The development of extensive pack ice fields on the Great Lakes significantly influences lake-effect storms and local airmass modification, as well as the regional hydrologic cycle and lake water levels. The evolution of the ice fields and their impacts on the atmospheric boundary layer complicates weather forecasters’ ability to accurately predict late-season lake-effect snows. The Great Lakes Ice Cover–Atmospheric Flux (GLICAF) experiment was conducted over Lake Erie during February 2004 to investigate the surface–atmosphere exchanges that occur over midlatitude ice-covered lakes. GLICAF observations taken by the University of Wyoming King Air on 26 February 2004 show a strong mesoscale thermal link between the lake surface and the overlying atmospheric boundary layer. Mesoscale atmospheric variations that developed over the lake in turn influenced heat exchanges with the surface. Boundary layer sensible and latent heat fluxes exhibited different relationships to variations in surface pack ice concentration. Turbulent sensible heat fluxes decreased nonlinearly with increases in underlying lake-surface ice concentration such that the largest decreases occurred when ice concentrations were greater than 70%. Latent heat fluxes tended to decrease linearly with increasing ice concentration and had a reduced correlation. Most current operational numerical weather prediction models use simple algorithms to represent the influence of heterogeneous ice cover on heat and moisture fluxes. The GLICAF findings from 26 February 2004 suggest that some currently used and planned approaches in numerical weather prediction models may significantly underestimate sensible heat fluxes in regions of high-concentration ice cover, leading to underpredictions of the local modification of air masses and lake-effect snows.
The presence of substantial pack ice cover on the Great Lakes significantly modifies the local and large-scale atmospheric response to the lakes (e.g., Niziol 1987). Although the presence of pack ice reduces the transfer of heat and moisture from the lake surface to the atmosphere, lake-effect clouds and precipitation have occurred during conditions of extensive ice coverage (R. LaPlante, Cleveland National Weather Service Forecast Office, 2003, personal communication; see also Buffalo National Weather Service Forecast Office 2005 at http://www.erh.noaa.gov/buf/lakeffect/indexlk.html; Laird and Kristovich 2004). A recent study by Cordeira and Laird (2005) examined the evolution of snowfall regions and ice-cover conditions for two noteworthy lake-effect snowfall events over the eastern Great Lakes when ice concentrations were greater than 90% for most of Lake Erie. For example, during one of the events examined (28–31 January 2004), snowfall totals downwind of Lake Erie exceeded 30 cm. These observations indicate that substantial sensible and latent heat fluxes can still occur over Great Lakes that are substantially covered with pack ice.
Studies of Great Lakes winter lake-effect processes have typically examined the mesoscale and microphysical atmospheric boundary layer responses to surface diabatic forcing over ice-free regions (e.g., Kristovich et al. 1999, 2000, 2003; Young et al. 2000; Cooper et al. 2000; Mayor et al. 2003; Schroeder et al. 2006; Miles and Verlinde 2005a, b) and the large-scale collective influence of the lakes (e.g., Sousounis and Fritsch 1994; Angel and Isard 1997; Sousounis 1997, 1998; Weiss and Sousounis 1999). There have been only a few lake-effect studies that have investigated the atmospheric response to variations in lake-surface characteristics. For example, Kristovich and Laird (1998) indicated that spatial variations in Lake Michigan lake-surface temperature influenced the location of initial cloud development in the lake-effect boundary layer, suggesting that lake-surface temperature heterogeneities can noticeably influence lake–atmosphere heat exchange and convective boundary layer development over the Great Lakes. Kristovich et al. (2001) observed that local variations in lake-surface temperature influenced mesoscale patterns of surface heat fluxes. The presence of pack ice would be expected to influence both the surface temperature fields and roughness, in turn having a significant impact on heat and moisture fluxes.
Investigations examining surface heat fluxes and the boundary layer response over ice-covered water have largely been confined to high-latitude or Arctic regions (e.g., Andreas et al. 1979; Alam and Curry 1995, 1997; Pinto et al. 1995; Andreas and Cash 1999; Rouse et al. 2003; Zulauf and Krueger 2003) or marginal oceanic ice shelves (e.g., Renfrew and Moore 1999; Krahmann et al. 2003). Wintertime ice-cover conditions in Arctic regions tend to be characterized by extensive thick ice with near-zero surface heat fluxes and a few leads of open water encompassing a small percentage of the surface area, but associated with large surface heat fluxes. Wintertime sea–air temperature differences can range from 20° to 40°C over leads; thus, breaks in Arctic ice cover are a major contributor to the Arctic heat budget (e.g., Alam and Curry 1997). The magnitude of the influence of leads on the arctic boundary layer depends on numerous factors, including air–ocean temperature differences, lead size and orientation relative to the low-level wind direction, wave age, water surface cooling rates, atmospheric stability, and low-level wind speed and shear (e.g., Alam and Curry 1997; Pinto et al. 1995; Zulauf and Krueger 2003). A key difficulty in applying results from Arctic studies to midlatitudes is that there are significant differences in typical wintertime environmental and surface conditions (e.g., lake–air temperature difference, near-surface static stability, and thickness of ice fields). In addition, many of the processes known to influence surface–air exchanges depend on factors not routinely monitored and therefore are not readily available for incorporation into operational weather prediction models.
Most of the ice cover over the Great Lakes is composed of pack ice, which tends to peak in coverage in late February–early March (Assel 1999). The variability of weather conditions in the Great Lakes, with the passage of polar fronts and cyclones accompanied by high winds, precipitation, and air masses of varying origins, can result in considerable ice formation and loss. In midlake areas, pack ice movement, compaction, formation, and melt can result in highly transitory ice configurations. In addition, variability in atmospheric conditions not only affects ice characteristics, but also leads to a large range in surface heat fluxes (Laird and Kristovich 2002). The variability in both surface and atmospheric conditions plays an important role in cold-season weather conditions, underscoring the need for adequate representations of surface-interaction processes in mesoscale numerical weather prediction models.
Many numerical weather prediction models currently utilize a simplified treatment of Great Lakes ice cover and can have difficulties accurately predicting lake-effect events when extensive ice cover is present (R. LaPlante, Cleveland National Weather Service Office, 2003, personal communication; T. Niziol, Buffalo National Weather Service Office, 2004, personal communication). For instance, the North American Mesoscale (NAM) Model designates 12-km lake grid boxes with less than 50% ice concentration as open water (i.e., no lake ice). Grid boxes with greater than 50% ice concentration are considered to be fully covered with 1-m-thick ice (Meteorology Education & Training 2006). Previously unavailable field observations of the boundary layer response to Great Lakes ice cover are needed to facilitate improved winter weather forecasting in the Great Lakes region. The Great Lakes Ice Cover–Atmospheric Flux (GLICAF) experiment was conducted with a primary goal of using aircraft to collect unprecedented boundary layer observations over a pack ice–covered Lake Erie.
This paper describes the data collection during GLICAF and analysis techniques in section 2, presents the observed relationships between ice cover and boundary layer properties and heat fluxes in section 3, and discusses the findings in the context of previous studies and numerical weather prediction in section 4.
2. Data and methods
This study utilizes a unique dataset for the Great Lakes area to quantify the surface–atmosphere heat exchanges that occur over midlatitude pack ice–covered lakes. This section describes the field data collection and analysis techniques employed to understand these exchanges.
During February 2004, the University of Wyoming King Air aircraft conducted research flights over Lake Erie in support of the GLICAF experiment. GLICAF operations were centered in Toledo, Ohio. Five intensive operations periods (IOPs) were conducted. Flight stacks were flown approximately perpendicular to the mean boundary layer wind and were composed of one upper-level (near 500-m altitude) and two low-level (near 45-m altitude) flight legs. Upper-level flight legs were performed to survey ice-cover conditions using a digital video camera and a downward-pointing pyrometer, while the low-level flight legs (hereinafter referred to as “flux legs”) were utilized to collect turbulent heat flux measurements. Individual flight legs in each flight stack are identified chronologically (i.e., CD2 is the second flight leg in stack CD).
The current study focuses on the 26 February 2004 IOP, during which there was in-flight evidence of positive sensible and latent heat fluxes and generally good low-level visibilities (required for pack ice observations). Four flight stacks were performed in total, with two conducted during the morning and the remaining two occurring during the afternoon. The Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) image of the study area at 1815 UTC (Fig. 1a) shows the spatial ice-cover distribution and the locations of flight stacks performed on 26 February. The large-scale ice-cover features are consistent with the 26 February National Ice Center ice-cover analysis (see Great Lakes ice analysis 2004, available online at http://www.natice.noaa.gov; Fig. 1b), with high-concentration ice in the vicinity and east of flight stack CD, particularly in southern regions. Lower-concentration ice was located near stack AB, and ice-free water along the entire northern and western portions of the lake. A segment of flight stack AB had low-level stratus and fog, which obscured observations of surface pack ice. In addition, melting of the ice surface during the afternoon flight stacks (EF and GH) precluded their use for the present study. Portions of the flight stacks used for the current investigation are highlighted with solid black lines in Fig. 1a.
During the GLICAF experiment, observations of surface pack ice concentration and turbulent heat fluxes (sensible and latent) were obtained over regions of variable ice concentration. King Air data were available at frequencies of 1 and 25 Hz, corresponding to flight distances of approximately 80 and 3 m, respectively. Temperature and water vapor pressure measurements were collected by the Minco Element Reverse Flow thermometer and the Li-Cor, Inc., Model LI-6262 CO2/H2O Analyzer, correspondingly (University of Wyoming 2005). Flight-level vertical air motions were observed by the University of Wyoming King Air gust probe (see, e.g., Lenschow 1973).
Ice concentration estimates were retrieved primarily by a downward-pointing Heimann KT-19.85 pyrometer, with supporting data from a digital video camera. This pyrometer measures upward-directed infrared radiation from which surface temperatures can be inferred. The Heimann pyrometer operated with a spectral range of 9.6–11.5 μm, a field-of-view of 2°, and a measurement range from −50° to 400°C and possessed an adjustable response time that was set at 0.1 s for this experiment (Laursen 2005; University of Wyoming 2005). Comparisons between pyrometer surface temperature data and digital video imagery (collected using a JVC, Inc., GR-DV800U video camera) show that lake-surface ice and water, as well as boundaries between ice and water, were well depicted by the pyrometer during the morning flight (stacks AB and CD). Figure 2 shows an example of responses of pyrometer measurements collected over water and ice surface transitions. During the morning hours, areas of water and ice cover were generally associated with surface temperatures of greater than and less than −0.5°C, respectively. Examination of inferred pack ice concentrations using thresholds between −0.1° and −0.9°C resulted in changes in values of ice concentrations, but did not change the overall shapes of the relationships (e.g., linear versus nonlinear) between ice-cover concentration and heat fluxes. For the present study, lake-surface ice concentration was estimated by the percentage of 25-Hz Heimann pyrometer observations below −0.5°C over the same time periods used for flux calculations.
Turbulent sensible and latent heat fluxes were estimated using eddy-correlation techniques (e.g., Stull 1988). Sensible heat flux was calculated using
where ρ represents the density of dry air at 0°C temperature and 1000-hPa pressure (1.275 kg m−3), cp is the specific heat of dry air at constant pressure (1004 J K−1 kg−1), w′ represents perturbations in vertical wind speed, and θ′ represents perturbations in potential temperature. Similarly, latent heat flux was determined using
where Lυ is the latent heat of vaporization (2.5 × 106 J kg−1) and q′ represents perturbations in specific humidity. Similar methods have been successfully employed in several studies of positive heat fluxes in lake-effect situations using University of Wyoming King Air collected data (e.g., Kelly 1984; Chang and Braham 1991; Kristovich 1993; Kristovich and Braham 1998; Kristovich et al. 2003).
To calculate heat fluxes, it is necessary to determine perturbations in potential temperature and specific humidity using appropriate mean values. Various methods have been used to determine mean values, including linear detrending (e.g., Kristovich 1993; Kristovich and Braham 1998), mean values for blocks of time periods and running mean values (Sun et al. 1996), and high-pass filtering techniques (e.g., Chang and Braham 1991). Temperature and specific humidity varied nonlinearly along flux legs in flight stacks AB and CD, making linear-detrending and block-averaging techniques inappropriate. For this study, θ′ and q′ were calculated by subtracting 30-s moving averages (i.e., running means) from instantaneous values of θ and q. It is recognized that running means do not satisfy Reynolds averaging criteria because the mean values are different for each point along the time series. Despite this, however, Sun et al. (1996) found that the method was very useful in cases with highly nonlinear trends of state variables. Pass-mean sensible heat fluxes calculated with data detrended using running means agreed well with those estimated using bulk methods. Calculated latent heat fluxes tended to be of equal sign but less magnitude than expected using bulk methods.
a. Synoptic weather and ice-cover conditions
During the morning of 26 February, regional weather conditions were dominated by a 1039-hPa surface high pressure system located east of Hudson Bay in Quebec, Canada. At 850 hPa, an area of high geopotential heights was centered over southern Lake Huron, to the north of the study region. The associated anticyclonic flow around the regions of high pressure resulted in winds generally from the ENE between the surface and 850 hPa throughout the study day. Cloud cover was sparse, though high-level cirrus clouds occasionally drifted over the study area. Patches of dense fog were also present, particularly along the southern shore of Lake Erie. The presence of low clouds obscured the lake surface during portions of these flight legs, making it impossible to estimate lake-surface ice concentration. Approximately 25 and 32 km of data from the southern ends of the two flux legs in flight stack AB were not analyzed for this reason. Since fog was not detected along any other portions of the flux legs, all remaining data were retained.
Horizontal wind speeds observed by the King Air during flux legs ranged from 3 to 10 m s−1. Flux-level temperatures from stacks AB and CD mostly ranged from −3.5° to −1°C, resulting in flux leg–average differences in temperature between the lake surface and near-surface air temperature (estimated by dry adiabatic adjustment of flux-level temperatures) of around 0.3°C in stack AB and 0.8°C in stack CD. With moderate winds and positive lake–air temperature differences, weak upward surface heat fluxes were anticipated. Bulk estimates using aircraft observations collected during flux legs in flight stacks AB and CD ranged from 1.7 to 7.1 W m−2 for sensible heat fluxes and 3.9 to 11.6 W m−2 for latent heat fluxes (Gerbush 2005). Observed average sensible heat fluxes during the flux legs agreed well with bulk estimates, ranging from 1.8 to 6.1 W m−2. Latent heat fluxes, however, were lower than expected, ranging from 1.2 to 3.7 W m−2.
An ice-cover analysis provided by the U.S. National Ice Center (Fig. 1b) indicated that substantial ice cover (>90%) was present over the majority of Lake Erie on 26 February 2004. The ice-cover analysis presented a similar spatial ice-cover distribution as the MODIS image shown in Fig. 1a. The highest concentration ice cover was generally confined to eastern portions of the lake, with greater ice concentration variability in the central and western basin where regions of ice-free water were present, especially along the southern and northern shores of the lake.
b. Relationships between lake-surface and low-level atmospheric conditions
Turbulent heat fluxes depend on both surface characteristics and near-surface atmospheric thermodynamic properties. Atmospheric conditions observed during the King Air flux legs (near 45-m altitude) were used to estimate near-surface temperature and humidity conditions to better understand their relationship with changes in the lake surface as well as to give background information necessary for understanding observations of heat fluxes discussed later.
Figure 3 shows lake-surface temperature, as determined by the Heimann pyrometer, and air temperatures observed by the King Air during a flux leg in each of the flight stacks AB and CD. Similar spatial patterns were seen in the other flux legs. Spatial patterns in lake-surface and 45-m air temperature display strong similarities, especially with respect to mesoscale (>5 km) variations. The general increasing trend in lake-surface temperatures from south to north along flight leg AB1 is matched by a similar northward increase in 45-m air temperatures (Fig. 3a). A significant mesoscale trend in lake-surface temperature (increase of around 2.5°C from south to north) was also present along flight leg CD2 (Fig. 3b). Mesoscale variations in lake-surface characteristics appear to play a dominant role in determining mesoscale variations in low-level atmospheric temperatures, despite the presence of extensive ice cover, as discussed later.
Alternatively, small-scale (<5 km) variations in lake-surface temperatures were not well correlated with flux-level temperature variations. For instance, small-scale variations in lake-surface temperatures (abrupt surface temperature fluctuations of 1.5°–2°C in less than 1 km) along the southern half of flight leg AB1 did not appear to correlate with similar variations in 45-m air temperatures (Fig. 3a). In flight leg CD2 (Fig. 3b), fewer small-scale surface temperature fluctuations were evident, but those present were also not consistently reflected in air temperatures at a height of 45 m.
Figure 4 shows estimates of lake-surface water vapor mixing ratio and 45-m atmospheric mixing ratio observed during flux legs AB1 and CD2. Surface mixing ratios were approximated by calculating the saturation mixing ratio (over a plane water surface) as a function of lake-surface temperature under the assumption that the air at the lake surface was the same temperature as the surface and saturated with respect to water. The saturation mixing ratio over ice cover should be reasonably approximated by the saturation mixing ratio estimated assuming a water surface since pyrometer-observed lake-surface temperatures were generally close to 0°C.
Large-scale variations in surface mixing ratio showed a relatively weak correlation with 45-m mixing ratio. In flight leg AB1, where surface mixing ratios generally increased from south to north, flux-level mixing ratio observations exhibited a slight decreasing trend (Fig. 4a). For flight leg CD2, the mixing ratio observed at 45-m height (Fig. 4b) exhibited an overall south-to-north increasing trend (approximately 0.25 g kg−1), but at a much slower rate of increase than at the surface (around 0.60 g kg−1). As was the case for temperature, the small-scale variations in surface mixing ratio in both stacks were not observed in the 45-m mixing ratio observations. The stronger relationship between lake-surface and 45-m temperatures relative to the correspondence in mixing ratios at both heights suggests that sensible heat fluxes may possess a stronger relationship with variations in surface ice concentration than latent heat fluxes for the present case.
c. Relationships between surface ice concentrations and heat fluxes
Since low-level atmospheric conditions were related to mesoscale lake-surface characteristics, it is anticipated that heat fluxes would also be influenced by lake-surface pack ice concentrations. Figure 5 gives time series of 30-s-average (approximately 2.4-km flight distance) heat fluxes and surface temperatures as the King Air flew from south to north in flight leg CD2. Both sensible and latent heat fluxes increased from the colder southern regions with high ice concentrations to the warmer northern regions with lower ice concentration. This overall south-to-north increasing trend illustrates the influence of regional variations in ice concentration on sensible heat fluxes. However, there is considerable spatial variability in both heat flux and surface temperature observations along the flight track.
Figure 6 shows the relationships between sensible heat fluxes and surface pack ice concentrations for both flux legs in flight stack CD, calculated over intervals of 15 s (about 1.2-km flight length), 30 s (2.4 km), and 45 s (3.6 km). Although substantial variability is exhibited regardless of averaging interval, decreases in ice concentration were generally associated with increases in sensible heat fluxes. For larger flux-averaging scales (30 and 45 s), the scatter among flux values is reduced.
Data shown in Fig. 6, particularly for larger averaging distances (30 and 45 s), suggest that the relationships between ice concentration and turbulent sensible heat fluxes are best represented by a nonlinear model. If a linear model were used to fit the 30-s average data (Fig. 6b), there would be a tendency for positive residuals between ice concentrations of about 50% and 90% (not shown); this violates assumptions for the applicability of linear regression. As ice concentration decreased from 100% to 70%, sensible heat flux magnitudes increased and remained nearly unchanged for concentrations less than 70%. To quantify the degree of nonlinearity, we fit a one-phase exponential association model to the sensible heat flux–ice concentration observations. The basic form of the model equation can be written as
where X and Y are the independent and dependent variables, respectively, ymax is the plateau of the curve, ymin is the base of the curve, and b is the exponential constant (Motulsky and Christopoulos 2003). In the present case, X is defined as the lake-surface water concentration (100% minus ice concentration), Y is the sensible heat flux, ymax is the maximum sensible heat flux (flux at 0% ice concentration), and ymin is the minimum sensible heat flux (flux at 100% ice concentration). Rewriting Eq. (3) in terms of ice concentration and sensible heat flux yields
where H0% and H100% are the sensible heat fluxes at 0% and 100% ice concentrations, respectively, and I is the lake-surface ice concentration (in percent). It is important to note that with the use of this equation for fitting the observed values, we do not preclude the possibility of a linear relationship (which was found in some cases, as discussed later).
Resulting regressions for a range of averaging scales between 15 and 60 s for stacks AB and CD are shown in Fig. 7. The regressions for stack AB are similar to each other, although the curves for smaller averaging scales (such as 15 s) approach a more linear relationship. Regressions for stack CD show nonlinear relationships for the range of averaging scales. The general agreement between regressions for different flux-averaging scales within each stack suggests the choice of flux-averaging scale does not strongly affect the nonlinear representation of the relationships. Therefore, this study reports the findings using regressions based on 45-s flux averaging (approximately 3.6-km flight distance) for each stack, a scaling period that represents the approximate mean characteristics of the ensemble of regressions. Regression parameters for the 45-s averaging scale for stacks AB and CD are given in Table 1.
The model parameters listed in Table 1 show substantial differences between the ice concentration–sensible heat flux relationships in stacks AB and CD. Both H0% and H100% values are larger in stack CD; H100% for stack AB is close to zero, implying that little heat transfer from the lake to the atmosphere occurred at 100% ice concentration. For stack CD, a positive H100% value suggests that weak upward sensible heat fluxes were present for ice concentrations of 100%. An examination of sensible heat flux observations in Fig. 6 shows that sensible heat fluxes in stack CD were often greater than 0 W m−2 at 100% ice concentration. The larger value of b in stack CD than stack AB indicates a larger increase in sensible heat fluxes with decreasing ice concentration, especially for ice concentrations greater than 70% in stack CD (Fig. 7). Possible reasons for the differences in the sensible heat flux and ice concentration relations between stacks are discussed in section 4.
Measurements of latent heat fluxes over different ice concentrations indicate that they increased steadily with decreasing ice concentration (Fig. 8). Latent heat flux values from stack CD generally clustered between 1 and 4 W m−2 for ice concentrations above 90%. Latent heat flux values at the highest observed ice concentrations in stack AB (not shown) tended to be clustered around 1 W m−2, but were more widely scattered at lower ice concentrations. For both flight stacks AB and CD, latent heat fluxes exhibited more variability at lower ice concentrations (below 90%). For the sake of remaining consistent with the methods used to quantify the ice concentration–sensible heat flux relationships, the same nonlinear model was initially applied to latent heat flux data. This approach, however, yielded near-linear regressions for most flux-averaging scales, suggesting that ice concentration–latent heat flux relationships in this case are better represented by a linear regression model.
Figure 9 shows best-fit regressions of ice concentration–latent heat flux relationships for stacks AB and CD for a range of averaging scales between 15 and 60 s. Latent heat flux relationships in stack AB indicate that minimal moisture exchange occurred at ice concentrations near 100%. For stack CD, the regressions show that estimated latent heat fluxes over regions of near-100% ice concentration were near 2 W m−2. Despite the overall larger latent heat fluxes observed in stack CD, the slopes of the regressions between flight stacks are very similar. Table 2 shows the linear regression model parameters from the 45-s flux-averaging scale, the most representative of the ice concentration–latent heat flux relationships shown in Fig. 9. Interpretation of the differences in pack ice concentration relationships between sensible and latent heat fluxes, and between flight stacks, are discussed below.
a. Mesoscale variations in surface and atmospheric conditions
Flux-level observations taken by the University of Wyoming King Air indicated a south-to-north increase in air temperature over Lake Erie (see, e.g., Fig. 3). Regional surface observations provide no indication of a south-to-north temperature gradient upwind of Lake Erie or over land areas north and south of the flight stacks. Modification of the air by the lake surface must have been responsible for the development of this overlake temperature gradient. As illustrated in Fig. 10, airflow along southern regions of Lake Erie experienced a fetch over areas of high pack ice concentration. The ice would be expected to limit east-to-west warming of the air as it flowed over the slightly warmer lake. Farther north, a fetch over lower ice concentrations would allow for more warming of the air, resulting in the observed south-to-north increase in temperature.
Atmospheric moisture content does not appear to relate to lake-surface characteristics as directly as temperature. Over southern regions, a greater overlake fetch would give more opportunity for moistening of the relatively dry air originating over upwind land areas. However, the smaller overlake fetch in northern areas may have been offset, to an extent, by the presence of abundant open water, resulting in more rapid moistening. The resulting gradients at 45-m height and the surface were not as strongly related for the mixing ratio (Fig. 4) as they were for temperature (Fig. 3). Regardless of the reason, this suggests that sensible heat fluxes would have a stronger relationship with ice concentration than latent heat fluxes in the current case.
Clearly, the spatial distribution of pack ice played a large role in influencing mesoscale variations in temperature and atmospheric moisture in the boundary layer. As will be discussed below, mesoscale air and lake-surface condition variability in turn appeared to influence small-scale variations in sensible and latent heat fluxes.
b. Contributions to heat fluxes
It is interesting to note that the magnitude of heat fluxes differs between flight stacks AB and CD. To a first approximation, sensible heat fluxes are proportional to near-surface wind speed and the temperature difference between the lake surface and the overlying air, ΔTlake–air (e.g., Garratt 1992). Stack-mean sensible heat fluxes in CD were approximately 2.4 times those in stack AB. Mean flux-level wind speeds were about 30% greater in stack CD than in stack AB, suggesting that differences in wind speed alone were not enough to explain the larger positive sensible heat fluxes observed during the stack CD. Differences between surface air temperature (estimated by adjusting 45-m air temperatures dry adiabatically to the surface) and pyrometer-detected lake-surface temperatures were also greater in stack CD than in AB. A mean ΔTlake–air of 0.80°C during stack CD was more than 2 times as large as the mean ΔTlake–air of 0.32°C in stack AB. Consequently, overall differences in ΔTlake–air between stack AB and CD were primarily responsible for the stronger pass-mean sensible heat fluxes observed in stack CD, with increases in wind speed playing a secondary role.
One factor that contributed to a larger mean ΔTlake–air in stack CD was a region of cold air and higher ice concentrations along the southern portion of the stack CD flux legs (Fig. 3b), which resulted in locally higher values of ΔTlake–air. A similar area of cold air was also present in the southern segment of stack AB, but is not evident in Fig. 3a because it was removed from the analyses because of the presence of low clouds along the flight path. The importance of this area of slightly cooler air on heat flux relationships highlights the strong sensitivity of surface–atmosphere exchanges to mesoscale variations in atmospheric thermodynamic conditions, common in the Great Lakes region during winter.
c. Ice concentration–heat flux relationships
A noteworthy finding from an analysis of data collected on 26 February 2004 during the GLICAF experiment was that sensible heat fluxes varied nonlinearly and latent heat fluxes varied linearly with surface pack ice concentration over Lake Erie. Turbulent sensible heat fluxes rapidly decreased as ice concentrations increased above about 70% and were nearly constant for lower ice concentrations. Latent heat fluxes tended to decrease linearly with increases in ice concentration, suggesting that latent heat fluxes did not respond strongly to small breaks in high ice concentration areas.
Despite the general relationships between heat fluxes and the surface, there is a great deal of scatter in the observations. This scatter might be interpreted as being due, in part, to random variations in turbulent eddies responsible for transporting heat and moisture vertically. However, some of the scatter likely reflects physical processes linking the surface to atmospheric conditions at the 45-m-altitude aircraft observations. As described in Hechtel et al. (1990), variations in heat and moisture transfers resulting from small-scale surface features would be expected to combine to create small-scale internal boundary layers that grow upward into the boundary layer.
To gain insight into whether such small-scale internal boundary layers from the surface influence the observations at 45-m height, Fig. 11 shows 1-Hz fluxes for areas of relatively cool and warm air temperatures measured along flux leg CD2. Similar findings were seen for CD3 and flux legs in stack AB (with the exception that the cooler area in stack AB was removed from consideration for this study because of low-level clouds). In the cooler region, 1-Hz sensible heat fluxes tended to exhibit peak values over and near regions of warmer surface temperatures (Fig. 11a). However, in the warmer area this relationship was not as obvious (Fig. 11c). We hypothesize that this is due to the locally larger lake–air temperature differences over leads in the cooler region as compared with the warmer region. Regardless of the reasons, these observations confirm differences in the relationship between sensible heat fluxes and the surface conditions in the cold, high ice concentration region when compared with the warm, low ice concentration regions, as suggested by the nonlinear relationships discussed in section 3c (and shown in Figs. 6, 7). Latent heat fluxes, on the other hand, did not appear to be closely related to small-scale peaks in surface temperatures in either the cool or warm regions. This is consistent with earlier findings that latent heat fluxes did not respond as strongly as sensible heat fluxes to small-scale areas of warm surface temperature (interpreted as leads) in regions of high-concentration pack ice.
While it is not possible to fully explain with this dataset why sensible and latent heat fluxes vary differently with surface ice concentrations, we speculate that nearly the same amount of water vapor would be available for vertical transport over water as over ice at air temperatures close to 0°C. For example, saturation vapor pressure over ice would be approximately 97%–99% of that over water over the typical range of lake-surface temperatures observed in this case (from −3° to −1°C). We speculate that even weak transfers of moisture between the lake and atmosphere over high concentration ice may decrease the local responses in latent heat fluxes over small breaks in ice cover, resulting in more linear relationships between fluxes and pack ice concentration.
d. Implications for late-winter lake-effect snow prediction
The nonlinear relationship between sensible heat fluxes and surface ice concentrations appears to be the result of mesoscale variations in both ice cover and atmospheric conditions along the flight legs, as well as the upwind surface–atmosphere interactions that influenced atmospheric conditions where aircraft observations were collected. If found to be a common feature, the nonlinear relationship would have important implications for mesoscale operational forecast models. For example, Fig. 12 schematically compares the relationships found from the analysis of the 26 February 2004 GLICAF measurements with the NAM Model’s representation of sensible heat fluxes as a function of ice concentration, assuming perfect agreement at 0% ice concentration. Currently, the NAM Model grid boxes with less than 50% ice concentration are assigned sensible heat fluxes representative of open water (i.e., no lake ice), while grid boxes with greater than 50% ice concentration are assumed to be fully covered with 1-m thick ice (Meteorology Education & Training 2006). The observed nonlinear relationship suggests that the sensible heat fluxes integrated over the entire range of ice concentrations would be underestimated using methods similar to the NAM Model. In fact, this underestimate in sensible heat fluxes would not be greatly improved if a linear relationship between ice concentration and sensible heat fluxes were employed in mesoscale models.
5. Summary and future work
Analyses of data collected by the University of Wyoming King Air on 26 February 2004 during the Great Lakes Ice Cover–Atmospheric Flux experiment revealed strong links between atmospheric conditions and characteristics of partially pack ice–covered Lake Erie. Observed low-level temperatures were found to correlate more closely than mixing ratio to lake-surface conditions. The close thermal link was also observed in sensible and latent heat flux relationships with surface ice concentration. It was shown that minor mesoscale variations in surface and atmospheric thermodynamic characteristics produced observable changes in heat fluxes.
Sensible heat fluxes, an important driving mechanism for lake-effect snow storms, were found to vary nonlinearly with surface ice concentrations. Near-open-water fluxes were observed for ice concentrations less than about 70%. This implies that if linear relationships are employed in numerical weather prediction models, the total heating of the atmosphere by lakes with considerable pack ice cover would be significantly underestimated.
It is critical to determine whether these observed ice concentration–heat flux relationships are common to the Great Lakes and to examine their potential influence on mesoscale atmospheric responses. Collection and analysis of field observations over a wide range of atmospheric and surface pack ice conditions will give necessary insight into both the statistical relationships and the processes responsible for these relationships. Given the observed links between atmospheric, lake, and pack ice processes, interdisciplinary observational field experiments and coupled modeling approaches are needed to fully understand the interactions.
The authors greatly appreciate the efforts of the staff and crew of the University of Wyoming King Air. Forecasting for the GLICAF experiment was carried out by lead forecasters Michael Kruk and Michael Spinar from the Illinois State Water Survey, the authors, and Stephen Jackman from the Department of Atmospheric Sciences, University of Illinois. Project assistance from Yarice Rodriguez, Department of Geography, University of Illinois, is also appreciated. Internal reviews of the manuscript were conducted by Jim Angel and Kenneth Kunkel from the Illinois State Water Survey. Reviews by three anonymous reviewers are also appreciated. GLICAF and analyses described here were funded by the National Science Foundation (NSF 02-02305 and NSF 05-12954). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation or the Illinois State Water Survey.
* Current affiliation: Office of the New Jersey State Climatologist, Rutgers, The State University of New Jersey, Piscataway, New Jersey
Corresponding author address: Dr. David A. R. Kristovich, 2204 Griffith Dr., Champaign, IL 61820-7495. Email: firstname.lastname@example.org