• Adams, D., and A. Comrie, 1997: The North American monsoon. Bull. Amer. Meteor. Soc., 78, 21972213.

  • Adler, R., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 11471167.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., L. Crevier, J. Mailhost, B. Bilodeau, and V. Delage, 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., R. Brown, J. Mailhost, B. Bilodeau, and L. Crevier, 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
  • Benoit, J., J. Cote, and J. Mailhot, 1989: Inclusion of the TKE boundary layer parameterization in the Canadian regional finite-element model. Mon. Wea. Rev., 117, 17261750.

    • Search Google Scholar
    • Export Citation
  • Bhumralkar, C., 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
  • Brasnett, B., 2008: The impact of satellite retrievals in a global sea-surface-temperature analysis. Quart. J. Roy. Meteor. Soc., 134, 17451760.

    • Search Google Scholar
    • Export Citation
  • Brier, G., 1950: Verification of forecasts expressed in terms of probability. Mon. Wea. Rev., 78, 13.

  • Brotzge, J., and K. Crawford, 2003: Examination of the surface energy budget: A comparison of eddy correlation and Browen ration measurement systems. J. Hydrometeor., 4, 160178.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., M. Miller, and T. Palmer, 1999: Stochastic representation of model uncertainties in the ECMWF ensemble prediction system. Quart. J. Roy. Meteor. Soc., 125, 28872908.

    • Search Google Scholar
    • Export Citation
  • Bukovsky, M., and D. Karoly, 2007: A brief evaluation of precipitation from the North American Regional Reanalysis. J. Hydrometeor., 8, 837846.

    • Search Google Scholar
    • Export Citation
  • Charron, M., G. Pellerin, L. Spacek, P. Houtekamer, N. Gagnon, H. Mitchell, and L. Michelin, 2010: Towards random sampling of model error in the Canadian ensemble prediction system. Mon. Wea. Rev., 138, 18771901.

    • Search Google Scholar
    • Export Citation
  • Chase, T., R. Pielke, T. Kittel, R. Nemani, and S. Running, 1996: Sensitivity of a general circulation model to global changes in leaf area index. J. Geophys. Res., 101 (D3), 73937408.

    • Search Google Scholar
    • Export Citation
  • Chelton, D., 2005: The impact of SST specification on ECMWF surface wind stress fields in the eastern tropical Pacific. J. Climate, 18, 530550.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and Coauthors, 1996: Modeling of land surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101 (D3), 72517268.

    • Search Google Scholar
    • Export Citation
  • Chen, F., Z. Janjic, and K. Michell, 1997: Impact of atmospheric surface-layer parametrizations in the new land-surface scheme of the NCEP mesoscale Eta model. Bound.-Layer Meteor., 85, 391421.

    • Search Google Scholar
    • Export Citation
  • Coté, J., A. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998a: The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395.

    • Search Google Scholar
    • Export Citation
  • Coté, J., J.-G. Desmarais, A. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998b: The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 13971418.

    • Search Google Scholar
    • Export Citation
  • Efron, B., and R. Tibshirani, 1993: An Introduction of the Bootstrap. Chapman and Hall, 460 pp.

  • Eltahir, E., 1998: A soil moisture-rainfall feedback mechanism. 1: Theory and observations. Water Resour. Res., 34, 765776.

  • Fujita, T., D. Stensrud, and D. Dowell, 2007: Surface data assimilation using an ensemble Kalman filter approach with initial condition and model physics uncertainties. Mon. Wea. Rev., 135, 18461868.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., 2000: Decomposition of the continuous ranked probability score for ensemble prediction systems. Wea. Forecasting, 15, 559570.

    • Search Google Scholar
    • Export Citation
  • Houtekamer, P., and H. Mitchell, 2005: Ensemble Kalman filtering. Quart. J. Roy. Meteor. Soc., 131, 32693289.

  • Jaquemin, B., and J. Noilhan, 1990: Sensitivity study and validation of a land surface parametrization using the HAPEX-MOBILHY data set. Bound.-Layer Meteor., 52, 93134.

    • Search Google Scholar
    • Export Citation
  • Jiang, X., G.-Y. Niu, and Z.-L. Yang, 2009: Impacts of vegetation and groundwater dynamics on warm season precipitation over the central United States. J. Geophys. Res., 114, D06109, doi:10.1029/2008JD010756.

    • Search Google Scholar
    • Export Citation
  • Kain, J., and J. Fritsch, 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 24, Amer. Meteor. Soc., 165170.

  • Koster, R., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305 (5687), 11381140, doi:10.1126/science.1100217.

    • Search Google Scholar
    • Export Citation
  • Kushnir, Y., R. Seager, M. Ting, N. Naig, and J. Nakamura, 2010: Mechanisms of tropical Atlantic SST influence on North American precipitation variability. J. Climate, 23, 56105628.

    • Search Google Scholar
    • Export Citation
  • Lee, M.-I., and Coauthors, 2007: An analysis of the warm-season diurnal cycle over the continental United States and northern Mexico in general circulation models. J. Hydrometeor., 8, 344366.

    • Search Google Scholar
    • Export Citation
  • Li, J., and H. Barker, 2005: A radiation algorithm with correlated-k distribution. Part I: Local thermal equilibrium. J. Atmos. Sci., 62, 286309.

    • Search Google Scholar
    • Export Citation
  • Li, X., M. Charron, L. Spacek, and G. Candille, 2008: A regional ensemble prediction system based on the moist targeted singular vectors and stochastic parameter perturbations. Mon. Wea. Rev., 136, 443462.

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

    • Search Google Scholar
    • Export Citation
  • Matheson, J., and R. Winkler, 1976: Scoring rules for continuous probability distributions. Manage. Sci., 22, 10871095.

  • McFarlane, N., 1987: The effect of orographically excited gravity wave drag on the general circulation of the lower stratosphere and troposphere. J. Atmos. Sci., 44, 17751800.

    • Search Google Scholar
    • Export Citation
  • McLaughlin, D., Y. Zhou, D. Entekhabi, and V. Chatdarong, 2006: Computational issues for large-scale land surface data assimilation problems. J. Hydrometeor., 7, 494510.

    • Search Google Scholar
    • Export Citation
  • Miguez-Machado, G., and J. Peagle, 2001: Sensitivity of North American numerical weather prediction to initial state uncertainty in selected upstream subdomains. Mon. Wea. Rev., 129, 20052022.

    • Search Google Scholar
    • Export Citation
  • Mo, K., 2000: Intraseasonal modulation of summer precipitation over North America. Mon. Wea. Rev., 128, 14901505.

  • Mo, K., and E. Kalnay, 1991: Impact of sea surface temperature anomalies on the skill of monthly forecasts. Mon. Wea. Rev., 119, 27712793.

    • Search Google Scholar
    • Export Citation
  • Mo, K., and J. Paegle, 2000: Influence of the sea surface temperature anomalies on the precipitation regimes over the southwest United States. J. Climate, 13, 35883598.

    • Search Google Scholar
    • Export Citation
  • Morcrette, J.-J., H. Barker, J. Cole, M. Iacono, and R. Pincus, 2008: Impact of a new radiation package, MCRAD, in the ECMWF integrated forecasting system. Mon. Wea. Rev., 136, 47734798.

    • Search Google Scholar
    • Export Citation
  • Niyogi, D., T. Holt, S. Zhong, P. Pyle, and J. Basara, 2006: Urban and land surface effects on the 30 July 2003 mesoscale convective system event observed in the Southern Great Plains. J. Geophys. Res., 111, D19107, doi:10.1029/2005JD006746.

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

    • Search Google Scholar
    • Export Citation
  • Nutter, P., D. Stensrud, and M. Xue, 2004: Effects of coarsely resolved and temporally interpolated lateral boundary conditions on the dispersion of limited-area ensemble forecast. Mon. Wea. Rev., 132, 23582377.

    • Search Google Scholar
    • Export Citation
  • Palmer, T., R. Buizza, F. Doblas-Reyes, T. Jung, M. Leutbecher, G. Shutts, M. Steinheimer, and A. Weisheimer, 2009: Stochastic parametrization and model uncertainty. ECMWF Tech. Memo. 598, Vol. 139, ECMWF, 42 pp.

  • Smith, C., M. Lakhtakia, W. Capehart, and T. Carlson, 1994: Initialization of soil-water content in regional-scale atmospheric prediction models. Bull. Amer. Meteor. Soc., 75, 585593.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D., and M. Wandishin, 2000: Correspondence and spread ratios in forecasts verification. Wea. Forecasting, 15, 593602.

  • Stensrud, D., and N. Yussouf, 2003: Short-range ensemble prediction of 2-m temperature and dewpoint temperature over New England. Mon. Wea. Rev., 131, 25102524.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D., H. Brooks, J. Du, M. Tracton, and E. Rogers, 1999: Using ensembles for short-range forecasting. Mon. Wea. Rev., 127, 433446.

    • Search Google Scholar
    • Export Citation
  • Sundqvist, H., E. Berge, and J. Kristjansson, 1989: Condensation and cloud parameterization studies with a mesoscale numerical weather prediction model. Mon. Wea. Rev., 117, 16411657.

    • Search Google Scholar
    • Export Citation
  • Talagrand, O., R. Vautard, and B. Strauss, 1999: Evaluation of probabilistic prediction systems. Proc. ECMWF Workshop on Predictability, Reading, United Kingdom, ECMWF, 1–25.

  • Treut, H. L., and Z. Li, 1991: Sensitivity of an atmospheric general circulation model to prescribed SST changes: Feedback effects associated with the simulation of cloud optical properties. Climate Dyn., 5, 175187.

    • Search Google Scholar
    • Export Citation
  • Vannitsem, S., and Z. Toth, 2002: Short-term dynamics of model errors. J. Atmos. Sci., 59, 25942604.

  • Wandishin, M., S. Mullen, D. Stensrud, and H. Brooks, 2001: Evaluation of a short-range multimodel ensemble system. Mon. Wea. Rev., 129, 729747.

    • Search Google Scholar
    • Export Citation
  • Yeh, K.-S., J. Coté, A. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 2002: The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part III: Nonhydrostatic formulation. Mon. Wea. Rev., 130, 339356.

    • Search Google Scholar
    • Export Citation
  • Yu, W., L. Garand, and A. Dastoor, 1997: Evaluation of model clouds and radiation at 100 km scale using goes data. Tellus, 49, 246262.

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

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

    Standard deviation of the initial values for (a) ALB, (b) LAI (m2 m−2), (c) SST (K), (d) VEGF (%), (e) LZ0 (m), (f) LSM, (g) SIF, and (h) SIT experiments (m).

  • View in gallery
    Fig. 2.

    Spread of the initial perturbations of soil temperature at (a) the superficial layer and (b) the root zone (K). (c),(d) The initial spread for soil moisture (m3 m−3) is shown for the same two layers.

  • View in gallery
    Fig. 3.

    Spread of screen-level air temperature (K) at 48-h forecast time for the most sensitive surface parameter experiments (production of the largest spread of the 2-m temperature): (a) ALB, (b) LAI, (c) SST, and (d) VEGF experiments.

  • View in gallery
    Fig. 4.

    Ensemble mean total precipitation (mm) accumulated for 48 h, averaged over the 20 summer periods in 2009.

  • View in gallery
    Fig. 5.

    Spread of precipitation (mm day−1) at 48-h forecast time for the most sensitive experiments: (a) the SST and (b) VEGF experiments.

  • View in gallery
    Fig. 6.

    Spread of cloud fraction (%) at 48-h forecast time for the most sensitive surface parameter experiments: (a) ALB, (b) LAI, (c) SST, and (d) VEGF experiments.

  • View in gallery
    Fig. 7.

    Spread of screen-level 10-m wind speed (m s−1) at 48-h forecast time for the most sensitive surface parameter experiments: (a) SST and (b) LZ0 experiments.

  • View in gallery
    Fig. 8.

    Standard deviation of (a) 2-m temperature (K), (b) 10-m wind speed (m s−1), (c) cloud fraction (%), and (d) precipitation (mm) in the SMT experiment, after 48 h.

  • View in gallery
    Fig. 9.

    Standard deviation of (a),(c),(e) 2-m temperature (K) and (b),(d),(f) precipitation (mm day−1) at 48 h: (a),(b) SURF, (c),(d) PTP, and (e),(f) LSF experiments.

  • View in gallery
    Fig. 10.

    (a) Location of the four areas where the impact is averaged and its temporal evolution is analyzed: (b) [1 in (a)] Quebec–Ontario, (c) [2 in (a)] Canadian Rockies, (d) [3 in (a)] U.S. Rockies, and (e) [4 in (a)] southern United States. Temporal evolution of the standard deviation of 2-m temperature (K) in the ATM experiment (black line), the PARA experiment (black line with crosses), the SMT experiment (black lines with squares), and the STT experiment (black line with diamonds). The gray line represents the evolution of the NEW experiment of the REPS. These evolutions are spatially averaged over the four areas and the equivalent local time is indicated at the bottom of each. The equivalent local time [local standard time (LST)] is indicated in the top of each panel.

  • View in gallery
    Fig. 11.

    Temporal evolution of the 2-m temperature standard deviation (K) averaged over North America for the SMT experiment. The black line indicates the deviation generated by the perturbations described in Table 2. For this study, five perturbation intensities have been used (from 0.006 to 0.096 m3 m−3 and from 0.004 to 0.064 m3 m−3 for the surface and root layer, respectively). Two additional experiments have been performed to analyze the sensitivity to the wavelength of the perturbations (in gray, between 250–750 and 1000–3000 km).

  • View in gallery
    Fig. 12.

    Standard deviation differences for (a) 2-m temperature (K), (b) 10-m wind speed (m s−1), (c) precipitation (mm s−1), and (d) cloud fraction (%), between the ATM experiment and the NEW experiment of the REPS.

  • View in gallery
    Fig. 13.

    Talagrand diagram of the observed 2-m temperature for the ATM (solid line) and the NEW experiments (dotted line). A total of 2320 stations are used.

  • View in gallery
    Fig. 14.

    (left) Temporal evolution of the difference NEW minus ATM (positive values are associated with an improvement of the forecast) of the CRPS over the North American continent. A total of 2320 stations are used for (a) 2-m temperature; (c),(e) 10-m wind speed (u and υ components, respectively); and (g) dewpoint temperature. A bootstrap significance test has been used to show the 90% confidence level (dotted lines). (right) RMSE (solid lines) and bias (dotted lines) of the ATM (blue) and NEW (red) experiments for (b) 2-m temperature; (d),(f) 10-m wind speed (u and υ components, respectively); and (h) dewpoint temperature.

  • View in gallery
    Fig. 15.

    (a) RMSE of the 2-m temperature (K) in the NEW experiment after 48 h of simulation. (b) RMSE difference (NEW minus ATM experiments) of the 2-m temperature (K) after 48 h of simulation. Blue points indicate improvement of the NEW when compared with ATM.

  • View in gallery
    Fig. 16.

    (a) Brier skill score for daily precipitation (accumulation between 24 and 48 h of simulation) in the ATM and NEW experiments (blue and red lines, respectively) following the precipitation amounts. (b) As in (a), but for the Brier score difference for ATM minus NEW experiments (negative values indicate improvement of the forecast), dotted lines show the 90% confidence level based on the bootstrap method.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 187 91 12
PDF Downloads 109 54 2

Impact of Surface Parameter Uncertainties within the Canadian Regional Ensemble Prediction System

Christophe LavaysseAtmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada

Search for other papers by Christophe Lavaysse in
Current site
Google Scholar
PubMed
Close
,
Marco CarreraMeteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Marco Carrera 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
,
Normand GagnonMeteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Normand Gagnon in
Current site
Google Scholar
PubMed
Close
,
Ronald FrenetteEnvironment Canada, Montreal, Quebec, Canada

Search for other papers by Ronald Frenette in
Current site
Google Scholar
PubMed
Close
,
Martin CharronMeteorological Research Division, Environment Canada, Montreal, Quebec, Canada

Search for other papers by Martin Charron in
Current site
Google Scholar
PubMed
Close
, and
M. K. YauAtmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada

Search for other papers by M. K. Yau in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

The aim of this study is to assess the impact of uncertainties in surface parameter and initial conditions on numerical prediction with the Canadian Regional Ensemble Prediction System (REPS). As part of this study, the Canadian version of the Interactions between Soil–Biosphere–Atmosphere (ISBA) land surface scheme has been coupled to Environment Canada’s numerical weather prediction model within the REPS. For 20 summer periods in 2009, stochastic perturbations of surface parameters have been generated in several experiments. Each experiment corresponds to 20 simulations differing by the perturbations at the initial time of one or several surface parameters or prognostic variables. The sensitivity to these perturbations is quantified especially for 2-m temperature, 10-m wind speed, cloud fraction, and precipitation up to 48-h lead time. Spatial variability of these sensitivities over the North American continent shows that soil moisture, albedo, leaf area index, and SST have the largest impacts on the screen-level variables. The temporal evolution of these sensitivities appears to be closely linked to the diurnal cycle of the boundary layer. The surface perturbations are shown to increase the ensemble spread of the REPS for all screen-level variables especially for 2-m temperature and 10-m wind speed during daytime. A preliminary study of the impact on the ensemble forecast has shown that the inclusion of the surface perturbations tends to significantly increase the 2-m temperature and 10-m wind speed skill.

Corresponding author address: Christophe Lavaysse, Atmospheric and Oceanic Sciences, McGill University, 805 Sherbrooke St. West, Montreal, QC H3A 2K6, Canada. E-mail: christophe.lavaysse@mcgill.ca

Abstract

The aim of this study is to assess the impact of uncertainties in surface parameter and initial conditions on numerical prediction with the Canadian Regional Ensemble Prediction System (REPS). As part of this study, the Canadian version of the Interactions between Soil–Biosphere–Atmosphere (ISBA) land surface scheme has been coupled to Environment Canada’s numerical weather prediction model within the REPS. For 20 summer periods in 2009, stochastic perturbations of surface parameters have been generated in several experiments. Each experiment corresponds to 20 simulations differing by the perturbations at the initial time of one or several surface parameters or prognostic variables. The sensitivity to these perturbations is quantified especially for 2-m temperature, 10-m wind speed, cloud fraction, and precipitation up to 48-h lead time. Spatial variability of these sensitivities over the North American continent shows that soil moisture, albedo, leaf area index, and SST have the largest impacts on the screen-level variables. The temporal evolution of these sensitivities appears to be closely linked to the diurnal cycle of the boundary layer. The surface perturbations are shown to increase the ensemble spread of the REPS for all screen-level variables especially for 2-m temperature and 10-m wind speed during daytime. A preliminary study of the impact on the ensemble forecast has shown that the inclusion of the surface perturbations tends to significantly increase the 2-m temperature and 10-m wind speed skill.

Corresponding author address: Christophe Lavaysse, Atmospheric and Oceanic Sciences, McGill University, 805 Sherbrooke St. West, Montreal, QC H3A 2K6, Canada. E-mail: christophe.lavaysse@mcgill.ca

1. Introduction

Numerical approximations and inaccurate representation of the physical processes (Vannitsem and Toth 2002) in addition to uncertainties for lateral boundary and initial conditions (Nutter et al. 2004; Miguez-Machado and Peagle 2001) tend to limit the accuracy of deterministic high-resolution numerical weather predictions (NWP). A viable alternative method is to use a probabilistic approach. An ensemble prediction system can account for these uncertainties, whereby a single forecast of the atmosphere is replaced by an ensemble of forecasts from varying initial conditions. The dispersion of the ensemble runs provides quantitative information on forecast uncertainty. Environment Canada has an operational version of Global Ensemble Prediction System (GEPS) since 2005 (Houtekamer and Mitchell 2005), and a first version of a Regional Ensemble Prediction System (REPS).

The initial and boundary conditions of the REPS, and the inclusion of physical tendency perturbations (PTP; Charron et al. 2010), are the principal differences between each member. Preoperational tests have shown that this version of the REPS is underdispersed at the surface. This underdispersion has been noted in previous studies and is hypothesized to be a major cause for the low degree of correlation between the ensemble spread and accuracy of the mean forecast (Stensrud et al. 1999) resulting in the ensemble being unable to correctly represent the forecast error of the ensemble mean. The goal of this study is thus to integrate surface perturbations into Environment Canada’s REPS to account for uncertainties in surface conditions and to increase the ensemble spread at and near the surface.

The accurate prediction of screen-level variables is important for human activities, prediction of extreme events (thunderstorms, convective events, etc.; Niyogi et al. 2006), and for power production (wind power, solar power, etc.). However these variables are especially difficult to forecast (Stensrud and Yussouf 2003), even in short-range regional models with relatively high horizontal resolution and accurate surface schemes. One of the reasons for the large errors in predicting surface variables is associated with the imperfect knowledge of surface initial conditions (e.g., like soil moisture) and characteristics (e.g., vegetation fraction).

To improve the ability and skill of the REPS, it is necessary to represent the surface parameter uncertainties in the model initial conditions. Whereas model dynamics, model physics, and initial condition uncertainties are known to be important in ensemble forecasting (Stensrud and Wandishin 2000; Wandishin et al. 2001), surface condition uncertainties are generally not taken into account in ensemble systems, and ensemble forecasts typically use the same surface conditions for all members. This is a problem because the initial conditions at the surface are often an approximation of the observed surface state, owing to the inherent assumptions in the surface modeling schemes. Furthermore, climatological fields of surface parameters are often used during the initialization. These fields may not represent the real state of the surface, as vegetation properties such as albedo, leaf area index (LAI), or vegetation fraction differ from climatology during floods or intense dry or wet periods.

Forecasts of temperature and humidity at the surface are more sensitive to the land surface scheme than to convection, planetary boundary layer, or radiation schemes (Fujita et al. 2007). Several studies have pointed out the impact of the surface parameters on atmospheric forecasts. These parameters are principally related to vegetation such as LAI (Chase et al. 1996), albedo (Morcrette et al. 2008), vegetation fraction (Chen et al. 1996), SST (Chelton 2005), and soil moisture (Eltahir 1998; Smith et al. 1994). As well, more accurate surface schemes have been shown to improve precipitation forecast over the central United States (Jiang et al. 2009).

Several studies have described the effects of surface conditions. Chase et al. (1996) for instance examined the model sensitivity in two experiments with different LAIs. They found strong effects during winter at high latitudes, with an increase of surface temperature and a decrease of precipitation. Numerical models that used land surface parameterizations have also shown sensitivity of the predicted fluxes to the vegetation fraction (Jaquemin and Noilhan 1990). Unfortunately the seasonal evolution of the vegetation fraction is not well known, so that predetermined tabulated values based on ground observations over different vegetation types are typically used to specify this important parameter (Chen et al. 1996). In the National Centers for Environmental Prediction (NCEP) mesoscale Eta Model, Chen et al. (1997) have shown the high sensitivity of the model to the choice of the total roughness length, particularly for heat transfer.

Through the ocean and atmospheric heat budgets and the albedo, the sea ice thickness and sea ice fraction are important for the accurate prediction of surface temperature. The SST is known to have a large impact on the atmospheric circulation. North American summer precipitation is sensitive to the central and tropical Pacific SSTs (Kushnir et al. 2010). This sensitivity is particularly strong over the U.S. Southwest (Mo and Paegle 2000), even at intraseasonal time scales (Mo 2000). This results in monthly forecast skills that are sensitive to the SST (Mo and Kalnay 1991). The SST also impacts the low-level winds in the eastern tropical Pacific, close to the Central American coast (Chelton 2005), and the cloud cover especially at low- and midlevels (Treut and Li 1991).

Locally, the soil moisture field strongly influences convection and precipitation (Koster et al. 2004), especially over the southern United States. Eltahir (1998) has shown that wet soil conditions over a large region could be associated with relatively large boundary layer moist static energy, which favors the occurrence of precipitation. While soil moisture is one of the most important variables for land surface initialization (Smith et al. 1994), the role of soil temperature in the evolution of the lower atmosphere, especially for short-range forecasts should not be underestimated. Longwave radiation and ground heat flux are functions of soil temperature and have an influence on the growth of the boundary layer as well as turbulence (Brotzge and Crawford 2003).

Owing to the importance of surface parameters identified in previous studies, this paper aims to identify, quantify, and analyze the impact of the surface parameter perturbations in the Canadian REPS configuration. Section 2 presents the design of the limited-area model used with a focus on the land surface scheme. The methodology to generate the perturbations of the prognostic variables and of the surface parameters is explained in section 3. In section 4, the impacts of these perturbations on 2-m temperature, 10-m wind speed, precipitation, and cloud cover are examined. After the identification of specific areas over North America, the temporal evolution of these impacts on screen-level variables will be discussed. Section 5 contains a discussion of the sensitivity of the atmospheric variables to the characterization of the initial perturbation fields and the net impact of the surface perturbations. Conclusions are given in section 6.

2. Model design

The Canadian limited-area model version of the Global Environment Multiscale configuration (GEM-LAM) is a nonhydrostatic gridpoint model based on the implicit and semi-Lagrangian techniques (Coté et al. 1998a,b; Yeh et al. 2002). Within the REPS, the GEM-LAM model is integrated over a domain of 267 × 208 horizontal grid points with a 33-km grid spacing. The domain essentially includes North America, as it extends from Central America to the northern part of Canada. In the vertical, the model uses a hybrid coordinate with 28 levels extending from the surface to 10 hPa. No motion is allowed to cross the surface and top levels. In this study, only initial conditions at 0000 UTC are considered. The model is integrated to 48 h with a 15-min time step.

The initial and lateral boundary conditions of the REPS are a source of differences between the members. The initial atmospheric conditions are provided by the global ensemble assimilation system at 100-km resolution. The initial atmospheric conditions are provided by the GEPS at 100-km resolution and are generated using an ensemble Kalman filter (Houtekamer and Mitchell 2005). A total of 96 sets of initial conditions are generated and a subset of 20 are interpolated to 33 km to initialize the members of the REPS. The lateral boundary conditions are given by the global ensemble prediction system (Charron et al. 2010) and are updated every 3 h.

Another source of differences between the REPS members is the inclusion of PTP (Charron et al. 2010). These stochastic perturbations are inspired by Buizza et al. (1999) and Palmer et al. (2009), and were introduced to partly compensate for the loss of ensemble spread due to the use of a single dynamical core used in the REPS.

The physical parameterizations used in this study are as follows: a radiation scheme including infrared and solar components interactive with clouds (Yu et al. 1997; Li and Barker 2005), a turbulent kinetic energy (TKE)-based boundary layer vertical diffusion scheme (Benoit et al. 1989), a gravity wave drag parameterization scheme (McFarlane 1987), a low-level blocking parameterization (Zadra et al. 2003), a large-scale condensation scheme including prognostic variables for cloud water/ice content (Sundqvist et al. 1989), and a Kain–Fritsch deep-convection scheme (Kain and Fritsch 1993). A daily analysis of SST is provided to the model (Brasnett 2008) and remains constant throughout the model forecast.

The land surface scheme is a modified version of the Interactions between Soil–Biosphere–Atmosphere (ISBA) scheme (Bélair et al. 2003a,b), first developed at Météo-France (Noilhan and Planton 1989). In ISBA, the prognostic equations for surface and near-surface temperatures and soil moisture follow the force-restore method (Bhumralkar 1975; Bélair et al. 2003a), and are computed for two layers of depth: 10 cm and a root zone whose depths varies as a function of vegetation type. Other prognostic variables include snow characteristics (mass, albedo, density, liquid water content, etc.) and liquid water retained in the canopy; however, these variables are not perturbed in this study.

The initial conditions for ISBA’s surface temperature and soil moisture variables are obtained through a sequential assimilation of screen-level air temperature and humidity, based on an optimal interpolation technique (Bélair et al. 2003a).

3. Perturbation methodology

a. Stochastic perturbations

In this study, eight parameters, which are kept constant during the simulation, are perturbed. These parameters are vegetation fraction, LAI, albedo, roughness length, land–sea mask, SST, and both the fraction and thickness of sea ice. Additionally two prognostic variables in the ISBA scheme, namely soil moisture and soil temperature, are also perturbed based upon their possible impact on the atmosphere following previous studies as mentioned in the introduction.

Sensitivity experiments are performed to determine the effect of surface perturbations on screen-level variables in the ensemble system. This effect on screen-level variables, termed “impact” in this paper, is quantified by the ensemble spread using the standard deviation of the members in these experiments. Each parameter (or variable) is perturbed independently in a series of experiments using an ensemble of 20 members, which differ only in the perturbation of the identified parameter. Here, 20 summer periods, from 2 June to 21 August, have been used as initial conditions. On average, 4 days separate two simulations to avoid persistent circulation regimes. The list of experiments is provided in Table 1. Eleven experiments have been performed to analyze the sensitivity of the REPS to the surface parameter and variable perturbations (from VEGF to STT in Table 1). In these experiments the control member from the GEPS is used to provide the same boundary and initial conditions for all the members. Also, PTP is switched off in the REPS.

Table 1.

List of experiments where each experiment corresponds to a set of 20 simulations, differing by the perturbations at the initial time, for 20 summer periods in 2009. See text for additional details.

Table 1.

Markov chain processes are used for generating the surface perturbations. A simple method used in Li et al. (2008) and Charron et al. (2010) is applied here, with a two-dimensional random function on the sphere correlated in space, with a probability density function symmetric around the mean.

The spatial decorrelation length scales are equivalent to a regional wavelength embedded between 500 and 1000 km on a global 800 × 400 grid, and represent regional anomalies of surface conditions. Random and independent perturbation patterns are provided for all members to avoid biases. As explained in Li et al. (2008) and Charron et al. (2010), a stretch function ensures that the random field lies within specified bounds for each variable.

b. Parameter perturbation strategies

For the eight parameters identified, the type and intensity of the perturbations are given in Table 2. SST and land–sea mask have been perturbed by means of additive perturbations. This choice is based on the assumption that the errors are not proportional to SST or to the percentage of sea area for the land–sea mask. This is justified in Brasnett (2008), where the author shows that the largest root-mean-square errors for the Canadian SST analyses, used in this study, occur between 30° and 45°N, associated with transient eddies and large temperature gradients. The intensity of the perturbations is in accordance with the root-mean-square error (0.5°C) and the size is based on the regional patterns of the errors. Furthermore, Brasnett (2008) shows that indeed the errors in the SST analyses are more closely associated with areas of large temperature gradients rather than the absolute value of the SST. For this reason we used additive perturbations. Using this approach, while SST is perturbed over the open ocean and coastal areas, the land–sea mask is perturbed only when the mask is different from 0 (entirely land) or 1 (entirely ocean). This means that the perturbations can occur over grid points that contain coastal areas or inland lakes. Note that for land–sea mask values lower than 0.05 or larger than 0.95 a correction index is added, which has the effect of reducing the magnitude of the perturbations and constraining values of the mask between 0 and 1. The average of the perturbed parameters remains close to the initial value and no bias is added.

Table 2.

List of perturbed parameters. The second column indicates the type of perturbation, either additive (+) or multiplicative (×); while the third and fourth columns show the amplitudes and numerical bounds of the perturbations, respectively.

Table 2.

All other variables are perturbed multiplicatively. For the vegetation fraction, albedo and sea ice fraction, perturbations are assumed to be low when values are close to the limits (0 or 1). For these reasons, and to avoid a modification of the mean value of the parameters, a symmetric perturbations centered at 0.5 is used. Perturbations to LAI, roughness length and sea ice thickness are considered to be dependent on the initial values implying an increase of the perturbation intensities proportional to the value.

The perturbation intensities for each parameter have been estimated in relation to the uncertainties in the model. It remains difficult to estimate the parameter uncertainties between a real value and a value provided by a preestablished table in the model, as discussed in Bélair et al. (2003a). Moreover, this kind of estimation is new and few studies provide references. For example, McLaughlin et al. (2006) have generated some perturbations for a few surface parameters, such as soil type fractions, LAI (with spatially uncorrelated multiplicative uniform noise from 0.85 to 1.15), initial volumetric water content (with spatially uncorrelated additive Gaussian noise ±0.06), and soil temperature (with spatially uncorrelated additive Gaussian noise ±4 K). The sensitivity of the atmosphere to the surface perturbations is related to their intensities, which are defined by the estimation of the parameter uncertainties. This point will be addressed in the discussion section.

In this study the ensemble spread is quantified by the standard deviation of the members. Even if the standard deviations of the perturbation fields are homogeneous, there is a large spatial variability for each parameter depending on its own characteristics. As a consequence, the standard deviations of the perturbed fields appear inhomogeneous. The variability of these standard deviations is shown in Fig. 1. For the parameters with multiplicative perturbation type, these standard deviations are linked to the distribution of the mean value. This is especially true for the ALB (Fig. 1a), LAI (Fig. 1b), LZ0 (Fig. 1e), and SIF and SIT experiments (Figs. 1g,h; see Table 1 for a description of all experiments). Because of the limits imposed on its perturbations, the spatial variability of the standard deviation for the vegetation fraction is not correlated to the variability of its mean value (Fig. 1d). For LSM and SST experiments, the additive perturbations generate a homogeneous standard deviation (Figs. 1c,f). These perturbations are different for each member and each day but constant in time during the 48 h of simulation. The SST perturbations are generated over the entire domain to perturb SST and inland water. Note that, the high values of SIT, SIF, and LSM perturbations in the northern part of Canada are associated with inland water perturbations.

Fig. 1.
Fig. 1.

Standard deviation of the initial values for (a) ALB, (b) LAI (m2 m−2), (c) SST (K), (d) VEGF (%), (e) LZ0 (m), (f) LSM, (g) SIF, and (h) SIT experiments (m).

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

c. Perturbations of the prognostics variables

The same perturbation strategy at the initial time was applied to the model prognostic variables and their uncertainties can evolve in the model throughout the 48-h simulation time. To be consistent, the sign of the perturbations is the same for the two layers of the ISBA scheme. However, owing to the higher spatial and temporal variabilities, the intensity of perturbations is larger for the superficial layer (see Table 3). For each member, the initial conditions for soil temperature and soil moisture of a given layer are calculated by adding the respective perturbation fields. Note that the perturbation fields are different for soil moisture and soil temperature. This choice is based on the fact that soil moisture and soil temperature can be independent at the initial time. The standard deviation of the initial soil moisture and temperature for both layers is shown in Fig. 2. Owing to the initial perturbation intensities, the standard deviation of the temperature perturbations is larger in the superficial layer than in the root zone. For the superficial layer, the zonal gradient of the soil temperature is explained by the first pass into the ISBA scheme used to generate the initial conditions. This first time step is used to ensure coherency between all the geophysical fields and occurs at 0000 UTC every day. The solar radiation is still important in the western part of the continent (1700 local solar time). As a consequence, this incoming radiation has the effect of decreasing the surface temperature spread over the western United States.

Table 3.

List of variables perturbed in superficial (“Sup”) and root (“Root”) zones. The third column indicates the type of perturbation, either additive (+) or multiplicative (×), with the amplitude and numerical bounds of the perturbations shown in columns four and five, respectively.

Table 3.
Fig. 2.
Fig. 2.

Spread of the initial perturbations of soil temperature at (a) the superficial layer and (b) the root zone (K). (c),(d) The initial spread for soil moisture (m3 m−3) is shown for the same two layers.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

In contrast to soil temperature, and as can be expected from the use of additive perturbations, the standard deviation amplitudes of the soil moisture are homogeneous. This means that there is no effect of the first pass into the ISBA scheme. The extremely low values of soil moisture in the northwestern Mexico desert and the southwestern United States act to dampen the standard deviation of the perturbation, which remains close to zero over the region. In this case, the spatial variability is not associated with the variability of soil moisture.

4. Impacts of the perturbations on the screen-level variables

a. Impact of parameter perturbations

In this section the sensitivity of the atmosphere to the surface perturbations after 48 h of simulation is discussed by mean of sensitivity experiments. In particular, parameters that have a significant influence on the 2-m temperature, 10-m wind speed, cloud fraction, and precipitation are investigated.

The four perturbed parameters that generate the largest 2-m temperature standard deviation are shown in Fig. 3. The standard deviation is especially strong over Mexico in the ALB experiment (Fig. 3a). This regional maximum occurs over an area where the perturbation of the albedo is relatively large (Fig. 1a). The spatial patterns of the 2-m temperature sensitivity display, however, some differences when compared with the initial albedo perturbations shown in Fig. 1a. The large albedo perturbations over northern Canada are not associated with a large temperature standard deviation. The higher albedo of snow and ice would act to create a more homogeneous surface and dampen the impacts of radiative forcing. In the LAI experiment (Fig. 3b), the distributions of the 2-m temperature standard deviation and of the LAI perturbations are in agreement only over the southeastern United States and northwestern Mexico (Fig. 1b). This suggests that the intensities of LAI perturbations do not lead to large perturbations in the surface energy budget through evapotranspiration. Note that the LAI perturbations were applied to the final aggregated values that represent a weighted averaged over the vegetation types.

Fig. 3.
Fig. 3.

Spread of screen-level air temperature (K) at 48-h forecast time for the most sensitive surface parameter experiments (production of the largest spread of the 2-m temperature): (a) ALB, (b) LAI, (c) SST, and (d) VEGF experiments.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The SST perturbations generate a uniform spread of the 2-m temperature over the ocean (Fig. 3c). Over the continent, the influence, depicted by the spread intensity, is weak owing to the short lead time of the numerical integrations. The vegetation fraction, especially important in the calculation of the snow cover area (Bélair et al. 2003a), acts to modify the surface energy balance. As a consequence, the greatest influence on the 2-m temperature occurs in northern Canada (Fig. 3d), where the perturbations are strongest. In contrast, over Mexico, the spread on 2-m temperature is associated with small perturbation intensities, as shown Fig. 1d. This indicates a local increase of the sensitivity of the 2-m temperature to surface conditions and suggests that over arid areas, a small modification of the vegetation can have a relatively large impact on the evapotranspiration and therefore on screen-level temperature.

To better understand the impact of these experiments, the ensemble mean 48-h accumulation precipitation accumulated over the 20 simulation periods is shown (Fig. 4). In agreement with the spatial pattern of the climatological precipitation (Adler et al. 2003), the largest amounts of precipitation are seen over the tropical Atlantic and the southeastern United States. Comparison of Figs. 4 and 3 reveals that areas of large precipitation (over the southern and central part of the United States) are associated with the largest 2-m temperature standard deviation, especially in the LAI and VEGF experiments. The triggering of the Kain–Fritsch scheme used in the REPS is strongly influenced by the low-level temperature (Kain and Fritsch 1993); the variability thus induced in the deep convection in turn influences the screen-level variables through subgrid-scale precipitation and downdrafts. Thus, over these areas the atmospheric sensitivity is larger and the perturbations could generate a larger spread, as observed in the ALB experiment over Mexico (Fig. 3a), than over areas where no precipitation occurs.

Fig. 4.
Fig. 4.

Ensemble mean total precipitation (mm) accumulated for 48 h, averaged over the 20 summer periods in 2009.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The SST experiment, in which the surface evaporation is inherently modified, exhibits the largest ensemble spread on precipitation (Fig. 5a) over the Atlantic (more than 1.6 mm day−1). As shown in Fig. 4, this area is also associated with the largest precipitation accumulations during the summer period. In the VEGF experiment, the impacts on precipitation are still not well understood, but could be associated with the modification induced by evapotranspiration.

Fig. 5.
Fig. 5.

Spread of precipitation (mm day−1) at 48-h forecast time for the most sensitive experiments: (a) the SST and (b) VEGF experiments.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The spread of cloud fraction for the LAI, ALB, SST, and VEGF experiments are similar to the results for the 2-m temperature (Fig. 6). The VEGF experiment increases the standard deviation over the northern part of Canada in accordance with the perturbation variability (Fig. 1d). The SST experiment is associated with a global and homogeneous increase of the standard deviation over the oceans (Fig. 6d).

Fig. 6.
Fig. 6.

Spread of cloud fraction (%) at 48-h forecast time for the most sensitive surface parameter experiments: (a) ALB, (b) LAI, (c) SST, and (d) VEGF experiments.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The SST and LZ0 experiments are important in generating 10-m wind speed standard deviation (Fig. 7), with the impacts of the SST perturbations stronger over precipitating areas (Fig. 7a). The modification or change of the precipitation intensity exerts a control over the local 10-m wind speed and wind orientation. The impact of the LZ0 experiment (Fig. 7b) is found over the continent; however, the origin of the near-surface wind variability appears to be more complex, as the response appear to be near uniform and not related to either the variability of the perturbation intensities (Fig. 1e), nor to the precipitating region (Fig. 4). For this variable, the sensitivity could be more associated with meteorological conditions during the 20 summer periods.

Fig. 7.
Fig. 7.

Spread of screen-level 10-m wind speed (m s−1) at 48-h forecast time for the most sensitive surface parameter experiments: (a) SST and (b) LZ0 experiments.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

b. Impact of variable perturbations

As could have been expected, the SMT experiment has generated the largest 2-m temperature standard deviation between the members (Fig. 8). The southern U.S. region is associated with the largest sensitivity for 2-m temperature, 10-m wind speed, and precipitation (Figs. 8a,b,d) in agreement with the findings of Koster et al. (2004). As suggested by Koster et al. (2004), the areas associated with strong soil moisture impacts are generally located in transition zones between wet and dry climates. In this study, the mean soil moisture in the superficial layer over this area is 0.14 m3 m−3 (not shown). This area is surrounded by dry soil moisture in the southwestern United States (less than 0.1 m3 m−3) and relatively wetter soil moisture in the southern and eastern United States (sometimes larger than 0.2 m3 m−3). The largest sensitivities are found over the soil moisture transition zones (Koster et al. 2004), and thus no strong spatial correlation is noted between soil moisture in both the superficial and root zone and the regions of strongest precipitation sensitivity.

Fig. 8.
Fig. 8.

Standard deviation of (a) 2-m temperature (K), (b) 10-m wind speed (m s−1), (c) cloud fraction (%), and (d) precipitation (mm) in the SMT experiment, after 48 h.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

On the other hand, the perturbation of the surface temperature in the STT experiment does not lead to a large spread of the screen-level variables, especially after 48 h of simulation (not shown). For example, the 2-m temperature standard deviation is less than 0.3°C on average over the continent suggesting that the radiative forcing acts to dampen the initial soil temperature perturbations.

c. Comparison with the impact of atmospheric perturbations

The impact of the surface perturbations can be compared with those resulting from atmospheric perturbations such as the LSF and PTP experiments, features that are already present in the REPS. Uncertainty associated with large-scale forcing arises from different initial atmospheric conditions and lateral boundary forcings, as these are provided by the 20 members of the GEPS, and updated every 3 h for the lateral boundary conditions (Charron et al. 2010). The PTP perturbations to the total momentum and temperature tendencies are produced by the subgrid-scale physical parameterizations of each ensemble member using different random fields (Charron et al. 2010). This acts to increase the ensemble standard deviation and thus helps in improving the representation of uncertainty associated with the subgrid-scale parameterization.

The SURF experiment is performed to quantify the atmospheric impact when all surface parameters and variables are perturbed simultaneously driven by the same atmospheric boundary conditions. Note that for PARA, SURF, and NEW experiments, although some of the parameters are closely related, different perturbations are applied to the individual fields to maintain coherent values for each variable and to be consistent with the ensemble perturbation methodology. Nonetheless, individual members can have perturbations that can complement or counteract each other, which can act to randomly increase or decrease the net surface perturbations.

The 2-m temperature and precipitation standard deviations for the SURF experiment are shown in Figs. 9a,b after 48 h of simulation. For much of the continent a standard deviation close to or larger than 1 K is found (Fig. 9a). The PTP generates a similar spatial structure for the 2-m temperature standard deviation as in the SURF experiments (Fig. 9c), but the entire continent is impacted. The sensitivity of the 2-m temperature is strongest in the LSF experiment (Fig. 9e). This sensitivity can reach standard deviations greater than 2.5 K over some areas (Mexico, the southern United States, the northern part of Canada, and Greenland).

Fig. 9.
Fig. 9.

Standard deviation of (a),(c),(e) 2-m temperature (K) and (b),(d),(f) precipitation (mm day−1) at 48 h: (a),(b) SURF, (c),(d) PTP, and (e),(f) LSF experiments.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The impact on precipitation shows different signatures when compared to 2-m temperature. The areas with the largest precipitation standard deviations are closely linked to areas where the largest precipitation occurs (over the tropical Atlantic Ocean; Fig. 4). Among the surface perturbations (Fig. 9b), the largest impact is seen over the Atlantic Ocean and is explained mostly by the perturbations of the SST. Over the continent, perturbations to soil moisture and vegetation fraction are associated with specific areas of large precipitation standard deviations. The intensities are lower when compared to the PTP experiment (Fig. 9d). In the LSF experiment (Fig. 9f), the magnitudes of the precipitation standard deviations are low but the areas concerned are larger than in other experiments. This result could be due to the differences in initial conditions that tend to modify the synoptic-scale circulation whereas SURF and PTP experiments are more local in scale.

Overall, the impact of the surface perturbations is of the same order of magnitude as the PTP experiment for 2-m temperature and 10-m wind speed (not shown). For 2-m temperature, the standard deviation maximum is collocated with the maximum in the individual perturbation experiments, also it is related to the sensitivity of the model to convective precipitation or areas with strong initial 2-m temperature standard deviation. Nevertheless, these areas with large variability are also associated with NWP forecasting errors and needs to be improved. The SURF experiment is associated with a significant increase in the standard deviation of 2-m temperature and precipitation over the central United States and Canada. The impact in the SURF experiment for the 10-m wind speed and cloud fraction (not shown) is similar to the impact for 2-m temperature and precipitation, respectively.

d. Temporal evolution of the standard deviations

The temporal evolution of the impact (standard deviations) on 2-m temperature through the 48-h forecast cycle is presented over four areas that have been identified owing to distinct behaviors to surface perturbations (Fig. 10a).

Fig. 10.
Fig. 10.

(a) Location of the four areas where the impact is averaged and its temporal evolution is analyzed: (b) [1 in (a)] Quebec–Ontario, (c) [2 in (a)] Canadian Rockies, (d) [3 in (a)] U.S. Rockies, and (e) [4 in (a)] southern United States. Temporal evolution of the standard deviation of 2-m temperature (K) in the ATM experiment (black line), the PARA experiment (black line with crosses), the SMT experiment (black lines with squares), and the STT experiment (black line with diamonds). The gray line represents the evolution of the NEW experiment of the REPS. These evolutions are spatially averaged over the four areas and the equivalent local time is indicated at the bottom of each. The equivalent local time [local standard time (LST)] is indicated in the top of each panel.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The Quebec–Ontario area is chosen to isolate the impact of LAI and land–sea mask upon screen-level variables during the daytime. The area extends over the domain from 80° to 70°W and from 45° to 50°N. It is a relatively flat area covered by numerous lakes and water bodies. As shown in Figs. 1b,f, albedo and LAI perturbations are large over this area, which impact the 2-m temperature standard deviation (Fig. 3b).

The temporal evolution of the 2-m temperature standard deviation of the ATM version of the REPS displays a bidiurnal cycle (black solid line in Fig. 10b). Two minimums are observed during the sunrise period after 11 and 35 h of simulation (0700 local time) and two during the sunset periods after 25 and 48 h of simulation (at 2100 local time). The maximum during nighttime is associated with the synoptic forcing difference provided by the GEPS, also observed in the LSF experiment. During early morning, the radiative forcing increases over an area with a stable boundary layer, which tends to converge all 2-m temperatures, and generates the minimum in standard deviation. During daytime, the boundary layer becomes turbulent and increases the ensemble standard deviation. A maximum occurs during daytime (1200 local time), not related to the convection cycle. Indeed, the maximum of precipitation variance occurs around 2000 local time, which is in accordance with the convective cycle maximum at 1800 UTC (1400 local time, Bukovsky and Karoly (2007)) in the observations. Finally, the 2-m temperature standard deviation decreases during the evening associated with a decrease in the low-level turbulence.

In the PARA experiment (black line with crosses in Fig. 10b), the standard deviation increases with time after 12 h of simulation, but remains lower than 0.6 K. The delay in the onset of the 2-m temperature standard deviation is due to the nighttime and it is only at 1200 UTC (0800 local time) when some perturbed surface parameters, such as the vegetation fraction and the LAI, begin to modify the radiation budget. Decomposing the eight individual surface parameters, the LAI and the roughness length perturbations (associated with a change in the vegetation fraction) mainly explain the ensemble standard deviation (not shown). The impact of the SMT experiment (black line with open squares in Fig. 10b) appears to be of the same magnitude as the PARA experiment and is larger during daytime. The impact of STT (black line with full diamond in Fig. 10a), is greatest during the first hours of simulation, then quickly decreases and remains constant just after the first sunset. This implies that soil temperature is strongly controlled by solar radiation and its perturbation does not influence the atmosphere.

To analyze the impact over mountainous terrain, the Canadian Rockies (55°–60°N, 120°–110°W; Fig. 10c) and the U.S. Rockies (35°–40°N, 120°–110°W; Fig. 10d) have been selected. The maximum of total roughness length and standard deviation are located over the Canadian Rockies (Fig. 1e). Over these two areas, other surface parameter perturbations are large, LAI, land–sea mask (associated with a large number of inland lakes) for the Canadian Rockies and albedo for the U.S. Rockies.

The temporal evolutions of the 2-m temperature variance over these two areas more closely resemble the temporal evolutions of the LSF experiment as compared to the SURF experiment. Even if the roughness length perturbations are large, the impact of the SMT experiment is still more prominent over the Canadian Rockies. The SMT and SURF experiments have the largest impact, and display a small increase during the second nighttime period. This peak is not related to the convection maximum, which occurs later in the afternoon (1500 local time; Lee et al. 2007) over central Mexico and the southwest United States, also called the North American monsoon area (Adams and Comrie 1997). As a conclusion, the contrasting temporal evolutions between the two areas stem from the LSF and SMT experiments and are not modified much by the PTP and SURF experiments.

The last area, located in the central United States (35°–40°N, 105°–95°W), is associated with the largest impact of the surface perturbations (Fig. 9a), primarily those associated with the soil moisture perturbations (Fig. 8a). The total ensemble standard deviation increases strongly from 0600 local time and reaches a maximum at 1700 local time for all experiments except for the temperature perturbations (Fig. 10e). Over this last area, the soil moisture is the predominant parameter with the largest standard deviation (larger than 1.5 K around 1700 local time). The intensity of the standard deviation associated with both the SMT and STT experiments suggests the prominent effect of the boundary layer turbulence. The origin of this sensitivity does not appear to be related to precipitation. The maximum of precipitation occurs during the night (around 0100 local time), whereas the model generally produces more precipitation during the daytime (not shown), which is in accordance with Lee et al. (2007). Nevertheless, the number of precipitation events is too low to extract a significant diurnal cycle of precipitation characteristics.

As suggested by these four time evolutions of the 2-m temperature standard deviations, the main impact of the surface perturbations are generally observed during the day owing to the modification of the radiative balance and/or convective precipitation cycle. Nevertheless, the behavior of the ensemble standard deviation differ greatly over each area, depending on surface characteristics, sensitivity to the large-scale forcing, or on the impact of the boundary layer turbulence strength and cycle. This induces a distinct diurnal evolution over each area.

5. Discussion

a. Sensitivity to the characteristics of the perturbation fields

To test the robustness of the perturbation methodology, several sensitivity tests have been performed using five different perturbation intensities (see Table 4). In the SMT experiment, the relationship between the standard deviation of 2-m temperature and the perturbation intensities is linear after a threshold (Fig. 11). For the smallest perturbations, the resulting 2-m temperature standard deviations are very low. A clear diurnal cycle appears when increasing the intensity of the perturbations (green line in Fig. 11) and it increases linearly with perturbation intensity. Using average values over smaller areas, the same linear increase is observed. This means that even at small scales the sensitivity is similar. On the other hand, the temporal evolution of the 2-m temperature standard deviation does not seem to depend strongly on the intensity used.

Table 4.

Maximum perturbation intensities used in the sensitivity tests for the follow parameters or variables. For soil moisture and temperature, the maximum perturbation intensities are indicated for the superficial layer and for the root zone (in parentheses).

Table 4.
Fig. 11.
Fig. 11.

Temporal evolution of the 2-m temperature standard deviation (K) averaged over North America for the SMT experiment. The black line indicates the deviation generated by the perturbations described in Table 2. For this study, five perturbation intensities have been used (from 0.006 to 0.096 m3 m−3 and from 0.004 to 0.064 m3 m−3 for the surface and root layer, respectively). Two additional experiments have been performed to analyze the sensitivity to the wavelength of the perturbations (in gray, between 250–750 and 1000–3000 km).

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The sensitivity to spatial scale was also examined (in gray, Fig. 11). Using smaller-scale perturbations with wavelengths of 250–750 km instead of 500–1000 km, the 2-m temperature standard deviations decrease by about 20%. In contrast, an increase of the perturbation wavelength to the synoptic scale (from 1000 to 3000 km, gray circle line Fig. 11) acts to increase the 2-m temperature standard deviation by an intensity equivalent to the 20% increase of perturbation intensities (orange curve in Fig. 11).

The same test for the other surface parameters shows that the sensitivity generally appears linear to the perturbation intensities. Only SST is sensitive to the perturbation scales with larger scale and coherent perturbations tending to have a stronger impact on the atmosphere. In contrast, in the ALB and LZ0 experiments, the modifications of the perturbations do not impact the pattern and the intensity of the atmospheric response (not shown).

b. Net effect of the surface perturbations

In this section, the additive impact of including the surface perturbations within the REPS is evaluated by comparing the REPS with and without surface perturbations, thus between the NEW and ATM experiments.

Over the continent, the net increase of the 2-m temperature standard deviation is largest over the central United States (+0.75 K; Fig. 12a), which represents a relative increase of around 30% when compared to the ATM. This result is encouraging as the 2-m temperature standard deviations were relatively low over this region in the ATM version. Even if the absolute standard deviation is increased everywhere, the Pacific Ocean undergoes the largest relative increase (larger than 100%). This is explained by the weak impact of the large-scale and PTP forcings in the previous version of the REPS and by the inclusion of the SST perturbations in the NEW version.

Fig. 12.
Fig. 12.

Standard deviation differences for (a) 2-m temperature (K), (b) 10-m wind speed (m s−1), (c) precipitation (mm s−1), and (d) cloud fraction (%), between the ATM experiment and the NEW experiment of the REPS.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The 10-m wind speed standard deviation (Fig. 12b) is clearly increased over the continent, owing largely to the LZ0 and ALB experiments. Over the ocean, although the SST perturbations influence the 10-m wind speed standard deviation, the impact of the surface perturbations is weak.

For precipitation and cloud fraction the changes to the ensemble spread due to the inclusion of the surface perturbations is more complex (Figs. 12c,d). Results do not indicate a clear large-scale tendency for the region of positive and negative standard deviation anomalies. Nonetheless, areas associated with greater precipitation are generally associated with increases in the precipitation standard deviations in the NEW experiment.

c. Preliminary results on forecast impacts

The increase of the ensemble spread observed in section 5b does not necessarily imply that the ensemble forecast is superior. A preliminary study of the forecast improvements has been conducted by comparing the NEW and ATM experiments over a dense network of 2320 stations throughout North America for 2-m temperature and 10-m wind speed. Additionally, a total of 1782 daily rain gauges are used to score the precipitation forecasts (as used in Mahfouf et al. 2007).

Talagrand diagrams are used to display the distribution of the observations within the sorted series of the forecast ensemble (Talagrand et al. 1999). The diagram, when applied to the sorted series of the ensemble 2-m temperature forecasts at 48-h integration (Fig. 13), shows that ATM is clearly associated with an underdispersive shape of the distribution (U shape) with an excessive amount of observations located in the extreme ranks. In the NEW experiment, the distribution is closer to a uniform distribution, resulting from a decrease in the number of observations in the extreme ranks. This suggests that the addition of the surface perturbations tends to decrease the underdispersive nature of the REPS for 2-m temperature at 48-h integration.

Fig. 13.
Fig. 13.

Talagrand diagram of the observed 2-m temperature for the ATM (solid line) and the NEW experiments (dotted line). A total of 2320 stations are used.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

Specific scores are available to characterize the quality of an ensemble forecast. The continuous ranked probability score (CRPS; Matheson and Winkler 1976) measures the sum of the squared differences in the cumulative probability space. Figure 14 depicts the CRPS for experiments NEW and ATM for the entire simulation period.

Fig. 14.
Fig. 14.

(left) Temporal evolution of the difference NEW minus ATM (positive values are associated with an improvement of the forecast) of the CRPS over the North American continent. A total of 2320 stations are used for (a) 2-m temperature; (c),(e) 10-m wind speed (u and υ components, respectively); and (g) dewpoint temperature. A bootstrap significance test has been used to show the 90% confidence level (dotted lines). (right) RMSE (solid lines) and bias (dotted lines) of the ATM (blue) and NEW (red) experiments for (b) 2-m temperature; (d),(f) 10-m wind speed (u and υ components, respectively); and (h) dewpoint temperature.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The diurnal cycle of the CRPS for 2-m temperature, 10-m wind, and dewpoint temperature (not shown) suggest a better skill of the ensemble forecasts around 0600 than 1800 LST. The scores for the 2-m temperature in the NEW experiment are significantly improved throughout the simulation period (Fig. 14a), based on the 90% significance using the bootstrap method (Efron and Tibshirani 1993). This increase is more pronounced after 1800 LST, in accordance with the significant decrease of the bias (defined as the difference between forecast minus observation, Fig. 14b), which could be associated with the increase of the standard deviation in the NEW experiment shown in Fig. 10, and with the impact of the soil moisture over the North American continent in Fig. 11. Note that, this improvement tends to decrease a cold bias of the 2-m temperature observed in the REPS in summer during the night. The improvement of the 10-m wind speed forecasts also appears significant (Figs. 14c,e) even if the RMSE and the bias are still the same for the two experiments (Figs. 14d,f). Using the decomposition of the CRPS in terms of reliability and resolution (Hersbach 2000), the increase of the scores for both 10-m wind speed and 2-m temperature are explained by an increase of the reliability (not shown). The dewpoint temperature forecasts display the same temporal evolution, with the strongest CRPS improvement after 1800 LST (Fig. 14g). RMSE and bias are significantly decreased during the same period of the day (Fig. 14h).

These improvements appear significant only for the screen-layer variables. Vertically, the impacts decrease quickly. In the boundary layer, 2-m temperature, and dewpoint temperature are slightly influenced, but the differences are still not significant (not shown). Above 850 hPa, the impacts are located over southeastern United States, where the main impact on the precipitation is observed (e.g., Fig. 12c). Nevertheless, the mean impact of the surface perturbations over North America above 850 hPa is negligible in comparison with the differences generated by the atmospheric advection, perturbation of the physical tendencies, and or the differences in the large-scale circulation.

The improvement of the 2-m temperature forecasts in the REPS is not homogeneous over the continent. They are located over the central and southern United States (around 100°W) and over the western part of the Canada (Fig. 15). These areas are also associated with the largest increase of the spread as shown in Fig. 12a. Nevertheless, stations centered around 40°N, 90°W possess large increases of the spread but a degradation of the forecasts. Even if this area presents generally a high RMSE during the summer of 2009 (Fig. 15a), this result suggests that the increase of the spread is not systematically associated with a forecast improvement. The largest decrease of the bias (not shown) occurs where the decrease of the RMSE is the strongest. Nevertheless, the spatial variability of the bias over each station generates a noisy signal elsewhere. Note that, this preliminary validation has been calculated using only 20 summer periods from the same year. A longer time series should be analyzed to quantify these improvements.

Fig. 15.
Fig. 15.

(a) RMSE of the 2-m temperature (K) in the NEW experiment after 48 h of simulation. (b) RMSE difference (NEW minus ATM experiments) of the 2-m temperature (K) after 48 h of simulation. Blue points indicate improvement of the NEW when compared with ATM.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

The Brier skill score (BSS; Brier 1950), which measures the mean squared probability error, has been calculated for the accumulation of precipitation for the last 24-h period over 1782 stations. The BSS are very close for both NEW and ATM experiments (Fig. 16a), with a small improvement for precipitation for quantities less than 10 mm and larger than 18 mm (Fig. 16b). Nevertheless, using the bootstrap method, the differences of these scores appear to be not significant. This could be due to the limited number of days used in this study. A longer study period would be required to better conclude if precipitation forecasts improve using surface perturbations.

Fig. 16.
Fig. 16.

(a) Brier skill score for daily precipitation (accumulation between 24 and 48 h of simulation) in the ATM and NEW experiments (blue and red lines, respectively) following the precipitation amounts. (b) As in (a), but for the Brier score difference for ATM minus NEW experiments (negative values indicate improvement of the forecast), dotted lines show the 90% confidence level based on the bootstrap method.

Citation: Monthly Weather Review 141, 5; 10.1175/MWR-D-11-00354.1

6. Conclusions

This study is a first effort to integrate surface uncertainties within the Canadian REPS. For 20 summer periods in 2009, stochastic perturbations to the surface parameters have been generated in several experiments. Each experiment corresponds to 20 simulations differing by the perturbations at the initial time of one or several surface parameters (e.g., vegetation fraction, LAI, sea ice fraction, etc.) or prognostic variables (e.g., soil moisture, soil temperature in different layers, etc.).

When examining the impact of the surface perturbations on the 2-m temperature, 10-m wind speed, cloud fraction, and precipitation, the results show that the southern part of the United States is very sensitive especially to soil moisture and to albedo perturbations. Table 5 summarizes the impacts of the various surface parameter perturbations upon the screen-level variable. When the standard deviations are larger than 10% of the ATM experiment, the impact is considered significant. This table indicates that several parameter perturbations (land–sea mask, sea ice thickness, and fraction) have negligible influence on the atmosphere during the summer period. In Table 5, it is also shown that soil temperature perturbations significantly impact 2-m temperatures particularly as a result of the large impact during the first hours of simulation. Albedo, LAI, and roughness length perturbations impact only one particular screen-level variables. Vegetation fraction perturbations, which tend to modify other vegetation characteristics such as LAI and albedo, are equally important for all screen-level variables except for the wind speed. The 10-m wind speed, similar to precipitation, is relatively insensitive to a variety of surface perturbations. Finally, only SST and soil moisture perturbations influence all the screen-level variables and cloud fraction.

Table 5.

Summary table of the impact of each parameter and variable on the 2-m temperature, 10-m wind speed, cloud fraction, and precipitation during the summer season averaged over the North American region. Check marks indicate an impact of the perturbations on screen-level variable larger than 10% of the ATM experiment.

Table 5.

The time evolution of these surface perturbations over the North American continent has also been studied. Whereas the behavior of the standard deviation evolution is typically controlled by the LSF experiment, surface characteristic perturbations can increase the ensemble spread by more than 30% for 2-m temperature over specific areas. These impacts are generally larger during the daytime as compared to the nighttime.

Using ensemble scores, a preliminary study based on the same summer cases in 2009 has been done to assess the impact of the implementation of the surface perturbations in the Canadian REPS. The 2-m temperature and wind forecasts appear significantly improved. These improvements are greatest at the end of the day, when the system is known to have the largest errors. A larger number of precipitation events in both summer and winter are needed before firm conclusions can be stated. In winter, the sea ice and snow should generate a stronger impact upon the atmosphere. Also, future work will evaluate the impact of surface perturbations on the overall performance of the REPS based on objective evaluation against observations. This will focus upon the integration of surface perturbations into a new version of the Canadian REPS with higher horizontal (20 km instead of 33 km) and vertical resolutions. This horizontal resolution improvement will allow for the generation of larger surface perturbations associated with the larger variability of surface conditions. However, the modification of the vertical resolution, especially in the boundary layer, may modify the surface–atmosphere interactions.

REFERENCES

  • Adams, D., and A. Comrie, 1997: The North American monsoon. Bull. Amer. Meteor. Soc., 78, 21972213.

  • Adler, R., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 11471167.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., L. Crevier, J. Mailhost, B. Bilodeau, and V. Delage, 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., R. Brown, J. Mailhost, B. Bilodeau, and L. Crevier, 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
  • Benoit, J., J. Cote, and J. Mailhot, 1989: Inclusion of the TKE boundary layer parameterization in the Canadian regional finite-element model. Mon. Wea. Rev., 117, 17261750.

    • Search Google Scholar
    • Export Citation
  • Bhumralkar, C., 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
  • Brasnett, B., 2008: The impact of satellite retrievals in a global sea-surface-temperature analysis. Quart. J. Roy. Meteor. Soc., 134, 17451760.

    • Search Google Scholar
    • Export Citation
  • Brier, G., 1950: Verification of forecasts expressed in terms of probability. Mon. Wea. Rev., 78, 13.

  • Brotzge, J., and K. Crawford, 2003: Examination of the surface energy budget: A comparison of eddy correlation and Browen ration measurement systems. J. Hydrometeor., 4, 160178.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., M. Miller, and T. Palmer, 1999: Stochastic representation of model uncertainties in the ECMWF ensemble prediction system. Quart. J. Roy. Meteor. Soc., 125, 28872908.

    • Search Google Scholar
    • Export Citation
  • Bukovsky, M., and D. Karoly, 2007: A brief evaluation of precipitation from the North American Regional Reanalysis. J. Hydrometeor., 8, 837846.

    • Search Google Scholar
    • Export Citation
  • Charron, M., G. Pellerin, L. Spacek, P. Houtekamer, N. Gagnon, H. Mitchell, and L. Michelin, 2010: Towards random sampling of model error in the Canadian ensemble prediction system. Mon. Wea. Rev., 138, 18771901.

    • Search Google Scholar
    • Export Citation
  • Chase, T., R. Pielke, T. Kittel, R. Nemani, and S. Running, 1996: Sensitivity of a general circulation model to global changes in leaf area index. J. Geophys. Res., 101 (D3), 73937408.

    • Search Google Scholar
    • Export Citation
  • Chelton, D., 2005: The impact of SST specification on ECMWF surface wind stress fields in the eastern tropical Pacific. J. Climate, 18, 530550.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and Coauthors, 1996: Modeling of land surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101 (D3), 72517268.

    • Search Google Scholar
    • Export Citation
  • Chen, F., Z. Janjic, and K. Michell, 1997: Impact of atmospheric surface-layer parametrizations in the new land-surface scheme of the NCEP mesoscale Eta model. Bound.-Layer Meteor., 85, 391421.

    • Search Google Scholar
    • Export Citation
  • Coté, J., A. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998a: The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395.

    • Search Google Scholar
    • Export Citation
  • Coté, J., J.-G. Desmarais, A. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998b: The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 13971418.

    • Search Google Scholar
    • Export Citation
  • Efron, B., and R. Tibshirani, 1993: An Introduction of the Bootstrap. Chapman and Hall, 460 pp.

  • Eltahir, E., 1998: A soil moisture-rainfall feedback mechanism. 1: Theory and observations. Water Resour. Res., 34, 765776.

  • Fujita, T., D. Stensrud, and D. Dowell, 2007: Surface data assimilation using an ensemble Kalman filter approach with initial condition and model physics uncertainties. Mon. Wea. Rev., 135, 18461868.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., 2000: Decomposition of the continuous ranked probability score for ensemble prediction systems. Wea. Forecasting, 15, 559570.

    • Search Google Scholar
    • Export Citation
  • Houtekamer, P., and H. Mitchell, 2005: Ensemble Kalman filtering. Quart. J. Roy. Meteor. Soc., 131, 32693289.

  • Jaquemin, B., and J. Noilhan, 1990: Sensitivity study and validation of a land surface parametrization using the HAPEX-MOBILHY data set. Bound.-Layer Meteor., 52, 93134.

    • Search Google Scholar
    • Export Citation
  • Jiang, X., G.-Y. Niu, and Z.-L. Yang, 2009: Impacts of vegetation and groundwater dynamics on warm season precipitation over the central United States. J. Geophys. Res., 114, D06109, doi:10.1029/2008JD010756.

    • Search Google Scholar
    • Export Citation
  • Kain, J., and J. Fritsch, 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 24, Amer. Meteor. Soc., 165170.

  • Koster, R., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305 (5687), 11381140, doi:10.1126/science.1100217.

    • Search Google Scholar
    • Export Citation
  • Kushnir, Y., R. Seager, M. Ting, N. Naig, and J. Nakamura, 2010: Mechanisms of tropical Atlantic SST influence on North American precipitation variability. J. Climate, 23, 56105628.

    • Search Google Scholar
    • Export Citation
  • Lee, M.-I., and Coauthors, 2007: An analysis of the warm-season diurnal cycle over the continental United States and northern Mexico in general circulation models. J. Hydrometeor., 8, 344366.

    • Search Google Scholar
    • Export Citation
  • Li, J., and H. Barker, 2005: A radiation algorithm with correlated-k distribution. Part I: Local thermal equilibrium. J. Atmos. Sci., 62, 286309.

    • Search Google Scholar
    • Export Citation
  • Li, X., M. Charron, L. Spacek, and G. Candille, 2008: A regional ensemble prediction system based on the moist targeted singular vectors and stochastic parameter perturbations. Mon. Wea. Rev., 136, 443462.

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

    • Search Google Scholar
    • Export Citation
  • Matheson, J., and R. Winkler, 1976: Scoring rules for continuous probability distributions. Manage. Sci., 22, 10871095.

  • McFarlane, N., 1987: The effect of orographically excited gravity wave drag on the general circulation of the lower stratosphere and troposphere. J. Atmos. Sci., 44, 17751800.

    • Search Google Scholar
    • Export Citation
  • McLaughlin, D., Y. Zhou, D. Entekhabi, and V. Chatdarong, 2006: Computational issues for large-scale land surface data assimilation problems. J. Hydrometeor., 7, 494510.

    • Search Google Scholar
    • Export Citation
  • Miguez-Machado, G., and J. Peagle, 2001: Sensitivity of North American numerical weather prediction to initial state uncertainty in selected upstream subdomains. Mon. Wea. Rev., 129, 20052022.

    • Search Google Scholar
    • Export Citation
  • Mo, K., 2000: Intraseasonal modulation of summer precipitation over North America. Mon. Wea. Rev., 128, 14901505.

  • Mo, K., and E. Kalnay, 1991: Impact of sea surface temperature anomalies on the skill of monthly forecasts. Mon. Wea. Rev., 119, 27712793.

    • Search Google Scholar
    • Export Citation
  • Mo, K., and J. Paegle, 2000: Influence of the sea surface temperature anomalies on the precipitation regimes over the southwest United States. J. Climate, 13, 35883598.

    • Search Google Scholar
    • Export Citation
  • Morcrette, J.-J., H. Barker, J. Cole, M. Iacono, and R. Pincus, 2008: Impact of a new radiation package, MCRAD, in the ECMWF integrated forecasting system. Mon. Wea. Rev., 136, 47734798.

    • Search Google Scholar
    • Export Citation
  • Niyogi, D., T. Holt, S. Zhong, P. Pyle, and J. Basara, 2006: Urban and land surface effects on the 30 July 2003 mesoscale convective system event observed in the Southern Great Plains. J. Geophys. Res., 111, D19107, doi:10.1029/2005JD006746.

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

    • Search Google Scholar
    • Export Citation
  • Nutter, P., D. Stensrud, and M. Xue, 2004: Effects of coarsely resolved and temporally interpolated lateral boundary conditions on the dispersion of limited-area ensemble forecast. Mon. Wea. Rev., 132, 23582377.

    • Search Google Scholar
    • Export Citation
  • Palmer, T., R. Buizza, F. Doblas-Reyes, T. Jung, M. Leutbecher, G. Shutts, M. Steinheimer, and A. Weisheimer, 2009: Stochastic parametrization and model uncertainty. ECMWF Tech. Memo. 598, Vol. 139, ECMWF, 42 pp.

  • Smith, C., M. Lakhtakia, W. Capehart, and T. Carlson, 1994: Initialization of soil-water content in regional-scale atmospheric prediction models. Bull. Amer. Meteor. Soc., 75, 585593.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D., and M. Wandishin, 2000: Correspondence and spread ratios in forecasts verification. Wea. Forecasting, 15, 593602.

  • Stensrud, D., and N. Yussouf, 2003: Short-range ensemble prediction of 2-m temperature and dewpoint temperature over New England. Mon. Wea. Rev., 131, 25102524.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D., H. Brooks, J. Du, M. Tracton, and E. Rogers, 1999: Using ensembles for short-range forecasting. Mon. Wea. Rev., 127, 433446.

    • Search Google Scholar
    • Export Citation
  • Sundqvist, H., E. Berge, and J. Kristjansson, 1989: Condensation and cloud parameterization studies with a mesoscale numerical weather prediction model. Mon. Wea. Rev., 117, 16411657.

    • Search Google Scholar
    • Export Citation
  • Talagrand, O., R. Vautard, and B. Strauss, 1999: Evaluation of probabilistic prediction systems. Proc. ECMWF Workshop on Predictability, Reading, United Kingdom, ECMWF, 1–25.

  • Treut, H. L., and Z. Li, 1991: Sensitivity of an atmospheric general circulation model to prescribed SST changes: Feedback effects associated with the simulation of cloud optical properties. Climate Dyn., 5, 175187.

    • Search Google Scholar
    • Export Citation
  • Vannitsem, S., and Z. Toth, 2002: Short-term dynamics of model errors. J. Atmos. Sci., 59, 25942604.

  • Wandishin, M., S. Mullen, D. Stensrud, and H. Brooks, 2001: Evaluation of a short-range multimodel ensemble system. Mon. Wea. Rev., 129, 729747.

    • Search Google Scholar
    • Export Citation
  • Yeh, K.-S., J. Coté, A. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 2002: The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part III: Nonhydrostatic formulation. Mon. Wea. Rev., 130, 339356.

    • Search Google Scholar
    • Export Citation
  • Yu, W., L. Garand, and A. Dastoor, 1997: Evaluation of model clouds and radiation at 100 km scale using goes data. Tellus, 49, 246262.

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

    • Search Google Scholar
    • Export Citation
Save