Predicting the Net Basin Supply to the Great Lakes with a Hydrometeorological Model

Daniel Deacu Recherche en prévision numérique environnementale, Meteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Daniel Deacu in
Current site
Google Scholar
PubMed
Close
,
Vincent Fortin Recherche en prévision numérique environnementale, Meteorological Research Division, Environment Canada, Dorval, Quebec, Canada

Search for other papers by Vincent Fortin in
Current site
Google Scholar
PubMed
Close
,
Erika Klyszejko Water Survey of Canada, Environment Canada, Ottawa, Ontario, Canada

Search for other papers by Erika Klyszejko in
Current site
Google Scholar
PubMed
Close
,
Christopher Spence Aquatic Ecosystem Impacts Research Division, Environment Canada, Saskatoon, Saskatchewan, Canada

Search for other papers by Christopher Spence in
Current site
Google Scholar
PubMed
Close
, and
Peter D. Blanken Department of Geography, University of Colorado, Boulder, Colorado

Search for other papers by Peter D. Blanken in
Current site
Google Scholar
PubMed
Close
Full access

We are aware of a technical issue preventing figures and tables from showing in some newly published articles in the full-text HTML view.
While we are resolving the problem, please use the online PDF version of these articles to view figures and tables.

Abstract

The paper presents the incremental improvement of the prediction of the Great Lakes net basin supply (NBS) with the hydrometeorological model Modélisation Environmentale–Surface et Hydrologie (MESH) by increasing the accuracy of the simulated NBS components (overlake precipitation, lake evaporation, and runoff into the lake). This was achieved through a series of experiments with MESH and its parent numerical weather prediction model [the Canadian Global Environmental Multiscale model in its regional configuration (GEM Regional)]. With forcing extracted from operational GEM Regional forecasts, MESH underestimated the NBS in fall and winter. The underestimation increased when the GEM precipitation was replaced with its corrected version provided by the Canadian Precipitation Analysis. This pointed to overestimated lake evaporation and prompted the revision of the parameterization of the surface turbulent fluxes over water used both in MESH and GEM. The revised parameterization was validated against turbulent fluxes measured at a point on Lake Superior. Its use in MESH reduced the lake evaporation and largely corrected the NBS underestimation. However, the Lake Superior NBS became overestimated, signaling an inconsistency between the reduced lake evaporation and the prescribed precipitation. To remove the inconsistency, a new forcing dataset (including precipitation) was generated with the GEM model using the revised flux parameterization. A major NBS simulation improvement was obtained with the new atmospheric forcing reflecting the atmospheric response to the modified surface fluxes over the lakes. Additional improvements resulted by correcting the runoff with a modified snowmelt rate and by insertion of observed streamflows. The study shows that accurate lake evaporation simulation is crucial for accurate NBS prediction.

Corresponding author address: Vincent Fortin, Meteorological Research Division, Environment Canada, 2121 Trans-Canada Highway, 5th floor, Dorval QC H9P 1J3, Canada. E-mail: vincent.fortin@ec.gc.ca

Abstract

The paper presents the incremental improvement of the prediction of the Great Lakes net basin supply (NBS) with the hydrometeorological model Modélisation Environmentale–Surface et Hydrologie (MESH) by increasing the accuracy of the simulated NBS components (overlake precipitation, lake evaporation, and runoff into the lake). This was achieved through a series of experiments with MESH and its parent numerical weather prediction model [the Canadian Global Environmental Multiscale model in its regional configuration (GEM Regional)]. With forcing extracted from operational GEM Regional forecasts, MESH underestimated the NBS in fall and winter. The underestimation increased when the GEM precipitation was replaced with its corrected version provided by the Canadian Precipitation Analysis. This pointed to overestimated lake evaporation and prompted the revision of the parameterization of the surface turbulent fluxes over water used both in MESH and GEM. The revised parameterization was validated against turbulent fluxes measured at a point on Lake Superior. Its use in MESH reduced the lake evaporation and largely corrected the NBS underestimation. However, the Lake Superior NBS became overestimated, signaling an inconsistency between the reduced lake evaporation and the prescribed precipitation. To remove the inconsistency, a new forcing dataset (including precipitation) was generated with the GEM model using the revised flux parameterization. A major NBS simulation improvement was obtained with the new atmospheric forcing reflecting the atmospheric response to the modified surface fluxes over the lakes. Additional improvements resulted by correcting the runoff with a modified snowmelt rate and by insertion of observed streamflows. The study shows that accurate lake evaporation simulation is crucial for accurate NBS prediction.

Corresponding author address: Vincent Fortin, Meteorological Research Division, Environment Canada, 2121 Trans-Canada Highway, 5th floor, Dorval QC H9P 1J3, Canada. E-mail: vincent.fortin@ec.gc.ca

1. Introduction

There is a growing need for accurate prediction of water level changes in the Great Lakes (Fig. 1) for the purposes of managing their water resources (Neff and Nicholas 2005). The physical processes over a lake basin that contribute to the water level change of that lake include the overlake precipitation (Plake), lake evaporation (Elake), and runoff into the lake from its drainage basin (R). The net basin supply (NBS), defined by
e1
is a measure commonly applied to evaluate prediction skill on seasonal to annual time scales.
Fig. 1.
Fig. 1.

The Great Lakes and their drainage basin boundaries (marked in red). The drainage basins of Lakes Michigan, Huron, and Georgian Bay are merged in the figure. The Great Lakes are connected to each other by a system of channels. The topography (m) of the GEM Regional model for this area is shown in shades of gray and blue. The closed black circle in Lake Superior marks the location of the Stannard Rock Light.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

A peculiarity of the Great Lakes is their proximity to the cold continental polar air mass forming over northern Canada in late fall and winter and to a major extratropical storm track. In late fall and winter, it is common for northeastward-traveling low-pressure systems to advect cold air of northern origin over the much warmer Great Lakes water, causing very high lake evaporation rates. This leads to intense convection in the overlying atmospheric boundary layer and to a strong local atmospheric response manifested through lake-effect precipitation (e.g., Laird and Kristovich 2002; Liu and Moore 2004). Examples of the typical synoptic atmospheric conditions prevalent during such cold-air outbreaks and of the ensuing lake-effect clouds are shown in Figs. 2 and 3. Cold-air outbreaks have an important effect on the NBS both through the net overlake precipitation, which can become even negative, and runoff generated from lake-effect precipitation. In this context, the NBS can be viewed as a spatial and temporal integrator of atmospheric, surface, and subsurface hydrologic processes taking place within the lake watershed. Therefore, a model capable of simulating all of these processes, including the surface–atmosphere interaction and water recycling, such as coupled land–lake–ice–atmosphere models, would be ideal for the prediction of the Great Lakes NBS.

Fig. 2.
Fig. 2.

Surface analysis valid at 1800 UTC 15 Dec 2008 (adapted from the chart produced by the NOAA/NWS Hydrometeorological Prediction Center and valid at the same time). The sea level pressure (hPa) is contoured in dark red every 4 hPa.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

Fig. 3.
Fig. 3.

Aqua MODIS 2-km resolution images (Bands 7–2–1) at (a) 1900 UTC 15 Dec and b) 1900 UTC 31 Dec 2008 showing lake-effect cloud streets (convective roll clouds) that developed over Lakes Superior, Michigan, and Huron during cold-air outbreaks. The red plus sign marks the location of the Stannard Rock Light. Image credit: NASA Goddard Space Flight Center (GSFC), MODIS Rapid Response (http://rapidfire.sci.gsfc.nasa.gov).

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

The National Oceanic and Atmospheric Administration (NOAA) Great Lakes Environmental Research Laboratory (GLERL) has a long tradition in developing estimation methods for the Great Lakes NBS (Quinn and Kelley 1983) based on a classical hydrometeorological approach that relies on point observations for the estimation of Plake and Elake. Inconsistencies between these two independently estimated NBS components occur because they are estimated using observed data that are scarce in time and space. In other words, the accuracy of these conventional NBS estimates can be compromised by data scarcity.

Most hydrometeorological models lack the capability to reproduce the water budget of a certain region by simulating the complete water cycle in the same way global coupled atmosphere–ocean–land models equipped with a land hydrology component do. Instead, these models, including the Modélisation Environmentale–Surface et Hydrologie (MESH) model (Pietroniro et al. 2007) used in this study, are forced with prescribed atmospheric fields derived from atmospheric analyses or forecasts. A problem arising in the effort to improve their skill is that changes in surface fluxes simulated by modified versions of the same model are not accompanied by consistent atmospheric responses, and hence the atmospheric feedback to the surface is precluded. The effect of this limitation on the model simulations remains practically unknown.

The purpose of this paper is to present and discuss the improvement of the skill of the hydrometeorological model MESH in predicting the Great Lakes NBS. Model skill is evaluated against monthly NBS estimates obtained from hydrological measurements. The aim is to demonstrate incremental improvements to the accuracy of the simulated NBS components over a series of experiments. The effect of a corrected precipitation forcing provided by a precipitation analysis is first assessed alone and then in combination with the effect of a corrected lake evaporation. This evaporation is calculated by a modified parameterization of the surface fluxes over the lakes, which is validated against observed fluxes at a point on Lake Superior. The effect of increasing the consistency between the corrected lake evaporation and the prescribed overlake precipitation, especially during lake-effect precipitation events, is next addressed. For this, the MESH model is forced with the precipitation simulated by a numerical weather prediction model [the Canadian Global Environmental Multiscale model in its regional configuration (GEM Regional)] that uses the same modified surface flux parameterization and yields very similar lake evaporation to MESH. Finally, the effect on the simulated NBS of corrections of the runoff to the lakes produced by two modifications of the snow parameterization and by replacing simulated with observed river flows is assessed.

The motivation for future use of coupled land–lake–ice–atmosphere models for the prediction of Great Lakes NBS is improved consistency between precipitation, lake evaporation, and basin runoff. Improvements will be hindered without incorporating such interaction in hydrometeorological models. Therefore, the present study could be placed alongside those that show the beneficial effect of the surface–atmosphere coupling on the surface model development. For instance, Liu et al. (2003) found a deficiency of the parameterization of canopy evaporation when they coupled their land surface model with a single-column atmospheric model—the coupling largely amplified the effect of the deficiency on the surface energy partitioning. Dutra et al. (2010) evaluated their improved snow scheme for the ECMWF land surface model in offline mode but acknowledged that complete validation can only be obtained in coupled surface–atmosphere simulations. Van den Hurk et al. (2002) describe the sensitivity of the precipitation produced by a regional climate model covering the Baltic Sea catchment to the choice of the land surface in their study and argue that some of the effects of these surface models may not be reproduced in uncoupled evaluations. It has even been argued that the neglect of the land–atmosphere interactions and feedbacks in modeling studies may distort our understanding of the water and energy cycle processes (Santanello et al. 2009; Lyon et al. 2008; Betts 2009).

2. Models

a. The GEM Regional model

From 18 May 2004 to 20 October 2010, the numerical weather prediction (NWP) model used at the Canadian Meteorological Centre (CMC) as part of Environment Canada’s regional deterministic prediction system (RDPS) was a stretched-grid configuration of the Global Environmental Multiscale model. This model configuration, referred to as the GEM Regional model, had an average horizontal grid spacing of ~15 km over North America (see Mailhot et al. 2006). The role of the GEM Regional model in this study is twofold: to test modified versions of its parameterization of the turbulent surface fluxes over the water surface and to provide atmospheric forcing to the MESH model (see next section).

b. The MESH model

1) Description

MESH is a hydrometeorological model developed by the Meteorological Research Division of Environment Canada (Pietroniro et al. 2007; Carrera et al. 2010). It is composed of a surface model derived from the surface component of the GEM Regional model and a river routing model (Fig. 4). The primary role of the surface model is to calculate the surface turbulent momentum, sensible heat, and latent heat fluxes, which represent the exchange of momentum, heat, and moisture between the surface and the atmosphere. These fluxes are obtained by aggregation of fluxes calculated over four possible surface types (land, glaciers and ice sheets, and open water and sea ice) within a grid cell. For each surface type, except for open water, there is a corresponding column model representing a slab of vegetated land, glacier–ice sheet, or sea ice. In the absence of a lake–ocean model to simulate the evolution of the open water surface temperature, this temperature remains unchanged during the integration from the initial conditions provided by an analysis of the same field. For vegetated land, the user can opt for a model based on a simple force–restore scheme, the Canadian Land Surface Scheme (CLASS; Verseghy 2000), or the Canadian implementation of Interactions between Soil, Biosphere, and Atmosphere (ISBA) (Bélair et al. 2003a), which will be referred to simply as ISBA. ISBA has been used both in the GEM Regional and MESH simulations presented herein.

Fig. 4.
Fig. 4.

The components of the GEM Regional and MESH models. The arrows indicate the direction of the information flow between model components.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

ISBA is a two-layer force–restore model representing the energy and water budget of a composite soil–vegetation–snow medium, and has been used both in the GEM Regional and MESH model simulations presented herein. ISBA includes a subgrid surface runoff scheme based on the Variable Infiltration Capacity (VIC) model that takes into consideration the subgrid variability in soil moisture in the calculation of a saturation excess runoff (Wood et al. 1992; Liang et al. 1994; Decharme and Douville 2006). The scheme partitions the water flux reaching the land surface into surface runoff and infiltration. ISBA also calculates interflow (lateral flow occurring in the upper part of the unsaturated zone) and gravitational drainage based on a subgrid-scale parameterization proposed by Soulis et al. (2011).

The model used for gridded river routing is WATROUTE, which is the routing component of the hydrological model WATFLOOD (Kouwen 2010). WATROUTE and ISBA share the same grid. The drainage simulated by ISBA is directly transferred to WATROUTE, whereas the surface runoff is first delayed over the grid cell by the kinematic wave method to simulate the overland flow (flow over the land surface following its slope) and then added to the interflow before being transferred. WATROUTE converts the gravitational drainage to baseflow (channel flow contribution of seepage from groundwater aquifers) using a reservoir with a power law release rule. The baseflow is then combined with the overland flow and interflow to obtain a total lateral flow. This lateral flow is then routed through the channel network using a storage routing technique based on continuity and the Manning formula that provides a streamflow estimate for each grid cell. The WATROUTE parameter values used in this study originate from previous studies of WATROUTE developers and users focusing on rivers on the Canadian side of the Great Lakes drainage basin (e.g., Pietroniro et al. 2007).

A MESH run starts with a series of 24-h surface model runs in offline mode with a half-hour time step and started at 0000 UTC of each day. The first run in the series is initialized with fields simulated at the end of a continuous multiday run, whereas each subsequent run is initialized with fields simulated at the end of the previous 24-h run. As an exception, the lake water temperature and ice-related fields come from daily CMC surface analyses, as MESH does not currently include a lake model. The specified forcing fields are near-surface (~40 m) wind and air temperature and humidity, downward shortwave and longwave radiation fluxes at the surface, surface pressure, and hourly precipitation accumulation. Once the surface model runs are complete, the fields to be transferred to WATROUTE are prepared and a series of 24-h WATROUTE runs with a time step of 1 h is performed. The MESH model calculates the NBS from the prescribed overlake precipitation, the lake evaporation simulated by its surface model, and the runoff into the lake calculated by WATROUTE.

2) Configuration and forcing

The MESH model domain covers the Great Lakes basin (Fig. 5) with a ⅙° latitude–longitude grid, having an average grid spacing of ~15 km. Since the model configuration is based on that used in the study described in Pietroniro et al. (2007), details on the determination of the vegetation types, soil texture, and parameters used in ISBA and on the physiographic parameters required by WATROUTE are not repeated here. The forcing fields used in our study cover the period June 2004–May 2009 and originate from hourly forecasts in the 0600–1800-h range from twice-daily runs of the GEM Regional model initialized at 0000 and 1200 UTC. The 0600–1800-h range was chosen to avoid errors due to model spinup, such as an underestimation of the precipitation amounts (Mahfouf et al. 2007). Because the horizontal grid spacing of about 15 km of the GEM Regional model is close to the average grid spacing of the MESH, downscaling correction of the forcing fields was not required.

Fig. 5.
Fig. 5.

The domain of the MESH model covers the Great Lakes basin and a surrounding area (light gray). The horizontal resolution of the model is ⅙°. Shown in light blue are the lake areas used by WATROUTE. The drainage basin of Lake St. Clair (not considered in this study) is shown in dark gray. The other colors depict the areas of the drainage basin that were gauged a percentage of the time between June 2004 and December 2008 limited by the thresholds shown in the legend. The watersheds of the Trent and Spanish Rivers are marked TRW and SRW, respectively.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

Precipitation forcing extracted from the Canadian Precipitation Analysis (CaPA) was also tested in the study. CaPA uses optimal interpolation to combine precipitation reports from synoptic stations with a first estimate of the precipitation (background) field provided by 6-h accumulated precipitation fields predicted by the GEM Regional model in the 0600–1800-h forecast range. Therefore, the CaPA precipitation can be seen as a corrected GEM Regional precipitation. MESH is forced with hourly precipitation accumulations resulting from the disaggregation of the 6-h CaPA precipitation accumulation based on fractions derived from the disaggregation of the 6-h GEM Regional precipitation accumulation into hourly values. More details about the CaPA can be found in Mahfouf et al. (2007) and Carrera et al. (2010). The latter paper points to deficiencies in CaPA, which were fixed to avoid introducing biases in the analysis process.

3. Model parameterization modifications

a. Modifications to the parameterization of the surface turbulent heat fluxes over water

In early tests (not shown), MESH generally underestimated the monthly NBS in winter, indicating a possible overestimation of the lake evaporation. A closer look at the model’s surface flux parameterization and settings has revealed two potential sources of errors in the calculation of the surface turbulent fluxes: the scalar roughness length of the water surface and the specified value of the Prandtl number. Both are discussed below.

Originally, both in the MESH and GEM Regional models, the heat and moisture roughness lengths, z0h and z0q, were set equal to the momentum roughness length (z0m) given by the Charnock relation for the sea surface roughness length (Charnock 1955):
e2
where u* denotes the friction velocity and g the gravitational acceleration. The sea state–dependent Charnock coefficient β is set here to a constant value of 0.018, corresponding to a saturated wave state. Powell et al. (2003) give the range 0.015–0.035 for β. Equation (2) yields z0m values that are commonly accepted for winds below hurricane strength (e.g., Powell et al. 2003). However, the same equation gives z0h and z0q increasing with the near-surface wind speed, whereas a decrease of these quantities with increasing near-surface wind speed was inferred from observations for moderate and high winds (DeCosmo et al. 1996; Fairall et al. 2003). Since higher z0h and z0q translate into higher sensible heat and moisture fluxes, overestimation of these fluxes is likely—more so with increasing wind speed. Therefore, we have tested the following formulation of the scalar roughness length of the water surface:
e3
with ν (the kinematic viscosity of air) set equal to 1.5 × 10−5 m2 s−1. The expression used for u* > 0.2 m s−1 was taken from the European Centre for Medium-Range Weather Forecasts (ECMWF) model (ECMWF 2011), but our coefficient (1.333) is larger.

As with the original z0h and z0q, the use of a prescribed constant value of 0.85 for the turbulent Prandtl number (Prt) in the surface flux parameterization of the model could also lead to an overestimation of the surface turbulent heat fluxes in unstable conditions, although to a smaller degree. Since this constant value does not always match the correct and variable Prt value resulting from the ratio of the stability functions for heat and momentum, it actually loses its physical meaning and intended role in our parameterization. The specified value remains, however, necessary in the model’s turbulent mixing parameterization to set the ratio of the momentum diffusivity to the scalar (heat and moisture) diffusivity. Changing the specified Prt value from 0.85 to 1, so that it loses its direct impact on the surface fluxes, has provided a temporary solution to the problem.

b. Modifications to the snow parameterization

ISBA includes the simple but realistic one-layer snow model of Douville et al. (1995), which was implemented in a modified form in the GEM Regional model (Bélair et al. 2003b) and consequently became part of MESH. The prognostic variables of the model are the snow mass, liquid water within the snowpack, snow density, and snow albedo. Below, we describe the model modifications that improved the NBS prediction in our MESH experiments.

The original snowmelt rate in ISBA (excluding the rain-on-snowmelt contribution) entering the snow mass budget equation is given by (Bélair et al. 2003b)
e4
where psn is the gridcell fraction covered by snow, Tn is a diagnostic snow temperature, T0 = 273.16 K, Lf is the latent heat of vaporization, and Δt is the time step. The quantity Cs is a thermal coefficient for snow depending on the thermal conductivity and heat capacity of snow, both varying with the snow density. The first change of the snowmelt rate consisted in the modification of the snow temperature over all cells with a combined fractional coverage of needleleaf and mixed forests less than 0.2. This cutoff value was considered reasonable for the delimitation of the needleleaf and mixed forest within the Great Lakes basin, given the uncertainty of the fractional coverage used in our model. The approximate percentage of the grid cells in the drainage basin affected by the modification was 100% for Lake Erie, 80% for Lake Ontario, 50% for Lake Michigan–Huron–Georgian Bay (MHG), and 5% for Lake Superior. Over these grid cells, the new snow temperature is given by
e5
where Ts is the surface temperature resulting from the energy budget equation; TA is the air temperature at about 40 m, which is part of the atmospheric forcing; and α = 0.7. This snow temperature modification is meant to enhance the snowmelt during warm air advection episodes when the temperature of the overlying air is greater than the surface temperature Ts. The Tn is left unchanged over the rest of the domain, where it is set to a weighted average of the surface temperature Ts and a deep-soil temperature, which is ISBA’s other prognostic temperature. The melt rate with the new snow temperature becomes directly dependent on the low-level air temperature and resembles the melt rates provided by the simplistic but very successful temperature index models (Hock 2003). Ohmura (2001) explains that what makes the low-level air temperature a very effective parameter for melt rates is the high correlation of this temperature with the main energy sources for melt (in the order of importance): longwave atmospheric radiation, shortwave radiation, and sensible heat flux.
The second modification consists of replacing with in Eq. (4), which turns the snowmelt rate expression into that originally used in ISBA by Douville et al. [1995, their Eq. (12)]. The quantity CT is the composite thermal coefficient that appears in the energy budget equation and includes the effect of bare soil, vegetation, and snow in our model (Bélair et al. 2003b):
e6
where veg is the grid cell fraction covered by vegetation; psng and psnυ are the fraction of bare soil and vegetation covered by snow, respectively; and Cg and Cυ are thermal coefficients for bare soil and vegetation. The Cg depends on the soil texture and moisture content, whereas Cυ depends on the type of vegetation (high or low) and its leaf area index. The effect of this second modification was also to increase the snowmelt, but it was smaller than the effect of the modified snow temperature [Eq. (5)].

4. Data

a. Turbulent heat fluxes observed at Stannard Rock Light

Turbulent latent heat fluxes (LHF) and sensible heat fluxes (SHF) calculated from eddy-covariance data using a 30-min averaging period were used to verify the surface LHF and SHF over water simulated by the GEM Regional and MESH models for December 2008. Their time series are shown in Fig. 6. The eddy-covariance measurements were made at 32.4 m above the water surface of Lake Superior from a mast fastened to the top of the Stannard Rock Light (47.183506°N, 87.22511°W; see Fig. 1)—a historic lighthouse located on a shoal 39 km from the nearest shore (the Keweenaw Peninsula). Their postprocessing is discussed in detail in Blanken et al. (2011) and Spence et al. (2011). The error in eddy-covariance-based turbulent fluxes is commonly quantified based on the quantification of energy balance closure, which requires calculation of the heat storage term for lakes. Blanken et al. (2011) showed that the heat storage term estimated as a residual of the energy balance in Lake Superior compared well with those calculated by Schertzer (1978) using measured water temperature profiles, giving us confidence that the errors in the LHF and SHF are likely within 10%–20% of the actual values within the turbulent flux measurement footprint.

Fig. 6.
Fig. 6.

Observed and simulated (a) LHF and (b) SHF for the month of December 2008. The observed fluxes (black dots) are turbulent fluxes derived from eddy-covariance measurements using a 30-min averaging period. The measurements were taken at the Stannard Rock Light at 32.5-m height. The observed data gaps >2 h are 0430–0700 UTC 15 Dec, 1530–1930 UTC 15 Dec, 0800 UTC 23 Dec–1730 UTC 25 Dec, 2300 UTC 27 Dec–0500 UTC 28 Dec, and 1200 UTC 28 Dec–1730 UTC 31 Dec. Plotted in color are the surface LHF and SHF simulated in the experiments with the GEM Regional model (described in the text) at each hour, at the nearest grid point to the Stannard Rock Light.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

b. Measured flows

WATROUTE can be configured to substitute the flow measured at a station for the simulated outflow from a grid cell associated with that station, which is the channel runoff from the entire upstream drainage area. This substitution was performed in one of the MESH experiments (introduced in section 5b) with the goal of correcting the simulated runoff into the lakes. In that experiment, WATROUTE was provided with streamflow observations from 182 stations around the Great Lakes. Most of the stations had complete records over the study period. The rest had data gaps, especially in last year of the study period. The resulting percentage of time that various areas were gauged at their outlets is illustrated in Fig. 5, which shows that most grid cells belonged to the gauged areas. The runoff from areas that were gauged all of the time was estimated from observations alone. Many areas adjacent to the lakes were ungauged because gauges are generally located upstream of the river outlet to the lake. Streamflow observations were not available for two large Canadian rivers, the Trent River, and the Spanish River (their watersheds are marked in Fig. 5) that flow into Lake Ontario and Georgian Bay, respectively. Therefore, the correction of the runoff into Lakes Ontario and MHG was affected to an unknown degree.

A similar strategy was used to insert the Nipigon reservoir outflow, which passes through the Alexander generating station before reaching Lake Superior. Estimates of the outflow from the Alexander generating station provided by Ontario Power Generation were used to replace the runoff simulated by WATROUTE for the Nipigon reservoir watershed. This strategy was used for all of the MESH experiments. It is important to note that the outflow from the Alexander generating station includes water that is diverted into the Great Lakes watershed from the Ogoki River. By definition, the net basin supply should only include runoff from the original Great Lakes watershed, before man-made diversions changed the drainage basin. For this reason, an estimate of the monthly flow diverted from the Ogoki River has been subtracted from the simulated NBS to Lake Superior in all of the experiments.

c. The component NBS estimated by GLERL and the residual NBS

The direct method for NBS estimation based on Eq. (1) and consisting in the estimation of the NBS components (Plake, Elake, and R) is called the component method. The resulting NBS is called component NBS. By calculating the NBS components, MESH calculates a component NBS and so does GLERL. GLERL’s approach is to employ observations of atmospheric variables such as precipitation, temperature, humidity, and wind as well as observed streamflows to estimate the NBS components, which are then used to produce NBS hindcasts. The NBS estimated by GLERL using their lake precipitation estimation method (GLERL 2011) is compared to the NBS calculated by MESH in section 6.

Alternatively, the NBS to a lake can be estimated indirectly as a residual in the water budget equation of the lake:
e7
where ΔS is the storage change, I is the inflow from the upstream lake, O is the outflow to the downstream lake, and D represents the diversions into and out of the lake. The resulting NBS is called residual NBS and both the U.S. Army Corps of Engineers and Environment Canada estimate provisional monthly values of this quantity. These values are subsequently agreed upon by Canada and the United States and become values of the so-called coordinated residual NBS, herein referred to simply as the residual NBS. To the authors’ knowledge, a consistent, quantitative uncertainty analysis of monthly residual NBS estimates for each of the Great Lakes has never been performed. A number of studies have shown that the connecting-channel flows (I and O) and the storage change (ΔS) terms in Eq. (7) are the greatest sources of uncertainty, as these terms are very large in comparison to the NBS (Neff and Nicholas 2005; Quinn 2009; Bruxer 2010). The effect of connecting-channel flows on residual NBS uncertainty is amplified as you move downstream through the system and flows increase. Furthermore, Bruxer (2010) found that the uncertainty in monthly residual NBS estimates for Lake Erie varied greatly from month to month. In some months the uncertainty was found to be greater than the actual NBS estimate itself.

The residual NBS (IUGLS 2012), which does not include the Ogoki diversion contribution to the Lake Superior NBS, is used herein to verify the NBS simulated by MESH. The units for the NBS and its components used in this paper are mm month−1 and represent a monthly change in lake water volume divided by the lake’s surface area. Values of the residual NBS and GLERL’s component NBS for the months January–May 2009 were not available at the time of the writing of the paper.

5. Experiment design

a. Experiments with the GEM Regional model

Three experiments have been performed with the GEM Regional model to test the impact of the modified scalar roughness length and Prandtl number mentioned in section 3a on the simulated surface turbulent heat fluxes over water. The model was run in its original (operational) configuration in a control experiment (GEM-Reg1) and with the new roughness length given by Eq. (3) in the second experiment (GEM-Reg2). In addition to the modification made in GEM-Reg2, the Prandtl number was changed from 0.85 to 1 in the third experiment (GEM-Reg3). Each experiment consisted of a series of 24-h runs initialized twice daily (at 0000 and 1200 UTC) for the entire month of December 2008. This particular month was chosen because some of the largest latent and sensible heat fluxes over the lakes were simulated then and observations were available for their verification (presented in section 6a).

b. Experiments with the MESH model

Six MESH model experiments covering the study period (June 2004–May 2009) were performed (Table 1). They are reported in chronological order to emphasize the factors considered to have the greatest potential for the NBS simulation improvement at different stages of the study. In the first experiment (MESH1), MESH was forced with a dataset extracted from short-range forecasts of the operational GEM Regional model and archived at the CMC, thought at the time to be the best hourly forcing dataset with a horizontal resolution close to that of MESH. In the second experiment (MESH2), the precipitation was replaced with its corrected version provided by CaPA [section 2b(2)]. In both MESH1 and MESH2, the parameterization of the surface fluxes over water in MESH was identical to that of the operational GEM Regional model, and used in experiment GEM-Reg1. The surface flux parameterization modifications described in section 3a and tested in the GEM Regional model in experiment GEM-Reg3 were also tested in MESH in the third experiment, MESH3, in which the forcing was the same as in MESH2. Consequently, the lake evaporation simulated by MESH changed in MESH3 compared to MESH1 and MESH2. The prescribed precipitation was insensitive to this change and thus inconsistent with the new lake evaporation.

Table 1.

Overview of the MESH experiments. Complete forcing datasets were obtained from operational forecasts with the GEM Regional model using the original parameterization of the turbulent surface fluxes over water (opGEM) and from forecasts with GEM Regional using the modified version of this parameterization described in section 3a (modGEM). The opGEM precipitation served as the background field for the CaPA precipitation analysis. The original and new lake evaporation was simulated by MESH with the original and modified flux parameterization, respectively.

Table 1.

An appropriate response of the atmosphere to changes in surface fluxes, including precipitation consistent with evaporation, can only be obtained in a coupled surface–atmosphere model. Because coupling of the GEM Regional model to MESH is technically complicated, we chose to mimic this coupling in the MESH experiment MESH4. For this, the surface flux parameterization of MESH in MESH3 was kept in MESH4 and the model was forced with a new dataset obtained from reforecasts with the GEM Regional model with the same (modified) surface flux parameterization (as in experiment GEM-Reg3). The reforecasts were obtained in a series of 24-h runs initialized at 0000 and 1200 UTC for the period June 2004–May 2009.

The new forcing was also used in the next two experiments, MESH5 and MESH6. The snow parameterization changes described in section 3b were implemented in the MESH version used in MESH4, and their effect on the NBS was investigated in experiment MESH5. Finally, experiment MESH6 was performed with the MESH configuration used in MESH5 modified to allow for runoff correction through replacement of simulated with observed river flows (see section 4b). Results from all experiments are summarized in the next section.

6. Results

a. Surface heat fluxes at Stannard Rock Light

The surface LHF and SHF simulated in the GEM Regional model experiments at the nearest grid point to the Stannard Rock Light at each hour in the 0600–1800-h forecast range are plotted along with the LHF and SHF observed at Stannard Rock Light (section 4a) in Fig. 6. Before comparing the simulated with the observed fluxes, we comment on the time series of these fluxes, focusing on the atmospheric convection associated with the large surface heat fluxes and on the representativeness of the observed fluxes.

With a few exceptions, the LHF and SHF values plotted in Fig. 6 are positive, indicating an upward moisture and sensible heat transfer and convective atmospheric conditions most of the time. The temporal variability of these fluxes is mainly modulated by synoptic weather events. For instance, the large increase in magnitude on 15 December 2008 occurred during the passage of a cyclone. Over the 12-h period from 0600 to 1800 UTC on that day, the temperature observed at Stannard Rock Light (at 32.4 m) dropped sharply from 5° to −14°C, as the warm and moist air (of southerly origin) in the warm sector of the cyclone was replaced by cold and dry air carried by northwesterly winds behind the cold front associated with the cyclone. During the same time, the observed wind speed increased from 11 to 20 m s−1. The advection of cold and dry air over the warmer open water of Lake Superior during this cold-air outbreak created unstable conditions and large surface heat fluxes, with observed fluxes peaking at about 400 W m−2 (Figs. 6 and 7). The NOAA/National Weather Service (NWS) Hydrometeorological Prediction Center (HPC) surface analysis at 1800 UTC 15 December 2008 (Fig. 2) shows a low-pressure center northeast of the Great Lakes, the cold front associated with the low extending north–south and located east of Lake Erie, and a high-pressure center located over central United States. This synoptic surface weather pattern is very similar to that identified by Laird and Kristovich (2002) to be characteristic for the strong northwesterly near-surface wind and cold-air advection events resulting in large surface heat fluxes over Lake Huron in the fall of the period of their study (April–November 1984). The location of the low-pressure system and its northeastward trajectory on 15–16 December 2008 also agree with the typical synoptic forcing found by Liu and Moore (2004) to be associated with cold-air outbreaks over the Great Lakes leading to intense lake-effect snowstorms over Southern Ontario (east of Lake Huron).

Fig. 7.
Fig. 7.

As in Fig. 6, but for the period 15–16 Dec 2008. The observed data gaps are 0430–0700 and 1530–1930 UTC 15 Dec, the latter for SHF only. Also plotted are the surface LHF and SHF simulated in experiments GEM-Reg3 (closed red circles), MESH3 (open blue circles), and MESH4 (open red circles) at each hour, at the nearest grid point of each model to the Stannard Rock Light. The modified surface flux parameterization over water was used in the three experiments.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

Characteristic to cold-air outbreaks over the western Great Lakes is the development of shallow atmospheric convection that organizes into longitudinal counterrotating rolls (Kristovich and Steve 1995). These rolls are nearly parallel to the mean wind and fill the atmospheric boundary layer. A proof of their existence on 15 December 2008 is given by the Moderate Resolution Imaging Spectroradiometer (MODIS) Aqua satellite image taken at 1900 UTC on the same day (Fig. 3a), in which the cloud bands developing in the updraft region of the rolls are clearly visible. Similar cloud bands can also be seen in MODIS Aqua and Terra satellite images for 7, 16, 22, 30, and 31 December 2008, all of these being days with relatively large simulated and observed fluxes (Fig. 6). Shown here is only an image taken at 1900 UTC 31 December 2008 (Fig. 3b).

The atmospheric boundary layer becomes horizontally inhomogeneous in the presence of longitudinal rolls. Such rolls have a typical horizontal width of a few kilometers during cold-air outbreaks over the Great Lakes (e.g., LeMone 1973), and their coherent updrafts and downdrafts induce considerable spatial variability in wind, temperature, and humidity and, in turn, in surface fluxes (Young et al. 2002). Consequently, the spatial representativeness of the heat fluxes derived from the single-tower measurements is likely insufficient for an area of the size of the model’s grid cell [~(15 km × 15 km)]. Moreover, even at the tower, the local heat fluxes may not be assessed accurately by the eddy-covariance method, given the important contribution of the mean vertical advection resulting from the low-level horizontal convergence or divergence of the flow in the cross-roll direction (Steinfeld et al. 2007). The inability of single-tower measurements to provide regional (area averaged) surface fluxes when coherent structures develop in the boundary layer is well known, and aircraft measurements are usually made in such conditions to obtain more accurate estimates (Mahrt 1998).

The limited representativeness of the observed fluxes makes it difficult for a proper validation of heat fluxes simulated by the GEM Regional model, which represents gridcell-averaged fluxes. Nonetheless, we consider that the observations give a good indication of the simulated fluxes, based on the results that will be presented next. Figure 6 illustrates two expected results: a severe overestimation (by as much as 100%) of the largest heat flux values in GEM-Reg1, and a drastic reduction of this overestimation in GEM-Reg3. The reduction resulted mainly from the use of the new scalar roughness length, according to the results from GEM-Reg2. The beneficial effect on the NBS of the changes made in GEM-Reg3, presented in section 6b, gives another strong indication that more realistic fluxes are simulated in this experiment. Moreover, we have obtained similar positive results by implementing the same changes in the GEM model (at 15-km resolution), which was then run over a domain covering the Gulf of Saint Lawrence both in stand-alone mode and coupled to an ice–ocean model. Much more realistic 2-m temperature and dewpoint values were simulated in extreme conditions created by cold-air outbreaks over the Gulf of Saint Lawrence (similar to those occurring over the Great Lakes) and a large reduction in the positive bias of the same variables was obtained in winter. The results obtained with the modified surface flux parameterization in the GEM Regional model presented herein prompted further testing at the CMC, followed by the recommendation that the modifications be implemented in the next operational version of the model. We note that an inappropriate scalar roughness length over water was also found responsible for the overestimation of the surface sensible and latent heat fluxes by the NCEP operational model during cold-air outbreaks (over the Labrador Sea) by Renfrew et al. (2002).

b. MESH simulations of the Great Lakes NBS

The monthly component NBS to the Great Lakes calculated in experiments MESH1–MESH4 (see section 5b) is shown alongside the observation-based monthly residual NBS (see section 4c) in Fig. 8. The cumulative NBS (the cumulative sum of the monthly NBS) plotted in the same figure is useful to identify the tendency of the model to underestimate or overestimate the NBS in the long term relative to the residual NBS. The Nash–Sutcliffe efficiency defined as
e8
with Oi residual NBS values with the mean and Pi predicted NBS values used as a measure of model skill over the period June 2004–December 2008. NSE values between 0 and 1 indicate acceptable model skill (increasing as NSE approaches 1), while 1 indicates a perfect model. The NSE values obtained in the MESH experiments are given in Table 2.
Fig. 8.
Fig. 8.

(left) Monthly residual NBS (ResidNBS), the difference between the simulated monthly component NBS (CompNBS) and ResidNBS, and (right) the cumulative ResidNBS and CompNBS to (a) Lake Superior, (b) Michigan–Huron–Georgian Bay, (c) Erie, and (d) Ontario in experiments MESH1 (orange), MESH2 (green), MESH3 (blue), and MESH4 (red). The residual NBS and its cumulative sum are shown in black.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

Table 2.

NSE obtained in the MESH experiments and for GLERL’s NBS predicted using their lake precipitation estimation method (GLERL 2011).

Table 2.

1) Effect of corrected precipitation provided by the CaPA analysis

A prominent characteristic of MESH1 (the control MESH experiment) is the NBS underestimation in many of the fall and winter months, which is common to all of the lakes (Fig. 8). Notable are, for instance, the large underestimations of low NBS values for Lake Superior in October 2006, Lake MHG in November 2007 and October 2008, and Lake Erie in October 2008. The cumulative effect of this underestimation is reflected in the drift of the cumulative NBS away from the residual NBS and toward low values for each of the lakes (Fig. 8). Precipitation was believed at the time to have the highest potential to improve the poor model performance in MESH1. Of concern was the ability of the GEM Regional model, which predicted the precipitation used in MESH1, to produce sufficiently accurate quantitative precipitation forecasts, especially during lake-effect storms when intense and localized precipitation occurs.

A corrected GEM Regional precipitation became available from an experimental CaPA analysis and this was tested in experiment MESH2. Surprisingly, the corrected precipitation increased the NBS underestimation in fall and winter in MESH2 relative to MESH1 (Fig. 8), particularly for Lakes Erie and Ontario. This was the result of reduced precipitation amounts provided by the corrected precipitation. Precipitation bias correction was largest for Lakes Erie and Ontario and smallest for Lake Superior. Coincidentally, the gauge density is also greatest over the basins of Lakes Erie and Ontario. The relatively large decrease in cumulative NBS to Lakes Erie and Ontario in MESH2 relative to MESH1 is mirrored by a decrease in NSE (and model skill) only for Lake Erie; the NSE for Lake Ontario increased from 0.40 to 0.46. Figure 8 shows a generally less (more) accurate Lake Ontario NBS in MESH2 in fall and winter (spring and summer) when lake evaporation is high (low).

2) Effect of corrected lake evaporation

The higher NBS underestimation produced by the corrected precipitation in the months with high lake evaporation prompted the revision of the parameterization of the surface fluxes over water. The revised parameterization was tested in MESH in experiment MESH3, in which the atmospheric forcing was the same as in MESH2. The successful verification of the latent heat fluxes simulated with the revised parameterization by the GEM Regional model in GEM-Reg3 (Figs. 6 and 7) and by MESH in MESH3 (Fig. 7) provided increased confidence in the simulated lake evaporation in these experiments, as the evaporative flux is the latent heat flux divided by the latent heat of vaporization.

The NBS simulation has improved considerably in MESH3 relative to MESH2 and MESH1 for Lakes MHG and Erie (Fig. 8 and Table 2). For instance, the reduced bias of the Lake MHG NBS in the fall and early winter months is clearly visible in Fig. 8b. The NSE increased for all of the lakes—for Lake MHG from 0.66 to 0.83. The cumulative effect of the decreased loss of lake water through evaporation in MESH3 is reflected in a greatly reduced drift of the cumulative NBS from the cumulative residual NBS relative to MESH2 for Lakes MHG, Erie, and Ontario (Figs. 8b–d). The same drift also decreased relative to that in MESH1 for Lakes MHG and Erie, whereas it was almost the same for Lake Ontario.

On the other hand, the simulation of reduced evaporation from Lake Superior in MESH3 led to an overestimation of the NBS in late fall and winter (Figs. 8a). Although the result cast some doubt on the accuracy of the simulated lake evaporation at the lake scale, further modification of the flux parameterization was considered unnecessary, given the good results obtained for the other lakes. Instead, we switched our focus back to the precipitation, recalling that it resulted from a relatively minor correction of the GEM Regional precipitation over Lake Superior compared to the other lakes. The winter pattern of the precipitation produced by the GEM Regional model (and used in MESH1) over the Lake Superior basin showed the precipitation largely concentrated over the lake (not shown). A similar pattern was obtained by Li et al. (2010) for Lake Superior for the winter season from the North American Regional Reanalysis. This suggests a strong dependence of the overlake precipitation on the atmospheric convection occurring over Lake Superior in winter. A large contribution to this precipitation is expected to be produced during strong boundary-layer convection developing over the lake after the passage of cold fronts (lake-effect precipitation). Thus, the overlake precipitation ultimately depends on the lake evaporation. Therefore, the inconsistency created in MESH3 between the (reduced) simulated lake evaporation and the prescribed overlake precipitation was likely to impact the NBS.

3) Effect of increased consistency between corrected lake evaporation and precipitation

Consistency between precipitation and surface evaporation (evapotranspiration over vegetated land) can only be achieved in simulations with a model that allows for interaction between the surface and the atmosphere, such as the GEM Regional model. This fact naturally led to the need for generating a new consistent atmospheric forcing dataset for MESH from reforecasts produced by the GEM Regional model with the revised surface flux parameterization (see section 5b), which was then tested in experiment MESH4. The seasonal mean precipitation from the new forcing dataset is shown in Fig. 9 and the previously mentioned concentration of the precipitation over Lake Superior in winter can be readily seen in Fig. 9d. Although the forcing fields are not perfectly consistent with the fluxes simulated in MESH4, because of the different time step and resolution of the MESH model, they do reflect to a large extent the response of the atmosphere to the fluxes simulated with the modified flux parameterization over water.

Fig. 9.
Fig. 9.

Seasonal mean precipitation (mm day−1) predicted in the GEM-Reg3 runs and used as forcing in experiments MESH4, MESH5, and MESH6: (a) spring (March–May), (b) summer (June–August), (c) fall (September–November), and (d) winter (December–February).

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

As expected, the Lake Superior NBS decreased in MESH4 relative to MESH3 in late fall and winter and this contributed to a large reduction in cumulative NBS drift (Fig. 8a). Interestingly, the drift of the cumulative NBS to Lakes MHG, Erie, and Ontario was also corrected in MESH4 (Figs. 8b–d). This is a very important result because the simulation of an unbiased cumulative NBS is crucial when the NBS is to be used for long-term prediction of the Great Lakes levels. On the negative side, an increased overestimation of the NBS to Lake Ontario resulted in April (Fig. 8d). The NSE for Lakes Superior and Erie increased in MESH4 compared to MESH3 from 0.90 to 0.80 and 0.92 to 0.82, respectively, whereas the NSE for Lakes MHG and Ontario decreased from 0.83 to 0.60 and 0.77 to 0.38, respectively.

The monthly values of Elake, PlakeElake, and NBS simulated in MESH4 for Lake Superior are plotted in Fig. 10a. The NBS reaches its lowest values in late fall and winter, and can become negative when the overlake precipitation and runoff cannot compensate for the loss of lake water through evaporation. The runoff is small compared to both Plake and Elake in the same period, becoming as low as 3 mm month−1 in January and February 2007, which represents less than 5% of Plake and Elake for these months (Fig. 10a). The difference PlakeElake is also relatively small in late fall and winter, indicating a delicate balance between Elake and Plake. These results explain the sensitivity of the simulated Lake Superior NBS to the changes in overlake precipitation and lake evaporation in the MESH experiments and the importance of having Plake consistent with Elake.

Fig. 10.
Fig. 10.

The monthly values of Elake (plotted as negative in blue), PlakeElake (green), and component NBS (red) for the Great Lakes in experiment MESH4. the quantity Plake and the simulated runoff can be seen as the vertical distance between the points representing Elake and PlakeElake, and PlakeElake and the component NBS, respectively. The residual NBS is shown in black.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

Figure 11 shows the monthly values of Plake, Elake, and PlakeElake in experiments MESH1–MESH4. The large reduction in Elake obtained in MESH4 and MESH3 in late fall and winter relative to MESH2 and MESH1 can be readily seen in Figs. 11a–d. Also visible are the lower Plake values in MESH4 compared to MESH3 (and MESH2) for Lake Superior (Fig. 11a) as a result of Plake becoming consistent with the reduced Elake. The above-mentioned balance between Plake and Elake for the same lake in late fall and winter is illustrated by the relatively small PlakeElake values in Fig. 11e.

Fig. 11.
Fig. 11.

(left) Monthly Plake and Elake and (right) their difference PlakeElake for the Great Lakes simulated in experiments MESH1–MESH4. In each of the panels on the left, the Plake values are plotted above the x axis, whereas the Elake values with changed sign are shown below the same axis.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

4) Effect of corrected runoff into the lakes

A systematic overestimation of the NBS to Lakes MHG, Erie, and Ontario plagues experiments MESH1–MESH4 in April—the month with some of the largest residual NBS values (Fig. 8). The maximum NBS overestimation of about 80% obtained in MESH4 in April 2005 for Lake Ontario (Fig. 10d) provided a good starting point for the investigation of the systematic error. Accounting for 84% of the simulated NBS in this case, the runoff (Fig. 10d) was the first NBS component deserving a closer look. Tracing back to the surface runoff and drainage produced by ISBA, their peaks in April (not shown) explained the peak in runoff in the same month. Consequently, the land surface component (ISBA) rather than the routing was found to be the main source of error in the monthly runoff. The subsequent identification of the snowmelt as the main problem in ISBA (in MESH) was suggested by the fact that April is the month with intense snowmelt. The need for further evidence prompted the modifications of the snow parameterization in ISBA aimed at enhancing the snowmelt rate during warm air advection episodes, tested in experiment MESH5.

The effect of the modified snowmelt was the simulation of an increased (decreased) meltwater volume in the winter months and March (April and May) in MESH5 compared to MESH4. Consequently, the underestimation (overestimation) in winter and March (April and May) of the NBS to Lakes MHG, Erie, and Ontario in MESH4 was partly corrected in MESH5 (Figs. 12b–d). Remarkable is the correction of the April peaks in NBS. The NSE values for the same lakes increased in MESH5 relative to MESH4 (Table 2). The largest improvement in model skill was obtained for Lake Ontario (NSE increased from 0.38 to 0.75). The reduced cumulative NBS in MESH5 relative to MESH4 (Figs. 12b–d) shows another effect of the modified snowmelt—namely, a loss of water for the three lakes. Plots of the monthly evapotranspiration over the drainage basins of the lakes (not shown) explain this loss through a net increase in evapotranspiration. In MESH5, the increased evapotranspiration in winter and March due to increased availability of water during snowmelt events is not compensated by the decreased evapotranspiration occurring in the following months.

Fig. 12.
Fig. 12.

Similar to Fig. 8 but with the component NBS simulated in experiments MESH4 (orange), MESH5 (green), and MESH6 (red), and the component NBS predicted by GLERL using their overlake precipitation estimation method (blue).

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

Further NBS simulation improvement was obtained in MESH6 with the runoff corrected with observed streamflows (Fig. 12). The NBS components in MESH6 are shown in Fig. 13. The important correction of the runoff underestimation (overestimation) in winter and March (April and May) for Lakes MHG, Erie, and Ontario is reflected in the improved accuracy of the simulated NBS for these months. The result increases the confidence in the accuracy of the PlakeElake difference for the same lakes. Of all of the MESH experiments, the highest NSE values for each of the lakes were obtained in MESH6 (Table 2).

Fig. 13.
Fig. 13.

Similar to Fig. 10 but for experiment MESH6.

Citation: Journal of Hydrometeorology 13, 6; 10.1175/JHM-D-11-0151.1

The use of observed streamflows raises some concerns, however. An example is the perturbation of the lake water budget caused by the inconsistency created between the corrected runoff and the precipitation forcing and simulated evaporation. There is also the risk of double counting of the runoff when the drainage basins are only partly gauged (as in MESH6). This effect occurs when the runoff is overestimated over the ungauged areas as the result of temporal and/or spatial shifts of the simulated precipitation or snowmelt patterns compared to the actual ones. The effect may partly explain the overestimation of the cumulative NBS for Lake Ontario in MESH6 despite the increased accuracy of the monthly NBS. The incomplete correction of the April peaks in NBS to this lake could be partly caused by an unrealistically high runoff contribution of the ungauged portion of the lake’s drainage basin due to the simulation of a delayed snowmelt and potentially shifted precipitation fields. Of particular concern is the area to the east of Lake Ontario that experiences intense precipitation, which is orographically enhanced over the Tug Hill Plateau (Fig. 9)—the location of the plateau is shown in Fig. 1. This area straddles the border between the gauged and ungauged portions of the drainage basin (Fig. 5). Water management is another factor contributing to the mismatch between the observed and modeled runoff.

The overall superior skill of MESH in simulating the monthly NBS in experiment MESH6 relative to the GLERL model can be seen in Fig. 12 and Table 2. Most notable is the relatively little drift of the cumulative NBS in MESH6 compared to that of GLERL.

7. General discussion

An important aspect emerging from this study is the need for consistency among the NBS components. The delicate balance between Elake and Plake for Lake Superior in late fall and winter addressed in section 6b(3) shows the need for coupled land–lake–atmosphere model simulations. Accurate simulation of lake evaporation in such models is crucial because this evaporation can be a dominant component of the NBS (Fig. 10) and influence directly Plake (e.g., lake-effect precipitation) and indirectly the runoff (through precipitation).

A change in Plake mirroring (without matching) the change in Elake similar to that obtained for Lake Superior was not obtained for the downstream lakes (especially for Lakes Erie and Ontario) in experiment MESH4 (Fig. 11). One reason for the apparently more complex dependence of Plake on Elake for any of these lakes could be the influence on Plake of the atmospheric convection over the upwind lakes (e.g., Sousounis and Mann 2000). For instance, for the typical northwesterly winds associated with cold-air outbreaks conducive to strong convection over the Great Lakes, wind-parallel lake-effect cloud and snowbands originating over Lake Superior can extend over the downwind lakes while experiencing continuous growth (Rodriguez et al. 2007). The resulting lake-to-lake bands can produce enhanced lake-effect precipitation over the downwind lakes, and hence link the precipitation over one lake to the evaporation over that lake and the upwind lakes. Such effects may have been reproduced by the GEM Regional model for some of the simulated lake-effect events.

The effect of lack of atmospheric feedback in the MESH model can be subtle. Let us look at the surface heat fluxes obtained with the same modified surface flux parameterization in GEM-Reg3, MESH4, and MESH3 shown in Fig. 7. Because MESH4 was forced with fields derived from the GEM Regional output in GEM-Reg3, it is no surprise that the fluxes simulated in the two experiments are close to each other. The differences are partly explained by the fact that the grid points where the fluxes were extracted were not collocated and by the different time steps of the models. While the surface fluxes simulated in MESH4 are nearly consistent with the prescribed atmospheric fields, being obtained in a quasi-coupled simulation (as if MESH were coupled to the GEM Regional model), the same cannot be said about MESH3. The lower LHF and SHF values obtained in MESH3 compared to MESH4 and shown in Fig. 7 resulted from higher near-surface air humidity and temperature values prescribed in MESH3. These higher values originated from GEM Regional model predictions with the original parameterization that overestimated the surface heat fluxes over the lakes and led to the simulation of a warmer and moister atmospheric boundary layer. The results reveal more than the unsurprising effect of changing the prescribed atmospheric fields on the fluxes calculated by the same flux parameterization in a surface model run in offline mode. They show that a realistic effect of a change in the flux parameterization implies not only accurate fluxes, but also an appropriate more general atmospheric response that feeds back on the surface through air temperature and humidity, wind, precipitation, and radiation.

The modified snowmelt rate ultimately changed the land surface fluxes simulated by MESH. Therefore, the impact of this modification on the surface water and energy budget simulated by MESH in stand-alone (offline) mode would be different from that simulated by MESH coupled to an atmospheric model. This is because the coupled model would allow for land–atmosphere feedbacks. Thus, MESH5 and the experiments with modified surface fluxes over water show that improving a surface model when the model testing can only be done in offline mode can be a difficult task. There will always be a need to compensate in some way for the missing interaction with the atmosphere to obtain a better match to observations. Worrisome is the fact that this need could dictate the modifications required by the model and steer the model improvement effort in the wrong direction.

The superior NBS prediction skill of MESH (in MESH6) relative to GLERL’s estimation method points to an advantage of MESH forced with atmospheric fields that are consistent both internally and with the simulated surface fluxes. Better yet would be to simulate the NBS with a coupled land–lake–atmospheric model with sufficiently high horizontal resolution. Thus, the NBS would act as a spatial and temporal integrator of important atmospheric, lake, and land processes in the presence of complex feedbacks allowed by the model.

8. Summary and conclusions

This paper presents the incremental improvement of the skill of the hydrometeorological model MESH in predicting the monthly NBS to the Great Lakes, which was verified against an observation-based residual NBS. The NBS was underestimated in late fall and winter by MESH forced with atmospheric fields extracted from forecasts produced by an operational NWP model (GEM Regional). The underestimation increased when the GEM Regional precipitation was replaced with its corrected version provided by the Canadian Precipitation Analysis. This result prompted the investigation of the simulated lake evaporation and the revision of the parameterization of the surface turbulent fluxes over water used both in the MESH and the GEM regional models. The revised surface flux parameterization was validated against turbulent fluxes measured at a point on Lake Superior. Its use in MESH largely corrected the NBS underestimation through the simulation of reduced lake evaporation. However, the Lake Superior NBS was now overestimated, signaling an inconsistency between the (unchanged) prescribed precipitation and the reduced lake evaporation. Since the simulation of an atmospheric response reflecting the change in modified surface fluxes over the lakes was possible in the GEM Regional model, this model was run with the modified parameterization to generate a new atmospheric forcing dataset. Major NBS improvements, including a reduced overestimation of Lake Superior NBS, resulted when MESH was forced with the new atmospheric forcing. Additional improvements resulted by correcting the runoff into the lakes with a modified snowmelt rate and by insertion of observed streamflows.

Accurate lake evaporation estimation was found to be crucial for accurate NBS simulation because this evaporation was a dominant NBS component in many of the fall and winter months and impacted the other two NBS components: the overlake precipitation and runoff (through precipitation). The need for capturing the atmospheric response to the modified lake evaporation, requiring surface–atmosphere coupling, has also emerged. In general, any modification of a surface model ultimately impacts the simulated surface turbulent fluxes, and such fluxes result from the surface–atmosphere interaction in reality. Therefore, the need to compensate for errors arising from the missing surface–atmosphere interaction in offline simulations could hinder the surface model improvement effort. Finally, the success of using the NBS in identifying major shortcomings of the MESH model shows the utility of the NBS in improving not only this model but also its parent numerical weather prediction model.

Acknowledgments

This study was made possible thanks to continued support by the International Upper Great Lakes Study of the International Joint Commission, including financial support and feedback from members of the technical working groups. We wish to thank in particular Alain Pietroniro, David Fay, Rob Caldwell, Frank Seglenieks, and Tim Hunter. Our confidence in the performance of the modified turbulent flux parameterization presented in this paper increased sharply following numerous tests performed by the first author over the Gulf of St. Lawrence with a coupled atmosphere–ocean forecasting system. We wish to thank Manon Faucher and François Roy for their interest in these tests and help with setting up and running this complex system. We also thank Gregory Smith, Pierre Pellerin, and three anonymous reviewers for recommendations that greatly improved the clarity of the paper.

REFERENCES

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

    • Search Google Scholar
    • Export Citation
  • Bélair, S., Brown R. , Mailhot J. , 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.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., 2009: Land-surface-atmosphere coupling in observations and models. J. Adv. Model. Earth Syst., 1, 1–18.

  • Blanken, P. D., Spence C. , Hedstrom N. , and Lenters J. D. , 2011: Evaporation from Lake Superior: 1. Physical controls and processes. J. Great Lakes Res., 37, 707716.

    • Search Google Scholar
    • Export Citation
  • Bruxer, J., 2010: Uncertainty analysis of Lake Erie net basin supplies as computed using the residual method. M.S. thesis, Dept. of Civil Engineering, McMaster University, 238 pp. [Available online at http://digitalcommons.mcmaster.ca/opendissertations/4987.]

  • Carrera, M. L., 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.

    • Search Google Scholar
    • Export Citation
  • Charnock, H., 1955: Wind stress on a water surface. Quart. J. Roy. Meteor. Soc., 81, 639640.

  • Decharme, B., and Douville H. , 2006: Introduction of a sub-grid hydrology in the ISBA land surface model. Climate Dyn., 26, 6578.

  • DeCosmo, J., Katsaros K. B. , Smith S. D. , Anderson R. J. , Oost W. A. , Bumke K. , and Chadwick H. , 1996: Air–sea exchange of water vapor and sensible heat: The humidity exchange over the sea (HEXOS) results. J. Geophys. Res., 101 (C5), 12 00112 016.

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

    • Search Google Scholar
    • Export Citation
  • Dutra, E., Balsamo G. , Viterbo P. , Miranda P. M. A. , Beljaars A. , Schär C. , and Elder K. , 2010: An improved snow scheme for the ECMWF land surface model: Description and offline validation. J. Hydrometeor., 11, 899916.

    • Search Google Scholar
    • Export Citation
  • ECMWF, 2011: Part IV: Physical processes. IFS Doc. Cy33r1, 162 pp. [Available online at http://www.ecmwf.int/research/ifsdocs/CY33r1/PHYSICS/IFSPart4.pdf.]

  • Fairall, C. W., Bradley E. F. , Hare J. E. , Grachev A. A. , and Edson J. B. , 2003: Bulk parameterization of air–sea fluxes: Updates and verification for the COARE algorithm. J. Climate, 16, 571591.

    • Search Google Scholar
    • Export Citation
  • GLERL, cited 2011: GLERL Great Lakes monthly hydrologic data (1860-2008). [Available online at http://www.glerl.noaa.gov/data/arc/hydro/mnth-hydro.html.]

  • Hock, R., 2003: Temperature index melt modelling in mountain areas. J. Hydrol., 282, 104115.

  • IUGLS, 2012: Lake Superior regulation: Addressing uncertainty in upper Great Lakes water levels. Final Rep. to the International Joint Commission, 214 pp. [Available online at http://www.ijc.org/iuglsreport/wp-content/report-pdfs/Lake_Superior_Regulation_Full_Report.pdf.]

  • Kouwen, N., 2010: WATFLOOD/WATROUTE hydrological model routing and flow forecasting system. Department of Civil Engineering, University of Waterloo, 247 pp.

  • Kristovich, D. A. R., and Steve R. A. , 1995: A satellite study of cloud-band frequencies over the Great Lakes. J. Appl. Meteor., 34, 20832090.

    • Search Google Scholar
    • Export Citation
  • Laird, N. F., and Kristovich D. A. R. , 2002: Variations of sensible and latent heat fluxes from a Great Lakes buoy and associated synoptic weather patterns. J. Hydrometeor., 3, 312.

    • Search Google Scholar
    • Export Citation
  • LeMone, M. A., 1973: The structure and dynamics of horizontal roll vortices in the planetary boundary layer. J. Atmos. Sci., 30, 10771091.

    • Search Google Scholar
    • Export Citation
  • Li, X., Zhong S. , Bian X. , Heilman W. E. , Luo Y. , and Dong W. , 2010: Hydroclimate and variability in the Great Lakes region as derived from the North American Regional Reanalysis. J. Geophys. Res.,115, D12104, doi:10.1029/2009JD012756.

  • Liang, X., Lettenmaier D. P. , Wood E. F. , and Burges S. J. , 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res.,99 (D7), 14 415–14 428.

  • Liu, A. Q., and Moore G. W. K. , 2004: Lake-effect snowstorms over Southern Ontario, Canada, and their associated synoptic-scale environment. Mon. Wea. Rev., 132, 25952609.

    • Search Google Scholar
    • Export Citation
  • Liu, Y., Bastidas L. A. , Gupta H. V. , and Sorooshian S. , 2003: Impacts of a parameterization deficiency on offline and coupled land surface model simulations. J. Hydrometeor., 4, 901914.

    • Search Google Scholar
    • Export Citation
  • Lyon, S. W., and Coauthors, 2008: Coupling terrestrial and atmospheric water dynamics to improve prediction in a changing environment. Bull. Amer. Meteor. Soc., 89, 12751279.

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

    • Search Google Scholar
    • Export Citation
  • Mahrt, L., 1998: Flux sampling errors for aircraft and towers. J. Atmos. Oceanic Technol., 15, 416429.

  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian Regional Forecast System. Atmos.–Ocean, 44, 133149.

  • Neff, B. P., and Nicholas J. R. , 2005: Uncertainty in the Great Lakes water balance. U.S. Geological Survey Scientific Investigations Rep. 2004-5100, 42 pp.

  • Ohmura, A., 2001: Physical basis for the temperature-based melt-index method. J. Appl. Meteor., 40, 753761.

  • Pietroniro, A., and Coauthors, 2007: Development of the MESH modelling system for hydrological ensemble forecasting of the Laurentian Great Lakes at the regional scale. Hydrol. Earth Syst. Sci., 11, 12791294.

    • Search Google Scholar
    • Export Citation
  • Powell, M. D., Vickery P. J. , and Reinhold T. A. , 2003: Reduced drag coefficient for high wind speeds in tropical cyclones. Nature, 422, 279283.

    • Search Google Scholar
    • Export Citation
  • Quinn, F. H., 2009: Net basin supply comparison analysis. Hydroclimate Technical Work Group Task 2.2, St. Clair River Task Team, International Upper Great Lakes Study, 76 pp. [Available online at http://pub.iugls.org/en/St_Clair_Reports/Hydroclimatic/Hydroclimate-03.pdf.]

  • Quinn, F. H., and Kelley R. N. , 1983: Great Lakes monthly hydrologic data. NOAA Data Rep. ERL GLERL-26, 79 pp.

  • Renfrew, I. A., Moore G. W. K. , Guest P. S. , and Bumke K. , 2002: A comparison of surface layer and surface turbulent flux observations over the Labrador Sea with ECMWF analyses and NCEP reanalyses. J. Phys. Oceanogr., 32, 383400.

    • Search Google Scholar
    • Export Citation
  • Rodriguez, Y., Kristovich D. A. R. , and Hjelmfelt M. R. , 2007: Lake-to-lake cloud bands: Frequencies and locations. Mon. Wea. Rev., 135, 42024213.

    • Search Google Scholar
    • Export Citation
  • Santanello, J. A., Peters-Lidard C. D. , Kumar S. V. , Alonge C. , and Tao W.-K. , 2009: A modeling and observational framework for diagnosing local land–atmosphere coupling on diurnal time scales. J. Hydrometeor., 10, 577599.

    • Search Google Scholar
    • Export Citation
  • Schertzer, W. M., 1978: Energy budget and monthly evaporation estimates for Lake Superior, 1973. J. Great Lakes Res., 4, 320330.

  • Soulis, E. D., Craig J. R. , Fortin V. , and Liu G. , 2011: A simple expression for the bulk field capacity of a sloping soil horizon. Hydrol. Processes, 25, 112116.

    • Search Google Scholar
    • Export Citation
  • Sousounis, P. J., and Mann G. E. , 2000: Lake-aggregate mesoscale disturbances. Part V: Impacts on lake-effect precipitation. Mon. Wea. Rev., 128, 728745.

    • Search Google Scholar
    • Export Citation
  • Spence, C., Blanken P. D. , Hedstrom N. , Fortin V. , and Wilson H. , 2011: Evaporation from Lake Superior: 2. Spatial distribution and variability. J. Great Lakes Res., 37, 717724.

    • Search Google Scholar
    • Export Citation
  • Steinfeld, G., Letzel M. O. , Raasch S. , Kanda M. , and Inagaki A. , 2007: Spatial representativeness of single tower measurements and the imbalance problem with eddy-covariance fluxes: Results of a large-eddy simulation study. Bound.-Layer Meteor., 123, 7798.

    • Search Google Scholar
    • Export Citation
  • Van den Hurk, B. J. J. M., Graham L. P. , and Viterbo P. , 2002: Comparison of land surface hydrology in regional climate simulations of the Baltic Sea catchment. J. Hydrol., 255, 169193.

    • Search Google Scholar
    • Export Citation
  • Verseghy, D. L., 2000: The Canadian Land Surface Scheme (CLASS): Its history and future. Atmos.–Ocean, 38, 113.

  • Wood, E., Lettenmaier D. , and Zartarian V. , 1992: A land-surface hydrology parameterization with sub-grid variability for general circulation models. J. Geophys. Res., 97 (D3), 27172728.

    • Search Google Scholar
    • Export Citation
  • Young, G. S., Kristovich D. A. R. , Hjelmfelt M. R. , and Foster R. C. , 2002: Rolls, streets, waves, and more: A review of quasi-two-dimensional structures in the atmospheric boundary layer. Bull. Amer. Meteor. Soc., 83, 9971001.

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

    • Search Google Scholar
    • Export Citation
  • Bélair, S., Brown R. , Mailhot J. , 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.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., 2009: Land-surface-atmosphere coupling in observations and models. J. Adv. Model. Earth Syst., 1, 1–18.

  • Blanken, P. D., Spence C. , Hedstrom N. , and Lenters J. D. , 2011: Evaporation from Lake Superior: 1. Physical controls and processes. J. Great Lakes Res., 37, 707716.

    • Search Google Scholar
    • Export Citation
  • Bruxer, J., 2010: Uncertainty analysis of Lake Erie net basin supplies as computed using the residual method. M.S. thesis, Dept. of Civil Engineering, McMaster University, 238 pp. [Available online at http://digitalcommons.mcmaster.ca/opendissertations/4987.]

  • Carrera, M. L., 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.

    • Search Google Scholar
    • Export Citation
  • Charnock, H., 1955: Wind stress on a water surface. Quart. J. Roy. Meteor. Soc., 81, 639640.

  • Decharme, B., and Douville H. , 2006: Introduction of a sub-grid hydrology in the ISBA land surface model. Climate Dyn., 26, 6578.

  • DeCosmo, J., Katsaros K. B. , Smith S. D. , Anderson R. J. , Oost W. A. , Bumke K. , and Chadwick H. , 1996: Air–sea exchange of water vapor and sensible heat: The humidity exchange over the sea (HEXOS) results. J. Geophys. Res., 101 (C5), 12 00112 016.

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

    • Search Google Scholar
    • Export Citation
  • Dutra, E., Balsamo G. , Viterbo P. , Miranda P. M. A. , Beljaars A. , Schär C. , and Elder K. , 2010: An improved snow scheme for the ECMWF land surface model: Description and offline validation. J. Hydrometeor., 11, 899916.

    • Search Google Scholar
    • Export Citation
  • ECMWF, 2011: Part IV: Physical processes. IFS Doc. Cy33r1, 162 pp. [Available online at http://www.ecmwf.int/research/ifsdocs/CY33r1/PHYSICS/IFSPart4.pdf.]

  • Fairall, C. W., Bradley E. F. , Hare J. E. , Grachev A. A. , and Edson J. B. , 2003: Bulk parameterization of air–sea fluxes: Updates and verification for the COARE algorithm. J. Climate, 16, 571591.

    • Search Google Scholar
    • Export Citation
  • GLERL, cited 2011: GLERL Great Lakes monthly hydrologic data (1860-2008). [Available online at http://www.glerl.noaa.gov/data/arc/hydro/mnth-hydro.html.]

  • Hock, R., 2003: Temperature index melt modelling in mountain areas. J. Hydrol., 282, 104115.

  • IUGLS, 2012: Lake Superior regulation: Addressing uncertainty in upper Great Lakes water levels. Final Rep. to the International Joint Commission, 214 pp. [Available online at http://www.ijc.org/iuglsreport/wp-content/report-pdfs/Lake_Superior_Regulation_Full_Report.pdf.]

  • Kouwen, N., 2010: WATFLOOD/WATROUTE hydrological model routing and flow forecasting system. Department of Civil Engineering, University of Waterloo, 247 pp.

  • Kristovich, D. A. R., and Steve R. A. , 1995: A satellite study of cloud-band frequencies over the Great Lakes. J. Appl. Meteor., 34, 20832090.

    • Search Google Scholar
    • Export Citation
  • Laird, N. F., and Kristovich D. A. R. , 2002: Variations of sensible and latent heat fluxes from a Great Lakes buoy and associated synoptic weather patterns. J. Hydrometeor., 3, 312.

    • Search Google Scholar
    • Export Citation
  • LeMone, M. A., 1973: The structure and dynamics of horizontal roll vortices in the planetary boundary layer. J. Atmos. Sci., 30, 10771091.

    • Search Google Scholar
    • Export Citation
  • Li, X., Zhong S. , Bian X. , Heilman W. E. , Luo Y. , and Dong W. , 2010: Hydroclimate and variability in the Great Lakes region as derived from the North American Regional Reanalysis. J. Geophys. Res.,115, D12104, doi:10.1029/2009JD012756.

  • Liang, X., Lettenmaier D. P. , Wood E. F. , and Burges S. J. , 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res.,99 (D7), 14 415–14 428.

  • Liu, A. Q., and Moore G. W. K. , 2004: Lake-effect snowstorms over Southern Ontario, Canada, and their associated synoptic-scale environment. Mon. Wea. Rev., 132, 25952609.

    • Search Google Scholar
    • Export Citation
  • Liu, Y., Bastidas L. A. , Gupta H. V. , and Sorooshian S. , 2003: Impacts of a parameterization deficiency on offline and coupled land surface model simulations. J. Hydrometeor., 4, 901914.

    • Search Google Scholar
    • Export Citation
  • Lyon, S. W., and Coauthors, 2008: Coupling terrestrial and atmospheric water dynamics to improve prediction in a changing environment. Bull. Amer. Meteor. Soc., 89, 12751279.

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

    • Search Google Scholar
    • Export Citation
  • Mahrt, L., 1998: Flux sampling errors for aircraft and towers. J. Atmos. Oceanic Technol., 15, 416429.

  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian Regional Forecast System. Atmos.–Ocean, 44, 133149.

  • Neff, B. P., and Nicholas J. R. , 2005: Uncertainty in the Great Lakes water balance. U.S. Geological Survey Scientific Investigations Rep. 2004-5100, 42 pp.

  • Ohmura, A., 2001: Physical basis for the temperature-based melt-index method. J. Appl. Meteor., 40, 753761.

  • Pietroniro, A., and Coauthors, 2007: Development of the MESH modelling system for hydrological ensemble forecasting of the Laurentian Great Lakes at the regional scale. Hydrol. Earth Syst. Sci., 11, 12791294.

    • Search Google Scholar
    • Export Citation
  • Powell, M. D., Vickery P. J. , and Reinhold T. A. , 2003: Reduced drag coefficient for high wind speeds in tropical cyclones. Nature, 422, 279283.

    • Search Google Scholar
    • Export Citation
  • Quinn, F. H., 2009: Net basin supply comparison analysis. Hydroclimate Technical Work Group Task 2.2, St. Clair River Task Team, International Upper Great Lakes Study, 76 pp. [Available online at http://pub.iugls.org/en/St_Clair_Reports/Hydroclimatic/Hydroclimate-03.pdf.]

  • Quinn, F. H., and Kelley R. N. , 1983: Great Lakes monthly hydrologic data. NOAA Data Rep. ERL GLERL-26, 79 pp.

  • Renfrew, I. A., Moore G. W. K. , Guest P. S. , and Bumke K. , 2002: A comparison of surface layer and surface turbulent flux observations over the Labrador Sea with ECMWF analyses and NCEP reanalyses. J. Phys. Oceanogr., 32, 383400.

    • Search Google Scholar
    • Export Citation
  • Rodriguez, Y., Kristovich D. A. R. , and Hjelmfelt M. R. , 2007: Lake-to-lake cloud bands: Frequencies and locations. Mon. Wea. Rev., 135, 42024213.

    • Search Google Scholar
    • Export Citation
  • Santanello, J. A., Peters-Lidard C. D. , Kumar S. V. , Alonge C. , and Tao W.-K. , 2009: A modeling and observational framework for diagnosing local land–atmosphere coupling on diurnal time scales. J. Hydrometeor., 10, 577599.

    • Search Google Scholar
    • Export Citation
  • Schertzer, W. M., 1978: Energy budget and monthly evaporation estimates for Lake Superior, 1973. J. Great Lakes Res., 4, 320330.

  • Soulis, E. D., Craig J. R. , Fortin V. , and Liu G. , 2011: A simple expression for the bulk field capacity of a sloping soil horizon. Hydrol. Processes, 25, 112116.

    • Search Google Scholar
    • Export Citation
  • Sousounis, P. J., and Mann G. E. , 2000: Lake-aggregate mesoscale disturbances. Part V: Impacts on lake-effect precipitation. Mon. Wea. Rev., 128, 728745.

    • Search Google Scholar
    • Export Citation
  • Spence, C., Blanken P. D. , Hedstrom N. , Fortin V. , and Wilson H. , 2011: Evaporation from Lake Superior: 2. Spatial distribution and variability. J. Great Lakes Res., 37, 717724.

    • Search Google Scholar
    • Export Citation
  • Steinfeld, G., Letzel M. O. , Raasch S. , Kanda M. , and Inagaki A. , 2007: Spatial representativeness of single tower measurements and the imbalance problem with eddy-covariance fluxes: Results of a large-eddy simulation study. Bound.-Layer Meteor., 123, 7798.

    • Search Google Scholar
    • Export Citation
  • Van den Hurk, B. J. J. M., Graham L. P. , and Viterbo P. , 2002: Comparison of land surface hydrology in regional climate simulations of the Baltic Sea catchment. J. Hydrol., 255, 169193.

    • Search Google Scholar
    • Export Citation
  • Verseghy, D. L., 2000: The Canadian Land Surface Scheme (CLASS): Its history and future. Atmos.–Ocean, 38, 113.

  • Wood, E., Lettenmaier D. , and Zartarian V. , 1992: A land-surface hydrology parameterization with sub-grid variability for general circulation models. J. Geophys. Res., 97 (D3), 27172728.

    • Search Google Scholar
    • Export Citation
  • Young, G. S., Kristovich D. A. R. , Hjelmfelt M. R. , and Foster R. C. , 2002: Rolls, streets, waves, and more: A review of quasi-two-dimensional structures in the atmospheric boundary layer. Bull. Amer. Meteor. Soc., 83, 9971001.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    The Great Lakes and their drainage basin boundaries (marked in red). The drainage basins of Lakes Michigan, Huron, and Georgian Bay are merged in the figure. The Great Lakes are connected to each other by a system of channels. The topography (m) of the GEM Regional model for this area is shown in shades of gray and blue. The closed black circle in Lake Superior marks the location of the Stannard Rock Light.

  • Fig. 2.

    Surface analysis valid at 1800 UTC 15 Dec 2008 (adapted from the chart produced by the NOAA/NWS Hydrometeorological Prediction Center and valid at the same time). The sea level pressure (hPa) is contoured in dark red every 4 hPa.

  • Fig. 3.

    Aqua MODIS 2-km resolution images (Bands 7–2–1) at (a) 1900 UTC 15 Dec and b) 1900 UTC 31 Dec 2008 showing lake-effect cloud streets (convective roll clouds) that developed over Lakes Superior, Michigan, and Huron during cold-air outbreaks. The red plus sign marks the location of the Stannard Rock Light. Image credit: NASA Goddard Space Flight Center (GSFC), MODIS Rapid Response (http://rapidfire.sci.gsfc.nasa.gov).

  • Fig. 4.

    The components of the GEM Regional and MESH models. The arrows indicate the direction of the information flow between model components.

  • Fig. 5.

    The domain of the MESH model covers the Great Lakes basin and a surrounding area (light gray). The horizontal resolution of the model is ⅙°. Shown in light blue are the lake areas used by WATROUTE. The drainage basin of Lake St. Clair (not considered in this study) is shown in dark gray. The other colors depict the areas of the drainage basin that were gauged a percentage of the time between June 2004 and December 2008 limited by the thresholds shown in the legend. The watersheds of the Trent and Spanish Rivers are marked TRW and SRW, respectively.

  • Fig. 6.

    Observed and simulated (a) LHF and (b) SHF for the month of December 2008. The observed fluxes (black dots) are turbulent fluxes derived from eddy-covariance measurements using a 30-min averaging period. The measurements were taken at the Stannard Rock Light at 32.5-m height. The observed data gaps >2 h are 0430–0700 UTC 15 Dec, 1530–1930 UTC 15 Dec, 0800 UTC 23 Dec–1730 UTC 25 Dec, 2300 UTC 27 Dec–0500 UTC 28 Dec, and 1200 UTC 28 Dec–1730 UTC 31 Dec. Plotted in color are the surface LHF and SHF simulated in the experiments with the GEM Regional model (described in the text) at each hour, at the nearest grid point to the Stannard Rock Light.

  • Fig. 7.

    As in Fig. 6, but for the period 15–16 Dec 2008. The observed data gaps are 0430–0700 and 1530–1930 UTC 15 Dec, the latter for SHF only. Also plotted are the surface LHF and SHF simulated in experiments GEM-Reg3 (closed red circles), MESH3 (open blue circles), and MESH4 (open red circles) at each hour, at the nearest grid point of each model to the Stannard Rock Light. The modified surface flux parameterization over water was used in the three experiments.

  • Fig. 8.

    (left) Monthly residual NBS (ResidNBS), the difference between the simulated monthly component NBS (CompNBS) and ResidNBS, and (right) the cumulative ResidNBS and CompNBS to (a) Lake Superior, (b) Michigan–Huron–Georgian Bay, (c) Erie, and (d) Ontario in experiments MESH1 (orange), MESH2 (green), MESH3 (blue), and MESH4 (red). The residual NBS and its cumulative sum are shown in black.

  • Fig. 9.

    Seasonal mean precipitation (mm day−1) predicted in the GEM-Reg3 runs and used as forcing in experiments MESH4, MESH5, and MESH6: (a) spring (March–May), (b) summer (June–August), (c) fall (September–November), and (d) winter (December–February).

  • Fig. 10.

    The monthly values of Elake (plotted as negative in blue), PlakeElake (green), and component NBS (red) for the Great Lakes in experiment MESH4. the quantity Plake and the simulated runoff can be seen as the vertical distance between the points representing Elake and PlakeElake, and PlakeElake and the component NBS, respectively. The residual NBS is shown in black.

  • Fig. 11.

    (left) Monthly Plake and Elake and (right) their difference PlakeElake for the Great Lakes simulated in experiments MESH1–MESH4. In each of the panels on the left, the Plake values are plotted above the x axis, whereas the Elake values with changed sign are shown below the same axis.

  • Fig. 12.

    Similar to Fig. 8 but with the component NBS simulated in experiments MESH4 (orange), MESH5 (green), and MESH6 (red), and the component NBS predicted by GLERL using their overlake precipitation estimation method (blue).

  • Fig. 13.

    Similar to Fig. 10 but for experiment MESH6.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 1808 582 123
PDF Downloads 978 177 37