100 m. Black dots are for all other possible pairs (e.g., the correlation between an alpine and a near–sea level station).

  • View in gallery

    Mean temperature difference (Nov 2007 to May 2009) between all possible pairs of screen-level temperature records. The black dots are for observations. The red and blue dots are for GEM-15km and -33km forecasts of screen-level temperature, respectively. (top) All observations and forecasts over the period Nov 2007 to May 2009. The other panels are for selected subsets (winter only, summer only, days only, and nights only). The black line is the lapse rate (γ = 6.0 K km−1) chosen for the downscaling of forcing fields (section 4).

  • View in gallery

    Observed and forecast screen-level relative humidity (%) at station VOC (1 Jan to 1 Apr 2009). (top),(middle) Observations are shown in black. The collated first 24-h GEM-15km forecasts are shown in blue in (top). In (middle), the screen-level humidity forecast with GEM-SURF is shown in red. (bottom) The forecast errors for GEM-15km and GEM-SURF are shown in blue and red, respectively.

  • View in gallery

    Histograms of observed and forecast temperatures and temperature daily extremes (K) about their daily mean at (left) VOA, a high-elevation station, and (right) WVF, a station at sea level (note the change in x axis scale). The observed temperatures are shown in black and the forecast GEM-15km temperatures are in red. (top) Hourly values, (middle) daily maxima about their daily mean, and (bottom) the same, but for minima.

  • View in gallery

    Observed and forecast hourly screen-level temperature (K) and surface pressure (hPa) at Squamish Airport. The black lines show observations. The blue lines show (top row),(third row) GEM-33km and (second row),(bottom row) GEM-15km operational forecasts. The red lines show the downscaled operational forecasts (i.e., corrected for elevation differences; see section 4 for details).

  • View in gallery

    (top) Elevation (m), (middle) screen-level temperature (°C), and (bottom) surface pressure (hPa) fields. (left) The low-resolution GEM-15km forecast fields valid at 00 UTC 3 Nov 2008 interpolated to the high-resolution grid, (middle) the high-resolution downscaled fields, and (right) differences (high minus low resolution).

  • View in gallery

    Observed and forecast snow depth (cm) from 1 Nov 2008 to 1 Apr 2009 for stations that are part of the (top) Olympics network (Table 1) and (middle) climate network (Table 2). (bottom) Results for (left) snow courses and (right) snow pillow stations (Table 3). The black stars are for observations. The blue and green lines are for the operational forecasts from GEM-15km and -33km, respectively. GEM-SURF forecasts are plotted in red.

  • View in gallery

    Snow cover and snow depth forecasts valid 1 Nov 2008, 1 Dec 2008, 1 Jan 2009, 1 Feb 2009, 1 Mar 2009, and 1 Apr 2009. The color bar is elevation (ME; m) or, when present, snow depth (SD; m). The dots show the station location for time series plotted in Fig. 9. The Olympic network stations are shown in red, the standard climate stations are shown in green, and the snow pillow stations are shown in yellow.

  • View in gallery

    Histograms of (left) screen-level air temperature (K) and (right) relative humidity (%) forecast errors. Day 1 means all 1–24-h forecasts valid at the time of the observations. Day 2 is for hours 25–48, day 3 for hours 49–72, and day 4 for hours 73–96.

  • View in gallery

    Observed and forecast screen-level temperatures at station VOB. (top) Observations are in black, GEM-15km in blue, and GEM-SURF in red. (bottom) Forecast error for GEM-15km and GEM-SURF.

  • View in gallery

    (top) Snow depth (cm), (middle) snow density (kg m−3), and (bottom) snow albedo forecasts valid at 00 UTC 30 Mar 2009 for (left) GEM-15km and (right) GEM-SURF.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 23 23 5
PDF Downloads 21 21 4

Near-Surface and Land Surface Forecast System of the Vancouver 2010 Winter Olympic and Paralympic Games

View More View Less
  • 1 Meteorological Research Division, Dorval, Quebec, Canada
  • | 2 Meteorological Service of Canada, Dorval, Quebec, Canada
© Get Permissions
Full access

Abstract

A high-resolution 2D near-surface and land surface model was developed to produce snow and temperature forecasts over the complex alpine region of the Vancouver 2010 Winter Olympic and Paralympic Games. The model is driven by downscaled operational outputs from the Meteorological Service of Canada’s regional and global forecast models. Downscaling is applied to correct forcings for elevation differences between the operational forecast models and the high-resolution surface model. The high-resolution near-surface and land surface model is then used to further refine the forecasts. The model was validated against temperature and snow depth observations. The largest improvements were found in regions where low-resolution (i.e., on the order of 10 km or more) operational models typically lack the spatial resolution to capture rapid elevation changes. The model was found to better reproduce the intermittent snow cover at low-lying stations and to reduce snow depth error by as much as 3 m at alpine stations.

Corresponding author address: Natacha B. Bernier, Meteorological Research Division, Science and Technology Branch, Environment Canada, 2121 Trans-Canada Highway, 5th Floor, Dorval QC H9P 1J3, Canada. E-mail: natacha.bernier@ec.gc.ca

Abstract

A high-resolution 2D near-surface and land surface model was developed to produce snow and temperature forecasts over the complex alpine region of the Vancouver 2010 Winter Olympic and Paralympic Games. The model is driven by downscaled operational outputs from the Meteorological Service of Canada’s regional and global forecast models. Downscaling is applied to correct forcings for elevation differences between the operational forecast models and the high-resolution surface model. The high-resolution near-surface and land surface model is then used to further refine the forecasts. The model was validated against temperature and snow depth observations. The largest improvements were found in regions where low-resolution (i.e., on the order of 10 km or more) operational models typically lack the spatial resolution to capture rapid elevation changes. The model was found to better reproduce the intermittent snow cover at low-lying stations and to reduce snow depth error by as much as 3 m at alpine stations.

Corresponding author address: Natacha B. Bernier, Meteorological Research Division, Science and Technology Branch, Environment Canada, 2121 Trans-Canada Highway, 5th Floor, Dorval QC H9P 1J3, Canada. E-mail: natacha.bernier@ec.gc.ca

1. Introduction

The 2010 Winter Olympic and Paralympic Games (VO2010) were held in Vancouver, Canada from 12 to 28 February 2010 and from 12 to 21 March 2010, respectively. In view of these Games, Environment Canada was asked to provide timely and accurate forecasts over the Games’ venue region (Fig. 1). To fulfill these requirements, Environment Canada made use of existing operational and new experimental forecast products. An overview of the suite of VO2010 specially designed and experimentally implemented products is available in Mailhot et al. (2010). In this study, we present the high-resolution near-surface and surface forecast systems developed and implemented for VO2010.

Fig. 1.
Fig. 1.

Domain of the near-surface and land surface forecast system GEM-SURF. Observation stations are marked with colored dots. The Olympic network is shown in red, standard climate stations are shown in green, and snow courses and snow pillow stations are shown in yellow (section 2 and Tables 13). The color bar indicates elevation in meters. The top and bottom right panels are close-ups of the two principal alpine VO2010 venue sites delimited by boxes in the left panel.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

Over the past decades, significant improvements have been made to global and regional forecast systems. However, it remains a challenge to forecast near-surface and land surface conditions using large-scale forecast models. Three important limitations to the ability of general circulation and other operational forecast models—such as the Global Environmental Multiscale 15 km (GEM-15km) and global 33 km (GEM-33km) forecast models (Mailhot et al. 2006; Bélair et al. 2009)—to predict land and near-surface processes are the generally crude representation of land processes, the crude parameterization of the subgrid land–atmosphere interactions that result, and the relatively low resolution of these models compared to the scales at which the surface balance is known to evolve. The latter results from the sensitivity of the land surface dynamics to local conditions such as elevation, soil types, vegetation types, and snow coverage (e.g., Loth et al. 1993; Douville et al. 1995; Hartman et al. 1999; Liston 2004; Strack et al. 2004; Tribbeck et al. 2006).

In the region of interest, the lack of resolution (over the complex and rapidly changing terrain) can lead to systematic surface temperature errors that can reach several degrees, while forecast yearly peak snow depth can be off by as much as three meters. These large forecast errors occur at the level at which athletes and the public alike experience weather and are thus of high impact. Any improvement of the forecasts would be of high value in this region, where only low-resolution operational forecasts are available together with sparse observation records and snow cover images only [e.g., Moderate Resolution Imaging Spectroradiometer (MODIS)]. The goal of this project was thus to develop and implement a robust yet inexpensive system to improve on current operational near-surface and land surface forecast capabilities, in particular snow variables, in the complex alpine region of the 2010 Winter Olympic Games.

At the Meteorological Service of Canada (MSC), operational forecast systems are currently computationally too expensive (in terms of storage and operational run time requirement) to explicitly resolve local effects over large areas (e.g., the Olympics region at a few hundred meters grid spacing). There is therefore a need to decouple, at least partially, the two systems such that land surface processes are allowed to evolve separately in terms of resolution and time increments. Within the operational constraints of the MSC, two methods can be used to allow independent temporal and spatial resolution. In the first method, low-level atmospheric fields from any standard operational forecast model can be used to drive an independent (hence external) high-resolution near-surface and land surface model. The result is that forecasts above the lowest level of the standard operational model remain identical, whereas near-surface and surface forecasts can now be the result of high-resolution integration. In addition, fluxes from the high-resolution surface model can be fed back into the standard operational model, thus allowing two-way coupling between the different resolutions. This constitutes the second method.

For the Olympics and Paralympics Games, and as an initial step toward a pan-Canadian fully coupled modeling system, the chosen strategy was to add an external near-surface and land surface model (GEM-SURF) to the standard operational forecast models, set the mesh to resolve local effects, and drive GEM-SURF using low-atmosphere operational forecasts (Fig. 2).

Fig. 2.
Fig. 2.

Schematics of the Vancouver Olympics high-resolution forecast system. GEM-SURF is driven by hourly downscaled forecast fields issued by MSC operational forecast models. The first 48 h of forcings are obtained by downscaling GEM-15km forecasts; hours 49–96 are driven using downscaled GEM-33km forecasts. Forcing fields are temperature (T), humidity (q), and winds (U, V) at the atmospheric model’s first level (~40 m above the surface) as well as surface pressure, radiation, and rate of precipitation.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

This paper describes the VO2010 near-surface and land surface forecast system. The outline of this paper is as follows. In section 2, we introduce the observations used for validation and discuss the geophysical fields used to describe the local conditions. The new high-resolution (100-m grid spacing) model is presented in section 3. The downscaling method applied to correct the low-resolution forcing fields is described and validated in section 4. In section 5, we use snow depth observations to validate winter 2008/09 forecasts and compare GEM-SURF forecasts to MSC’s operational outputs. In section 6, we briefly demonstrate the effect of the refinement on other fields. The results of this study are summarized and discussed in section 7.

2. Observations and surface fields

In preparation for VO2010, several automated weather stations were deployed at outdoor competition sites (Fig. 1, red dots; Table 1). The stations are standardly equipped to record temperature, wind, humidity, and pressure. In addition, a few stations are equipped with snow depth and visibility sensors. Standard weather stations are also available in the vicinity of and within the Olympics region. Hourly records of temperature (Fig. 1, red dots; Table 1), daily minimum and maximum temperatures, and once-daily snow depth (Fig. 1, green dots; Table 2) are recorded at these standard climate observation stations and are available online (http://climate.weatheroffice.ec.gc.ca/climateData/canada_e.html).

Table 1.

The 2010 Vancouver Olympics observation network. The top part is for stations located at or near competition sites. The bottom part is for other hourly recording stations located within the Olympic region. WMO ID = World Meteorological Organization identification.

Table 1.
Table 2.

WMO data stations within the Olympic region. Stations report daily min and max temperatures. Some stations also report snow depth once daily.

Table 2.

Several snow courses and snow pillow station observations are also available within the model domain (Fig. 1, yellow dots; Table 3). At snow courses, measurements are performed manually using standard snow sampling tubes approximately monthly during winter months. Measurements are available online from the River Forecast Centre of the British Columbia Ministry of Environment (www.env.gov.bc.ca/rfc/index.htm). At snow pillow stations, the recording of observations is automated, hourly, and includes variables such as temperature and humidity. Exact station coordinates and hourly data for the snow pillow stations were obtained through the River Forecast Centre data services (S. Jackson 2009, personal communication).

Table 3.

Snow courses and snow pillow stations. Station names ending with “P” indicate automated stations that report hourly (variables such as temperature are also available at automated stations). Other stations are manual data collection sites where snow depth is measured several times per winter.

Table 3.

a. Observed snow depths

The spatial signature of the snowpack is shown in Fig. 3, where dots show correlations between pairs of stations with elevation of 500 m or more as a function of station separation (horizontal and vertical). (The discussion and results presented in this paragraph are based on all observations available from November 2007 to May 2009; note this is the only time we will use and discuss snow depth data outside of the forecast period of 1 November 2008 to 1 April 2009.) As expected, low-level stations where below-freezing temperature periods are short and do not persist long enough to allow the growth of a snowpack are not well correlated. Observed snowpack depths within the alpine region (defined here as stations located above 500 m) are horizontally and vertically correlated (Fig. 3, red dots).

Fig. 3.
Fig. 3.

Correlation R between pairs of snow depth observation records (Nov 2007 to Jan 2009). (top) Correlation as a function of elevation separation (m), and (bottom) horizontal separation (km). Red dots are correlations calculated between pairs of stations located above 500 m. Blue dots are for pairs of stations at elevations between 100 and 500 m. The green dots are stations with elevation less than 100 m. Black dots are for all other possible pairs (e.g., the correlation between an alpine and a near–sea level station).

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

Well-defined snow seasons are clearly evident at all alpine stations (not shown) with 2009 peak snow depth between 77 cm (Blackcomb Mountain Base) and 292 cm (Cypress Bowl North). Interannual variability can be large in alpine regions. In the study region, snow packs for winter 2008 (not shown) were deeper than those observed in winter 2009. At coastal stations (e.g., Squamish Airport and West Vancouver), winter temperatures are milder and snow typically melts rapidly following storm events or short accumulation periods (not shown). For these stations, 2009 maximum snow depths exceeded those recorded in 2008. Note that the observation records of April and May 2009 were examined to verify that maximum snow depth had been reached by 1 April 2009 at all stations. Maximum snow depths at all stations discussed in this paper are thus the maximum attained for the fall 2008 to spring 2009 snow season.

b. Observed screen-level temperatures

The horizontal separation between a number of observation stations is small enough that they fall within the same grid box of the coarse GEM-15km or -33km forecast fields used to drive the near-surface and land surface model. The local variability is thus not resolved. An important contributor to this small-scale variability is the vertical separation between horizontally adjacent pairs of observation records. The mean temperature differences (using all observations available from November 2007 to May 2009; note this is the only time we discuss temperature data outside of the forecast period of 1 November 2008 to 1 April 2009) as a function of elevation differences between all possible pairs of stations is shown in Fig. 4. The differences were also calculated for winter-, summer-, day-, and night-only subsets of observations. A lapse rate of 6.0 K km−1 (Fig. 4, black line) was found to best fit the observed lapse rates for all conditions. We will return to this point and discuss the method used to downscale the low-resolution (GEM-15km and -33km) forcing fields in section 4.

Fig. 4.
Fig. 4.

Mean temperature difference (Nov 2007 to May 2009) between all possible pairs of screen-level temperature records. The black dots are for observations. The red and blue dots are for GEM-15km and -33km forecasts of screen-level temperature, respectively. (top) All observations and forecasts over the period Nov 2007 to May 2009. The other panels are for selected subsets (winter only, summer only, days only, and nights only). The black line is the lapse rate (γ = 6.0 K km−1) chosen for the downscaling of forcing fields (section 4).

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

c. Relative humidity

Wet-bulb temperatures are also recorded at most Olympic venue stations. They are locally compiled and transmitted as relative humidity. Time series of screen-level humidity for 1 January to 1 April 2009 are shown in Fig. 5 for Blackcomb Base, a venue station. The observations are shown in black. Note the rapidly changing signal.

Fig. 5.
Fig. 5.

Observed and forecast screen-level relative humidity (%) at station VOC (1 Jan to 1 Apr 2009). (top),(middle) Observations are shown in black. The collated first 24-h GEM-15km forecasts are shown in blue in (top). In (middle), the screen-level humidity forecast with GEM-SURF is shown in red. (bottom) The forecast errors for GEM-15km and GEM-SURF are shown in blue and red, respectively.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

d. Geophysical fields

Geophysical fields refer to the set of variables used to describe the surface characteristics that are not expected to change over the span of a few winters. They include (at each GEM-SURF grid point) information such as elevation, vegetation type, soil texture, and land–water mask. The information is primarily based on the Canadian Digital Elevation Data (CDED50) database for orography, a classification based on Landsat-7 data for vegetation type, and the CanVec database for the land water mask (http://geogratis.gc.ca/geogratis/en/index.html).

Several surface fields are influenced by the geophysical characteristics (e.g., albedo, thermal and hydraulic properties of the soil, leaf area index, and roughness length) and can be modified by time-dependent local conditions (e.g., the albedo of snow-covered ground is higher than that of a grass field, and roughness length is influenced by the orography, vegetation, and state of ground cover).

The roughness length is used to parameterize the surface stress (e.g., Grant and Mason 1990). Ultimately, it parameterizes the degree of coupling between the atmosphere and the land surface. In this study, GEM-SURF is driven by large-scale GEM-15km and -33km forecasts. The forcing fields are thus produced by low-resolution models, which are influenced by large-scale orography, as opposed to the local orography influencing GEM-SURF. The result of sensitivity runs is that GEM-SURF roughness lengths must be set to those of the driving model (GEM-33km values were chosen). Failure to do so was found to severely limit vertical mixing, particularly at night, and resulted in overnight temperature minima far colder than observed. Adding a local component to the surface roughness (i.e., the part that results namely from resolving local vegetation) was considered. However, given the high resolution used for GEM-SURF, subgrid orography is small compared to the subgrid orography of the lower resolution used for GEM-33km. Test runs showed that the local contributions to roughness length can be neglected without leading to a significant change in the forecast fields.

3. Near-surface and land surface forecast system

GEM-SURF, the near-surface and land surface forecast system, is forced with hourly operational (i.e., GEM-15km or -33km) forecasts that are internally adjusted for orographic differences between forcing and integration grids. The model is integrated for 96 h (4 days). The results are high-resolution hourly forecasts of near-surface and surface fields such as screen-level air temperature and snow conditions (e.g., depth, density, and albedo). This section describes GEM-SURF and the fields used to drive and initialize it. The spinup and configuration of the system are also briefly discussed.

a. GEM-SURF schemes

There are a number of land surface models of varying complexity in use (see Slater et al. 2001 for a thorough review). In this study, the surface schemes are middle-range complexity and based on Interactions between Soil, Biosphere, and Atmosphere (ISBA; Noilhan and Planton 1989; Douville et al. 1995; Bélair et al. 2003a,b). The mesh is set to 100-m grid spacing over the VO2010 region (Fig. 1). GEM-SURF schemes have the advantage of being complex enough to allow for a prognostic evolution of snow properties (e.g., albedo, thermal conductivity, heat capacity, and snow density) and yet simple enough to be integrated several times daily with high spatial and temporal resolution. In GEM-SURF, surface temperatures and soil water contents evolve according to the force–restore approach (Bhumralkar 1975). Evapotranspiration is influenced by the presence of vegetation through a surface resistance that depends on both atmospheric factors and soil moisture availability (Noilhan and Planton 1989). GEM-SURF also includes a snow submodel that features prognostic equations for snow temperature (skin and deep), snow mass, snow surface albedo, mean snow density, and liquid water retained in the snow canopy (see Bélair et al. 2003b for details).

b. Forcing fields

The forcing fields [hourly rate of precipitation, incident radiation, surface pressure, and low-level (~40 m) hourly temperature, wind, and humidity] are obtained from MSC’s operational forecasts. The first 48 h are GEM-15km, the 15-km regional forecast model (Mailhot et al. 2006). The remainder of the period is covered by GEM-33km, the 33-km operational global forecast model (Bélair et al. 2009), as illustrated in Fig. 2. We note that the operational center also runs a 2.5-km GEM limited-area model over the region. However, these high-resolution integrations are still considered experimental at MSC, in addition to being too short (in forecast range) and not being available early enough (latency) to be used as forcing for GEM-SURF. We note that orography differences between the models (e.g., between GEM-15km and GEM-SURF) can exceed 1 km. This implies that some forcing fields are not representative of the local conditions. Thus, they need to be adapted (i.e., downscaled or corrected) for elevation differences between the two models. Temperature, humidity, and pressure are thus downscaled and phase of precipitation adjusted before they are used to drive GEM-SURF. (The downscaling method is described and demonstrated in section 4.)

In this paper, the operational model forecasts for surface and near-surface variables are evaluated alongside GEM-SURF forecasts. Note that the operational model level used for the evaluation is not the same as the level used to drive GEM-SURF. The operational model surface forecasts are computed using their own configuration of the land surface model. These forecasts are adapted before being compared to GEM-SURF forecasts to isolate the contribution of the high-resolution near-surface and land surface model.

c. Initialization and spinup

At the first time step of GEM-SURF (i.e., for a cold start), variables such as soil water content and deep ground temperature are initialized using GEM-15km fields. Since the first-guess fields used to initialize GEM-SURF are the product of an operational model with a sequential assimilation scheme that has been running continuously since 2001 (Bélair et al. 2003b), they can be regarded as having reached equilibrium (at low resolution). The spin-up time of the high-resolution grid was determined based on the comparison of test runs with a long integration. Thus, a long-term integration (benchmark) of the high-resolution model (over a representative subdomain) was launched for September 2007 and integrated until April 2009. Test runs were also launched biweekly from August 2008 to December 2008 and compared with the benchmark integration launched in September 2007. The spin-up time was found to be approximately 6 weeks with high-altitude points reaching equilibrium with the benchmark faster than coastal points. The rapid transition toward a solid soil water state at high elevations during the fall likely leads to the reduced spin-up time. It is expected that a more classical cold start of GEM-SURF would require much longer spin-up times (e.g., de Goncalves et al. 2006).

The GEM-SURF forecast run presented in this paper was initialized on 1 September 2008 and the spinup carried on until 1 November 2008. Over the course of these two months, daily 24-h integrations were driven using GEM-15km forecast fields in an open loop manner (i.e., each new 24-h integration is simply a continuation of the previous forecast with no input, data assimilation, or forcing other than the forcing fields described in section 3b). All model variables such as snow depth, soil temperatures, and soil water content thus evolve internally (i.e., independently from the operational system).

d. Configuration of the continuous forecast system

GEM-SURF forecast period began November 2008 following the spinup described in section 3c. It was then run in a continuous open loop, forced with hourly forecasts, and corrected for elevation (section 4) until 1 April 2009. Hence, 96-h forecasts were produced for each day using the previous 24th-hour forecast for initial conditions. The system described in this paper was transferred to production and was run for the winter 2009/10. Real-time production and delivery to the Olympics and Paralympics forecast team of daily 96-h forecasts began 1 November 2009, following a two-month spinup, and continued until 1 April 2010.

4. Elevation correction of the forcing fields

Large-scale models typically lack the resolution to capture local variability in topography, and thus to represent variations in elevation-dependent fields. This type of problem is well documented and remains a challenge in many fields of study that must rely on low-resolution forecast fields to study the plausible impacts of climate change on ecosystems (e.g., Trivedi et al. 2008). Thus, it remains difficult to evaluate how ecosystems may react to a warming atmosphere because of the large differences in scale between climate forecasts and the scale at which ecosystems are affected, especially in complex montane environments where landscape-driven variability can be important (e.g., Fridley 2009).

The adaptation method used in this study is basic and allows for rapid and thus inexpensive downscaling of large-scale low-resolution operational or climate forecast to finescale forecast. The method can be used in complex terrain provided the following two conditions are met. First, the low-resolution forecast model and the observation records should have roughly the same variation of temperature with elevation differences (i.e., lapse rate). Second, the probability density distribution of the observed and large-scale forecast temperatures should be similar where the downscaling is performed. That is, forecasts of low-resolution (i.e., operational or climate forecasts) low-level temperature at high-elevation grid points should have a probability density distribution with shape (i.e., standard deviation, skewness, and kurtosis) similar to those observed at high-elevation stations (within the region). The same should hold for low-lying observation and forecast records.

Hourly GEM-15km and -33km mean temperature difference with elevation separation are shown in Fig. 4 along with the observed mean difference. The black line, the lapse rate of 6.0 K km−1 fitted to the observation record, agrees well with the forecast GEM-15km and -33km values. Thus, the first condition is met, given hourly data is used to set the lapse rate.

GEM-15km and -33km screen-level temperature forecasts both reproduce observed temperature patterns (i.e., standard deviation, skewness, and kurtosis) within a given elevation range. This is illustrated in Fig. 6. The left column, top panel, shows histograms of the hourly observed and GEM-15km forecast (1 November 2008 to 1 April 2009) screen-level temperatures at Whistler, a high-elevation station. The right column shows the same but for Sandheads CS, a low-elevation station. Since both the distributions and the mean temperature differences with elevation separation of the low-resolution forcing fields and observations are similar, the lapse rate of 6.0 K km−1 should explain the bulk of the differences between the observed and forecast values and meets the second condition. We caution that it will not account for small-scale variations associated with landscape factors such as shading from vegetation or soil water content. We use GEM-SURF, the system described in section 3, to further our local refinement and also account for local variability driven by factors other than orography (i.e., surface characteristics associated with land use/land cover such as albedo, thermal characteristics, and roughness).

Fig. 6.
Fig. 6.

Histograms of observed and forecast temperatures and temperature daily extremes (K) about their daily mean at (left) VOA, a high-elevation station, and (right) WVF, a station at sea level (note the change in x axis scale). The observed temperatures are shown in black and the forecast GEM-15km temperatures are in red. (top) Hourly values, (middle) daily maxima about their daily mean, and (bottom) the same, but for minima.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

We also note that several studies have made use of lapse rates fitted to daily extremes for downscaling purposes (e.g., Dodson and Marks 1997; Stahl et al. 2006). In the complex VO2010 mountainous region, the range of observed surface hourly temperatures can be very different between near–sea level and high-elevation stations. This is illustrated in Fig. 6, with histograms of hourly temperature plotted about each daily mean temperature (i.e., for each day, the daily mean is removed from each hourly observation). Note how the range of minimum and maximum temperatures at the high-elevation station are several degrees larger than those at the low-level station. This altitudinal variability is typical within the region of interest and prevents us from fitting our lapse rates to daily extreme values. The result is that lapse rates fitted to daily extremes are not portable to our hourly data and cannot be used for our region of interest.

a. Downscaling method

We apply an adjustment for elevation differences (a basic form of downscaling) to the temperature forcing fields as follows:
e1
where Thr is the adapted temperature for the high-resolution grid; Tlr is the temperature from the low-resolution GEM-15km or -33km model, cubically interpolated to the high-resolution grid; δz is the elevation difference; and γ is the lapse rate. This type of height-based correction is simple but is known to work well. For example, similar lapse-rate corrections were used until very recently to postprocess operational temperature forecasts at the Met Office (e.g., Sheridan et al. 2010). Other recent studies (e.g., Barstad et al. 2009) have also used this type of elevation-based correction. We note that comparison of improvements that stem from the lapse-rate correction and improvements in other studies is not carried out since it amounts to comparing model errors plus or minus differences in lapse rate (under comparable differences between low- and high-resolution orography). Finally, we stress that the lapse-rate correction, combined with other corrections mentioned below, only concern the preparation of forcing fields used to drive the high-resolution near-surface and land surface model developed to refine forecasts for the Olympic Games. Hence, the final near-surface and land surface forecasts are the product of a numerical integration.
Pressure is adapted as follows:
e2
where Phr is the adapted high-resolution pressure, Plr is the pressure from the low-resolution models cubically interpolated to the high-resolution grid, g is the gravitational constant, and R = 287.05 J kg−1 K−1 is the gas constant for dry air.

Relative humidity is cubically interpolated to the high-resolution grid and subsequently converted to specific humidity using the newly corrected temperature and pressure fields.

Precipitation rate is linearly interpolated to the high-resolution grid and filtered to smooth the transition between adjacent grid boxes. The precipitation phase is adjusted as a function of temperature: for below-freezing temperatures, precipitation is snow, and for above-freezing temperatures, precipitation is rain. Thus as a valley (or peak) becomes better resolved, a given land surface area may find itself below (or above) the freeze–thaw line and receive rain (snow) instead of snow (rain), therefore leading to thawing (snow accumulation) as opposed to snow accumulation (thawing).

Incident radiation is linearly interpolated to the high-resolution grid. At the time this work was produced, the direct and diffused components of the solar radiation incident at the surface were not available from MSC’s operational models and were thus not corrected for elevation changes. These components have since become available as separate variables. Work is under way to include them in the downscaling process, allowing for elevation, slope, and shading effects.

Winds are cubically interpolated to the high-resolution grid. Obviously, channeling effects in valleys should be considered when interpolating the wind fields to the high-resolution model grid. At this point, they are not. This limitation essentially results from the gridpoint downscaling approach in place. To account for channeling effects, downscaling should be applied as a result of domain changes as opposed to local (i.e., grid point) differences, which allow no information from surrounding points. Another limiting factor is that our driving model wind fields suffer from orographic blocking (Zadra et al. 2003), a form of drag applied to low-level winds to mimic underlying mountain drag. The blocking leads to improved short- and medium-range forecasts downstream of the blocking, especially in winter (Zadra et al. 2003). However, once the first model-level winds are interpolated to the surface based on surface layer stability functions, resulting surface winds are too weak. This worsens a known problem of decoupling between the surface and the atmosphere and favors cold temperature biases (particularly at night under clear-sky conditions). In an effort to minimize this numerical problem, the minimum 10-m wind speeds are set to 2.5 m s−1 (the same value is used in the operational models). Work is under way to improve surface winds by adding vertical levels to the driving model and running it at a much higher resolution (test cases are currently conducted using 1-km and 250-m grid spacing, albeit over much smaller domains and shorter forecast range because of resource constraints).

b. Evaluation of the downscaling method

In this section, we evaluate GEM-15km and -33km forecasts of screen-level temperature and pressure at the surface (pre- and postdownscaling). Examples of the applied corrections are presented in Fig. 7. The top two panels show the observed (black), forecast (blue), and downscaled forecast (red) temperatures at Squamish Airport (WSK). At this location, both forecast models have station elevation differences on the order of 900 m (which corresponds to forecast temperature corrections of roughly 5.6 K). The resulting downscaled temperatures (Fig. 7, top two panels) are much closer than those initially forecast. The bottom two panels of Fig. 7 show the observed (black), forecast (blue), and downscaled forecast (red) pressures. Here, the pressure correction exceeds 100 hPa. The resulting pressure fits the observation records well, with pressure biases reduced by 104 and 109 hPa for GEM-15km and -33km, respectively.

Fig. 7.
Fig. 7.

Observed and forecast hourly screen-level temperature (K) and surface pressure (hPa) at Squamish Airport. The black lines show observations. The blue lines show (top row),(third row) GEM-33km and (second row),(bottom row) GEM-15km operational forecasts. The red lines show the downscaled operational forecasts (i.e., corrected for elevation differences; see section 4 for details).

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

The spatial variability of the correction applied across the entire grid is illustrated in Fig. 8. The fields shown are valid at 0000 UTC, 3 November 2008. The columns show the low-resolution fields (GEM-15km), the high-resolution fields, and their differences. The rows are orography, temperature, and pressure. The difference in orography between the two models is typically several hundred meters but can exceed 1000 m around peaks and valleys. The resulting changes in the temperature fields (second row) are thus typically of a few degrees except near peaks and valleys where temperature changes can exceed 5°C. Similarly, pressure changes (bottom row) are typically a few tens of hPa but can exceed 100 hPa near peaks and valleys.

Fig. 8.
Fig. 8.

(top) Elevation (m), (middle) screen-level temperature (°C), and (bottom) surface pressure (hPa) fields. (left) The low-resolution GEM-15km forecast fields valid at 00 UTC 3 Nov 2008 interpolated to the high-resolution grid, (middle) the high-resolution downscaled fields, and (right) differences (high minus low resolution).

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

The spatial signature of the correction highlights the importance of the adaptation in regions where peaks and valleys are close enough that the low-resolution model cannot resolve the underlying variability in orography (Fig. 8, top panels). It is in the vicinity of these areas that orographic departures are largest. The temperature bias correction is typically smaller at alpine stations and larger in low-lying valleys and at coastal stations located close to the mountains. In the alpine region (i.e., for stations located above 500 m), the correction is roughly 0.5°C for both models, whereas the low-lying areas have corrections of roughly 1.5° and 2.2°C, respectively. Mean biases and standard deviations of the errors based on all available stations (i.e., Olympic network, standard climate, and snow pillow stations) are listed in Table 4. In this table, results for Day 1 are for GEM-15km forecasts for hours 1 to 24, put back to back to cover the forecast period of 1 November 2008 to 1 April 2009. Day 2 indicates that observations are compared to forecast hours 25 to 48 valid at observation times. Day 3 and Day 4 are the same but for GEM-33km and forecast hours 49–72 and 73–96, respectively.

Table 4.

Screen-level temperature (K), relative humidity (%), and pressure (hPa) forecast error (observed minus forecast) statistics. The columns indicate (left to right) forecast model, mean of forecast errors for days 1 to 4, and mean of standard deviation of the forecast error for days 1 to 4.

Table 4.

The pressure bias is also greatly improved by the downscaling process. Again, the correction is largest at low-lying stations where mean biases are improved by about 40 and 50 hPa for GEM-15km and -33km, respectively. At most stations, the leftover pressure bias is a few hPa, thereby validating the orography of the high-resolution model. Mean error statistics computed using all available stations are given in Table 4.

The humidity error statistics are slightly improved by the downscaling process. GEM-15km mean bias errors are reduced from roughly −0.65% to 0.33% over the first day and −0.95% to 0.03% for the second day. GEM-33km mean bias errors are reduced by about 1.2% for days 3 and 4 (Table 4).

Snow depth forecasts are not directly improved by the downscaling method discussed in this section. In the next section, GEM-SURF, the high-resolution near-surface and land surface forecast model, is used to carry out the further refinements necessary to improve snow forecasts.

5. Validation of GEM-SURF forecasts

In this section, the high-resolution near-surface and land surface model forecasts are validated against observations. GEM-SURF forecasts are also compared to downscaled operational forecasts (i.e., not the same as those used to drive GEM-SURF; see Fig. 2). Thus, the comparison of these surface forecasts is effectively a measure of differences due to the refinement of the surface.

a. Snow depth

Improved snow depth forecasts was the main objective of the high-resolution forecast system presented here. GEM-SURF snow forecasts were found to greatly improve on current operational systems. This is illustrated in Fig. 9 for six representative stations. The left column is low-lying stations, and the right column is alpine stations. The top row shows stations from the Olympic network where measurements are hourly. The middle row shows stations from the climate network where measurements are daily. The bottom row shows results at a snow course and a snow pillow station where measurements were made a few times a season and hourly, respectively. At low-lying stations (Fig. 9, left column), it is the intermittency of the presence of a snow cover that is better captured by GEM-SURF. There, the refinement of the surface leads to more realistic periods of accumulation and melt. The gain in forecast skill is also evident within the alpine region where snow depth errors at the end of the forecast period are systematically reduced and can exceed 3 m at some stations (e.g., Orchid Lake; Fig. 9, bottom right).

Fig. 9.
Fig. 9.

Observed and forecast snow depth (cm) from 1 Nov 2008 to 1 Apr 2009 for stations that are part of the (top) Olympics network (Table 1) and (middle) climate network (Table 2). (bottom) Results for (left) snow courses and (right) snow pillow stations (Table 3). The black stars are for observations. The blue and green lines are for the operational forecasts from GEM-15km and -33km, respectively. GEM-SURF forecasts are plotted in red.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

We were encouraged by the results at individual stations. This prompted us to produce maps of the snow cover and snow depth for each month of the winter 2008/09 (Fig. 10). In Fig. 10, stations shown in Fig. 9 are marked with the colored dots. In November, low-lying areas are snow free while high-elevation areas are already covered by snow. In low-lying areas, the intermittent nature of the snow cover is illustrated by the evolution of the snow cover from January to March. At high altitudes, snow covers most areas and slowly extends to valleys. Snow depths also increase throughout the winter until late March, when maximum snow depth for the winter discussed here were attained (based on observation records available for April and May). Thereafter, the melt season starts and valleys begin to clear up (Fig. 10, April, bottom right).

Fig. 10.
Fig. 10.

Snow cover and snow depth forecasts valid 1 Nov 2008, 1 Dec 2008, 1 Jan 2009, 1 Feb 2009, 1 Mar 2009, and 1 Apr 2009. The color bar is elevation (ME; m) or, when present, snow depth (SD; m). The dots show the station location for time series plotted in Fig. 9. The Olympic network stations are shown in red, the standard climate stations are shown in green, and the snow pillow stations are shown in yellow.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

b. Temperature forecasts

Mean temperature forecast errors were calculated. Figure 11 (left column) shows histograms of observed minus forecast values for 35 stations (the 31 stations listed in Table 1 and the 4 snow pillow stations with automated weather stations listed in Table 3). Day 1 forecasts are the first 24 h of the 96 h forecasts put back to back. Day 2 are forecasts valid from 25 to 48 h. Days 3 and 4 are for hours 49–72 and 73–96, respectively. GEM-15km and GEM-SURF show very similar error statistics, indicating that the gain results primarily from the downscaling process.

Fig. 11.
Fig. 11.

Histograms of (left) screen-level air temperature (K) and (right) relative humidity (%) forecast errors. Day 1 means all 1–24-h forecasts valid at the time of the observations. Day 2 is for hours 25–48, day 3 for hours 49–72, and day 4 for hours 73–96.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

At this point, it becomes important to mention a correction allowed within GEM-15km’s processing of screen-level temperatures: an imposed maximum lapse rate is used to limit surface temperature minima. The correction is only applied at the surface (i.e., it is not applied to the forcing fields). GEM-SURF has no imposed limitation on the evolution of the surface temperature fields. The result is an overall cold bias that is slightly larger than that of GEM-15km. This is illustrated in Fig. 12. The observations are shown in black for the period 1 November 2008 to 1 April 2009. The forecasts are collated 24-h forecasts. The blue line is the downscaled GEM-15km forecasts and the red line is GEM-SURF forecasts. The top panel shows that the frequency-dependent structure of the forecast screen-level temperature compares well with the observations for the slowly varying component of the signal and generally reproduces the high-frequency signal well, although we note a tendency to overestimate the daily temperature range. Short periods of much colder temperature minima are forecast with GEM-SURF. They occur at times when the constraint is imposed on the operational surface forecast temperatures. The mean and the standard deviation of GEM-15km forecast errors (Fig. 12, bottom panel) are both slightly smaller than those of GEM-SURF. However, if periods when the correction to GEM-15km is applied are neglected, it is GEM-SURF that slightly improves on GEM-15km. The same is true for the daily minima. The result is not surprising (since improvements were not made to the model itself) although encouraging since, unlike GEM-15km no data assimilation is allowed in GEM-SURF. Hence, fields such as ground temperatures and soil moisture are allowed to depart from analyses. It is expected that improvements will be noted once downscaling of the radiative fluxes are applied, data assimilation is implemented, and improvements of the forcing wind fields remove the constraints of orographic blocking.

Fig. 12.
Fig. 12.

Observed and forecast screen-level temperatures at station VOB. (top) Observations are in black, GEM-15km in blue, and GEM-SURF in red. (bottom) Forecast error for GEM-15km and GEM-SURF.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

c. Relative humidity

Figure 5 shows time series of the observed and forecast relative humidity at Blackcomb Base. The forecast error for GEM-SURF and GEM-15km is shown in the bottom panel. In the alpine region, over snow-covered areas, both GEM-SURF and GEM-15km tend to forecast dryer-than-observed conditions. Both models also have a tendency to predict more humid conditions during the melting period. In the alpine regions, GEM-SURF outperforms GEM-15km, whereas in low-lying and coastal areas, it is GEM-15km that performs best over GEM-SURF (not shown). It is suspected that much of this is due to coastal GEM-SURF tiles being defined as 100% land as a result of the high resolution with no information about the surrounding waters (GEM-15km has partial water and land tiles) over most of the low-lying-land-only tiles of GEM-SURF for which we have observations. Mean humidity forecast errors were also calculated. Figure 11 (right column) shows histograms of observed minus forecast values for 35 stations. GEM-15km and GEM-SURF show very similar error statistics, indicating that the gain in humidity results primarily from the downscaling process.

6. Beyond snow forecasts

This section provides a brief overview of the high-resolution information a forecast system, such as the one presented in this paper, can provide on other surface fields for which we often have little to no reliable observation readily available. We also briefly discuss how improvements in our modeling skills may impact low-resolution models once they are allowed to evolve in a coupled mode using three snow fields (snow depth, snow density, and snow albedo) for illustration purposes (Fig. 13). They are the fields forecast by GEM-15km (left column) and GEM-SURF (right column), valid for 00 UTC 30 March 2009.

Fig. 13.
Fig. 13.

(top) Snow depth (cm), (middle) snow density (kg m−3), and (bottom) snow albedo forecasts valid at 00 UTC 30 Mar 2009 for (left) GEM-15km and (right) GEM-SURF.

Citation: Journal of Hydrometeorology 12, 4; 10.1175/2011JHM1250.1

As expected, snow depths (Fig. 13, top row) are fairly uniform in the low-resolution model where peaks and valleys are not resolved. In contrast, the snow depths of the high-resolution model follow orographic features and are known to much better represent the observed snow cover (section 5a) based on evaluation of forecast snow depth at all observation-record locations.

Surface fields such as snow density (Fig. 13, middle row), liquid water stored in the snow, and snow albedo (Fig. 13, bottom row, which is temperature, density, and age dependent) are similarly expected to be better represented by the high-resolution model. The elevational variation in snow is also reflected in the albedo. Differences between the low- and high-resolution versions are considerable. It is likely that allowing for a more spatially representative albedo in the low-resolution models (based on the high-resolution runs) could contribute to improved low-resolution forecasts.

The ability to forecast snowpack spatial variability (including differences between mountain tops and valleys and the evolution of the snowpack in adjacent forested and open-site areas) is of interest for a region plagued with pine beetle outbreaks where clear cuts are performed in an effort to control the invasion (Jackson and Prowse 2009). One interesting aspect of the model presented here is the ability to easily modify the underlying vegetation type. Large clear-cut areas and large recently burned areas can rapidly be included in the high-resolution model such that forecasts and snow energy budgets can be adapted to the changing conditions as they happen.

Finally, we note that in a region where snowmelt contributes some 50%–80% of the annual streamflow (Stewart et al. 2004), massive changes in total forecast snow quantities (such as those forecasts by the low- versus high-resolution models) can have a major impact on estimates of liquid water stored in the form of snow and thus on our ability to forecast the hydrological cycle, especially at lower elevation where high-elevation snowmelt is the primary water source. It is hoped that improvements to hydrological modeling in this complex region will follow from this high-resolution near-surface modeling effort.

7. Discussion and future work

A new 2D near-surface and land surface forecast system, GEM-SURF, was used to refine the low-resolution forecast, thus allowing for improved representation of small-scale structures. The model was driven by the forecast fields issued for winter 2008/09. Results showed that the surface model is particularly useful for snow forecast and orographic adjustment of screen-level air temperatures and pressures.

The simple downscaling method for orographic adjustments between low-resolution forecasts and high-resolution GEM-SURF grid was presented and validated using surface weather stations. Results show that the simple downscaling method is capable of refining low-resolution fields enough to correct for the bulk of biases that result from smoothed orography in fields that are highly altitudinally sensitive (e.g., temperature). The technique was particularly useful in the complex alpine region where peaks and valleys were often not represented by the low-resolution forecast models. Screen-level temperatures and surface pressures, following downscaling, were found to agree well with observed values. Given that the downscaling method was successfully validated for screen-level temperatures and surface pressures, maps of the extent of the correction over the Olympics region were produced (Fig. 8). The maps highlight the importance of the correction along rapidly changing orography where small areas may include large elevation differences. Results demonstrate that the operational model outputs for this region should be upgraded to at least include GEM-SURF-based postprocessing in the form of downscaling.

Snow depth forecasts were shown to be improved in both the alpine and the low-lying regions. The improvements are primarily due to the elevation refinement of the surface temperature fields, which allows a better localization of the freeze–thaw line and thus of the precipitation phase (i.e., rain or snow) and of the growth or melt periods. This significant improvement has prompted us to produce high-resolution snow coverage and snow depth maps for the region. The maps show the intermittent presence of the snow cover at low elevations. At high elevations, the maps show thinner depths in valleys and large areas with peak snow depths in excess of 3 m. This agrees well with the high spatial correlation found in the observation records for stations located at high altitudes.

Maps of the spatial variability of snow density, snow albedo, and snow depth were shown for the low-resolution operational model and GEM-SURF (Fig. 13). The differences are quite significant. It is expected that allowing for a feedback into GEM-15km of the snow albedo from the subgrid information available through GEM-SURF could also help further reduce operational forecast errors by improving the radiative budget. Similarly, it is expected that such high-resolution forecasts will prove valuable for hydrological modeling by providing more realistic estimates of available water stored in the form of snow. The high-resolution snow fields will also soon be used as first guess for the assimilation of terrestrial snow in the new land data assimilation system.

At the Meteorological Service of Canada, updateable model output statistics (UMOS) systems are used to statistically adjust numerical forecasts for specific locations (Wilson and Vallée 2003). Hence, given observations are available, it is possible to use UMOS to correct for elevational differences between the forecast model and observation station. The system unfortunately typically requires extensive training with observations before it can optimally be used to correct forecast fields (e.g., GEM-15km fields) and can only be used at observation station locations (i.e., it can only correct a point forecast as opposed to the full field). GEM-SURF does not require such training. Except for 6 weeks of spin-up time, it can be used to produce site-specific forecasts anywhere, regardless of the availability of observations. Another advantage of GEM-SURF over UMOS is that it provides local information on nonobserved variables.

The decoupling of the surface from the atmospheric model allows both to run at their own time steps and relevant resolutions. Given the improvements presented in this paper, the investigation of two-way interaction between the surface and the driving atmospheric models in the form of fluxes from below and forcing from above is now being planned.

Work is also currently underway to include and test snow surface temperature algorithms and downscaling radiative fluxes and the inclusion of slopes and shadow effects to account for mountain orientation. Another current research avenue is the development of a locally nested model to achieve a higher vertical and horizontal resolution of the driving forecast model. The primary goal of this research is the improvement of near-surface wind forecasts.

Acknowledgments

The authors wish to thank Alexandre Leroux, who has been instrumental in acquiring the high-resolution geophysical fields, and Maria Abrahamowicz for her contribution to the quality control of the Olympic network observation records. Special thanks are extended to Paul Vaillancourt for the internal review of this document. The authors also thank Jocelyn Mailhot and Ayrton Zadra for their constructive comments. Finally, thanks to the reviewers whose comments were very valuable.

REFERENCES

  • Barstad, I., , Sorteberg A. , , Flatøy F. , , and Déqué M. , 2009: Precipitation, temperature and wind in Norway: Dynamical downscaling of ERA40. Climate Dyn., 33, 769776, doi:10.1007/s00382-008-0476-5.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., , Crevier L.-P. , , Mailhot J. , , Bilodeau B. , , and Delage Y. , 2003a: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part I: Warm season results. J. Hydrometeor., 4, 352370.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., , Brown R. , , Mailhot J. , , and Bilodeau B. , 2003b: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part II: Cold season results. J. Hydrometeor., 4, 371386.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., , Roch M. , , Leduc A.-M. , , Vaillancourt P. , , Laroche S. , , and Mailhot J. , 2009: Medium-range quantitative precipitation forecasts from Canada’s new 33-km deterministic global operational system. Wea. Forecasting, 24, 690708.

    • Search Google Scholar
    • Export Citation
  • Bhumralkar, C. M., 1975: Numerical experiments on the computation of ground surface temperature in an atmospheric general circulation model. J. Appl. Meteor., 14, 12461258.

    • Search Google Scholar
    • Export Citation
  • de Goncalves, L. G. G., , Shuttleworth W. J. , , Burke E. J. , , Houser P. , , Toll D. L. , , Rodell M. , , and Arsenault K. , 2006: Toward a South America land data assimilation system: Aspects of land surface model spin-up using the simplified simple biosphere. J. Geophys. Res., 111, D17110, doi:10.1029/2005JD006297.

    • Search Google Scholar
    • Export Citation
  • Dodson, R., , and Marks D. , 1997: Daily air temperature interpolated at high spatial resolution over a large mountainous region. Climate Res., 8, 120.

    • Search Google Scholar
    • Export Citation
  • Douville, H., , Royer J.-F. , , and Mahfouf J.-F. , 1995: A new snow parameterization for the Météo-France climate model. Part I: Validation in stand-alone experiments. Climate Dyn., 12, 2135.

    • Search Google Scholar
    • Export Citation
  • Fridley, J. D., 2009: Downscaling climate over complex terrain: High finescale (<1000 m) spatial variation of near-ground temperatures in a montane forested landscape (Great Smokey Mountains). J. Appl. Meteor. Climatol., 48, 10331049.

    • Search Google Scholar
    • Export Citation
  • Grant, A., , and Mason P. , 1990: Observations of boundary-layer structure over complex terrain. Quart. J. Roy. Meteor. Soc., 116, 159186.

    • Search Google Scholar
    • Export Citation
  • Hartman, M. D., , Baron J. S. , , Lammers R. B. , , Cline D. W. , , Band L. E. , , Liston G. E. , , and Tague C. , 1999: Simulations of snow distribution and hydrology in a mountain basin. Water Resour. Res., 35, 15871603.

    • Search Google Scholar
    • Export Citation
  • Jackson, S. I., , and Prowse T. D. , 2009: Spatial variation of snowmelt and sublimation in a high-elevation semi-desert basin of western Canada. Hydrol. Processes, 23, 26112627.

    • Search Google Scholar
    • Export Citation
  • Liston, G. E., 2004: Representing subgrid snow cover heterogeneities in regional and global models. J. Climate, 17, 13811397.

  • Loth, B., , Graf H.-F. , , and Oberhuber J. M. , 1993: Snow cover model for global climate simulations. J. Geophys. Res., 98 (D6), 10 45110 464.

    • Search Google Scholar
    • Export Citation
  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44, 133149.

  • Mailhot, J., and Coauthors, 2010: Environment Canada’s experimental numerical weather prediction systems for the Vancouver 2010 Winter Olympic and Paralympic Games. Bull. Amer. Meteor. Soc., 91, 10731085.

    • Search Google Scholar
    • Export Citation
  • Noilhan, J., , and Planton S. , 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117, 536549.

    • Search Google Scholar
    • Export Citation
  • Sheridan, P., , Smith S. , , Brown A. , , and Vosper S. , 2010: A simple height-based correction for temperature downscaling in complex terrain. Meteor. Appl., 17, 329339, doi:10.1002/met.177.

    • Search Google Scholar
    • Export Citation
  • Slater, A. G., and Coauthors, 2001: The representation of snow in land surface schemes: Results from PILPS 2(d). J. Hydrometeor., 2, 725.

    • Search Google Scholar
    • Export Citation
  • Stahl, K., , Moore R. , , Floyer J. , , Asplin M. , , and McKendry I. , 2006: Comparison of approaches for spatial interpolation of daily air temperature in a large region with complex topography and highly variable station density. Agric. For. Meteor., 139, 224236.

    • Search Google Scholar
    • Export Citation
  • Stewart, I., , Cayan D. R. , , and Dettinger M. , 2004: Changes in snowmelt runoff timing in western North America under a ‘business as usual’ climate change scenario. Climatic Change, 62, 217232.

    • Search Google Scholar
    • Export Citation
  • Strack, J. E., , Liston G. E. , , and Pielke R. A. Sr., 2004: Modeling snow depth for improved simulation of snow–vegetation–atmosphere interactions. J. Hydrometeor., 5, 723734.

    • Search Google Scholar
    • Export Citation
  • Tribbeck, M. J., , Gurney R. J. , , and Morris E. M. , 2006: The radiative effect of a fir canopy on a snowpack. J. Hydrometeor., 7, 880895.

    • Search Google Scholar
    • Export Citation
  • Trivedi, M. R., , Berry P. M. , , Morecroft M. D. , , and Dawson T. P. , 2008: Spatial scale affects bioclimate model projections of climate change impacts on mountain plants. Global Change Biol., 14, 10891103.

    • Search Google Scholar
    • Export Citation
  • Wilson, L. J., , and Vallée M. , 2003: The Canadian Updateable Model Output Statistics (UMOS) system: Validation against perfect prog. Wea. Forecasting, 18, 288302.

    • Search Google Scholar
    • Export Citation
  • Zadra, A., , Roch M. , , Laroche S. , , and Charron M. , 2003: The subgrid-scale orographic blocking parametrization of the GEM model. Atmos.–Ocean, 41, 155170.

    • Search Google Scholar
    • Export Citation
Save