mesoscale boundary layer and heat flux variations over pack ice–covered lake...

Mesoscale Boundary Layer and Heat Flux Variations over Pack Ice–CoveredLake Erie

MATHIEU R. GERBUSH*

Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, Illinois

DAVID A. R. KRISTOVICH

Center for Atmospheric Science, Illinois State Water Survey, Illinois Department of Natural Resources, Champaign, Illinois

NEIL F. LAIRD

Department of Geoscience, Hobart and William Smith Colleges, Geneva, New York

(Manuscript received 17 April 2006, in final form 7 March 2007)

ABSTRACT

The development of extensive pack ice fields on the Great Lakes significantly influences lake-effectstorms and local airmass modification, as well as the regional hydrologic cycle and lake water levels. Theevolution of the ice fields and their impacts on the atmospheric boundary layer complicates weatherforecasters’ 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 investi-gate the surface–atmosphere exchanges that occur over midlatitude ice-covered lakes. GLICAF observa-tions taken by the University of Wyoming King Air on 26 February 2004 show a strong mesoscale thermallink between the lake surface and the overlying atmospheric boundary layer. Mesoscale atmospheric varia-tions that developed over the lake in turn influenced heat exchanges with the surface. Boundary layersensible and latent heat fluxes exhibited different relationships to variations in surface pack ice concentra-tion. Turbulent sensible heat fluxes decreased nonlinearly with increases in underlying lake-surface iceconcentration 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 corre-lation. Most current operational numerical weather prediction models use simple algorithms to representthe influence of heterogeneous ice cover on heat and moisture fluxes. The GLICAF findings from 26February 2004 suggest that some currently used and planned approaches in numerical weather predictionmodels 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.

1. Introduction

The presence of substantial pack ice cover on theGreat Lakes significantly modifies the local and large-scale atmospheric response to the lakes (e.g., Niziol

1987). Although the presence of pack ice reduces thetransfer of heat and moisture from the lake surface tothe atmosphere, lake-effect clouds and precipitationhave occurred during conditions of extensive ice cov-erage (R. LaPlante, Cleveland National Weather Ser-vice Forecast Office, 2003, personal communication;see also Buffalo National Weather Service Forecast Of-fice 2005 at http://www.erh.noaa.gov/buf/lakeffect/indexlk.html; Laird and Kristovich 2004). A recentstudy by Cordeira and Laird (2005) examined the evo-lution of snowfall regions and ice-cover conditions fortwo noteworthy lake-effect snowfall events over theeastern Great Lakes when ice concentrations were

* Current affiliation: Office of the New Jersey State Climatolo-gist, 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.E-mail: [email protected]

668 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 47

DOI: 10.1175/2007JAMC1479.1

© 2008 American Meteorological Society

JAM2623

greater than 90% for most of Lake Erie. For example,during one of the events examined (28–31 January2004), snowfall totals downwind of Lake Erie exceeded30 cm. These observations indicate that substantial sen-sible and latent heat fluxes can still occur over GreatLakes that are substantially covered with pack ice.

Studies of Great Lakes winter lake-effect processeshave typically examined the mesoscale and microphysi-cal atmospheric boundary layer responses to surfacediabatic forcing over ice-free regions (e.g., Kristovich etal. 1999, 2000, 2003; Young et al. 2000; Cooper et al.2000; Mayor et al. 2003; Schroeder et al. 2006; Miles andVerlinde 2005a,b) and the large-scale collective influ-ence of the lakes (e.g., Sousounis and Fritsch 1994; An-gel and Isard 1997; Sousounis 1997, 1998; Weiss andSousounis 1999). There have been only a few lake-ef-fect studies that have investigated the atmospheric re-sponse to variations in lake-surface characteristics. Forexample, Kristovich and Laird (1998) indicated thatspatial variations in Lake Michigan lake-surface tem-perature influenced the location of initial cloud devel-opment in the lake-effect boundary layer, suggestingthat lake-surface temperature heterogeneities can no-ticeably influence lake–atmosphere heat exchange andconvective boundary layer development over the GreatLakes. Kristovich et al. (2001) observed that local varia-tions in lake-surface temperature influenced mesoscalepatterns of surface heat fluxes. The presence of pack icewould be expected to influence both the surface tem-perature fields and roughness, in turn having a signifi-cant impact on heat and moisture fluxes.

Investigations examining surface heat fluxes and theboundary layer response over ice-covered water havelargely 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 iceshelves (e.g., Renfrew and Moore 1999; Krahmann etal. 2003). Wintertime ice-cover conditions in Arctic re-gions tend to be characterized by extensive thick icewith near-zero surface heat fluxes and a few leads ofopen water encompassing a small percentage of the sur-face area, but associated with large surface heat fluxes.Wintertime sea–air temperature differences can rangefrom 20° to 40°C over leads; thus, breaks in Arctic icecover are a major contributor to the Arctic heat budget(e.g., Alam and Curry 1997). The magnitude of the in-fluence of leads on the arctic boundary layer dependson numerous factors, including air–ocean temperaturedifferences, lead size and orientation relative to thelow-level wind direction, wave age, water surface cool-ing rates, atmospheric stability, and low-level windspeed and shear (e.g., Alam and Curry 1997; Pinto et al.

1995; Zulauf and Krueger 2003). A key difficulty inapplying results from Arctic studies to midlatitudes isthat there are significant differences in typical winter-time environmental and surface conditions (e.g., lake–air temperature difference, near-surface static stability,and thickness of ice fields). In addition, many of theprocesses known to influence surface–air exchanges de-pend on factors not routinely monitored and thereforeare not readily available for incorporation into opera-tional weather prediction models.

Most of the ice cover over the Great Lakes is com-posed of pack ice, which tends to peak in coverage inlate February–early March (Assel 1999). The variabilityof weather conditions in the Great Lakes, with the pas-sage of polar fronts and cyclones accompanied by highwinds, precipitation, and air masses of varying origins,can result in considerable ice formation and loss. Inmidlake areas, pack ice movement, compaction, forma-tion, and melt can result in highly transitory ice con-figurations. In addition, variability in atmospheric con-ditions not only affects ice characteristics, but also leadsto a large range in surface heat fluxes (Laird and Kris-tovich 2002). The variability in both surface and atmo-spheric conditions plays an important role in cold-season weather conditions, underscoring the need foradequate representations of surface-interaction pro-cesses in mesoscale numerical weather prediction mod-els.

Many numerical weather prediction models currentlyutilize a simplified treatment of Great Lakes icecover and can have difficulties accurately predictinglake-effect events when extensive ice cover is present(R. LaPlante, Cleveland National Weather Service Of-fice, 2003, personal communication; T. Niziol, BuffaloNational Weather Service Office, 2004, personal com-munication). For instance, the North American Meso-scale (NAM) Model designates 12-km lake grid boxeswith less than 50% ice concentration as open water (i.e.,no lake ice). Grid boxes with greater than 50% iceconcentration are considered to be fully covered with1-m-thick ice (Meteorology Education & Training2006). Previously unavailable field observations of theboundary layer response to Great Lakes ice cover areneeded to facilitate improved winter weather forecast-ing in the Great Lakes region. The Great Lakes IceCover–Atmospheric Flux (GLICAF) experiment wasconducted with a primary goal of using aircraft to col-lect unprecedented boundary layer observations over apack ice–covered Lake Erie.

This paper describes the data collection duringGLICAF and analysis techniques in section 2, presentsthe observed relationships between ice cover andboundary layer properties and heat fluxes in section 3,

FEBRUARY 2008 G E R B U S H E T A L . 669

and discusses the findings in the context of previousstudies and numerical weather prediction in section 4.

2. Data and methods

This study utilizes a unique dataset for the GreatLakes area to quantify the surface–atmosphere heat ex-changes that occur over midlatitude pack ice–coveredlakes. This section describes the field data collectionand analysis techniques employed to understand theseexchanges.

a. Data

During February 2004, the University of WyomingKing Air aircraft conducted research flights over LakeErie in support of the GLICAF experiment. GLICAFoperations were centered in Toledo, Ohio. Five inten-sive operations periods (IOPs) were conducted. Flightstacks were flown approximately perpendicular to themean boundary layer wind and were composed of oneupper-level (near 500-m altitude) and two low-level(near 45-m altitude) flight legs. Upper-level flight legswere performed to survey ice-cover conditions using adigital video camera and a downward-pointing pyrom-eter, while the low-level flight legs (hereinafter referredto as “flux legs”) were utilized to collect turbulent heatflux measurements. Individual flight legs in each flightstack are identified chronologically (i.e., CD2 is the sec-ond flight leg in stack CD).

The current study focuses on the 26 February 2004IOP, during which there was in-flight evidence of posi-tive sensible and latent heat fluxes and generally goodlow-level visibilities (required for pack ice observa-tions). Four flight stacks were performed in total, withtwo conducted during the morning and the remainingtwo occurring during the afternoon. The Aqua Moder-ate Resolution Imaging Spectroradiometer (MODIS)image of the study area at 1815 UTC (Fig. 1a) shows thespatial ice-cover distribution and the locations of flightstacks performed on 26 February. The large-scale ice-cover features are consistent with the 26 February Na-tional Ice Center ice-cover analysis (see Great Lakesice analysis 2004, available online at http://www.natice.noaa.gov; Fig. 1b), with high-concentration ice in thevicinity and east of flight stack CD, particularly insouthern regions. Lower-concentration ice was locatednear stack AB, and ice-free water along the entirenorthern and western portions of the lake. A segmentof flight stack AB had low-level stratus and fog, whichobscured observations of surface pack ice. In addition,melting of the ice surface during the afternoon flightstacks (EF and GH) precluded their use for the present

study. Portions of the flight stacks used for the currentinvestigation are highlighted with solid black lines inFig. 1a.

During the GLICAF experiment, observations ofsurface pack ice concentration and turbulent heatfluxes (sensible and latent) were obtained over regionsof variable ice concentration. King Air data were avail-able at frequencies of 1 and 25 Hz, corresponding toflight distances of approximately 80 and 3 m, respec-tively. Temperature and water vapor pressure measure-ments were collected by the Minco Element ReverseFlow thermometer and the Li-Cor, Inc., Model LI-6262CO2/H2O Analyzer, correspondingly (University ofWyoming 2005). Flight-level vertical air motions wereobserved by the University of Wyoming King Air gustprobe (see, e.g., Lenschow 1973).

Ice concentration estimates were retrieved primarilyby a downward-pointing Heimann KT-19.85 pyrom-eter, with supporting data from a digital video camera.This pyrometer measures upward-directed infrared ra-diation from which surface temperatures can be in-ferred. The Heimann pyrometer operated with a spec-tral range of 9.6–11.5 �m, a field-of-view of 2°, and ameasurement range from �50° to 400°C and possessedan adjustable response time that was set at 0.1 s for thisexperiment (Laursen 2005; University of Wyoming2005). Comparisons between pyrometer surface tem-perature data and digital video imagery (collected usinga JVC, Inc., GR-DV800U video camera) show thatlake-surface ice and water, as well as boundaries be-tween ice and water, were well depicted by the pyrom-eter during the morning flight (stacks AB and CD).Figure 2 shows an example of responses of pyrometermeasurements collected over water and ice surfacetransitions. During the morning hours, areas of waterand ice cover were generally associated with surfacetemperatures of greater than and less than �0.5°C, re-spectively. Examination of inferred pack ice concentra-tions using thresholds between �0.1° and �0.9°C re-sulted in changes in values of ice concentrations, but didnot change the overall shapes of the relationships (e.g.,linear versus nonlinear) between ice-cover concentra-tion and heat fluxes. For the present study, lake-surfaceice concentration was estimated by the percentage of25-Hz Heimann pyrometer observations below �0.5°Cover the same time periods used for flux calculations.

b. Methods

Turbulent sensible and latent heat fluxes were esti-mated using eddy-correlation techniques (e.g., Stull1988). Sensible heat flux was calculated using

H � �cp�w��, �1�


where � represents the density of dry air at 0°C tem-perature and 1000-hPa pressure (1.275 kg m�3), cp isthe specific heat of dry air at constant pressure (1004 JK�1 kg�1), w� represents perturbations in vertical windspeed, and �� represents perturbations in potential tem-perature. Similarly, latent heat flux was determined us-ing

HL � �L��w�q��, �2�

where L is the latent heat of vaporization (2.5 106 Jkg�1) and q� represents perturbations in specific hu-

midity. Similar methods have been successfully em-ployed in several studies of positive heat fluxes in lake-effect situations using University of Wyoming King Aircollected data (e.g., Kelly 1984; Chang and Braham1991; Kristovich 1993; Kristovich and Braham 1998;Kristovich et al. 2003).

To calculate heat fluxes, it is necessary to determineperturbations in potential temperature and specific hu-midity using appropriate mean values. Various meth-ods have been used to determine mean values, includ-ing linear detrending (e.g., Kristovich 1993; Kristovich

FIG. 1. (a) Aqua MODIS 250-m resolution image of ice cover on Lake Erie from 1815 UTC 26 Feb 2004.Locations of flight stacks AB, CD, EF, and GH, conducted by the University of Wyoming King Air on this date,are indicated. Thick, solid lines indicate portions of the flight stacks used in the current study. The approximate45-m wind direction observed by the King Air during flight stacks AB and CD is shown. (b) Portion of the NationalIce Center ice analysis for this date (see http://www.natice.noaa.gov).


and Braham 1998), mean values for blocks of time pe-riods and running mean values (Sun et al. 1996), andhigh-pass filtering techniques (e.g., Chang and Braham1991). Temperature and specific humidity varied non-linearly along flux legs in flight stacks AB and CD,making linear-detrending and block-averaging tech-niques inappropriate. For this study, �� and q� werecalculated by subtracting 30-s moving averages (i.e.,running means) from instantaneous values of � and q. Itis recognized that running means do not satisfy Reyn-olds averaging criteria because the mean values are dif-ferent for each point along the time series. Despite this,however, Sun et al. (1996) found that the method wasvery useful in cases with highly nonlinear trends of statevariables. Pass-mean sensible heat fluxes calculatedwith data detrended using running means agreed wellwith those estimated using bulk methods. Calculatedlatent heat fluxes tended to be of equal sign but lessmagnitude than expected using bulk methods.

3. Results

a. Synoptic weather and ice-cover conditions

During the morning of 26 February, regional weatherconditions were dominated by a 1039-hPa surface highpressure system located east of Hudson Bay in Quebec,Canada. At 850 hPa, an area of high geopotentialheights was centered over southern Lake Huron, to thenorth of the study region. The associated anticyclonicflow around the regions of high pressure resulted inwinds generally from the ENE between the surface and850 hPa throughout the study day. Cloud cover wassparse, though high-level cirrus clouds occasionallydrifted over the study area. Patches of dense fog werealso present, particularly along the southern shore ofLake Erie. The presence of low clouds obscured thelake surface during portions of these flight legs, makingit impossible to estimate lake-surface ice concentration.Approximately 25 and 32 km of data from the southernends of the two flux legs in flight stack AB were notanalyzed for this reason. Since fog was not detectedalong any other portions of the flux legs, all remainingdata were retained.

Horizontal wind speeds observed by the King Airduring flux legs ranged from 3 to 10 m s�1. Flux-leveltemperatures from stacks AB and CD mostly rangedfrom �3.5° to �1°C, resulting in flux leg–average dif-ferences in temperature between the lake surface andnear-surface air temperature (estimated by dry adia-batic adjustment of flux-level temperatures) of around0.3°C in stack AB and 0.8°C in stack CD. With mod-erate winds and positive lake–air temperature differ-ences, weak upward surface heat fluxes were anticipat-

FIG. 2. Example of (b) 25-Hz Heimann pyrometer measure-ments (°C) and (a), (c) digital video images over two lake-surfaceice water boundaries during flight leg CD1. The arrows in thepyrometer time series correspond to the times of the video imagesin (a) and (c). The King Air was flying from bottom to top in thevideo images. The King Air flew approximately 3.5 km over thetime interval shown.


ed. Bulk estimates using aircraft observations collectedduring flux legs in flight stacks AB and CD ranged from1.7 to 7.1 W m�2 for sensible heat fluxes and 3.9 to 11.6W m�2 for latent heat fluxes (Gerbush 2005). Observedaverage sensible heat fluxes during the flux legs agreedwell 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. NationalIce Center (Fig. 1b) indicated that substantial ice cover(�90%) was present over the majority of Lake Erie on26 February 2004. The ice-cover analysis presented asimilar spatial ice-cover distribution as the MODIS im-age shown in Fig. 1a. The highest concentration icecover was generally confined to eastern portions of thelake, with greater ice concentration variability in thecentral and western basin where regions of ice-free wa-ter were present, especially along the southern andnorthern shores of the lake.

b. Relationships between lake-surface and low-levelatmospheric conditions

Turbulent heat fluxes depend on both surface char-acteristics and near-surface atmospheric thermody-namic properties. Atmospheric conditions observedduring the King Air flux legs (near 45-m altitude) wereused to estimate near-surface temperature and humid-ity conditions to better understand their relationshipwith changes in the lake surface as well as to give back-ground information necessary for understanding obser-vations of heat fluxes discussed later.

Figure 3 shows lake-surface temperature, as deter-mined by the Heimann pyrometer, and air tempera-tures observed by the King Air during a flux leg in eachof the flight stacks AB and CD. Similar spatial patternswere seen in the other flux legs. Spatial patterns inlake-surface and 45-m air temperature display strongsimilarities, especially with respect to mesoscale (�5km) variations. The general increasing trend in lake-surface temperatures from south to north along flightleg AB1 is matched by a similar northward increase in45-m air temperatures (Fig. 3a). A significant mesoscaletrend in lake-surface temperature (increase of around2.5°C from south to north) was also present along flightleg CD2 (Fig. 3b). Mesoscale variations in lake-surfacecharacteristics appear to play a dominant role in deter-mining mesoscale variations in low-level atmospherictemperatures, despite the presence of extensive icecover, as discussed later.

Alternatively, small-scale (�5 km) variations in lake-surface temperatures were not well correlated withflux-level temperature variations. For instance, small-scale variations in lake-surface temperatures (abrupt

surface temperature fluctuations of 1.5°–2°C in lessthan 1 km) along the southern half of flight leg AB1 didnot appear to correlate with similar variations in 45-mair temperatures (Fig. 3a). In flight leg CD2 (Fig. 3b),fewer small-scale surface temperature fluctuations wereevident, but those present were also not consistentlyreflected in air temperatures at a height of 45 m.

Figure 4 shows estimates of lake-surface water vapormixing ratio and 45-m atmospheric mixing ratio ob-served during flux legs AB1 and CD2. Surface mixingratios were approximated by calculating the saturationmixing ratio (over a plane water surface) as a functionof lake-surface temperature under the assumption thatthe air at the lake surface was the same temperature asthe surface and saturated with respect to water. Thesaturation mixing ratio over ice cover should be rea-sonably approximated by the saturation mixing ratioestimated assuming a water surface since pyrometer-observed lake-surface temperatures were generallyclose to 0°C.

Large-scale variations in surface mixing ratio showed

FIG. 3. Time series of Heimann radiometric surface tempera-ture (black) and 45-m air temperature (gray) for flux legs (a) AB1and (b) CD2. Here, “S” and “N” denote the southern and north-ern endpoints of the passes, respectively. Time is in hours andminutes, UTC.


a relatively weak correlation with 45-m mixing ratio. Inflight leg AB1, where surface mixing ratios generallyincreased from south to north, flux-level mixing ratioobservations exhibited a slight decreasing trend (Fig.4a). For flight leg CD2, the mixing ratio observed at45-m height (Fig. 4b) exhibited an overall south-to-north increasing trend (approximately 0.25 g kg�1), butat 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 bothstacks were not observed in the 45-m mixing ratio ob-servations. The stronger relationship between lake-surface and 45-m temperatures relative to the corre-spondence in mixing ratios at both heights suggests thatsensible heat fluxes may possess a stronger relationshipwith variations in surface ice concentration than latentheat fluxes for the present case.

c. Relationships between surface ice concentrationsand heat fluxes

Since low-level atmospheric conditions were relatedto mesoscale lake-surface characteristics, it is antici-

pated that heat fluxes would also be influenced by lake-surface pack ice concentrations. Figure 5 gives time se-ries of 30-s-average (approximately 2.4-km flight dis-tance) heat fluxes and surface temperatures as the KingAir flew from south to north in flight leg CD2. Bothsensible and latent heat fluxes increased from thecolder southern regions with high ice concentrations tothe warmer northern regions with lower ice concentra-tion. This overall south-to-north increasing trend illus-trates the influence of regional variations in ice concen-tration on sensible heat fluxes. However, there is con-siderable spatial variability in both heat flux andsurface temperature observations along the flight track.

Figure 6 shows the relationships between sensibleheat fluxes and surface pack ice concentrations for bothflux legs in flight stack CD, calculated over intervalsof 15 s (about 1.2-km flight length), 30 s (2.4 km), and45 s (3.6 km). Although substantial variability is exhib-ited regardless of averaging interval, decreases in iceconcentration were generally associated with increasesin 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

FIG. 4. Time series of estimated saturation mixing ratio (black),calculated from Heimann radiometric temperature observations,and 45-m mixing ratio (gray) for flux legs (a) AB1 and (b) CD2.Here, “S” and “N” denote the southern and northern endpoints ofthe passes, respectively.

FIG. 5. Time series of 30-s average heat fluxes (open squares)and lake-surface temperatures calculated from Heimann radio-metric observations (closed dots) taken during flight leg CD2, 26Feb 2004. (a) Sensible heat fluxes; (b) latent heat fluxes. Here, “S”and “N” denote the southern and northern endpoints of thepasses, respectively. Each 30-s data point represents approxi-mately a 2.4-km flight distance.


distances (30 and 45 s), suggest that the relationshipsbetween ice concentration and turbulent sensible heatfluxes are best represented by a nonlinear model. If alinear model were used to fit the 30-s average data (Fig.6b), there would be a tendency for positive residualsbetween ice concentrations of about 50% and 90% (notshown); this violates assumptions for the applicabilityof linear regression. As ice concentration decreased

from 100% to 70%, sensible heat flux magnitudes in-creased and remained nearly unchanged for concentra-tions less than 70%. To quantify the degree of nonlin-earity, we fit a one-phase exponential associationmodel to the sensible heat flux–ice concentration ob-servations. The basic form of the model equation can bewritten as

Y � ymin �ymax � ymin��1 � e�bX�, �3�

where X and Y are the independent and dependentvariables, respectively, ymax is the plateau of the curve,ymin is the base of the curve, and b is the exponentialconstant (Motulsky and Christopoulos 2003). In thepresent case, X is defined as the lake-surface waterconcentration (100% minus ice concentration), Y is thesensible heat flux, ymax is the maximum sensible heatflux (flux at 0% ice concentration), and ymin is the mini-mum sensible heat flux (flux at 100% ice concentra-tion). Rewriting Eq. (3) in terms of ice concentrationand sensible heat flux yields

H � H100% �H0% � H100%��1 � e�b�100�I��,

�4�

where H0% and H100% are the sensible heat fluxes at0% and 100% ice concentrations, respectively, and I isthe lake-surface ice concentration (in percent). It is im-portant to note that with the use of this equation forfitting the observed values, we do not preclude the pos-sibility of a linear relationship (which was found insome cases, as discussed later).

Resulting regressions for a range of averaging scalesbetween 15 and 60 s for stacks AB and CD are shownin Fig. 7. The regressions for stack AB are similar toeach other, although the curves for smaller averagingscales (such as 15 s) approach a more linear relation-ship. Regressions for stack CD show nonlinear relation-ships for the range of averaging scales. The generalagreement between regressions for different flux-aver-aging scales within each stack suggests the choice offlux-averaging scale does not strongly affect the non-linear representation of the relationships. Therefore,this study reports the findings using regressions basedon 45-s flux averaging (approximately 3.6-km flight dis-tance) for each stack, a scaling period that representsthe approximate mean characteristics of the ensembleof regressions. Regression parameters for the 45-saveraging scale for stacks AB and CD are given inTable 1.

The model parameters listed in Table 1 show sub-stantial differences between the ice concentration–sen-sible heat flux relationships in stacks AB and CD. BothH0% and H100% values are larger in stack CD; H100% for

FIG. 6. Turbulent sensible heat fluxes from flux legs in stack CDplotted as a function of lake-surface ice concentration for fluxaveraging scales of (a) 15, (b) 30, and (c) 45 s. Best-fit nonlinearregression curves are also shown in each example.


stack AB is close to zero, implying that little heat trans-fer from the lake to the atmosphere occurred at 100%ice concentration. For stack CD, a positive H100% valuesuggests that weak upward sensible heat fluxes werepresent for ice concentrations of 100%. An examina-tion of sensible heat flux observations in Fig. 6 showsthat sensible heat fluxes in stack CD were often greaterthan 0 W m�2 at 100% ice concentration. The largervalue of b in stack CD than stack AB indicates a largerincrease in sensible heat fluxes with decreasing ice con-centration, especially for ice concentrations greaterthan 70% in stack CD (Fig. 7). Possible reasons for thedifferences in the sensible heat flux and ice concentra-tion relations between stacks are discussed in section 4.

Measurements of latent heat fluxes over different iceconcentrations indicate that they increased steadilywith decreasing ice concentration (Fig. 8). Latent heatflux values from stack CD generally clustered between1 and 4 W m�2 for ice concentrations above 90%. La-tent heat flux values at the highest observed ice con-centrations in stack AB (not shown) tended to be clus-tered around 1 W m�2, but were more widely scatteredat lower ice concentrations. For both flight stacks ABand CD, latent heat fluxes exhibited more variability atlower ice concentrations (below 90%). For the sake ofremaining consistent with the methods used to quantifythe ice concentration–sensible heat flux relationships,the same nonlinear model was initially applied to latentheat flux data. This approach, however, yielded near-linear regressions for most flux-averaging scales, sug-

gesting that ice concentration–latent heat flux relation-ships in this case are better represented by a linearregression model.

Figure 9 shows best-fit regressions of ice concentra-tion–latent heat flux relationships for stacks AB andCD for a range of averaging scales between 15 and 60 s.Latent heat flux relationships in stack AB indicate thatminimal moisture exchange occurred at ice concentra-tions near 100%. For stack CD, the regressions showthat estimated latent heat fluxes over regions of near-100% ice concentration were near 2 W m�2. Despitethe overall larger latent heat fluxes observed in stackCD, the slopes of the regressions between flight stacksare very similar. Table 2 shows the linear regressionmodel parameters from the 45-s flux-averaging scale,the most representative of the ice concentration–latentheat flux relationships shown in Fig. 9. Interpretation ofthe differences in pack ice concentration relationshipsbetween sensible and latent heat fluxes, and betweenflight stacks, are discussed below.

TABLE 1. Nonlinear regression model parameters for flightstacks AB and CD derived from turbulent sensible heat fluxescalculated using a flux-averaging scale of 45 s. These parametersare for the model Eq. (4).

AB CD

H100% (W m�2) �0.24 1.12H0% (W m�2) 3.1 5.9b 0.031 0.077

FIG. 7. Best-fit nonlinear regressions of sensible heat flux as a function of ice concentrationfor flux-averaging scales ranging from 15 to 60 s. Flight stack (AB or CD) and flux-averagingscale for each curve are indicated. In addition, R2 values are given for each averaging scale.


4. Discussion

a. Mesoscale variations in surface and atmosphericconditions

Flux-level observations taken by the University ofWyoming King Air indicated a south-to-north increasein air temperature over Lake Erie (see, e.g., Fig. 3).Regional surface observations provide no indication ofa south-to-north temperature gradient upwind of Lake

Erie or over land areas north and south of the flightstacks. Modification of the air by the lake surface musthave been responsible for the development of this over-lake temperature gradient. As illustrated in Fig. 10, air-flow along southern regions of Lake Erie experienced afetch over areas of high pack ice concentration. The icewould be expected to limit east-to-west warming of theair as it flowed over the slightly warmer lake. Farthernorth, a fetch over lower ice concentrations would al-low for more warming of the air, resulting in the ob-served south-to-north increase in temperature.

Atmospheric moisture content does not appear torelate to lake-surface characteristics as directly as tem-perature. Over southern regions, a greater overlakefetch would give more opportunity for moistening ofthe relatively dry air originating over upwind land ar-eas. However, the smaller overlake fetch in northernareas may have been offset, to an extent, by the pres-ence of abundant open water, resulting in more rapidmoistening. The resulting gradients at 45-m height andthe surface were not as strongly related for the mixingratio (Fig. 4) as they were for temperature (Fig. 3).Regardless of the reason, this suggests that sensibleheat fluxes would have a stronger relationship with iceconcentration than latent heat fluxes in the currentcase.

Clearly, the spatial distribution of pack ice played alarge role in influencing mesoscale variations in tem-perature and atmospheric moisture in the boundarylayer. As will be discussed below, mesoscale air andlake-surface condition variability in turn appeared toinfluence small-scale variations in sensible and latentheat fluxes.

b. Contributions to heat fluxes

It is interesting to note that the magnitude of heatfluxes differs between flight stacks AB and CD. To afirst approximation, sensible heat fluxes are propor-tional to near-surface wind speed and the temperaturedifference between the lake surface and the overlyingair, �Tlake–air (e.g., Garratt 1992). Stack-mean sensibleheat fluxes in CD were approximately 2.4 times those instack AB. Mean flux-level wind speeds were about 30%greater in stack CD than in stack AB, suggesting thatdifferences in wind speed alone were not enough toexplain the larger positive sensible heat fluxes observedduring the stack CD. Differences between surface airtemperature (estimated by adjusting 45-m air tempera-tures dry adiabatically to the surface) and pyrometer-detected lake-surface temperatures were also greater instack CD than in AB. A mean �Tlake–air of 0.80°C dur-ing stack CD was more than 2 times as large as the

FIG. 8. Turbulent latent heat fluxes from flux legs in stack CDplotted as a function of lake-surface ice concentration for fluxaveraging scales of (a) 15, (b) 30, and (c) 45 s. Best-fit linearregression lines are also shown in each example.


mean �Tlake–air of 0.32°C in stack AB. Consequently,overall differences in �Tlake–air between stack AB andCD were primarily responsible for the stronger pass-mean sensible heat fluxes observed in stack CD, withincreases in wind speed playing a secondary role.

One factor that contributed to a larger mean �Tlake–air

in stack CD was a region of cold air and higher iceconcentrations along the southern portion of the stackCD flux legs (Fig. 3b), which resulted in locally highervalues of �Tlake–air. A similar area of cold air was alsopresent in the southern segment of stack AB, but is notevident in Fig. 3a because it was removed from theanalyses because of the presence of low clouds alongthe flight path. The importance of this area of slightlycooler air on heat flux relationships highlights thestrong sensitivity of surface–atmosphere exchanges tomesoscale variations in atmospheric thermodynamicconditions, common in the Great Lakes region duringwinter.

c. Ice concentration–heat flux relationships

A noteworthy finding from an analysis of data col-lected on 26 February 2004 during the GLICAF experi-ment was that sensible heat fluxes varied nonlinearlyand latent heat fluxes varied linearly with surface packice concentration over Lake Erie. Turbulent sensible

heat fluxes rapidly decreased as ice concentrations in-creased above about 70% and were nearly constant forlower ice concentrations. Latent heat fluxes tended todecrease linearly with increases in ice concentration,suggesting that latent heat fluxes did not respondstrongly to small breaks in high ice concentration areas.

Despite the general relationships between heat fluxesand the surface, there is a great deal of scatter in theobservations. This scatter might be interpreted as beingdue, in part, to random variations in turbulent eddiesresponsible for transporting heat and moisture verti-cally. However, some of the scatter likely reflects physi-cal processes linking the surface to atmospheric condi-tions at the 45-m-altitude aircraft observations. As de-scribed in Hechtel et al. (1990), variations in heat andmoisture transfers resulting from small-scale surfacefeatures would be expected to combine to create small-scale internal boundary layers that grow upward intothe boundary layer.

TABLE 2. Linear regression model parameters for flight stacksAB and CD derived from turbulent latent heat fluxes calculatedusing a flux-averaging scale of 45 s.

AB CD

Slope �0.027 �0.034Y intercept 2.78 5.26

FIG. 9. Best-fit linear regressions of latent heat flux as a function of ice concentration forflux-averaging scales ranging from 15 to 60 s. Flight stack (AB or CD) and flux-averaging scalefor each curve are indicated. In addition, R2 values are given for each averaging scale.


To gain insight into whether such small-scale internalboundary layers from the surface influence the obser-vations at 45-m height, Fig. 11 shows 1-Hz fluxes forareas of relatively cool and warm air temperatures mea-sured along flux leg CD2. Similar findings were seen for

CD3 and flux legs in stack AB (with the exception thatthe cooler area in stack AB was removed from consid-eration for this study because of low-level clouds). Inthe cooler region, 1-Hz sensible heat fluxes tended toexhibit peak values over and near regions of warmer

FIG. 10. Schematic representation of mesoscale regions of warmer and colder air in relationto surface land, water, and pack ice locations. Approximate locations of data used from flightstacks AB and CD are indicated. Surface imagery is from MODIS, as described in Fig. 1.

FIG. 11. One-second average fluxes (open boxes) and lake-surface temperatures (solid diamonds) observed during King Air pass CD2on 26 Feb 2004. From a (top) portion of the pass with relatively cool air temperatures and (bottom) warmer portion of the pass. Airtemperatures for the entire pass are shown in Fig. 3b. The 1-s average fluxes with the largest magnitudes are circled. One minute isapproximately equal to a 4.8-km flight distance.


surface temperatures (Fig. 11a). However, in thewarmer area this relationship was not as obvious (Fig.11c). We hypothesize that this is due to the locallylarger lake–air temperature differences over leads inthe cooler region as compared with the warmer region.Regardless of the reasons, these observations confirmdifferences in the relationship between sensible heatfluxes and the surface conditions in the cold, high iceconcentration region when compared with the warm,low ice concentration regions, as suggested by the non-linear relationships discussed in section 3c (and shownin Figs. 6, 7). Latent heat fluxes, on the other hand, didnot appear to be closely related to small-scale peaks insurface temperatures in either the cool or warm re-gions. This is consistent with earlier findings that latentheat fluxes did not respond as strongly as sensible heatfluxes to small-scale areas of warm surface temperature(interpreted as leads) in regions of high-concentrationpack ice.

While it is not possible to fully explain with thisdataset why sensible and latent heat fluxes vary differ-ently with surface ice concentrations, we speculate thatnearly the same amount of water vapor would be avail-able for vertical transport over water as over ice at airtemperatures close to 0°C. For example, saturation va-por 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 mois-ture between the lake and atmosphere over high con-centration ice may decrease the local responses in la-tent heat fluxes over small breaks in ice cover, resultingin more linear relationships between fluxes and packice concentration.

d. Implications for late-winter lake-effect snowprediction

The nonlinear relationship between sensible heatfluxes and surface ice concentrations appears to be theresult of mesoscale variations in both ice cover andatmospheric conditions along the flight legs, as well asthe upwind surface–atmosphere interactions that influ-enced atmospheric conditions where aircraft observa-tions were collected. If found to be a common feature,the nonlinear relationship would have important impli-cations for mesoscale operational forecast models. Forexample, Fig. 12 schematically compares the relation-ships found from the analysis of the 26 February 2004GLICAF measurements with the NAM Model’s repre-sentation of sensible heat fluxes as a function of iceconcentration, assuming perfect agreement at 0% iceconcentration. Currently, the NAM Model grid boxeswith less than 50% ice concentration are assigned sen-

sible heat fluxes representative of open water (i.e., nolake ice), while grid boxes with greater than 50% iceconcentration are assumed to be fully covered with 1-mthick ice (Meteorology Education & Training 2006).The observed nonlinear relationship suggests that thesensible heat fluxes integrated over the entire range ofice concentrations would be underestimated usingmethods similar to the NAM Model. In fact, this un-derestimate in sensible heat fluxes would not be greatlyimproved if a linear relationship between ice concen-tration and sensible heat fluxes were employed in me-soscale models.

5. Summary and future work

Analyses of data collected by the University of Wyo-ming King Air on 26 February 2004 during the GreatLakes Ice Cover–Atmospheric Flux experiment re-vealed strong links between atmospheric conditionsand characteristics of partially pack ice–covered LakeErie. Observed low-level temperatures were found tocorrelate more closely than mixing ratio to lake-surfaceconditions. The close thermal link was also observed insensible and latent heat flux relationships with surfaceice concentration. It was shown that minor mesoscalevariations in surface and atmospheric thermodynamiccharacteristics produced observable changes in heatfluxes.

Sensible heat fluxes, an important driving mechanismfor lake-effect snow storms, were found to vary nonlin-early with surface ice concentrations. Near-open-waterfluxes were observed for ice concentrations less thanabout 70%. This implies that if linear relationships areemployed in numerical weather prediction models, the

FIG. 12. Flux observations averaged over 45-s intervals in flightstack CD (gray dots), best-fit nonlinear curve (black, solid line),and schematic representation of fluxes in some NWP models(black, dotted line).


total heating of the atmosphere by lakes with consid-erable pack ice cover would be significantly underesti-mated.

It is critical to determine whether these observed iceconcentration–heat flux relationships are common tothe Great Lakes and to examine their potential influ-ence on mesoscale atmospheric responses. Collectionand analysis of field observations over a wide range ofatmospheric and surface pack ice conditions will givenecessary insight into both the statistical relationshipsand the processes responsible for these relationships.Given the observed links between atmospheric, lake,and pack ice processes, interdisciplinary observationalfield experiments and coupled modeling approachesare needed to fully understand the interactions.

Acknowledgments. The authors greatly appreciatethe efforts of the staff and crew of the University ofWyoming King Air. Forecasting for the GLICAF ex-periment was carried out by lead forecasters MichaelKruk and Michael Spinar from the Illinois State WaterSurvey, the authors, and Stephen Jackman from theDepartment of Atmospheric Sciences, University of Il-linois. Project assistance from Yarice Rodriguez, De-partment of Geography, University of Illinois, is alsoappreciated. Internal reviews of the manuscript wereconducted by Jim Angel and Kenneth Kunkel from theIllinois State Water Survey. Reviews by three anony-mous reviewers are also appreciated. GLICAF andanalyses described here were funded by the NationalScience Foundation (NSF 02-02305 and NSF 05-12954).Any opinions, findings, and conclusions or recommen-dations expressed in this material are those of the au-thors and do not necessarily reflect the views of theNational Science Foundation or the Illinois State WaterSurvey.

REFERENCES

Alam, A., and J. A. Curry, 1995: Lead-induced atmospheric cir-culations. J. Geophys. Res., 100, 4643–4651.

——, and ——, 1997: Determination of surface turbulent fluxesover leads in Arctic sea ice. J. Geophys. Res., 102, 3331–3343.

Andreas, E. L, and B. A. Cash, 1999: Convective heat transferover wintertime leads and polynyas. J. Geophys. Res., 104,25 721–25 734.

——, C. A. Paulson, R. M. Williams, R. W. Lindsay, and J. A.Businger, 1979: The turbulent heat flux from Arctic leads.Bound.-Layer Meteor., 17, 57–91.

Angel, J. R., and S. A. Isard, 1997: An observational study of theinfluence of the Great Lakes on the speed and intensity ofpassing cyclones. Mon. Wea. Rev., 125, 2228–2237.

Assel, R. A., 1999: Great Lakes ice cover. Potential ClimateChange Effects on Great Lakes Hydrodynamics and WaterQuality, D. C. L. Lam and W. M. Schertzer, Eds., AmericanSociety of Civil Engineers, 1–21.

Chang, S. S., and R. R. Braham Jr., 1991: Observational study ofa convective internal boundary layer over Lake Michigan. J.Atmos. Sci., 48, 2265–2279.

Cooper, K. A., M. R. Hjelmfelt, R. G. Derickson, D. A. R. Kris-tovich, and N. F. Laird, 2000: Numerical simulation of tran-sitions in boundary layer convective structures in a lake-effect snow event. Mon. Wea. Rev., 128, 3283–3295.

Cordeira, J. M., and N. F. Laird, 2005: The impact of ice cover ontwo lake-effect snow events in the eastern Great Lakes re-gion. Preprints, 11th Conf. on Mesoscale Processes, Albu-querque, NM, Amer. Meteor. Soc., P3M.4.

Garratt, J. R., 1992: The Atmospheric Boundary Layer. Cam-bridge University Press, 316 pp.

Gerbush, M. R., 2005: Surface heat fluxes over Lake Erie pack icefields. M.S. thesis, Department of Atmospheric Sciences,University of Illinois at Urbana–Champaign, 145 pp.

Hechtel, L. M., C.-H. Moeng, and R. B. Stull, 1990: The effects ofnonhomogeneous surface fluxes on the convective boundarylayer: A case study using large-eddy simulation. J. Atmos.Sci., 47, 1721–1741.

Kelly, R. D., 1984: Horizontal roll and boundary-layer interrela-tionships observed over Lake Michigan. J. Atmos. Sci., 41,1816–1826.

Krahmann, G., and Coauthors, 2003: The Laborador Sea DeepConvection Experiment data collection. Geochem. Geophys.Geosyst., 4, 1091, doi:10.1029/2003GC000536.

Kristovich, D. A. R., 1993: Mean circulations of boundary-layerrolls in lake-effect snow storms. Bound.-Layer Meteor., 63,293–315.

——, and R. R. Braham Jr., 1998: Profiles of moisture fluxes insnow-filled boundary layers. Bound.-Layer Meteor., 87, 195–215.

——, and N. F. Laird, 1998: Observations of widespread lake-effect cloudiness: Influences of lake surface temperatures andupwind conditions. Wea. Forecasting, 13, 811–821.

——, ——, M. R. Hjelmfelt, R. G. Derickson, and K. A. Cooper,1999: Transitions in boundary layer meso-� convective struc-tures: An observational case study. Mon. Wea. Rev., 127,2895–2909.

——, and Coauthors, 2000: The Lake-Induced Convection Ex-periment and the Snowband Dynamics Project. Bull. Amer.Meteor. Soc., 81, 519–542.

——, M. C. Peters, and M. S. Timlin, 2001: Spatial variations ofsurface temperatures and heat fluxes over Lake Michigan.Preprints, 11th Conf. on Interaction of the Sea and Atmo-sphere, San Diego, CA, Amer. Meteor. Soc., 142–145.

——, N. F. Laird, and M. R. Hjelmfelt, 2003: Convective evolu-tion across Lake Michigan in a lake-effect snow event. Mon.Wea. Rev., 131, 643–655.

Laird, N. F., and D. A. R. Kristovich, 2002: Variations of sensibleand latent heat fluxes from a Great Lakes buoy and associ-ated synoptic weather patterns. J. Hydrometeor, 3, 3–12.

——, and ——, 2004: Comparison of observations with idealizedmodel results for a method to resolve winter lake-effect me-soscale morphology. Mon. Wea. Rev., 132, 1093–1103.

Laursen, K., cited 2005: Passive broadband and spectral radiomet-ric measurements available on NSF/NCAR research aircraft.NCAR Research Aviation Facility (RAF) Bulletin 25.[Available online at http://www.eol.ucar.edu/raf/Bulletins/bulletin25.html.]

Lenschow, D. H., 1973: Two examples of planetary boundarylayer modification over the Great Lakes. J. Atmos. Sci., 30,568–581.


Mayor, S. D., G. J. Tripoli, and E. W. Eloranta, 2003: Evaluatinglarge-eddy simulations using volume imaging lidar data. Mon.Wea. Rev., 131, 1428–1452.

Meteorology Education & Training, cited 2006: Operational mod-els matrix: Characteristics of operational NWP models.[Available online at http://meted.ucar.edu/nwp/pcu2/index.htm.]

Miles, N. L., and J. Verlinde, 2005a: Observations of transientlinear organization and nonlinear scale interactions in lake-effect clouds. Part I: Transient linear organization. Mon.Wea. Rev., 133, 677–691.

——, and ——, 2005b: Observations of transient linear organiza-tion and nonlinear scale interactions in lake-effect clouds.Part II: Nonlinear scale interactions. Mon. Wea. Rev., 133,692–706.

Motulsky, H. J., and A. Christopoulos, 2003: Fitting models tobiological data using linear and nonlinear regression. Graph-Pad Prism version 4.0 user’s guide, 351 pp. [Availableonline at http://www.graphpad.com/manuals/prism4/RegressionBook.pdf.]

Niziol, T. A., 1987: Operational forecasting of lake effect snowfallin western and central New York. Wea. Forecasting, 2, 310–321.

Pinto, J. O., J. A. Curry, and K. L. McInnes, 1995: Atmosphericconvective plumes emanating from leads 1. Thermodynamicstructure. J. Geophys. Res., 100, 4621–4631.

Renfrew, I. A., and G. W. K. Moore, 1999: An extreme cold-airoutbreak over the Labrador Sea: Roll vortices and air–seainteraction. Mon. Wea. Rev., 127, 2379–2394.

Rouse, W. R., C. M. Oswald, J. Binyamin, P. D. Blanken, W. M.Schertzer, and C. Spence, 2003: Interannual and seasonalvariability of the surface energy balance and temperature ofcentral Great Slave Lake. J. Hydrometeor., 4, 720–730.

Schroeder, J. J., D. A. R. Kristovich, and M. R. Hjelmfelt, 2006:Boundary layer and microphysical influences of natural cloudseeding on a lake-effect snow storm. Mon. Wea. Rev., 134,1842–1858.

Sousounis, P. J., 1997: Lake-aggregate mesoscale disturbances.Part III: Description of a mesoscale aggregate vortex. Mon.Wea. Rev., 125, 1111–1134.

——, 1998: Lake-aggregate mesoscale disturbances. Part IV: De-velopment of a mesoscale aggregate vortex. Mon. Wea. Rev.,126, 3169–3188.

——, and J. M. Fritsch, 1994: Lake-aggregate mesoscale distur-bances. Part II: A case study of the effects on regional andsynoptic-scale weather systems. Bull. Amer. Meteor. Soc., 75,1793–1811.

Stull, R. B., 1988: An Introduction to Boundary Layer Meteorol-ogy. Kluwer Academic, 666 pp.

Sun, J., J. F. Howell, S. K. Esbensen, L. Mahrt, C. M. Greb, R.Grossman, and M. A. LeMone, 1996: Scale dependence ofair–sea fluxes over the western equatorial Pacific. J. Atmos.Sci., 53, 2997–3012.

University of Wyoming, cited 2005: University of Wyoming KingAir variable list. [Available online at http://flights.uwyo.edu/ka_doc/index.html.]

Weiss, C. C., and P. J. Sousounis, 1999: A climatology of collectivelake disturbances. Mon. Wea. Rev., 127, 565–574.

Young, G. S., B. K. Cameron, and E. E. Hebble, 2000: Observa-tions of the entrainment zone in a rapidly entraining bound-ary layer. J. Atmos. Sci., 57, 3145–3160.

Zulauf, M. A., and S. K. Krueger, 2003: Two-dimensional cloud-resolving modeling of the atmospheric effects of Arctic leadsbased upon midwinter conditions at the surface heat budgetof the Arctic Ocean ice camp. J. Geophys. Res., 108, 4312,doi:10.1029/2002JD002643.


mesoscale boundary layer and heat flux variations over pack ice–covered lake...

Documents