Figures and Tables not currently available for the Web.

ABSTRACT

Despite recent advances in the classification of synoptic-scale events, there remains the need for development of a simple, automated, continental-scale air mass-based procedure. We present a new method of analysis to identify resident air masses at first-order weather stations to facilitate inter-site comparison of daily air masses across a very large region.

The "spatial synoptic classification" (SSC) requires initial identification of the major air masses and their typical meteorological characteristics at each site. "Seed days", which exhibit these typical characteristics for each air mass, are used as input for a linear discriminant function analysis, which produces a daily categorization of air masses with spatially continuous results. A second discriminant function analysis is used to determine whether a day is to be considered transitional between air masses.

Using the SSC, air mass frequencies were developed for winter across the United States east of the Rockies, and rates of modification were determined as each air mass traversed the region. The impact of snow cover on modification rates was determined by comparing two winter months with very different snow cover characteristics across the area. Keeping other factors constant, it appears that extended snow cover can reduce the temperature of a dry polar air mass by up to 11øC.

Synoptic approaches can generally be subdivided by the type of climatological phenomena evaluated, as well as the spatial applicability of each approach (Table 1). "Weather typing" defines synoptic groups by pressure or wind fields (Davis and Walker, 1992); resulting categories represent distinct flow regimes which influence pollution concentration, oceanic wave patterns, and a multitude of other environmental factors. Weather type approaches are most useful when the evaluation requires detail pertaining to atmospheric transport mechanisms; thus, pesticide spray drift problems, pollution dispersal, and atmospherically-forced insect migration patterns are conveniently analyzed with the use of weather types.

Although homogeneous from a hydrodynamic standpoint, weather types are less useful when the investigation dictates that synoptic groups should be thermodynamically homogeneous. Certainly, if surface and upper level flow patterns are similar on two particular days, it is likely that the thermal and moisture characteristics of those days would also be similar. However, there are numerous instances when this is not the case. For example, an anticyclone located to the west of Memphis, Tennessee in winter, supported by negative vorticity advection aloft, could result in various meteorological scenarios at the surface. First, cold, dry air could be transported to the region, and skies could be clear with afternoon temperatures hovering near freezing. Second, modified Pacific air, warmed adiabatically on the lee side of the Rockies, might be in place, and afternoon temperatures could approach 15øC. Finally, the rare intrusion of dry, warm air from the desert Southwest is possible, with associated afternoon temperatures well above 20øC. Thus, for bioclimatological applications where thermal and moisture characteristics of the air are of particular importance, weather typing may prove insufficient as the resulting synoptic categories are not necessarily homogeneous.

Air mass-based approaches are designed to categorize days based on a variety of weather elements, and the resulting classification is less dependent on pressure patterns and wind fields. An air mass is a large volume of air which has acquired, "...characteristics of temperature and humidity related to the condition of the sea, land, or ice beneath it" (Crowe, 1971). Various thermal and moisture variables are commonly used to classify air masses, along with visibility, cloud cover, and sometimes precipitation. Thus, air masses can be defined by their distinctive thermodynamic character. Since the criteria for categorization is based on similarities in thermal and moisture characteristics, it is conceivable that pressure and wind could vary considerably among days within an air mass.

This definition varies somewhat from the classic description put forth by Bergeron (1930), which defines polar, tropical, continental, and maritime air masses according to their source regions. A similar definition was adopted by Schwartz (1991; 1995), whose methodology followed a "numerical limits with transition zones" definition of air masses. Source region delineation is less important here; instead the concern is the identification of "umbrellas of air" which traverse a region and possess distinctive meteorological characteristics. Environmental responses to air masses are not based on source region; rather, response is most frequently dependent on the meteorological character of the air at a place in time. Periods of change in meteorological character which occur when one air mass supplants another are considered to be "transition situations".

Much like weather typing, air mass-based approaches are used widely in environmental analysis. They are best-suited, but not limited to, bioclimatological problems such as phenological analyses (Schwartz and Marotz, 1986), as well as other instances where organisms respond to the character of the atmosphere rather than to pressure or wind patterns.

Synoptic classifications can also be defined by their spatial applicability (Table 1). Some are generally used as point indices, such as Muller's (1977) weather types and the air mass-based temporal synoptic index (TSI). In both of these cases, it is possible, but unwieldy, to expand use to the larger scale. Muller's weather typing evaluates daily map patterns at each site, requiring considerable time and effort. TSI is virtually impossible to develop beyond single point analysis, as the statistical procedures are not sensitive to similar categorizations developed at adjacent locations. Several regional synoptic approaches have been successfully developed, and these are most useful when describing weather types or air masses within an area up to a few thousand kilometers squared. For example, Schwartz (1991) devised an approach to evaluate January, April, July and October air mass frequencies at 15 sites in the north central U.S from 1958-1981. He successfully identified five air masses using procedures similar to the "partial collective method" proposed by Bryson (1966), and calculated air mass frequencies for the region. Although useful at a regional level, many of these procedures are less efficient on a continental scale.

Some progress has been achieved in the development of continental-scale weather typing schemes, but air mass-based procedures at this large scale have been an enigma to atmospheric scientists. Davis and Kalkstein (1990a), applying procedures similar to those of the TSI, developed a continental-scale spatial synoptic index using 1984 surface weather data for the United States. The resulting daily maps identified regions over which distinct air masses were present. This classification was used to analyze air quality variations; however, the procedure suffers from two important shortcomings. First, the technique must be developed on a day-to-day basis, which can be exceedingly time-consuming when treating a dataset of many years. Second, there were considerable problems tracking a given air mass across the country, especially if modification occurred during transport. Third, the procedure produced an unwieldy number of synoptic types which varied both seasonally and regionally.

Thus, despite recent advances in the classification of synoptic-scale events, there remains the need for development of a simpler, automated, large-scale air mass-based procedure. We propose a new method of analysis to identify resident air masses at hundreds of weather stations on a national scale. The proposed "spatial synoptic classification", or SSC, is less complex than previous procedures of this type, yet permits inter-site comparison of daily air mass types across a very large region. The resulting spatial air mass inventory could have important environmental applications, such as the determination of climatic limiting factors for certain biota and biotic communities, the rate of air mass modification across large areas, the impact of temporary surface features, such as snow, on modification rates, a large scale evaluation of climatic change based on variations in air mass frequencies and character, and an evaluation of precipitation characteristics for individual air masses. This paper will describe results derived from the SSC for 126 first-order weather stations east of the Rocky Mountains in the U.S. during a 30-year period (1961 through 1990) for winter (December, January, and February). Stations west of the Rockies have also been evaluated using the SSC, but the complex terrain has rendered trends in air mass frequency and character more difficult to discern. Thus, this initial paper will be limited to the eastern two-thirds of the U.S. Although discussion here is confined to winter, the procedure is designed to classify air masses year-round. 

a. Air mass identification
Unlike many existing air mass-based classification techniques, the SSC requires initial identification of the major air masses which traverse the region as well as their typical meteorological characteristics. The SSC categorizes days within the following six air mass types: dry polar (DP), moist polar (MP), dry temperate (DM), moist temperate (MM), dry tropical (DT), and moist tropical (MT). The nomenclature is designed to identify the character, rather than source region, of the air mass. Dry polar air is synonymous to continental polar; it is the coldest, and sometimes the driest, air mass in a region. Cloud cover is most often minimal. Dry temperate air is typically an adiabatically-warmed Pacific air mass which has descended the lee side of the Rockies. The air mass is associated with mild, dry conditions in the eastern and midwestern U.S., and most frequently intrudes when strong zonal flow exists aloft. Dry tropical defines the hot and very dry air mass which most often originates from the Desert Southwest or northern Mexico. It is frequently associated with the hottest and driest conditions, especially in the Midwest. Moist polar air is cool and humid, with overcast conditions, and frequently, easterly winds. In the East, this air mass is synonymous to maritime polar conditions around the northern flank of a mid-latitude cyclone. However, it is also common that overrunning, associated with a slow-moving front sometimes hundreds of kilometers south, produces virtually identical surface conditions. Thus, the two synoptic situations are grouped together. Moist temperate is also associated with overcast, humid conditions, but temperatures and dew points are much higher owing to the close proximity of the responsible front. Moist polar and moist temperate air masses may persist for many days over a locale if frontal movement is particularly lethargic. Moist tropical air masses, commonly recognized as maritime tropical, represent warm, humid conditions found frequently in the warm sector of an open wave cyclone or the western flank of a subtropical anticyclone. Atmospheric instability and convective activity are common within the air mass.

b. Selection of a classification procedure
Using the spatial synoptic classification proposed here, all days can be placed within one of these predetermined, readily identifiable, air mass categories. There are numerous means to classify meteorologically homogeneous days in an automated manner, but three general approaches seem most efficient.

First, hierarchical, agglomerative clustering procedures have been utilized successfully to categorize meteorologically homogeneous days (Sokal and Michener, 1958; Romesberg, 1984; Kalkstein et al., 1987; Schwartz and Skeeter, 1994). Clustering techniques, such as average linkage, are best-suited when the number of categories to be developed is not predetermined. Grouping is based on some measure of similarity between pairs of objects (days), and guidelines are provided to terminate clustering when dissimilar days are forced into the same group (Anderberg, 1973; Fovell and Fovell, 1993). As air mass categories are predetermined, hierarchical agglomerative procedures are not suitable here.

Second, non-hierarchical iterative clustering procedures, such as the frequently-utilized convergent k-means method, allow for the rearrangement of objects (days) after they have been classified into a group, thereby optimizing the final classification (Davis and Walker, 1992). In addition, such procedures select initial starting values, or seeds, to represent mean conditions for each predetermined category (air mass). Each object (day) is then assigned to its nearest category, based on its distance from mean "seed day" values. The process is iterative and new seed day means are computed as category membership is updated. The procedure terminates after that iteration which produces no new reassignment (Davis and Kalkstein, 1990b).

Although non-hierarchical procedures are advantageous because of their use of predetermined seed days, the recalculation of seed day means with each iteration represents a distinct shortcoming when it is desired that seeds be selected based on prior climatological knowledge about air masses. In the spatial synoptic classification described here, climatological expertise plays a significant role in determining air mass types and their typical representation at a particular locale through careful selection of seed days. Thus, the redevelopment of new group centroids in the k-means method as defined by the recalculation of seed day means within each iterative step creates an environment where the procedure is driven less on climatological expertise and more on mathematical rigor. Since the mathematical recalculation is performed on a site-by-site basis (totally independent of conditions at a neighboring locale), the resulting meteorological character of the seed days after the final iteration has lost a significant degree of spatial continuity. A third approach, discriminant function analysis, is appropriate when group structure is predetermined and recalculation of seed days is deemed undesirable (James, 1985). The term "discriminant analysis" refers to a wide range of statistical procedures which are designed to measure the differences between two or more groups of objects with respect to one or more variables simultaneously. The principal objective is the assignment of new objects to predetermined groups using developed classification rules. These rules, called discriminant functions, are calculated and used to identify the group to which an object belongs. Linear discriminant function analysis, used here, assumes multivariate normality and equal covariance matrices within and among groups (Klecka, 1980). Statistical evaluations of linear discriminant function analysis indicate that it is "quite satisfactory", even when the assumption of equal covariances is relaxed (Gilbert, 1969; Marks and Dunn, 1974). In addition, analyses which use bimodal variables (e.g., cloud cover) are more likely to violate the assumption of normality; however, this can occur without significantly affecting the results of the classification (Klecka, 1980). Thus, in the SSC, where groups (air masses) are predetermined and represented by seed days, linear discriminant analysis is a robust procedure which produces a daily categorization with spatially continuous results (refer to Miller, 1962 for a more complete description of the meteorological applications of discriminant analysis). This type of discriminant analysis is conceptually similar (but more mathematically rigorous) to the "partial collectives" approach devised by Bryson (1966).

Using the covariance matrix and mean values of the variables selected, discriminant analysis develops classification functions, which in turn are used to identify which group best fits the characteristics of an individual day. The discriminant analysis is based upon the development of a set of linear equations described as follows (based on Lachenbruch, 1975 and Klecka, 1980):

where hk is the value (score) of the discriminant function for group k, Xi is the value of the discriminating variable i (i.e., temperature, dew point, etc.) up to p number of variables, and bki represents coefficients which modify the function so that it closely resembles the true group variability. The functions are evaluated for each group for each case. These coefficients are derived so that the group scores are as different as possible. The coefficients for the discriminant classification functions are derived as follows:

where bki is the coefficient for variable i in the equation corresponding to group k, Xjk is the value of the discriminating variable, a*ij is an element from the inverse of the covariance matrix (A), and nt is the total number of cases over all groups (g). The elements of A are defined as:

where g is the number of groups, nk is the number of elements in group k, ik is the mean of values in the kth group, and Xikm is the value of variable i for case m in group k. The inverse of this matrix is then computed to determine the a*ij values. The constant term in equation (1), bk0, is defined as:

A separate discriminant function is derived for each group and evaluated for each day. The day is then classified into the group with the highest score (i.e., the largest hk).

c. Development of the procedure
The foundation for the development of the SSC is the proper selection of seed days (Figure 1). These days represent the typical meteorological character of each air mass at a location and are used to classify all other days. A number of seed days are used to develop a more robust sample of the typical character of each air mass at each location.

The seed day group for each air mass is selected by a specification of ranges in afternoon surface temperature, dew point, dew point depression, wind speed, wind direction, cloud cover, and diurnal temperature range. Afternoon observations are emphasized as it is during this time that air mass distinction is especially clear, particularly for surface temperature and dew point depression. In addition, four six-hourly dew point changes are evaluated for systematic trends to prevent those days which have undergone significant changes (such as frontal passages) from being selected as seed days.

Initial estimates of criteria for seed day selection are specified for each air mass at each location. This is accomplished through careful evaluation of surface meteorological data and maps from 1961-90; days which meet these criteria are selected to represent that air mass. The criteria are developed so that all seeds selected comprise a homogeneous representation of the air mass. For example, to be included in the Cleveland DP seed day pool, a seed day must possess, among other attributes, a 3 PM air temperature below -1øC, a dew point below -8øC, an afternoon cloud cover of less than five-tenths, a diurnal temperature range of at least 7øC, and a dew point depression of at least 7øC (Table 2). Every effort is made to maximize the number of seed days for each air mass (in most cases, at least 30 seed days are selected) without jeopardizing group homogeneity. However, when an air mass is rare at a locale, a lesser number of days must be designated.

An important aspect of the procedure is that the criteria for seed selection process can be changed if the resulting seed days for each air mass are deemed non-representative. For example, the maximum afternoon temperature can be raised if the days selected are too cold to represent the air mass at the particular location. This is determined by comparing seed days with those selected at adjacent locales and by adjusting for local climatic factors. Seed day attributes should not vary dramatically between adjacent sties, and criteria selection is based partially on the spatial consistency of seed days.

Following seed day selection, discriminant function analysis is used to generate a linear function for each air mass from its group of seed days (Figure 2). Each day is subsequently evaluated using all air mass discriminant functions to determine which group it most closely resembles. The result of this evaluation is a calendar which lists the air mass to which each day has been assigned.

The sensitivity of the procedure was tested by examining the effects of variations in the selection of seed days and in the choice of the variables used in the discrimination. Thus, variations in seed selection criteria, which resulted in different numbers of seed days, were analyzed at 15 sample cities. Two sets of seed days, one consisting of a narrow range of criteria with a set of only a few seed days ("A" seeds) and the other, a larger, less restrictive group with approximately double the number of seed days ("B" seeds), were compared. Results showed that the frequencies and character of the air masses varied little between the "A" seed and "B" seed discrimination (Table 3). For example, the overall frequency of winter DP varied by less than 1% in Bismarck. One of the largest differentials was at Bismarck for MP days; frequencies differed by 3.5% using the two different sets of seeds. However, this represented one of the largest differentials uncovered at all 15 tested cities. In almost all cases, variations using different combinations of seeds were less than 2 percent.

Another important consideration is the selection of variables used in the final determination of the discriminant functions. Over a dozen combinations of meteorological elements were evaluated to determine which would be most effective in distinguishing between air masses. The number of variables studied ranged from 5 to 21 (Table 3). For example, the 5-variable procedure included daily means of temperature, dew point, sea level pressure, and cloud cover, along with diurnal temperature range, while the 21-variable procedure included four-times-daily values for temperature, dew point, sea level pressure, a north-south wind scalar, an east-west wind scalar, and mean cloud cover. While there are some changes in the final frequencies of the different air masses, the differences are negligible. For example, the frequencies at Bismarck for DP air vary by less than 1%, and the values for St. Louis vary by only 1.5% among the different procedures tested. The resulting change in character is less than 0.25øC for afternoon temperature and dew point at Bismarck. Similar results are found for the other air masses as well as other locations.

Days which were assigned to different groups by the various element combinations were evaluated through surface weather map analysis. This evaluation indicated that a 12-variable procedure best represents the synoptic character of each day. The 12 variables are: six-hourly temperature and dew point, mean daily cloud cover and sea level pressure, and diurnal air temperature and dew point range. As thermal and moisture characteristics are important indicators for a particular air mass, the four-times-daily measurements of temperature and dew point are retained. However, cloud cover and sea level pressure are more conservative variables, so mean daily values are used for these variables. In addition, a significant component of the character of air masses is the diurnal nature of their temperature and moisture characteristics. Diurnal temperature range and dew point range are therefore included in the final set of variables.

A significant number of days are initially incorrectly classified because they represent transition situations. To account for this, the SSC performs a second discriminant function analysis to determine whether or not a day represents a transition situation. Two groups of seed days are used for this analysis: a group representing transition situations, and a group containing the combined seed days of all air masses at the location. The former group is selected using processes similar to the original seed identification; however, the selection variables differ. Two particularly conservative elements, diurnal dew point change and sea level pressure change, are evaluated for systematic trends through the day. If a significant diurnal change is observed for either element (at least twice the average diurnal change of all days), the day is used as a transition seed day.

This discriminant analysis produces a second calendar which designates each day as either transitional or non-transitional. All days classified as non-transitional retain their air mass designation assigned in the first discriminant analysis. All other days are classified as transitions from the air mass of the previous day to that of the present day, with two exceptions. If a transition day is assigned to the same air mass as that of the previous day in the first discriminant analysis, it is classified as a transition from the present day's air mass to that of the following day. When all three days (previous, present, and following) are classified as the same air mass in the first discriminant analysis, the present day is treated as non-transitional.

Seed day selection
Seed days were selected for all air masses at all locales based on predetermined criteria. Seed days are quite homogeneous within sites, but their meteorological character varies considerably among sites (Tables 4 and 5). For example, the 50 seed days from the 1961-90 time period selected for St. Louis DP air possess similar thermal and moisture characteristics, with minimal cloud cover and generally north winds. However, while early afternoon DP seed day temperatures at Bismarck average about -15øC, at St. Louis and New Orleans they are about 8 and 19øC higher, respectively. DP seed day mean afternoon dew points exhibit even more spatial variation; the St. Louis and New Orleans dew points are 18 and 28øC higher than those at Bismarck.

Not all air masses demonstrate this magnitude of spatial variability in seed day meteorological character. For example, DT seed day temperature means are much more similar at St. Louis and New Orleans, although the air mass occurs very infrequently at St. Louis in winter. No seeds were determined for either DT or MT air at Bismarck; hence, it was concluded that neither of these air masses reached Bismarck during the evaluated winters.

Air mass frequencies
Air mass frequency maps illustrate spatial trends in the occurrence of each air mass (Figure 3). The winter frequency of MT air decreases rapidly moving northward from Florida. While approximately one-half of all days are categorized as MT at the southern tip of the state, this frequency diminishes to approximately one-third near Daytona Beach. The gradient becomes less steep to the north and west of Florida, and northern Gulf Coast cities such as New Orleans and Houston average about 18 percent MT days. North of a line from Washington to St. Louis, frequencies are below 5 percent, and almost no MT air intrudes as far north as Buffalo and Boston. During the period of record, MT air masses never occurred in Flint, Chicago, Des Moines, Lincoln, or any points north of these locations.

The gradient of DP frequency is generally oriented in a north-south direction, but changes are less uniform than MT. DP frequencies range from less than 8 percent of winter days in southern Florida and Texas to greater than 26 percent in North Dakota and along the northeastern coast. Interestingly, DP frequency is higher in Norfolk, VA than in Minneapolis. However, a much larger proportion of days is classified as MP in Minneapolis, and if the DP and MP frequencies are combined to develop a "polar air mass index" (Figure 4), a more intuitive pattern becomes apparent. Cities within several hundred kilometers of the Great Lakes experience a reduced number of DP days and a comparatively high number of MP days. The greater cloud cover and concomitant low dew point depressions in this area increase the MP frequency at the expense of DP. However, along the mid-Atlantic coast, DP occurs with greater frequency as a consequence of downslope drying of polar air masses east of the Appalachians. The polar index demonstrates a steep latitudinal gradient across the country, and the 50 percent polar air mass line traverses southern Virginia, central Kentucky, and southern Illinois. It moves northwestward into southern Nebraska, due to an increase in DM air mass frequency toward the west. Downslope motion of air masses traversing the Rockies during zonal flow increases the frequency of DM air in the Great Plains.

About a quarter of all days in central and western Nebraska are categorized as DM, while this number diminishes to less than 15 percent at the same latitude further east. A strong southwest-to-northeast frequency gradient is apparent, and maxima are noted in the southern Great Plains, where polar air is less likely to intrude when zonal flow predominates aloft. DM air masses are relatively common over the southeastern U.S., where the frequency gradient takes on a south-north orientation. A southwest-to-northeast gradient is noted for DT air mass frequencies as well, but magnitudes are much lower than DM in winter. However, in western Texas, almost half of all winter days experience one of these two air masses.

The MM air mass demonstrates the most complex pattern, and a continuous gradient is less apparent. There is an increasing frequency toward the Gulf Coast, related to slow-moving fronts which often align themselves in the northern Gulf of Mexico during winter. This results in a pronounced overrunning situation with easterly surface flow, cloudiness, high precipitation probability, and comparatively mild temperatures along the Gulf Coast. The frequency of MM air diminishes in the western Great Plains. An axis of somewhat lower frequencies extends from Kansas to coastal Virginia and North Carolina, and an increase is noted toward the Canadian border. An intraseasonal frequency evaluation reveals that MM air is disproportionately common at the northern stations very early and very late in the winter season. Storm tracks are generally aligned further north at this time (as compared to their location during the middle of winter), permitting more frequent incursions of MM air near the Canadian border.

Frequencies of transition situations are remarkable for their spatial consistency (Figure 4). Transition frequencies average about 15 percent, with slightly lower values across the western Great Plains and slightly higher values in eastern North Carolina and Virginia. However, there is no discernable spatial trend in transition frequencies.

Air mass character
Mean meteorological characteristics were evaluated for the six air masses, and the resulting spatial variations indicate rates of air mass modification (DP, MT, and DM are illustrated in Figures 5a, 5b, and 5c, respectively). Modification with increasing distance from the source region occurs most rapidly within the DP air mass. For example, mean 3PM DP temperatures exhibit a very steep north-south gradient, varying from over 10øC in southern Florida to about -16øC in northern Minnesota. The most rapid modification appears to occur in the Midwest, where the afternoon temperature increases by about 2øC for every 100 km distance from the source region. This modification rate is twice that noted in the southern and extreme northern U.S. The rapid modification of DP air in the Midwest may be associated with the reduced frequency of snow cover which occurs southward in winter across the region. North Dakota, Minnesota, northern Wisconsin, and Michigan typically have continuous snow cover throughout much of the winter (Robinson and Hughes, 1991), and the homogeneous surface reduces DP air mass modification over this area. Conversely, continuous snow cover is a very rare event in the Southeast, which inhibits the modification rate there as well. However, the percentage of winter days with continuous snow cover decreases very rapidly southward from northern Nebraska and Iowa to northern Oklahoma and Arkansas (Leathers and Robinson, 1993), creating a non-homogeneous surface and potentially producing the greatest rate of modification.

The rates of modification of MT and DM air masses are somewhat less than DP. This is especially true for MT air in the Southeast, where mean 3 PM temperatures change at about 0.5øC for each 100 km. Although modification rates increase farther north (about 1øC for every 100 km; possibly due to increasing snow cover frequency), the gradient for MT is about half that for DP.

Spatial variations in dew points within the air masses are somewhat different than air temperature (Figure 6). For example, the DP dew point temperature gradient is less steep than that of air temperature; the range in the former from northern Minnesota to the Gulf Coast is about 17øC, while for the latter it is about 22øC. In addition, the DP dew point gradient is steeper in the northern U.S., especially in the Midwest, where snow cover is usually continuous in winter. This suggests that moisture ablates from snow-covered areas into the dry air, but as the air mass moves south over surfaces not covered with snow, less moisture is added from the surface. The MT dew point gradient is slightly greater than DP across the southern and central U.S., a pattern converse to the trend exhibited by air temperature. The gradient is steepest over Texas and Oklahoma, and this unstable air mass apparently loses moisture rapidly as it passes over a relatively dry surface. There is a pronounced southwest-northeast trend in the isodrosotherms; Oklahoma City and New York City possess similar mean MT dew point temperatures.

The DM gradient in dew point temperature is weaker than that of the previous two air masses. The mean DM dew point temperature is only about 7øC lower in northern Minnesota than along the Gulf Coast. There is an interesting north-south orientation of isolines within the Great Plains, which probably represents the constant moisture characteristics of this air mass as it descends the lee side of the Rockies. Thus, modification in this region occurs as the air mass moves eastward. Another interesting feature is the consistency in DM dew point temperature from northern Illinois and Indiana to northern Mississippi and Alabama. Little modification in DM moisture content is apparent here as distance from the source region increases. In addition, the dew point temperature gradient is particularly weak from North Carolina to northern New England. The difference in mean DM dew point temperature from Boston to Wilmington, NC is only about 3øC.

The mean diurnal temperature range of the air masses at selected East Coast locations (Portland, ME; New York, NY; Richmond, VA; Wilmington, NC; Charleston, SC; Jacksonville, FL; Miami, FL) is fairly consistent within air mass type (Figure 7). The greatest diurnal range is noted for DM air, and the mean minimum temperature for this air mass is less than 5øC for all the East Coast stations with the exception of Miami. For most of the locales, the diurnal temperature range is greater than 10øC. This is somewhat diminished at the two northernmost sites, where reduced insolation may be a contributing factor. The smallest diurnal ranges are noted for the two air masses most commonly associated with overcast conditions, MP and MM. New York City's diurnal range appears reduced for most of the air masses, especially the driest and clearest ones, possibly due to lowered atmospheric transparency related to urban impacts. Thus, it is possible that the SSC permits a quantitative evaluation of urban effects upon the thermal and moisture characteristics of each air mass type.

A Potential Application: The Impact of Snow Cover
In 1963, Namias pointed out that snow cover played, "...a vital role..." in affecting paths of cold polar anticyclones, as well as mid-latitude cyclones. Thus, a positive feedback mechanism existed where temperature distribution was significantly impacted by the extent of snow across the continent.

The impact of snow cover on the differential modification rates of various air masses is of similar interest, and the SSC permits a detailed examination of modification, as well as the impact of different surfaces on such modification. Thus, two winter months with distinctly different snow cover extents were evaluated to determine how differential surface conditions might influence air mass characteristics.

January, 1981 was noted for its general lack of snow cover across the central U.S., while January, 1979 was associated with virtually continuous snow cover north of Tennessee, central Arkansas and Oklahoma (Figure 8). Mean afternoon temperatures for each air mass were compared for these two months, and the largest differential was detected for DP air. During January, 1979, mean DP afternoon temperatures were considerably lower than those in January, 1981. However, the rate of DP modification for both months appeared to be closely related to snow cover extent. For example, the gradient of modification during January, 1981 was steepest through central North and South Dakota, northeastern Nebraska, and Iowa, which corresponds closely to the southern edge of snow cover for the month. During January, 1979, the steepest gradient was detected much further south (northern Arkansas, Kentucky, West Virginia, and eastern Pennsylvania), but remained near the southern edge of snow cover.

An evaluation of DP mean afternoon temperature differences between the two months clearly indicates that the differential was greatest within the zone where snow cover occurred in 1979 but was absent in 1981. In central Minnesota, where snow cover was present during both months, the temperature differential averaged about 5øC. A similar difference is noted in the south central U.S., where snow cover was absent for both months. However, through Missouri, Kansas, Nebraska, and eastern Wyoming, where snow cover was continuous in 1979 but absent in 1981, the DP afternoon temperature differential is about 6 to 15øC between the two months. Thus, although the air mass was generally colder by about 5øC due to factors other than surface characteristics, it appears that the extended snow cover in January, 1979 further reduced the temperature of DP air masses by at least several degrees and by up to 11øC in the region where snow cover was absent in 1981.

A comparison of air mass frequencies during the two months helps to distinguish them even further (Figure 9). The largest disparities between the months were in DM and MP frequencies. In January 1979, when snow cover was present over Kansas, Missouri, and northern Oklahoma, MP frequencies varied from about 33 percent in western Kansas to 50 percent in eastern Missouri. During January, 1981, when snow was largely absent from these regions, MP frequencies were generally below 15 percent. The "polar index" was 70 to 85 percent over much of this region in 1979 and was less than half of this in 1981. The preponderance of the MP air mass in 1979 apparently contributed to the abundant snow cover in the area.

The frequency of DM air was considerably higher in January, 1981. Over 50 percent of the days in western Kansas were classified as DM in 1981, as compared to 0 percent in 1979. In fact, no DM air intruded north of a line from central Kansas to Ohio in January, 1979, which corresponds closely to the southern snow boundary during that month. The zonal flow which permits intrusion of DM air inhibits precipitation development, and the mild temperatures associated with the air mass are more likely to be associated with a rain event. It is interesting to note that the MT frequency was only slightly higher during the warmer month of January, 1981 in the central U.S., although further east the frequency of MT air was considerably greater in January, 1979.

Potential Environmental Applications
The SSC has been designed to develop an improved continental-scale air mass climatology. However, the primary driving force behind development is to facilitate synoptic climatological impacts analysis. Although site-specific air mass-based indices such as the TSI are useful for synoptic climatological environmental analysis at a given locale, they are generally not designed for inter-site comparative analyses. The SSC retains the ability to distinguish air masses across a large area successfully, and it is relatively simple from a computational standpoint.

There are a number of problems which lend themselves well to SSC impacts analysis, and the ability of SSC to compare similar air masses across very large areas allows for a detailed determination of interregional responses. Among the most intriguing applications is the evaluation of climate-health relationships, and the potential development of weather/health "watch-warning systems". Two of the significant assets of synoptic analysis in climate/health research are (1) the identification of those synoptic situations which are associated with deteriorating health, and (2) the ability to distinguish climate impacts from those of other environmental factors, such as air pollution (Kalkstein, 1991). For example, the SSC has recently been applied to an analysis of weather-related mortality across the United States (Kalkstein, et al., in press). Results showed that certain air masses at particular locations were highly correlated with increases in acute mortality, specifically during the summer months. That study also determined which air masses are associated with high concentrations of air pollutants, such as total suspended particulates and sulfates. With the availability of EPA-monitored air pollution data from numerous sites around the U.S. (the Aerometric Information Retrieval System, or AIRS, administered by the Office of Air Quality Planning and Standards), the SSC will permit continental-scale analyses of air masses which are responsible for high concentrations of pollutants. A more detailed separation of the effects of pollutants and weather using the SSC is currently underway.

A number of agencies, such as NOAA, EPA, and city Public Health Departments are emphasizing the need for weather/health watch-warning systems, especially as they relate to human health or air pollution (Scheraga and Sussman, in press; Haines et al., 1993). The SSC has been designed for integration within a watch/warning system, since air mass type can be predicted for up to 72 hours in advance based on available weather forecasts. For example, if a certain air mass has the propensity to be associated with high daily mortality totals, a "health watch" could be issued if this air mass is forecast within the next day or two. Such a program was recently adopted by the Philadelphia, Pennsylvania Department of Public Health. Since June 21, 1995, the operational weather forecasts from the National Meteorological Center and the local National Weather Service office in Philadelphia are being used to forecast the occurrence of air masses which are associated with elevated mortality. Plans are underway to expand this system to other locales, and a national EPA/NOAA weather/health watch-warning system, based on the SSC, is presently in the planning stages. 

Co., Ltd., London, p. 555.
Bergeron, T. 1930. 'Richtlinien einer dynamischen klimatologie', Met. Zeit., 47, 246-262. Brazel, A. J., F. B. Chambers, and L. S. Kalkstein 1992. 'Summer energy balance on west Gulkana glacier, Alaska, and linkages to a temporal synoptic index', Z. Geomorph. N.F., Suppl.Bd. 86, 15-34.

Bryson, R. A. 1966. 'Air masses, streamlines and the Boreal forest', Geographical Bulletin, 8, 228-269. Christensen, W. I., Jr., and R. A. Bryson 1966. 'An investigation of the potential of component

analysis for weather classification', Monthly Weather Review, 94(12), 697-709. .Crowe, P. R. 1971. Concepts in Climatology. St. Martin's Press, New York, p. 589. Davis, R. E., and L. S. Kalkstein 1990a. 'Using a spatial synoptic climatological classification to

assess changes in atmospheric pollution concentrations', Physical Geography, 11(4), 320-342. Davis, R. E., and L. S. Kalkstein 1990b 'Development of an automated spatial synoptic

climatological classification', International Journal of Climatology, 10, 769-794.

Davis, R. E., and D. R. Walker 1992. 'An upper air synoptic climatology of the western United

States', Journal of Climate, 5, 1449-1467. Davis, R. E. 1991. 'A synoptic climatological analysis of winter visibility trends in the mideastern

United States', Atmospheric Environment, 25B, 165-175. Dilley, F. B. 1992. 'The statistical relationship between weather-type frequencies and corn (maize)

yields in southwestern Pennsylvania, USA', Agricultural and Forest Meteorology, 59, 149-164. Dunne, P. C. 1991. 'Determination of Climatic Change in the Canadian North American Arctic

Using a Synoptic Climatological Approach', Masters Thesis, University of Delaware, p. 132. Fovell, R. G., and Mei-Ying C. Fovell 1993. 'Climate zones of the conterminous United States

defined using cluster analysis', Journal of Climate, 6(11), 2103-2135. Gilbert, E. S. 1969. 'The effect of unequal variance-covariance matrices on Fisher's linear

discriminant function', Biometrics, 25, 505-516. Haines, A., P. R. Epstein, A. J. McMichael, L. S. Kalkstein (on behalf of an international panel) 1993. 'Global health watch: Monitoring impacts of environmental change', The Lancet, 342, 1464-1469.

James, M. 1985. Classification Algorithms, John Wiley and Sons, N.Y., p. 210. Kalkstein, L. S., and P. Corrigan 1986. 'A synoptic climatological approach for geographical analysis: Assessment of sulfur dioxide concentrations', Annals of the Association of American Geographers, 76(3), 381-395. Kalkstein, L. S., G. Tan, and J. Skindlov 1987. 'An evaluation of objective clustering procedures for use in Synoptic climatological classification', Journal of Climate and Applied Meteorology, 26(6), 717-730. Kalkstein, L. S., P. Dunne, and R. Vose 1990. 'Detection of climatic change in the western North American Arctic using a synoptic climatological approach', Journal of Climate, 3(10), 1153-1167. Kalkstein, L. S. 1991. 'A new approach to evaluate the impact of climate on human mortality', Environmental Health Perspectives, 96, 145-150. Kalkstein, L. S., P. Dunne, and H. Ye 1993. 'Climate change in the Arctic: The role of topographically-enhanced inversions: Response to Burn', Journal of Climate, 6(7), 1476-1478.

Kalkstein, L. S. 1993. 'Health and climate change - direct impacts in cities', The Lancet, 342, 1397-1399. Kalkstein, L. S., C. D. Barthel, H. Ye, K. S. Smoyer, S. Cheng, J. S. Greene, and M. C. Nichols, 'The impacts of weather and pollution on human mortality', EPA Office of Policy, Planning and Evaluation, Climate Change Division, Document #EPA 230-R-94-019, 145 pp (in press). Klecka, W. R. 1980. Discriminant Analysis. Sage University Paper Series on Quantitative Applications in the Social Sciences, Beverly Hills and London: Sage Pubs. p. 71. Kirchhofer, W. 1974. 'Classification of european 500 mb patterns', Schweizerische Meteorologische Anstalt, Institut Suisse de Meteorologie, Zurich, 43, 1-16.

Lachenbruch, P. A. 1975: Discriminant Analysis, Hafner Press, New York, 128 pp. Leathers, D. J., and D. A. Robinson 1993. 'The association between extremes in North American

snow cover extent and United States temperatures', Journal of Climate, 6(7), 1345-1355. Lund, I. A. 1963. 'Map-pattern classification by statistical methods', Journal of Applied

Meteorology, 2, 56-65.
Marks, S., and J. O. Dunn 1974. 'Discriminant function when covariance matrices are unequal',

Journal of the American Statistical Association, 69, 555-559. McCabe, G. J., L. E. Hay, L. S. Kalkstein, M. A. Ayers, and D. M. Wolock 1989. 'Simulation of precipitation by weather-type analysis', Hydraulic Engineering '89 Proceedings National Conf. on Hydraulic Engineering/HY Div/ASCE), New Orleans, LA, 679-684.

Miller, R. G. 1962. 'Statistical prediction by discriminant analysis'. Meteorol. Mono., 4, 54 pp. Muller, R. A. 1977. 'A synoptic climatology for environmental baseline analysis: New Orleans',

Journal of Applied Meteorology, 16, 20-33. Namias, J. 1963. 'Surface-Atmosphere Interactions as Fundamental Causes of Drought and Other Climatic Fluctuations', Arid Zone Research XX, Changes of Climate, Proceedings of Rome Symposium, UNESCO and WMO, 345-359. Robinson, D. A., and M. G. Hughes 1991. 'Snow cover variability on the northern and central Great Plains', Great Plains Research, A Journal of Natural and Social Sciences, 1(1), 93-113. Romesberg, H. C. 1984. Cluster Analysis for Researchers. Lifetime Learning Publications, Belmont, California, p. 334. Scheraga, J. S., and F. Sussman. 'Preliminary assessment of the benefits to the U.S. of avoiding, or adapting to climate change', Washington, DC: EPA Climate Change Division (in press). Schwartz, M. D. 1991. 'An integrated approach to air mass classification in the north central United States', The Professional Geographer, 43(1), 77-91. Schwartz, M. D., and G. A. Marotz 1986. 'An approach to examining regional atmosphere-plant interactions with phenological data', Journal of Biogeography, 13, 551-560. Schwartz, M. D., and B. R. Skeeter 1994. 'Linking air mass analysis to daily and monthly mid-tropospheric flow patterns', International Journal of Climatology, 14, 439-464. Schwartz, M. D. 1995. 'Detecting structual climate change: an air mass-based approach in the North Central United States, 1958-1992, Annals, Association of American Geographers, 85, 553-568. Sokal, R. R., and C. D. Michener 1958. 'A statistical method for evaluating systematic relationships', The University of Kansas Science Bulletin, 38, 1409-1438. Ye, H., D. J. Leathers, and L. S. Kalkstein 1994. 'Classification of upper level circulation patterns in the polar region and their relationship to surface air temperature', Middle States Geographer, 26, 1-6. 

Figure 1. Procedure to select seed days.

Figure 2. Steps in the development of the SSC.

Figure 3. Air mass frequencies (percent for 1961-90). a. dry polar; b. moist polar; c. dry temperate; d. moist temperate; e. dry tropical; f. moist tropical. Dots indicate all stations where a particular air mass was detected during the study period.

Figure 4. a. polar index frequencies; b. transition frequencies.

Figure 5. Mean 1500 LST temperatures (øC) for selected air masses. a. dry polar; b. moist

tropical; c. dry temperate.

Figure 6. Mean 1500 LST dew points (øC) for selected air masses. a. dry polar; b. moist

tropical; c. dry temperate.

Figure 7. Mean diurnal temperature ranges (øC) for the air masses. a. dry polar; b. moist polar;

c. dry temperate; d. moist temperate; e. moist tropical.

Figure 8. Mean 1500 LST dry polar temperatures (øC) for: a. January, 1979; b. January, 1981; c. temperature difference between the two months (1981 minus 1979). Shaded areas represent grid cells covered by snow for over two weeks during the month. For map c, shaded area represents region covered with snow during January, 1979 but free of snow during January, 1981.

Figure 9. Frequencies of selected air masses (percent) during January, 1979 and January, 1981.

January, 1979 moist polar; b. January, 1981 moist polar; c. January, 1979 polar index; d. January, 1981 polar index; e. January, 1979 moist tropical; f. January, 1981 moist tropical.