• Avissar, R., 1998: Which type of soil–vegetation–atmosphere transfer scheme is needed for general circulation models: A proposal for a higher-order scheme. J. Hydrol., 212–213, 136154, doi:10.1016/S0022-1694(98)00227-3.

    • Search Google Scholar
    • Export Citation
  • Avissar, R., and Schmidt T. , 1998: An evaluation of the scale at which ground-surface heat flux patchiness affects the convective boundary layer using large-eddy simulations. J. Atmos. Sci., 55, 26662689, doi:10.1175/1520-0469(1998)055<2666:AEOTSA>2.0.CO;2.

    • 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, doi:10.1175/1525-7541(2003)4<352:OIOTIL>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., Brown R. , Mailhot J. , Bilodeau B. , and Crevier L.-P. , 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, doi:10.1175/1525-7541(2003)4<371:OIOTIL>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Bernier, N., Bélair S. , Bilodeau B. , and Tong L. , 2011: Near-surface and land surface forecast system of the Vancouver 2010 winter Olympic and Paralympic games. J. Hydrometeor., 12, 508530, doi:10.1175/2011JHM1250.1.

    • Search Google Scholar
    • Export Citation
  • Best, M., Beljaars A. , Polcher J. , and Viterbo P. , 2004: A proposed structure for coupling tiled surfaces with the planetary boundary layer. J. Hydrometeor., 5, 12711278, doi:10.1175/JHM-382.1.

    • Search Google Scholar
    • Export Citation
  • Bicheron, P., and Coauthors, 2006: Globcover: A 300-m global land cover product for 2005 using ENVISAT MERIS time series. Proc. Second Int. Symp. on Recent Advances in Quantitative Remote Sensing, Valencia, Spain, University of Valencia–Global Change Unit, 538–542. [Available online at http://ipl.uv.es/raqrs/.]

  • Bontemps, S., Defourny P. , Bogaert E. V. , Arino O. , Kalogirou V. , and Perez J. , 2011: Globcover 2009: Products description and validation report. Tech. Rep., Université catholique de Louvain/European Space Agency, 53 pp. [Available online at http://due.esrin.esa.int/files/GLOBCOVER2009_Validation_Report_2.2.pdf.]

  • Carrera, M., Bélair S. , Fortin V. , Bilodeau B. , Charpentier D. , and Doré I. , 2010: Evaluation of snowpack simulations over the Canadian Rockies with an experimental hydrometeorological modeling system. J. Hydrometeor., 11, 11231140, doi:10.1175/2010JHM1274.1.

    • Search Google Scholar
    • Export Citation
  • Côté, J., Gravel S. , Méthot A. , Patoine A. , Roch M. , and Staniforth A. , 1998a: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395, doi:10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Côté, J., Desmarais J.-G. , Gravel S. , Méthot A. , Patoine A. , Roch M. , and Staniforth A. , 1998b: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 13971418, doi:10.1175/1520-0493(1998)126<1397:TOCMGE>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Dutra, E., Kotlarski S. , Viterbo P. , Balsamo G. , Miranda P. , Schär C. , Bissolli P. , and Jonas T. , 2011: Snow cover sensitivity to horizontal resolution, parameterizations, and atmospheric forcing in a land surface model. J. Geophys. Res., 116, D21109, doi:10.1029/2011JD016061.

    • Search Google Scholar
    • Export Citation
  • Erfani, A., Mailhot J. , Gravel S. , Desgagné M. , King P. , Sills D. , McLennan N. , and Jacob D. , 2005: The high resolution limited area version of the Global Environmental Multiscale Model (GEM-LAM) and its potential operational applications. 11th Conf. on Mesoscale Processes, Albuquerque, NM, Amer. Meteor. Soc., 1M.4. [Available online at http://ams.confex.com/ams/pdfpapers/97308.pdf.]

  • Essery, R., Best M. , Betts R. , Cox P. , and Taylor C. , 2003: Explicit representation of subgrid heterogeneity in a GCM land surface scheme. J. Hydrometeor., 4, 530543, doi:10.1175/1525-7541(2003)004<0530:EROSHI>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Fischer, E., Seneviratne S. , Vidale P. , Lüthi D. , and Schär C. , 2007: Soil moisture–atmosphere interactions during the 2003 European summer heat wave. J. Climate, 20, 50815099, doi:10.1175/JCLI4288.1.

    • Search Google Scholar
    • Export Citation
  • Girard, C., and Coauthors, 2014: Staggered vertical discretization of the Canadian Environmental Multiscale (GEM) model using a coordinate of the log-hydrostatic-pressure type. Mon. Wea. Rev., 142, 11831196, doi:10.1175/MWR-D-13-00255.1.

    • Search Google Scholar
    • Export Citation
  • Leroyer, S., Bélair S. , Mailhot J. , and Strachan I. , 2011: Microscale numerical prediction over Montreal with the Canadian external urban modeling system. J. Appl. Meteor. Climatol., 50, 24102428, doi:10.1175/JAMC-D-11-013.1.

    • Search Google Scholar
    • Export Citation
  • Leroyer, S., Bélair S. , Husain S. , and Mailhot J. , 2014: Subkilometer numerical weather prediction in an urban coastal area: A case study over the Vancouver metropolitan area. J. Appl. Meteor. Climatol., 53, 14331453, doi:10.1175/JAMC-D-13-0202.1.

    • Search Google Scholar
    • Export Citation
  • Mahfouf, J., Brasnett B. , and Gagnon S. , 2007: A Canadian Precipitation Analysis (CAPA) project: Description and preliminary results. Atmos.–Ocean, 45, 117, doi:10.3137/ao.450101.

    • Search Google Scholar
    • Export Citation
  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44, 133149, doi:10.3137/ao.440202.

    • Search Google Scholar
    • Export Citation
  • Marke, T., Mauser W. , Pfeiffer A. , and Zängl G. , 2011: A pragmatic approach for the downscaling and bias correction of regional climate simulations: Evaluation in hydrological modeling. Geosci. Model Dev., 4, 759770, doi:10.5194/gmd-4-759-2011.

    • Search Google Scholar
    • Export Citation
  • Masson, V., and Coauthors, 2013: The SURFEXv7.2 land and ocean surface platform for coupled or offline simulation of earth surface variables and fluxes. Geosci. Model Dev., 6, 929960, doi:10.5194/gmd-6-929-2013.

    • Search Google Scholar
    • Export Citation
  • Molod, A., and Salmun H. , 2002: A global assessment of the mosaic approach to modeling land surface heterogeneity. J. Geophys. Res., 107, 4217, doi:10.1029/2001JD000588.

    • 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, doi:10.1175/1520-0493(1989)117,0536:ASPOLS.2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Noilhan, J., and Mahfouf J.-F. , 1996: The ISBA land surface parameterization. Global Planet.Change, 13, 145159, doi:10.1016/0921-8181(95)00043-7.

    • Search Google Scholar
    • Export Citation
  • Pigeon, G., Moscicki M. , Voogt J. , and Masson V. , 2008: Simulation of fall and winter surface energy balance over a dense urban area using the TEB scheme. Meteor. Atmos. Phys., 102, 159171, doi:10.1007/s00703-008-0320-9.

    • Search Google Scholar
    • Export Citation
  • Polcher, J., and Coauthors, 1998: A proposal for a general interface between land-surface schemes and general circulation models. Global Planet. Change, 19, 261276, doi:10.1016/S0921-8181(98)00052-6.

    • Search Google Scholar
    • Export Citation
  • Salgado, R., and Moigne P. L. , 2010: Coupling of the FLake model to the Surfex externalized surface model. Boreal Env. Res., 15, 231244.

    • Search Google Scholar
    • Export Citation
  • Schomburg, A., Venema V. , Ament F. , and Simmer C. , 2012: Disaggregation of screen-level variables in a numerical weather prediction model with an explicit simulation of subgrid-scale land-surface heterogeneity. Meteor. Atmos. Phys., 116, 8194, doi:10.1007/s00703-012-0183-y.

    • Search Google Scholar
    • Export Citation
  • Seity, Y., Brousseau P. , Malardel S. , Hello G. , Bénard P. , Bouttier F. , Lac C. , and Masson V. , 2011: The AROME-France convective-scale operational model. Mon. Wea. Rev., 139, 976991, doi:10.1175/2010MWR3425.1.

    • Search Google Scholar
    • Export Citation
  • Seth, A., Giorgi F. , and Dickinson R. , 1994: Simulating fluxes from heterogeneous land surfaces: Explicit subgrid method employing the biosphere-atmosphere transfer scheme (BATS). J. Geophys. Res., 99, 18 65118 667, doi:10.1029/94JD01330.

    • Search Google Scholar
    • Export Citation
  • Yu, Z., 2000: Assessing the response of subgrid hydrologic processes to atmospheric forcing with a hydrologic model response. Global Planet. Change, 25, 117, doi:10.1016/S0921-8181(00)00018-7.

    • Search Google Scholar
    • Export Citation
  • Zabel, F., and Mauser W. , 2013: 2-way coupling the hydrological land surface model PROMET with the regional climate model MM5. Hydrol. Earth Syst. Sci., 17, 17051714, doi:10.5194/hess-17-1705-2013.

    • Search Google Scholar
    • Export Citation
  • Zabel, F., Mauser W. , Marke T. , Pfeiffer A. , Zängl G. , and Wastl C. , 2012: Inter-comparison of two-land surface models applied at different scales and their feedbacks while coupled with a regional climate model. Hydrol. Earth Syst. Sci., 16, 10171031, doi:10.5194/hess-16-1017-2012.

    • Search Google Scholar
    • Export Citation
  • View in gallery
    Fig. 1.

    Flowchart of the SPS. SPS is driven by hourly atmospheric fields (i.e., 40-m air humidity, air temperature, and wind; surface pressure; solar and longwave radiation; precipitation) produced by the 15-km version of the regional GEM model and downscaled from the lowest atmospheric level to the land surface. Initial condition for SPS is derived from the land surface analyses (e.g., soil moisture and surface temperature).

  • View in gallery
    Fig. 2.

    REG simulated with the SPS covering the central part of North America and colored by the 2.5-km terrain elevation field (m). White-lined boxes show the location of the local domains. (left) Mountains (Pacific Coast Ranges, British Columbia, LBC), (middle) prairies (Saskatchewan, LPR) and lakes (Manitoba, LLA), and (right) lowland forests (southern Quebec and Ontario, LQC).

  • View in gallery
    Fig. 3.

    Near-surface air temperature (K) simulated with SPS at 2.5-km horizontal resolution at 1200 UTC 27 Jul 2012. (top) The REG computational domain and (bottom) close-ups on the Canadian prairies and lakes (including the LPR and LLA local domains).

  • View in gallery
    Fig. 4.

    As in Fig. 3, but for the 25-km horizontal resolution case.

  • View in gallery
    Fig. 5.

    Schematic of the subgrid-scale variability for any surface variable (or any geophysical field) at a given forecast hour; comparison between the (left) 25-km reference case and the (right) 2.5-km high-resolution case. The large black dots (left) correspond to the grid points explicitly resolved at the 25-km resolution, the gray dots (right) correspond to the subgrid-scale grid points (explicitly resolved at the 2.5-km resolution), and the red dot corresponds to the mean value over the subgrid-scale grid points contained within the 25-km grid cell under consideration. The PDF of any surface variable over the subgrid-scale grid points can be statistically characterized by its first moments: mean value (red dot) and STD (spread of the gray dots).

  • View in gallery
    Fig. 6.

    Maps of (a) model deviation and (b) subgrid-scale STD for the simulated near-surface air temperature (K) over the REG computational domain at 1200 UTC 27 Jul 2012; comparison between the 2.5- (Fig. 3) and 25-km (Fig. 4) runs. Color bars correspond to air temperature variations (K).

  • View in gallery
    Fig. 7.

    Examples of factors influencing the near-surface air temperature model deviation shown in Fig. 6. (a) Model deviation for the terrain elevation (m) over the western part of the REG domain (including the LBC local domain). (b) The 40-m air temperature (°C) simulated by the regional version of GEM at 1200 UTC 27 Jul 2012 (including the REG computational domain).The terrain elevation scale runs from ±250 in increments of 50 m and the temperature scale runs from −12.1° to 32.8° in increments of ~1.0°C.

  • View in gallery
    Fig. 8.

    Time-evolving model deviation statistics at 1-h intervals and in the REG domain based on the 2.5-km run. Box plot over a 24-h period (27 Jul 2012) corresponding to (a) the near-surface air temperature (K) and (b) the upward surface sensible heat flux (W m−2). For one box, the span is 25% and 75% of the model deviation values, the vertical dashed line spans 5% and 95% of these values, and the solid line represents their median.

  • View in gallery
    Fig. 9.

    As in Fig. 8, but for the case of (a) the near-surface dewpoint temperature (K) and (b) the upward surface latent heat flux (W m−2).

  • View in gallery
    Fig. 10.

    Time-evolving subgrid-scale STD statistics at 1-h intervals from 25 to 29 Jul 2012 and in the REG computational domain based on the 2.5-km SPS run related to (a) the near-surface air temperature (K), (b) the surface sensible heat flux (W m−2), (c) the near-surface dewpoint temperature (K), and (d) the near-surface wind speed (m s−1). The black solid line represents the subgrid-scale STD median in REG, the red dashed line represents its mean value, and the black dashed lines correspond to one STD above and below the mean value.

  • View in gallery
    Fig. 11.

    As in Fig. 10, but for the simulated near-surface air temperature (K), in the following local domains: (a) LBC, (b) LPR, (c) LLA, and (d) LQC; comparison between the 25- and 2.5-km SPS runs.

  • View in gallery
    Fig. 12.

    As in Fig. 11, but for the case of the upward surface sensible heat flux (W m−2).

  • View in gallery
    Fig. 13.

    Time-evolving correlation of the near-surface temperature model deviation (K) with a subset of GlobCover/USGS geophysical field model deviation [i.e., terrain topography, roughness length, soil texture (percent of sand or clay in soil), glacier mask, land–water mask, LAI]; comparison between the 25- and 2.5-km cases in LBC, LLA, LPR, and LQC.

  • View in gallery
    Fig. 14.

    As in Fig. 13, but for the surface sensible heat flux model deviation (W m−2).

  • View in gallery
    Fig. 15.

    Time-evolving statistical metrics of the simulated near-surface air temperature (K), at 1-h intervals from 25 to 29 Jul 2012 and in the LBC local domain. Subgrid-scale STD statistics based on the (a) 2.5- and (b) 10-km runs, and model deviation statistics based on the (c) 2.5- and (d) 10-km runs (the 25-km run is the reference configuration for all these configurations). The black solid line represents the statistical metrics median, the red dashed line represents their mean value, and the black dashed lines correspond to one STD above and below the mean value.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 572 434 51
PDF Downloads 42 8 0

Subgrid-Scale Variability for Thermodynamic Variables in an Offline Land Surface Prediction System

Mélanie C. RochouxMeteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Mélanie C. Rochoux in
Current site
Google Scholar
PubMed
Close
,
Stéphane BélairMeteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Stéphane Bélair in
Current site
Google Scholar
PubMed
Close
,
Maria AbrahamowiczMeteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Maria Abrahamowicz in
Current site
Google Scholar
PubMed
Close
, and
Pierre PellerinMeteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Pierre Pellerin in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

This study presents a numerical analysis of the impact of the horizontal resolution on the forecast capability of the Canadian offline land surface prediction system (SPS; formerly known as GEM-Surf) forced by the 15-km Global Environmental Multiscale (GEM) atmospheric model. This system is used to quantify on a statistical basis the subgrid-scale variability of (near-)surface variables for 25-km grid spacing based on the 2.5- or 10-km SPS run at regional scale over the 2012 summer season. The model bias and the distributions characterizing the subgrid-scale variability drastically depend on the geographic areas as well as on the diurnal cycle. These results show the benefits of high-resolution land surface simulations to account for length scales that are more consistent with the scales at which the actual land surface balance is affected by the heterogeneous geophysical fields (i.e., roughness length, land–water mask, glacier mask, and soil texture). The model bias results highlight the potential of an SPS–GEM two-way coupling strategy for refining predictions near the surface through the upscaling of high-resolution surface heat fluxes to the coarser atmospheric grid spacing, with these fluxes being significantly different from those explicitly resolved at 25 km and featuring nonlinear behavior with respect to the horizontal resolution. Since the computational power of meteorological operational centers progressively increases, making it possible to run high-resolution limited-area models, solving the surface at high resolution in a surface–atmosphere fully coupled system becomes a key aspect for improving numerical weather and environmental forecast performance.

Denotes Open Access content.

Corresponding author address: Mélanie C. Rochoux, Meteorological Research Division, Environment Canada, 2121 Trans-Canada Highway, Dorval, QC H9P 1J3, Canada. E-mail: melanie.rochoux@graduates.centraliens.net

Abstract

This study presents a numerical analysis of the impact of the horizontal resolution on the forecast capability of the Canadian offline land surface prediction system (SPS; formerly known as GEM-Surf) forced by the 15-km Global Environmental Multiscale (GEM) atmospheric model. This system is used to quantify on a statistical basis the subgrid-scale variability of (near-)surface variables for 25-km grid spacing based on the 2.5- or 10-km SPS run at regional scale over the 2012 summer season. The model bias and the distributions characterizing the subgrid-scale variability drastically depend on the geographic areas as well as on the diurnal cycle. These results show the benefits of high-resolution land surface simulations to account for length scales that are more consistent with the scales at which the actual land surface balance is affected by the heterogeneous geophysical fields (i.e., roughness length, land–water mask, glacier mask, and soil texture). The model bias results highlight the potential of an SPS–GEM two-way coupling strategy for refining predictions near the surface through the upscaling of high-resolution surface heat fluxes to the coarser atmospheric grid spacing, with these fluxes being significantly different from those explicitly resolved at 25 km and featuring nonlinear behavior with respect to the horizontal resolution. Since the computational power of meteorological operational centers progressively increases, making it possible to run high-resolution limited-area models, solving the surface at high resolution in a surface–atmosphere fully coupled system becomes a key aspect for improving numerical weather and environmental forecast performance.

Denotes Open Access content.

Corresponding author address: Mélanie C. Rochoux, Meteorological Research Division, Environment Canada, 2121 Trans-Canada Highway, Dorval, QC H9P 1J3, Canada. E-mail: melanie.rochoux@graduates.centraliens.net

1. Introduction

In numerical environmental prediction (NEP) systems, adequate representation of the small-scale variability of meteorological features often observed near the earth’s surface requires considerable horizontal resolution (i.e., subkilometer-scale resolution), especially in mountains (Carrera et al. 2010; Bernier et al. 2011), in coastal and lake regions (Salgado and Moigne 2010), and over urban areas (Pigeon et al. 2008; Leroyer et al. 2011).

At the operational level, numerical weather prediction (NWP) systems rely on an inline surface–atmosphere coupled system (Polcher et al. 1998; Best et al. 2004), in which the surface and the atmosphere are explicitly resolved using the same horizontal grid spacing. Running fully three-dimensional atmospheric models at subkilometer scale for operational global and regional systems is, however, not feasible with the current and upcoming computational power. The horizontal grid spacing of the inline systems such as the Global Environmental Multiscale (GEM) model (Côté et al. 1998a,b; Girard et al. 2014) at Environment Canada (EC) remains on the order of 10–25 km for regional and global deterministic prediction systems. Only local limited-area models (LAM) are on the way to achieve kilometer-scale simulations; see EC’s 2.5-km GEM–LAM (Erfani et al. 2005), Météo-France’s 2.5-km Applications of Research to Operations at Mesoscale (AROME) system (Seity et al. 2011), or NOAA’s 3-km High-Resolution Rapid Refresh (HRRR) model.1 In these inline systems, a tiling approach is used as a first-order approximation to represent the subgrid-scale variability of the surface fluxes, partly resulting from a nonlinear response of the surface parameters to the subgrid-scale heterogeneity (Essery et al. 2003). While computationally efficient, this statistical approach does not account for the subgrid-scale distribution of the surface heterogeneity, which may be valuable for the representation of surface–atmosphere interactions.

A persisting challenge in meteorological and climate modeling applications remains to increase the resolution of the surface prediction system in order to properly capture both spatial and temporal heterogeneity in the different surface covers (i.e., land, urban, water, continental ice, and sea ice). At the research level, horizontal grid spacing on the order of 250 m has been recently achieved for large computational domains and relatively long time integration with external land surface modeling systems such as the Canadian offline land surface prediction system (SPS; Carrera et al. 2010; Bernier et al. 2011; Leroyer et al. 2011) at Environment Canada or the French Surface Externalisée (SURFEX; Masson et al. 2013) at Météo-France. In these external systems, the land surface scheme is solved at high resolution in an offline mode, that is, separate from the three-dimensional atmospheric model run at coarser resolution (in contrast to the inline mode used at operational level). It is then driven by meteorological fields (i.e., air temperature, humidity, and pressure; wind; precipitation; solar and infrared radiation) downscaled to the horizontal grid spacing of the surface prediction system based on the tiling approach. The improved performance of SPS compared to the inline regional NWP operational system has been demonstrated against in situ and remotely sensed observations, in both summer and winter conditions (Carrera et al. 2010; Bernier et al. 2011; Leroyer et al. 2011).

These offline systems are, however, currently limited to the one-way coupling approach; the surface fluxes of heat, moisture, and momentum not being fed back to the atmosphere. The response of the atmosphere [the development of the planetary boundary layer (PBL) is, for instance, affected by the land surface as well as cloud formation and the initiation of convective processes] and the potential feedback on the (near-)surface processes are therefore neglected. While an advanced two-way coupling approach based on an upscaling of the surface fluxes to the lowest atmospheric levels best answers this modeling limitation, a preliminary step consists of examining the subgrid-scale variability of the near-surface meteorological variables and the subsequent spatial patterns of the high-resolution surface fluxes at the scale of an atmospheric model grid cell (Seth et al. 1994). Characterizing and better understanding the underlying heterogeneity is believed to be of great value prior to designing the upscaling procedure in the forthcoming two-way mode.

No extensive numerical study has yet been performed to evaluate and analyze on a systematic and statistical basis the model bias that is induced by coarse horizontal resolution as well as the subgrid-scale variability that is explicitly resolved at high resolution but not accounted for at coarser resolution, partly because of a degraded representation of the geophysical properties. For instance, Dutra et al. (2011) show the drastic impact of horizontal resolution on snow cover in complex terrain regions, but the approach was not generalized to other land surface processes. Additionally, Leroyer et al. (2014) show the positive impact of subkilometer-scale surface–atmosphere representation to predict the finescale interactions between the sea-breeze propagation, the terrain orography, and the urban environment with the Canadian inline system over the local urban coastal area of Vancouver (British Columbia, Canada). Leroyer et al. (2014) also highlight the issue of representativity of single-point measurements when comparing model simulations with observations, emphasizing the need to better quantify the gain in information obtained when running the land surface scheme at higher horizontal resolution. In the context of regional climate simulations, Marke et al. (2011) demonstrate that accounting for the subgrid-scale variability reduces simulation bias in hydrological applications, partly because of a better representation of soil moisture and precipitation (Yu 2000; Fischer et al. 2007). Similar improvements are expected in NEP and NWP applications if the subgrid-scale variability at the land surface is explicitly resolved.

This study aims to quantify both the model bias and the subgrid-scale variability of the SPS near-surface meteorological variables (i.e., 2-m air temperature and 2-m dewpoint temperature) and upward surface heat fluxes (i.e., surface sensible and latent heat fluxes) due to different choices of horizontal grid spacing (25, 10, and 2.5 km) over a regional domain covering the central part of North America. These grid spacings were selected as representing the typical resolutions of current operational NWP systems at global and regional scales, since an improved surface description is expected to be worthwhile for medium- or long-range prediction. The only sources of differences between these three simulation cases are the resolution and associated spatial variability of the geophysical fields on the one hand, and the downscaling procedure of the atmospheric forcing on the other hand. In this regard, this study addresses two main scientific issues: 1) characterizing the dependency of the subgrid-scale variability on the geographic areas and on their specific geophysical properties, in order to identify the most sensitive forcing parameters in SPS with respect to the horizontal resolution, and 2) evaluating the potential contribution of high-resolution land surface modeling in view of the forthcoming two-way mode, that is, analyzing how subgrid-scale information induce changes in the surface heat fluxes considered by the atmospheric model.

To this end, the present analysis statistically characterizes the subgrid-scale probability density function (PDF) of the screen-level and surface variables for 25-km grid spacing based on the 2.5- or 10-km SPS simulations over the 2012 summer season, in terms of hourly, daily, and monthly evolution. The main objective is to identify typical patterns of the PDF moments over the diurnal (day–night) cycle, in humid and arid zones, in mountains and over lowlands, as well as the dominant geophysical properties responsible for these patterns. These patterns are representative of the information that is not captured by the current global and regional NWP systems used at operational level, and subsequently of the level of uncertainties that translates into the model outputs and propagates in the prediction systems over time. Note that this analysis does not go down to the 250-m grid spacing. Instead, the objective is to quantify the differences in the representation of the surface and in the representation of the PBL between the global (or regional) version of GEM and the 2.5-km GEM–LAM. Note also that this analysis focuses on summertime since the main changes in the thermodynamic variables found when increasing the horizontal resolution occur over the warm season, when the high evapotranspiration rate strongly affects atmospheric processes. In contrast, the snow cover is expected to smooth out the spatial variability of the land surface variables in wintertime.

The paper is organized as follows. The Canadian offline land surface prediction system is presented in section 2 along with the simulation case. Section 3 describes the statistical metrics used to analyze the model bias and the subgrid-scale variability. The effect of the horizontal grid spacing on these statistical metrics is studied in section 4 for the near-surface (or screen level) meteorological variables and the surface heat fluxes over the 2012 summer season.

2. SPS, a state-of-the-art land surface prediction system

EC’s Meteorological Research Division is currently developing a land surface prediction system (formerly known as GEM-Surf) operating in an offline mode (Carrera et al. 2010; Bernier et al. 2011; Leroyer et al. 2011). SPS consists of four main components: 1) geophysical land surface characteristics, 2) GEM atmospheric forcing (i.e., surface pressure, near-surface air temperature and humidity, wind, solar and longwave radiation, and precipitation), 3) initial conditions for the land surface model (e.g., soil moisture and surface temperature), and 4) the land surface scheme to simulate the temporal evolution of the soil and surface processes. This offline system currently relies on a one-way approach, whereby the atmospheric forcing interacts with the land surface processes without feedback on the state of the atmosphere.

a. Land surface modeling system

In SPS the surface heterogeneity is represented at subgrid scales as an aggregation of different covers (i.e., land, urban, water, continental ice, and sea ice). Each type of cover (or tile) features specific input parameters and parameterization schemes to simulate the related physicochemical processes (Avissar and Schmidt 1998; Molod and Salmun 2002; Essery et al. 2003). In this study, the land surface processes are represented through the well-established Interactions between Soil, Biosphere, and Atmosphere (ISBA) scheme (Noilhan and Planton 1989; Noilhan and Mahfouf 1996), in which the land tile parameters account for the contribution of the subgrid-scale terrain heterogeneity (e.g., bare soil, permanent snow, and vegetation); cities are not represented since the focus is on the natural land surface. The Canadian version of ISBA (Bélair et al. 2003a,b) solves the surface balance using a two-layer force-restore approach, in which the land surface consists of a soil superficial layer and a rooting-depth layer, in order to represent the thermal and hydrological exchanges between the soil and the atmosphere. It also solves the lower boundary conditions for the vertical diffusion of temperature, moisture, and momentum. The corresponding sensible, latent, and momentum surface fluxes are derived from an area-weighted aggregation of the fluxes over the different surface covers. This tile approach is based on flux averaging (rather than on parameter averaging) due to the strong nonlinearities in the land surface model, in particular with regard to the latent and sensible heat fluxes (Seth et al. 1994; Avissar 1998; Schomburg et al. 2012).

Note that all land surface schemes in SPS can be viewed as a one-dimensional vertical model, whereby the horizontal exchanges at the surface are neglected. This assumption implies that the surface balance at each grid point of the surface is solved independently from the others; all the grid points are only correlated through the spatial patterns of the atmospheric fields and of the land surface characteristics. This detail is of primary importance for the analysis of the subgrid-scale variability.

b. Land surface characteristics

The integration of the surface balance in SPS requires the specification of the land surface characteristics at each grid point of the computational domain, in particular information on the vegetation type (e.g., forest and grass), on the soil texture (e.g., sand and clay), and on several parameters specific to the type of vegetation [e.g., leaf area index (LAI)]. These geophysical fields are generated at each given horizontal resolution using the 300-m GlobCover land cover database2 (Bicheron et al. 2006; Bontemps et al. 2011). The 1-km global U.S. Geological Survey (USGS) database3 is also used to specify terrain orography. The choice of these geophysical databases is user defined; the present study relies (on purpose) on standard databases that have been largely tested over the past decade, in research and at operational level in a wide range of configurations and for a variety of surface covers. This choice allows a statistical analysis of the subgrid-scale variability representative of operational NWP systems.

c. Atmospheric forcing

The atmospheric forcing in SPS is based on the regional version of the GEM atmospheric model (Mailhot et al. 2006), using hourly forecasts of the atmospheric fields at 15-km resolution4 with assimilation 4 times a day (at 0000, 0600, 1200, and 1800 UTC); the precipitation rate is derived at an hourly rate from the Canadian Precipitation Analyses (CaPA) system (Mahfouf et al. 2007) using the regional model prediction as prior information. This regional version is based on a hydrostatic version of the model, integrated with a 450-s time step and a predefined 58-level vertical resolution (the first level being located at about 40 m above the canopy level). Through the downscaling procedure, the atmospheric forcing is corrected to account for elevation differences between GEM and SPS and thereby adjust SPS roughness lengths consistently with those of the three-dimensional atmospheric model in the calculation of the near-surface variables (Carrera et al. 2010; Bernier et al. 2011).

d. Simulation case and experimental setup

The flowchart of the present SPS simulation case, starting from land surface analyses and updated with hourly atmospheric forcing from the regional version of GEM, is presented in Fig. 1. This simulation case provides continuous forecasts of the land surface over a regional domain covering the central part of North America (see the REG computational domain colored by the 2.5-km terrain elevation field in Fig. 2), from 0000 UTC 1 January to 0000 UTC 29 August 2012, with a 10-min time step and a 6-month spinup carried out until 0000 UTC 1 July 2012. The forecasts of interest cover the 2012 warm season (July–August).

Fig. 1.
Fig. 1.

Flowchart of the SPS. SPS is driven by hourly atmospheric fields (i.e., 40-m air humidity, air temperature, and wind; surface pressure; solar and longwave radiation; precipitation) produced by the 15-km version of the regional GEM model and downscaled from the lowest atmospheric level to the land surface. Initial condition for SPS is derived from the land surface analyses (e.g., soil moisture and surface temperature).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

Fig. 2.
Fig. 2.

REG simulated with the SPS covering the central part of North America and colored by the 2.5-km terrain elevation field (m). White-lined boxes show the location of the local domains. (left) Mountains (Pacific Coast Ranges, British Columbia, LBC), (middle) prairies (Saskatchewan, LPR) and lakes (Manitoba, LLA), and (right) lowland forests (southern Quebec and Ontario, LQC).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

The same SPS configuration is used for varying horizontal grid spacing (25, 10, and 2.5 km). The land surface scheme (ISBA), the downscaling of the atmospheric fields and the geospatial databases (GlobCover, USGS) remain the same in all three configurations; the only sources of differences are the resolution of the geophysical fields and the end result of downscaling that reflects the orography associated with each configuration. Thus, the objective of this work is not to reevaluate the tiling approach, but to analyze the impact of the geophysical fields’ spatial variability on land surface modeling. Moving toward higher horizontal resolution induces a lower-order effect of the tiling approach, but undoubtedly this remains essential at kilometer scale in order to apply specific treatment to each type of subgrid-scale surface covers.

For example, Figs. 3 and 4 present the simulated near-surface air temperature at 1200 UTC 27 July 2012 for the 2.5- and 25-km configurations, respectively. They highlight at a given forecast hour the changes in the near-surface air temperature due to an increased horizontal resolution in the geophysical fields. As expected, the large-scale structures are consistent between both configurations. However, the 2.5-km configuration features small-scale structures unresolved at 25 km, in particular over the Manitoba lake region and the prairies. For instance, the 25-km case is unable to capture the local effect of rivers and the sharp temperature gradients, suggesting that the impact of these local geophysical features on the (near-)surface variables may be significantly underestimated or even neglected at typical resolution of operational NWP systems.

Fig. 3.
Fig. 3.

Near-surface air temperature (K) simulated with SPS at 2.5-km horizontal resolution at 1200 UTC 27 Jul 2012. (top) The REG computational domain and (bottom) close-ups on the Canadian prairies and lakes (including the LPR and LLA local domains).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

Fig. 4.
Fig. 4.

As in Fig. 3, but for the 25-km horizontal resolution case.

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

In the following, the sensitivity of the screen-level diagnostic variables and the surface prognostic variables to the horizontal grid spacing as well as the corresponding sources of subgrid-scale variability are analyzed over the main domain (REG) but also over some specific geographical areas: 1) mountains in British Columbia (LBC), 2) prairies in Saskatchewan (LPR), 3) lakes in Manitoba (LLA), and 4) lowland forests in southern Quebec and Ontario (LQC). The location of these local domains within the REG computational domain is shown in Fig. 2.

3. Subgrid-scale variability and statistical metrics

The impact of the horizontal resolution on the SPS performance is studied through the statistical characterization of the subgrid-scale variability in the surface thermodynamic variables (i.e., 2-m air temperature and dewpoint temperature; surface sensible and latent heat fluxes) for 25-km grid spacing model based on the 2.5-km (or 10 km) SPS run. In this study, the 25-km case is referred to as the reference (i.e., mimicking the resolution at which surface fluxes would be upscaled to GEM in a two-way mode); the 2.5- and 10-km resolution cases are considered as high resolution (i.e., mimicking the resolution at which SPS would be resolved in a two-way mode).

a. Definition of the subgrid-scale variability

The subgrid-scale variability for any given variable corresponds to the (unresolved) PDF of this variable within a reference grid cell of the computational domain. This concept is illustrated in Fig. 5 for a comparison between the 25-km reference case and the 2.5-km high-resolution case. The gray dots represent the set of grid points at which the variable under consideration is computed for the 2.5-km model resolution; these grid points are referred to as subgrid-scale grid points. Typically, 100 subgrid-scale grid points are located within one 25-km grid cell Thus, the subgrid-scale variability refers at a given forecast time to the statistical characterization of the variable PDF over these 100 subgrid-scale grid points, the subgrid-scale grid points being regarded as statistical sample points.

Fig. 5.
Fig. 5.

Schematic of the subgrid-scale variability for any surface variable (or any geophysical field) at a given forecast hour; comparison between the (left) 25-km reference case and the (right) 2.5-km high-resolution case. The large black dots (left) correspond to the grid points explicitly resolved at the 25-km resolution, the gray dots (right) correspond to the subgrid-scale grid points (explicitly resolved at the 2.5-km resolution), and the red dot corresponds to the mean value over the subgrid-scale grid points contained within the 25-km grid cell under consideration. The PDF of any surface variable over the subgrid-scale grid points can be statistically characterized by its first moments: mean value (red dot) and STD (spread of the gray dots).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

b. Characterization of the subgrid-scale variability

1) Model deviation

For any given 25-km grid cell, the subgrid-scale mean value of a given surface variable is computed by averaging the value of the variable over the subgrid-scale grid points (obtained through the 2.5- or 10-km SPS run). In Fig. 5, the red dot corresponds to the subgrid-scale mean value derived from the 2.5-km case. The model deviation can therefore be represented, locally, by the difference between the value simulated by the 25-km reference case and the subgrid-scale mean value derived from the high-resolution case, locally meaning at each grid point of the 25-km case. Note that the subgrid-scale mean value corresponds to a linear upscaling of the variable to the coarse grid resolution. Note also that this linear upscaling is consistent with the focus on thermodynamic variables (while beyond the scope of this study, the validity of the linear upscaling for the surface momentum flux needs to be assessed). In the following, the model deviation is regarded as the expression of the model bias when degrading the model resolution to 25 km.

2) Subgrid-scale standard deviation

The standard deviation (STD) of a given variable over the subgrid-scale grid points (obtained for the 2.5- or 10-km SPS run), also referred to as the subgrid-scale STD, is computed for each reference grid cell in order to quantify the unresolved subgrid-scale variability at the 25-km grid spacing. In the following, the subgrid-scale STD is regarded as the expression of the unresolved small-scale variability at the reference horizontal resolution.

3) Statistical metrics

To provide a complete and systematic characterization, the spatial distribution of the model deviation and subgrid-scale STD over the limited-area domains (REG, LBC, LPR, LLA, and LQC) is estimated, at each forecast hour, by its mean value and quantiles. This distribution is computed according to the local time at each grid point: for a given forecast hour, only the grid points with the local time corresponding to this forecast hour are retained in the limited-area domain under consideration. Thus, the impact of the horizontal grid spacing on the SPS forecast performance is measured through the temporal evolution of the subgrid-scale distribution computed over a given limited-area domain, at 1-h time intervals.

To rigorously identify the sources of subgrid-scale variability in a given surface variable, correlations of the variable model deviation with any of the geophysical field model deviation are computed for each local domain.

4. Results and discussion

The statistical metrics related to the model deviation and subgrid-scale STD are used to quantify in this section the model response to the horizontal grid spacing over the diurnal (day–night) cycle at regional scale (section 4a) and in specific geographic areas (section 4b); a comparison between the 2.5- and the 10-km high-resolution cases (with respect to the 25-km reference case) is provided in section 4c.

a. Regional impact

1) Spatial patterns

Two-dimensional maps of the model deviation and subgrid-scale STD are helpful to evaluate the nonuniform and nonconstant information gain when running SPS at higher resolution. For example, Fig. 6a presents the map of the near-surface air temperature model deviation that is obtained by comparing the 25- and 2.5-km cases in the REG domain at 1200 UTC 27 July 2012 (this model deviation represents the actual signed difference between Figs. 3 and 4). Figure 6b shows the counterpart of Fig. 6a for the near-surface air temperature subgrid-scale STD.

Fig. 6.
Fig. 6.

Maps of (a) model deviation and (b) subgrid-scale STD for the simulated near-surface air temperature (K) over the REG computational domain at 1200 UTC 27 Jul 2012; comparison between the 2.5- (Fig. 3) and 25-km (Fig. 4) runs. Color bars correspond to air temperature variations (K).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

A spatial alternation of positive and negative local model deviations (with an amplitude reaching up to 2 K) is present along lakeshores (e.g., Manitoba lake region) and seashores (e.g., Hudson Bay) as well as in the British Columbia Coast Ranges. The signal observed in the mountains is strongly correlated with the model deviation in the terrain elevation shown in Fig. 7a, which can reach up to 200 m. Decreasing the horizontal grid spacing from 2.5 to 25 km induces a considerable loss of information at the land surface; local sharp gradients of geophysical variables being averaged. Specific environments such as glaciers, lakeshores, and seashores thus feature a strong response to changes in the horizontal resolution, with a subgrid-scale STD above 2 K.

Fig. 7.
Fig. 7.

Examples of factors influencing the near-surface air temperature model deviation shown in Fig. 6. (a) Model deviation for the terrain elevation (m) over the western part of the REG domain (including the LBC local domain). (b) The 40-m air temperature (°C) simulated by the regional version of GEM at 1200 UTC 27 Jul 2012 (including the REG computational domain).The terrain elevation scale runs from ±250 in increments of 50 m and the temperature scale runs from −12.1° to 32.8° in increments of ~1.0°C.

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

A quasi-uniform negative model deviation (with magnitudes varying between 0 and 1 K) is also visible in the REG western area (corresponding to nighttime). Such a uniform signal is not necessarily due to a lack of resolution in the geophysical fields, but it could also be a result of the local and instantaneous atmospheric conditions. Figure 7b presents the map of the 40-m air temperature simulated by GEM at 1200 UTC 27 July 2012; the geographical area where the near-surface air temperature model deviation is negative is shown to be subject to a coastal regime. Thus, in the region under consideration and at the forecast hour under consideration (1200 UTC), the atmospheric conditions seem to prevail over the heterogeneity of the geophysical fields. The extent of the atmospheric forcing impact on the subgrid-scale variability strongly depends on the type of geographic area under consideration. Since these atmospheric conditions are time dependent, the response of the SPS system to the horizontal grid spacing must be studied over time.

Note that the near-surface air temperature model deviation and subgrid-scale STD is mostly nonzero in the REG domain: there is a background signal of 0.5-K magnitude (regardless of the geographic areas), which can be regarded as the minimum error introduced by the 25-km SPS run with respect to the 2.5-km grid spacing.

2) Sensitivity to the diurnal cycle

Figure 8 presents the evolution of the model deviation statistics for the near-surface air temperature and the upward surface sensible heat flux over a 24-h period (27 July 2012), resulting from the comparison between the 25- and 2.5-km cases. These statistics are represented at each forecast hour using a box plot, for which the box spans 25% and 75% of the model deviation values in the REG domain, the dashed line spans 5% and 95% of these values, and the solid line represents their median. Figure 9 presents the counterpart of Fig. 8 for the near-surface dewpoint temperature and the upward surface latent heat flux.

Fig. 8.
Fig. 8.

Time-evolving model deviation statistics at 1-h intervals and in the REG domain based on the 2.5-km run. Box plot over a 24-h period (27 Jul 2012) corresponding to (a) the near-surface air temperature (K) and (b) the upward surface sensible heat flux (W m−2). For one box, the span is 25% and 75% of the model deviation values, the vertical dashed line spans 5% and 95% of these values, and the solid line represents their median.

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

Fig. 9.
Fig. 9.

As in Fig. 8, but for the case of (a) the near-surface dewpoint temperature (K) and (b) the upward surface latent heat flux (W m−2).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

The degree of dispersion of the model deviation values in the REG domain is high for all variables under consideration, reaching (in terms of magnitude) 1 K for the air temperature, 100 W m−2 for the sensible heat flux, 1.2 K for the dewpoint temperature, and 60 W m−2 for the latent heat flux. Note that these amplitudes are significant for regional-scale statistics, in particular for the surface heat fluxes for which the model deviation maximum spread corresponds to about 50% of the variable range in REG and for which 10% of the subgrid-scale values feature a sensible (latent) heat flux model deviation higher than 75 W m−2 (40 W m−2) in daylight hours.

It is also found that the model deviation statistics are strongly affected by the diurnal (day–night) cycle. They follow the diurnal temperature variation: there is an inflection of the model deviation spread during the transitions between day and night cycles, which translates into a minimum value for the model deviation statistics of the air temperature and the dewpoint temperature during the morning transition. Then, there is a strong increase of the model deviation spread during daytime, with a maximum reached at the peak daily temperature. The model deviation statistics are larger by a factor of 5 in daylight hours (only the evolution of the air temperature statistics is limited to a factor of 2). At nighttime the surface sensible heat flux statistics feature a lower degree of dispersion than during daytime: there is less available energy and winds weaken, leading to a reduction in the surface fluxes’ variability. The spread of these statistics for the surface sensible heat flux (with an amplitude reaching 25 W m−2) can still be regarded as significant relative to the nominal values; this is not the case for the surface latent heat flux that is nearly zero in the REG domain during the night cycle. In contrast, the near-surface temperature model deviation statistics remain high at nighttime. The near-surface air temperature model deviation is even higher at nighttime than during daytime; the temperature inversions present at nighttime inducing a stronger dependence of the near-surface meteorology on the land surface and thereby a high spatial variability. The near-surface dewpoint temperature model deviation is still lower at nighttime than during daytime, the impact of the soil relative humidity inducing a stronger variability during daylight hours.

Except for the surface latent heat flux, the model deviation values in the REG domain are found not to be centered on zero, implying that the positive and negative values do not balance each other out and that there is no conservation. Since the model deviation is defined as the signed difference between the upscaled high-resolution simulation case and the reference simulation case, and since the high-resolution simulation case has been shown to be more accurate (see Carrera et al. 2010; Leroyer et al. 2014), the reference simulation case tends to underestimate (with respect to the high-resolution simulation) the near-surface air temperature during daytime. It also tends to overestimate the near-surface air temperature and the near-surface dewpoint temperature at nighttime. The amplitude of the surface sensible heat flux is overestimated over the entire diurnal cycle (the sensible heat flux features negative values at nighttime and positive values at daytime).

These results show the direct and strong impact of the diurnal cycle on the SPS model sensitivity to the horizontal grid spacing. The resulting model deviation statistics are controlled by the physics of surface–atmosphere interactions and (at least partially) by the surface radiation budget (similar behavior is obtained for the subgrid-scale STD statistics; not shown here).

3) Temporal pattern

Figure 10 presents the subgrid-scale STD statistics obtained in the REG domain over a 5-day period (from 25 to 29 July 2012) for the near-surface air temperature, the upward surface sensible heat flux, the near-surface dewpoint temperature, and the near-surface wind speed. These statistics result from the comparison between the 25- and 2.5-km cases. The red dashed line represents the temporal evolution of the variable mean value at 1-h intervals; the black solid line represents the counterpart for its median, and the black dashed lines correspond to its STD (i.e., one STD above and below the mean value).

Fig. 10.
Fig. 10.

Time-evolving subgrid-scale STD statistics at 1-h intervals from 25 to 29 Jul 2012 and in the REG computational domain based on the 2.5-km SPS run related to (a) the near-surface air temperature (K), (b) the surface sensible heat flux (W m−2), (c) the near-surface dewpoint temperature (K), and (d) the near-surface wind speed (m s−1). The black solid line represents the subgrid-scale STD median in REG, the red dashed line represents its mean value, and the black dashed lines correspond to one STD above and below the mean value.

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

For all (near-)surface variables, the subgrid-scale STD features nonzero median and spread. The median has values of at least 0.4 K for the air temperature and 0.2 m s−1 for the wind speed, while it is varying between 0.25 and 0.75 K for the dewpoint temperature and 5 and 40 W m−2 for the sensible heat flux. The spread of the subgrid-scale STD reaches up to 1.6 K for the air temperature, 2 K for the dewpoint temperature, 0.6 m s−1 for the wind speed, and 70 W m−2 for the sensible heat flux.

The time evolution of the subgrid-scale median and spread shows that the same temporal pattern is repeated daily, with a slight day-to-day fluctuation of the maximum amplitude occurring in daylight hours (similar behavior is obtained over the whole summer; not shown here). Since the geophysical fields do not change over time, this day-to-day fluctuation of the variable subgrid-scale STD is very likely due to the time-dependent meteorological conditions and to their nonlinear impact on the SPS forecast. This impact may be nonlinear in the sense that the model response to the horizontal grid spacing over a given geographical area depends on the length scales of the atmospheric patterns and on the land surface characteristics. Through the downscaling procedure there is a loss of information in the 25-km case compared to the 2.5-km case (the subgrid-scale STD statistics for the near-surface wind magnitude are indeed not negligible; see Fig. 10d). This loss of information is propagated over time, enhancing model bias and affecting the subgrid-scale STD (the impact of the geophysical characteristics being dependent on the diurnal cycle phase and on the ongoing weather regime).

Figure 10 also highlights the constant gap between the variable subgrid-scale STD mean value and median, implying that this statistical metric does not follow a Gaussian distribution and is skewed toward high values in the REG domain. There are no multiple modes in the distribution (microclimates cannot be distinguished, for instance); the large size of the computational domain smooths it out through the background signal. Similar behavior is obtained for the model deviation statistics (not shown here). This skewness, even stronger for surface heat fluxes than for screen-level meteorological variables, confirms the strong heterogeneity in the model deviation and the subgrid-scale STD in the REG domain. Local gradients of the geophysical properties combined with a particular instantaneous meteorological state can create conditions for which the horizontal resolution drastically modifies the state of the land surface (the local state nonlinearly depends on local gradients of the geophysical properties and may not be simply related to average gradients). For instance, the strong gradients in the geophysical fields such as in the terrain elevation (see Fig. 7a) induce a high level of dispersion in the statistical metrics during daytime. It is therefore important to identify each major contribution to the land surface subgrid-scale variability (even though it is difficult to separate the effects of the weather and land heterogeneity).

b. Sensitivity to geographic areas

This section analyzes the variations in the statistical metrics between the local limited-area domains and thus between significantly different environments (e.g., mountains, prairies, lakeshores, and lowland forests), with a focus on the near-surface air temperature and the upward surface sensible heat flux.

1) Spatial heterogeneity

Figures 11a–d present the statistics of the near-surface air temperature subgrid-scale STD computed over a 5-day period (from 25 to 29 July 2012) for the local domains LBC, LPR, LLA, and LQC, respectively (see Fig. 10a for comparison with the results obtained in the REG domain). Figure 12 presents the counterpart of Fig. 11 for the upward surface sensible heat flux.

Fig. 11.
Fig. 11.

As in Fig. 10, but for the simulated near-surface air temperature (K), in the following local domains: (a) LBC, (b) LPR, (c) LLA, and (d) LQC; comparison between the 25- and 2.5-km SPS runs.

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

Fig. 12.
Fig. 12.

As in Fig. 11, but for the case of the upward surface sensible heat flux (W m−2).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

The magnitude of the air temperature subgrid-scale STD statistics in each local domain is nonnegligible with a diurnal cycle of significant magnitude (similarly to the REG domain). However, they feature some particularities in each environment. First, the largest spread of the subgrid-scale STD (1.5–2.25 K) is obtained for mountains (LBC) and lakes (LLA); in contrast, the spread is limited to 1–1.5 K over the prairies (LPR) and lowland forests (LQC). The maximum amplitude of the spread is subject to day-to-day fluctuations as for the REG domain, these fluctuations being an indicator of the time-varying weather impact on the land surface. Second, the peak amplitude of the subgrid-scale STD is not systematically reached during daytime; the subgrid-scale STD statistics are larger at nighttime in LPR and LLA, or at least equivalent to the daytime statistics in LQC. It is also found that the minimum level of the subgrid-scale STD in LBC remains high over the successive diurnal cycles (above 1 K). In contrast, the subgrid-scale STD in LLA is reduced to 0.5 K, while its maximum amplitude is similar to that obtained in LBC.

The differences between the local domains in the surface sensible heat flux are not as strong as for the near-surface air temperature, the subgrid-scale STD median and mean value being relatively similar and constant around 40 W m−2 (see Fig. 12). Still, there are significant spatial variations in the subgrid-scale STD maximum amplitude reached during daylight hours: LLA features the largest spread (up to 90 W m−2), while this spread is almost halved in LPR because of their relatively uniform land surface and their higher sensitivity to the meteorological conditions (compared to LBC or LLA).

2) Correlation to geophysical properties

To identify which are the dominant sources of subgrid-scale variability during summertime, the model deviation correlations between the (near-)surface variables and the geophysical fields—1) glacier mask, 2) land–water mask, 3) roughness length, 4) terrain orography, 5) soil texture (percentage of clay and sand in soil), 6) LAI, 7) vegetation fraction, and 8) distribution of vegetation types (mixed shrubs, mixed wood forests, long grass, and evergreen or deciduous needleleaf)—are computed over a 5-day period (from 25 to 29 July 2012) in LBC, LPR, LLA, and LQC. Note that the correlation coefficient only provides a relative measure of the importance of each geophysical field (there is a constant competition between the different sources of subgrid-scale variability). Note also that at a given forecast hour, the weather regime is not the same in the whole REG computational domain, leading to changes in the amplitude of the statistics between the different local domains. Still, the trend observed over the warm season for LBC, LPR, LLA, and LQC is similar to the results presented in this study over a 5-day period, and the geophysical properties to which the land surface model is sensitive remain important for the whole warm season.

(i) Main sources of subgrid-scale variability

As an example, Table 1 presents the mean value of the model deviation correlation over the 5-day period under consideration for the near-surface air temperature, with an intercomparison between the previously mentioned geophysical fields and the different local domains. The criterion to be considered in this study as a major source of subgrid-scale variability is that the mean value of the model deviation correlation over the 5-day period must be above 0.2. Table 2 summarizes the dominant sources of subgrid-scale variability per local domain for the near-surface air temperature, the upward surface sensible heat flux, the near-surface dewpoint temperature, and the upward surface latent heat flux. Based on this criterion, it is found that the roughness length, the land–water mask, the glacier mask, the LAI, and the soil texture are the most important geophysical fields when moving toward higher surface resolution (interestingly, not the terrain orography). Specific behavior is noticed for each geophysical field as described below.

Table 1.

Intercomparison of the model deviation correlation between the near-surface air temperature and the geophysical fields for each local domain. Correlation factors in boldface correspond to those identified as the main factors of subgrid-scale variability in Table 2.

Table 1.
Table 2.

Dominant sources of variability in near-surface air temperature, upward surface sensible heat flux, near-surface dewpoint temperature, and upward surface latent heat flux for 25-km grid spacing based on the 2.5-km land surface forecast system for each local domain (LBC, LPR, LLA, and LQC); mean model deviation correlation is above 0.2 over the 5-day period (25–29 Jul 2012).

Table 2.

The roughness length and the LAI are common dominant sources of variability between the local domains, with the amplitude of the model deviation correlation reaching up to 0.8 for the roughness length in LBC and LPR as well as for the LAI in LBC, LLA, and LQC. Note that the roughness length subgrid-scale variability has a reduced impact over lowland forests (in particular for the air temperature with a mean correlation of 0.12, in contrast to 0.35 for LBC and 0.25 for LPR). Note also that the LAI subgrid-scale variability is not as dominant over the prairies since they are significantly affected by the meteorological conditions (with a mean correlation of 0.05 for LPR, in contrast to 0.22 for LBC and 0.26 for LLA).

The impact of the soil texture subgrid-scale variability is significant on the surface sensible heat flux (except in LPR). The soil texture is crucial in the representation of surface heat fluxes through the dependency of the soil thermal properties (e.g., albedo and emissivity) on the sand or clay soil texture. There is indeed a link with the soil moisture and the evaporation process. Thus, the prairies are less sensitive to the soil moisture variability than the other local domains; instead, they are mainly controlled by the near-surface meteorology. To a lesser extent, the mountainous regions are also less sensitive to the soil texture than lakes and lowland forests.

The near-surface air temperature model deviation in LBC is also strongly correlated to the glacier mask model deviation (with a mean correlation of 0.48) and to a lesser extent to that of the terrain orography (with a mean correlation of 0.37). The small-scale representation of the glacier delimitations is thus crucial to accurately simulate the near-surface meteorological conditions over complex alpine spine systems (the terrain orography is not the main geophysical property explaining the small-scale variability in LBC).

Lakes also have a strong impact on the near-surface air temperature and the surface sensible heat flux, in particular in LLA and LQC; these local domains feature a respective mean correlation of 0.50 and 0.47 for the near-surface air temperature, in contrast to 0.21 for LPR and 0.19 for LBC. The model deviation map shown in Fig. 6 highlights these differences in the surface predictions over the Manitoba lake region (included in the LLA domain). In general, the high amplitude of the model deviation statistics related to the land–water mask over all local domains demonstrates the significant effect of lakes on the surface prediction: when moving toward higher horizontal resolution, smaller lakes as well as narrower rivers are considered instead of land by the surface model. This change from the land tile to the water tile (or vice versa) at a given location implies a nonlinear impact on the local near-surface meteorology with respect to the horizontal resolution.

(ii) Sensitivity to diurnal cycle

Figures 13 and 14 present the time evolution of the model deviation correlations between the (near-)surface variables and the geophysical fields over the 5-day period under consideration. It is shown that the model deviation correlations are not constant over one diurnal cycle, implying that the dominant source of variability changes over the day–night cycle because of the land surface balance and because of the ongoing competition between geophysical properties and atmospheric conditions.

Fig. 13.
Fig. 13.

Time-evolving correlation of the near-surface temperature model deviation (K) with a subset of GlobCover/USGS geophysical field model deviation [i.e., terrain topography, roughness length, soil texture (percent of sand or clay in soil), glacier mask, land–water mask, LAI]; comparison between the 25- and 2.5-km cases in LBC, LLA, LPR, and LQC.

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

Fig. 14.
Fig. 14.

As in Fig. 13, but for the surface sensible heat flux model deviation (W m−2).

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

In the LBC domain, the roughness length is the main source of variability during daytime for both the near-surface air temperature and the surface sensible heat flux (with a correlation reaching up to 0.8 and −0.8 over each diurnal cycle, respectively). At nighttime the most influential factor is the presence of glaciers for the near-surface air temperature (with a correlation reaching up to −0.6) and the soil texture for the surface sensible heat flux (with a correlation reaching up to 0.6).

In the LLA and LQC domains, the influence of the roughness length is limited to daylight hours. However, the effect of the lakes and soil texture is strong over the whole diurnal cycle (even though the correlation coefficient periodically changes sign, the maximum amplitude of the signal is relatively constant). The time evolution of the (near-)surface conditions due to the land–water mask and to the soil texture is significantly underestimated at the 25-km resolution (with respect to the 2.5-km resolution). The behavior of the model deviation correlation also changes depending on the variable being analyzed: the representation of the land–water mask is more important for the near-surface air temperature (with correlations fluctuating between 0.8 and −0.8 over the diurnal cycle), while the representation of the soil texture is more important for the surface sensible heat flux (with correlations fluctuating between 0.8 and −0.8 over the diurnal cycle).

In the LPR domain, the roughness length is the main source of variability during daytime and at nighttime. To a lesser extent, terrain orography and the land–water mask are also important contributors to the surface variability at nighttime.

In summary, the high-resolution representation of geophysical properties such as the roughness length, the soil texture, the land–water mask, and the glacier mask is important to account for the length scales and temporal scales at which near-surface atmospheric conditions and surface heat fluxes evolve. The dominant sources of variability are not the same for the different local domains (even though this intercomparison between LBC, LPR, LLA, and LQC may be affected by the wide differences in terms of weather regimes among the geographical areas, the results shown here are consistent with the trends observed over the warm season). The dominant sources of variability are also significantly different for the near-surface temperature variables and the upward surface fluxes: the representation of the soil properties is a key component to improve the surface fluxes that are closely related to the soil relative humidity, and the variability of the roughness length has a strong impact on the near-surface meteorology since it is directly used to compute the surface layer diagnostics (i.e., the screen-level variables). Thus, the LLA and LQC domains seem to be mainly affected by the variability in the soil moisture, inducing strong variability in the surface heat fluxes. In contrast, the LPR and LBC domains are subject to strong winds and seem to be significantly impacted by the variability in the turbulent transfer, that is, why the roughness length is an essential geophysical property in these geographic areas.

c. Sensitivity to the land surface model resolution

The sensitivity of the statistical metrics to the horizontal resolution (2.5 vs 10 km) is highlighted for the near-surface air temperature in the LBC local domain in Fig. 15. Figures 15a and 15b present the corresponding subgrid-scale STD statistics; Figs. 15c and 15d present the model deviation statistics. Figures 15a and 15c present the statistical metrics based on the 2.5-km run, and Figs. 15b and 15d present those based on the 10-km run. While featuring similar temporal patterns for both 2.5- and 10-km high-resolution configurations, the subgrid-scale STD statistics have a reduced amplitude in the 10-km case. The amplitude of the median and of the spread is significantly reduced by about 0.5 K (a similar behavior is obtained in LLA and LQC). In contrast, there is no significant difference in LPR between the 2.5- and 10-km cases since prairies are strongly affected by meteorological conditions and the atmospheric forcing is the same for all high-resolution cases (not shown here). Thus, solving near-surface meteorological features at higher resolution reduces simulation bias, in particular over mountains and at the interface between land and inland lakes.

Fig. 15.
Fig. 15.

Time-evolving statistical metrics of the simulated near-surface air temperature (K), at 1-h intervals from 25 to 29 Jul 2012 and in the LBC local domain. Subgrid-scale STD statistics based on the (a) 2.5- and (b) 10-km runs, and model deviation statistics based on the (c) 2.5- and (d) 10-km runs (the 25-km run is the reference configuration for all these configurations). The black solid line represents the statistical metrics median, the red dashed line represents their mean value, and the black dashed lines correspond to one STD above and below the mean value.

Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0016.1

The temporal pattern of the model deviation statistics over one diurnal cycle significantly varies between the 2.5- and 10-km cases. The alternation of negative and positive model deviation in LBC is not symmetric for both cases. The model deviation distribution is more drastically skewed toward negative values for the 2.5-km case than for the 10-km case, with a negative model deviation going below −1.5 K (instead of −0.7 K for the 10-km case), in contrast to 1 K for the positive model deviation (instead of 1 K for the 10-km case). Similar behavior is obtained for the surface heat fluxes (not shown here). This highlights the nonconservation of the physical processes and therefore the highly nonlinear response of the land surface model when increasing the horizontal resolution (average surface variables are not simply related to average gradients of the geophysical fields, and the extent of the model response highly depends on the type of surface cover). Because of the strong nonlinearities in the statistics with respect to the horizontal grid spacing, explicitly solving for the small-scale near-surface meteorological features seems inevitable to address the challenge of NWP and NEP applications. Thus, there is a need to further investigate and characterize the land surface model response for varying environment when moving toward kilometer- and subkilometer-scale resolution, but also over long-term forecasts when fully coupled to the three-dimensional atmospheric model.

5. Conclusions

This paper is a preliminary step toward the development of an advanced surface–atmosphere two-way coupling approach, in which the surface state is explicitly resolved at high resolution and in which the surface heat fluxes are upscaled to a coarser-resolution atmospheric model grid. Based on previous studies highlighting the improved NWP forecast performance when moving toward kilometer- or subkilometer-scale land surface simulations (Carrera et al. 2010; Bernier et al. 2011; Leroyer et al. 2011), the objective here is twofold: 1) providing a more comprehensive understanding of the surface subgrid-scale variability sources among the geophysical properties on the one hand and 2) demonstrating the potential contribution of solving for the land surface at kilometer scale in order to improve the representation of surface–atmosphere interactions on the other hand.

For this purpose, an extensive numerical analysis of the model bias and of the unresolved subgrid-scale variability (induced by typical resolutions of current NWP systems) is performed during summertime over a regional domain covering the central part of North America. This analysis relies on the following system: the Canadian offline land surface prediction system (SPS), including the ISBA land surface scheme; atmospheric and precipitation forcing from the 15-km regional version of the Canadian three-dimensional atmospheric model (GEM) downscaled at the SPS resolution; and GlobCover/USGS geospatial databases. The subgrid-scale variability of the thermodynamic variables (i.e., near-surface meteorological variables and upward surface heat fluxes) is characterized using model deviation and subgrid-scale STD statistics over the regional domain, based on the comparison between the 25- and 2.5-km (or 10 km) SPS simulation cases. The sensitivity of the surface variable response to the horizontal grid spacing is studied over the diurnal cycle, for various environments (i.e., mountains, presence of inland lakes, and prairies).

It is found that the model deviation and the subgrid-scale STD are high over the whole domain for all variables being analyzed (even though this study is limited to the current grid spacing of the global and regional versions of operational NWP systems and does not consider subkilometer predictions). They feature a specific diurnal pattern, during which a minimum occurs at the day–night transitions and a maximum is reached at the peak daily temperature, the amplitude of the day-to-day statistics being controlled by the local and instantaneous meteorological conditions. It is also found that the amplitude of the statistical moments varies highly between the geographic areas and does not scale linearly with increased horizontal resolution. Thus, the model response to varying horizontal resolution features nonlinearities with respect to the local gradients of the geophysical characteristics (i.e., roughness length, glacier and land–water masks, LAI, and soil texture), meaning that average variables are not simply related to average gradients of the geophysical fields. This model response is particularly strong over mountains as well as along lakeshores and seashores, since a change of resolution may imply a change of surface scheme (from land tile to water tile, or vice versa) at a given location. The significant impact of the roughness length over prairies and mountains highlights the merits and potential benefits of better accounting for surface–atmosphere interactions. The soil texture (closely related to the LAI and the soil moisture) is also found to play an important role in the evaporation variability representation over lakes and lowland forests.

This numerical analysis also demonstrates that the mean subgrid-scale value (obtained by simulating SPS at 2.5- or 10-km resolution and by linear upscaling at 25-km resolution) is significantly different from the explicitly solved coarse-resolution value, regardless of the geographic areas. The subgrid-scale variability of the near-surface meteorological variables remains relatively controlled since the atmospheric forcing is identical for all horizontal resolutions and since SPS simulations are currently limited to the one-way forcing mode. However, this is not the case for upward surface heat fluxes that feature important spatial and temporal variations at subgrid scale. Thus, the information viewed by the three-dimensional atmospheric model in a two-way coupling mode is expected to be significantly modified by the increased resolution of the land surface model. The subsequent improved representation of the surface–atmosphere interactions will presumably induce significant changes in the state of the atmosphere near the surface and in the PBL and influence processes such as the growth of the PBL height, cloud formation, and precipitation (Zabel et al. 2012; Zabel and Mauser 2013). In turn, the modified atmospheric conditions may induce a nonnegligible feedback on the near-surface meteorological variables and further enhance the subgrid-scale variability at the land surface, in particular when moving toward kilometer- and subkilometer-scale simulations (Leroyer et al. 2014).

In this perspective, and given the increased computational power in meteorological operational centers, solving the land surface at high resolution in a surface–atmosphere fully coupled system becomes a key aspect to consider for improving forecast performance and capturing the subgrid-scale variability of environments as different as land, glaciers, and inland water. The nonlinear model response to the change of horizontal grid resolution implies that the benefits of a kilometer- or subkilometer-scale surface prediction system are difficult to quantify a priori and need to be studied in detail. A similar numerical analysis could be performed for the momentum flux, although the design of the upscaling technique is not straightforward (the linear upscaling technique may be questionable); the characterization of the momentum flux subgrid-scale variability is expected to provide further insight in mountains as well as urban environments (cities are not included in the present study). In regard to objective evaluation and data assimilation, the characterization of the subgrid-scale variability could be useful to improve the systematic comparison of the forecast simulations with (local and incomplete) observations as well as to enrich the prior information required by land surface data assimilation systems. Further investigation of the offline land surface prediction system and of the land surface–atmosphere interactions at high resolution is believed to be valuable for meteorological and climate modeling applications.

Acknowledgments

The financial support provided by Environment Canada through NSERC (Natural Sciences and Engineering Research Council of Canada) under the 2014–15 project grant entitled “Improving the coupling between the surface and the atmosphere” is greatly appreciated. High-performance computing resources from Environment Canada are also beneficial.

REFERENCES

  • Avissar, R., 1998: Which type of soil–vegetation–atmosphere transfer scheme is needed for general circulation models: A proposal for a higher-order scheme. J. Hydrol., 212–213, 136154, doi:10.1016/S0022-1694(98)00227-3.

    • Search Google Scholar
    • Export Citation
  • Avissar, R., and Schmidt T. , 1998: An evaluation of the scale at which ground-surface heat flux patchiness affects the convective boundary layer using large-eddy simulations. J. Atmos. Sci., 55, 26662689, doi:10.1175/1520-0469(1998)055<2666:AEOTSA>2.0.CO;2.

    • 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, doi:10.1175/1525-7541(2003)4<352:OIOTIL>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., Brown R. , Mailhot J. , Bilodeau B. , and Crevier L.-P. , 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, doi:10.1175/1525-7541(2003)4<371:OIOTIL>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Bernier, N., Bélair S. , Bilodeau B. , and Tong L. , 2011: Near-surface and land surface forecast system of the Vancouver 2010 winter Olympic and Paralympic games. J. Hydrometeor., 12, 508530, doi:10.1175/2011JHM1250.1.

    • Search Google Scholar
    • Export Citation
  • Best, M., Beljaars A. , Polcher J. , and Viterbo P. , 2004: A proposed structure for coupling tiled surfaces with the planetary boundary layer. J. Hydrometeor., 5, 12711278, doi:10.1175/JHM-382.1.

    • Search Google Scholar
    • Export Citation
  • Bicheron, P., and Coauthors, 2006: Globcover: A 300-m global land cover product for 2005 using ENVISAT MERIS time series. Proc. Second Int. Symp. on Recent Advances in Quantitative Remote Sensing, Valencia, Spain, University of Valencia–Global Change Unit, 538–542. [Available online at http://ipl.uv.es/raqrs/.]

  • Bontemps, S., Defourny P. , Bogaert E. V. , Arino O. , Kalogirou V. , and Perez J. , 2011: Globcover 2009: Products description and validation report. Tech. Rep., Université catholique de Louvain/European Space Agency, 53 pp. [Available online at http://due.esrin.esa.int/files/GLOBCOVER2009_Validation_Report_2.2.pdf.]

  • Carrera, M., Bélair S. , Fortin V. , Bilodeau B. , Charpentier D. , and Doré I. , 2010: Evaluation of snowpack simulations over the Canadian Rockies with an experimental hydrometeorological modeling system. J. Hydrometeor., 11, 11231140, doi:10.1175/2010JHM1274.1.

    • Search Google Scholar
    • Export Citation
  • Côté, J., Gravel S. , Méthot A. , Patoine A. , Roch M. , and Staniforth A. , 1998a: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395, doi:10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Côté, J., Desmarais J.-G. , Gravel S. , Méthot A. , Patoine A. , Roch M. , and Staniforth A. , 1998b: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 13971418, doi:10.1175/1520-0493(1998)126<1397:TOCMGE>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Dutra, E., Kotlarski S. , Viterbo P. , Balsamo G. , Miranda P. , Schär C. , Bissolli P. , and Jonas T. , 2011: Snow cover sensitivity to horizontal resolution, parameterizations, and atmospheric forcing in a land surface model. J. Geophys. Res., 116, D21109, doi:10.1029/2011JD016061.

    • Search Google Scholar
    • Export Citation
  • Erfani, A., Mailhot J. , Gravel S. , Desgagné M. , King P. , Sills D. , McLennan N. , and Jacob D. , 2005: The high resolution limited area version of the Global Environmental Multiscale Model (GEM-LAM) and its potential operational applications. 11th Conf. on Mesoscale Processes, Albuquerque, NM, Amer. Meteor. Soc., 1M.4. [Available online at http://ams.confex.com/ams/pdfpapers/97308.pdf.]

  • Essery, R., Best M. , Betts R. , Cox P. , and Taylor C. , 2003: Explicit representation of subgrid heterogeneity in a GCM land surface scheme. J. Hydrometeor., 4, 530543, doi:10.1175/1525-7541(2003)004<0530:EROSHI>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Fischer, E., Seneviratne S. , Vidale P. , Lüthi D. , and Schär C. , 2007: Soil moisture–atmosphere interactions during the 2003 European summer heat wave. J. Climate, 20, 50815099, doi:10.1175/JCLI4288.1.

    • Search Google Scholar
    • Export Citation
  • Girard, C., and Coauthors, 2014: Staggered vertical discretization of the Canadian Environmental Multiscale (GEM) model using a coordinate of the log-hydrostatic-pressure type. Mon. Wea. Rev., 142, 11831196, doi:10.1175/MWR-D-13-00255.1.

    • Search Google Scholar
    • Export Citation
  • Leroyer, S., Bélair S. , Mailhot J. , and Strachan I. , 2011: Microscale numerical prediction over Montreal with the Canadian external urban modeling system. J. Appl. Meteor. Climatol., 50, 24102428, doi:10.1175/JAMC-D-11-013.1.

    • Search Google Scholar
    • Export Citation
  • Leroyer, S., Bélair S. , Husain S. , and Mailhot J. , 2014: Subkilometer numerical weather prediction in an urban coastal area: A case study over the Vancouver metropolitan area. J. Appl. Meteor. Climatol., 53, 14331453, doi:10.1175/JAMC-D-13-0202.1.

    • Search Google Scholar
    • Export Citation
  • Mahfouf, J., Brasnett B. , and Gagnon S. , 2007: A Canadian Precipitation Analysis (CAPA) project: Description and preliminary results. Atmos.–Ocean, 45, 117, doi:10.3137/ao.450101.

    • Search Google Scholar
    • Export Citation
  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44, 133149, doi:10.3137/ao.440202.

    • Search Google Scholar
    • Export Citation
  • Marke, T., Mauser W. , Pfeiffer A. , and Zängl G. , 2011: A pragmatic approach for the downscaling and bias correction of regional climate simulations: Evaluation in hydrological modeling. Geosci. Model Dev., 4, 759770, doi:10.5194/gmd-4-759-2011.

    • Search Google Scholar
    • Export Citation
  • Masson, V., and Coauthors, 2013: The SURFEXv7.2 land and ocean surface platform for coupled or offline simulation of earth surface variables and fluxes. Geosci. Model Dev., 6, 929960, doi:10.5194/gmd-6-929-2013.

    • Search Google Scholar
    • Export Citation
  • Molod, A., and Salmun H. , 2002: A global assessment of the mosaic approach to modeling land surface heterogeneity. J. Geophys. Res., 107, 4217, doi:10.1029/2001JD000588.

    • 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, doi:10.1175/1520-0493(1989)117,0536:ASPOLS.2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Noilhan, J., and Mahfouf J.-F. , 1996: The ISBA land surface parameterization. Global Planet.Change, 13, 145159, doi:10.1016/0921-8181(95)00043-7.

    • Search Google Scholar
    • Export Citation
  • Pigeon, G., Moscicki M. , Voogt J. , and Masson V. , 2008: Simulation of fall and winter surface energy balance over a dense urban area using the TEB scheme. Meteor. Atmos. Phys., 102, 159171, doi:10.1007/s00703-008-0320-9.

    • Search Google Scholar
    • Export Citation
  • Polcher, J., and Coauthors, 1998: A proposal for a general interface between land-surface schemes and general circulation models. Global Planet. Change, 19, 261276, doi:10.1016/S0921-8181(98)00052-6.

    • Search Google Scholar
    • Export Citation
  • Salgado, R., and Moigne P. L. , 2010: Coupling of the FLake model to the Surfex externalized surface model. Boreal Env. Res., 15, 231244.

    • Search Google Scholar
    • Export Citation
  • Schomburg, A., Venema V. , Ament F. , and Simmer C. , 2012: Disaggregation of screen-level variables in a numerical weather prediction model with an explicit simulation of subgrid-scale land-surface heterogeneity. Meteor. Atmos. Phys., 116, 8194, doi:10.1007/s00703-012-0183-y.

    • Search Google Scholar
    • Export Citation
  • Seity, Y., Brousseau P. , Malardel S. , Hello G. , Bénard P. , Bouttier F. , Lac C. , and Masson V. , 2011: The AROME-France convective-scale operational model. Mon. Wea. Rev., 139, 976991, doi:10.1175/2010MWR3425.1.

    • Search Google Scholar
    • Export Citation
  • Seth, A., Giorgi F. , and Dickinson R. , 1994: Simulating fluxes from heterogeneous land surfaces: Explicit subgrid method employing the biosphere-atmosphere transfer scheme (BATS). J. Geophys. Res., 99, 18 65118 667, doi:10.1029/94JD01330.

    • Search Google Scholar
    • Export Citation
  • Yu, Z., 2000: Assessing the response of subgrid hydrologic processes to atmospheric forcing with a hydrologic model response. Global Planet. Change, 25, 117, doi:10.1016/S0921-8181(00)00018-7.

    • Search Google Scholar
    • Export Citation
  • Zabel, F., and Mauser W. , 2013: 2-way coupling the hydrological land surface model PROMET with the regional climate model MM5. Hydrol. Earth Syst. Sci., 17, 17051714, doi:10.5194/hess-17-1705-2013.

    • Search Google Scholar
    • Export Citation
  • Zabel, F., Mauser W. , Marke T. , Pfeiffer A. , Zängl G. , and Wastl C. , 2012: Inter-comparison of two-land surface models applied at different scales and their feedbacks while coupled with a regional climate model. Hydrol. Earth Syst. Sci., 16, 10171031, doi:10.5194/hess-16-1017-2012.

    • Search Google Scholar
    • Export Citation
4

The operational resolution during the 2012 summertime was 15 km (i.e., the time period of the present simulation case); the regional version of GEM is currently running with a 10-km horizontal grid spacing.

Save