1. Introduction

Recently, snow cover has taken a more prominent role in climate assessment and prediction. Snow cover has been shown to influence climatic variables such as temperature through increased albedo (Robinson and Kukla 1985; Walsh et al. 1985), increased thermal emissivity (Wagner 1973), decreased thermal conductivity (Cohen and Rind 1991), and by serving as a sink for latent heat (Cohen and Rind 1991). Snow cover influences both surface temperature as well as atmospheric thickness, a function of temperature. Mostly due to the influences of snow cover on temperature and subsequently thickness, numerous other variables are indirectly impacted by snow cover, such as precipitation (Karl et al. 1993; Namias 1985), energy balances (Ellis and Leathers 1998; Male and Granger 1981), midtropospheric flow (Frei and Robinson 1999; Heim and Dewey 1984), and storm track positioning (Namias 1978; Ross and Walsh 1986). It is often debated to what degree snow cover alters atmospheric circulation through temperature and thickness changes or to what extent do changes in circulation serve to alter the amount of snow cover.

An interesting snow cover–atmospheric circulation relationship that has been studied extensively is the relationship between Eurasian snow cover and the Indian monsoon. Through varying methods, scientists are generally in agreement that increased Eurasian snow cover is correlated with a decrease in the summertime precipitation across the Indian subcontinent (Bamzai and Shukla 1999; Dey and Bhanu Kumar 1983; Dickson 1984; Hahn and Shukla 1976; Vernekar and Zhou 1995). It is argued that the driving force for this relationship is the decreased strength of the thermal low over the landmass during high snow-cover years due to the cooling properties of snow as well as the increased soil moisture from a melting snowpack.

Similarly, the North American monsoon has been examined regarding its relationship with snow cover across the southern portion of the Rocky Mountains (Gutzler and Preston 1997; Gutzler 2000). Results have shown a strong correlation (r = −0.61) between spring snow cover and western New Mexican precipitation. The analysis worked well for the period 1961–90. Deterioration of the model before and after this time period was hypothesized to be due to a possible climatic shift. Both Gutzler and Preston (1997) and Gutzler (2000) hypothesized a thermally driven relationship, similar to the Indian monsoon, where increased snow cover in the Rockies served to decrease surface temperatures possibly through increased soil moisture.

A possible reason for the lack of stronger results to date between snow cover and the North American monsoon is that the circulation associated with the North American monsoon differs slightly from the circulation of the Indian monsoon. The North American monsoon flow pattern often establishes itself in early July with a shift in wind direction from the west to either the south or southeast. The monsoon typically lasts through mid-September and is focused on northwestern Mexico and portions of the southwestern United States (Adams and Comrie 1997). Arizona, for example, receives 40%–60% of its annual precipitation during the monsoon season (Jurwitz 1953). A large driving component of the North American monsoon is the location of the midtropospheric, subtropical ridge. This ridge has been associated with moisture surges from the Gulf of California (Hales 1972), “bursts” and “breaks” in the monsoon (Carleton 1986), and general monsoon strength (Carleton et al. 1990). This is to say that the North American monsoon is not solely a function of a regional thermal circulation.

It seems logical that snow cover's relationship with the subtropical ridge could possibly be a more important factor in assessing the monsoon rather than snow cover's thermal influence on the monsoon. To better ascertain this relationship, snow cover must be examined on a larger, synoptic scale. A cursory examination of synoptic scale snow cover extent and the North American monsoon showed that indeed large-scale August snow-cover extent appears to be correlated with August precipitation totals and frequency in the southwestern United States (Ellis and Hawkins 2001). Analyses revealed a distinct region in the southwestern United States, within which August precipitation was significantly correlated with summer snow-cover extent across western North America. Producing an areal average precipitation for this region increased the strength of the relationships between August snow cover over western North America and precipitation totals (r = −0.67) and frequency (r = −0.61) in the southwestern United States (Ellis and Hawkins 2001). Synoptic atmospheric composites showed that during low snow-cover extent years over western North America, 500-mb geopotential heights were higher over the United States and 850-mb specific humidity values were increased across the monsoon region.

Presented in this paper are the results of a more in depth examination of the relationship between North American snow-cover extent and the North American monsoon. Snow-cover extent was examined on a monthly and seasonal basis for the nine months preceding and coinciding with the monsoon season. In addition to precipitation, the monsoon was also defined using severe weather as an indicator of monsoon intensity. Finally, a cursory examination is presented to determine the potential applicability of large-scale snow-cover extent as a forecast tool for the North American monsoon.

2. Data and methodology

Snow-cover data were obtained from the National Oceanic and Atmospheric Administration (NOAA)/National Environmental Satellite, Data, and Information Service (NESDIS) snow-cover database. Data exist for 7921 grid cells in the Northern Hemisphere. Each grid cell has an associated latitude, longitude, area, and binary snow-cover variable. These data are available for each grid cell for each week from 1973 to present. For this study, the years 1973–97 were used to coincide with available precipitation data.

The snow-cover data allow for areal extents of snow cover to be calculated for any region of the Northern Hemisphere for any timescale greater than one week. For this study areal extents were calculated for western, eastern, and all of North America (Fig. 1). The first part of the study focused on western North America because eastern North America is atmospherically downstream of the monsoon region and therefore seemingly not critical in influencing the North American monsoon. The western North America snow-cover region was defined as the area 72°N to as far south as snow cover existed and 96° to 170°W. Regions north of 72°N were deemed less important as they are predominantly covered with snow or sea ice for the entire year. Eastern North America was defined as 72°N to as far south as snow cover existed and 30° to 96°W.

Weekly snow-cover data for each grid cell within western North America were transformed to monthly values of the mean areal extent of snow cover. For each month of the 25-yr study period, the fraction of the month during which snow cover was present within a particular grid cell was multiplied by the area of the cell. The areas obtained for each cell using this method were then summed. In other words, a month exhibiting a full period of snow cover in a particular grid cell contributed its entire area to an overall, total areal extent of snow cover for the region. Shown in Fig. 2 is an example of how snow-cover extent varies. Shown are the highest (1978) and lowest (1993) areal extents of snow cover for the month of August. During the summer, snow-cover extent varies mostly in the Canadian Rockies and the extreme north. Monthly areal extents of snow cover were also summed to create 2- and 3-month totals of areal snow-cover extents. All areas were converted to z scores by subtracting the average monthly snow-cover extent from each year's snow cover extent and dividing by the standard deviation.

Daily precipitation data were obtained from the National Climatic Data Center's (NCDC) Summary of the Day database. An overly large monsoon region was defined (Fig. 1) in order to establish any spatial patterns that may exist within the monsoon region. The region's boundaries were defined as the area 39°N to the United States–Mexico border and 102° to 116°W. All stations within this region contained at least 28 years of daily data, ending in 1997 and possessing 90% coverage over each station's history. With these restrictions, 384 stations were extracted from the database and used for this study.

For each station, precipitation totals were calculated by transforming daily precipitation into seasonal totals of precipitation for the monsoon season. In other words, for any given year, the daily precipitation that fell between the start and end date of the monsoon season for that year was summed. Precipitation frequency was calculated as the number of days that a given station reported precipitation during the monsoon season. Precipitation totals and frequency were calculated on a seasonal basis for the years 1973–97 and converted to z scores.

Start and end dates of the monsoon season were obtained from the Phoenix office of the National Weather Service (NWS). The start date is defined as the third consecutive day with an average Phoenix dewpoint temperature of 12.8°C (55°F) or higher. The end date is defined retrospectively by the NWS staff based on experience in forecasting the monsoon. The definitions of both the start and end dates are problematic. The start date is based on Phoenix dewpoints alone and consequently may not be geographically representative. The end date is far too subjective. More work needs to be done in redefining these dates. However, currently they are the standards. Annual monsoonal variation, therefore, is a function of both the intensity of the monsoon and its duration as defined by the NWS.

For each station, precipitation totals and frequency were linearly regressed with monthly and seasonal snow-cover extents. Isoline maps were created using these coefficients to determine where the relationships between snow-cover extent and the precipitation variables were the strongest. This procedure is identical to that described in Ellis and Hawkins (2001). However, in this case monsoon season precipitation values were used instead of simply August values. Also, the precipitation values were regressed with numerous areal extents of snow cover [January to September and January–February–March to July–August–September (JFM–JAS)].

Because of the convective nature of the monsoon, individual station data may not be representative of the rainfall patterns associated with the monsoon. Ellis and Hawkins (2001) defined a subregion to average precipitation stations to account for scattered precipitation events. For this study, a principal components analysis (PCA) was used to determine a monsoon region for which to average precipitation values.

The input data for the PCA were the precipitation totals (z scores) for the 384 stations contained within the monsoon region (Fig. 1). The data matrix for the PCA had 384 rows (stations) and 30 columns (1968–97). Note that five more years of monsoon season precipitation data were added to create a 30-yr climatology. Missing data were substituted with the corresponding value from the nearest station. If the nearest station was also missing that particular datum, then the average precipitation value for the entire monsoon region was substituted for that one missing datum.

An unrotated PCA was run and three components were selected based on an examination of the scree plot (not shown). For the three components, the eigenvalues ranged from 1.9 to 4.0. Combined, the three components accounted for 29.5% of the variance in the precipitation totals for the monsoon area. It should be noted that varimax rotation was applied to the PCA and the explained variance of an equal number of components was slightly less for the varimax rotated PCA as compared with the unrotated PCA.

Loadings, a representation of how well each year is represented by the individual components, had highest absolute values of 0.69 (1972), −0.55 (1978), and −0.53 (1988) for components one, two, and three, respectively. Communalities, a representation of how well each year was represented by all the components with eigenvalues greater than one, ranged from 0.43 (1969) to 0.71 (1983).

Component scores generated from the PCA for each station were used to create isoline maps. These maps showed the locations of wet (positive component scores) and dry (negative component scores) areas. They also showed the precipitation pattern represented by the components.

In order to assess how snow cover extent covaried with the precipitation patterns in the monsoon study area, the loadings for each year from 1973–97 were linearly regressed with the numerous snow-cover extent variables. Because the loadings show how well a year is represented by a spatial pattern, correlations between snow-cover extent and the loadings should represent how well snow-cover extent covaries with the different patterns.

Component 3 emerged as the “monsoon” component. The zero line on the isoline map for component three was used as a boundary to average all the precipitation values inside this line. These average precipitation totals and frequencies were regressed with monthly and seasonal snow-cover extent.

Severe weather was also examined for the monsoon region as an indicator of monsoon intensity. Severe weather data were obtained from NOAA's Storm Prediction Center (SPC). For the monsoon region (Fig. 1) during the monsoon season for each year from 1973–95, occurrences of tornadoes, wind events greater than 25.7 m s⁻¹ (50 kt), and hail events greater than 1.9 cm in diameter were calculated. Occurrences of hail, wind, and tornadoes were summed to give a total number of severe weather events per year during the monsoon season. The final result was four severe weather variables for each year: hail occurrences, wind occurrences, tornado occurrences; and the sum of the hail, wind, and tornado occurrences. The four severe weather variables were converted to z scores and linearly regressed with monthly and seasonal snow cover extents.

In order to begin to assess a physical explanation for the relationships established by the preceding analyses, synoptic composites of sea level pressure, 500-mb geopotential heights, and 850-mb specific humidity were examined. Data were obtained from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis project (Kalnay et al. 1996). Midtropospheric heights show the location of the subtropical ridge, which is known to be related to snow-cover extent and the strength of the monsoon. The location of the ridge influences the location of surface high pressure centers associated with monsoonal circulation. Low-level humidity values help to indicate the amount of moisture being advected into the region. Beyond mean composites for sea level pressure, 500-mb heights and 850-mb specific humidity, composites of the difference between the five highest June–July–August (JJA) snow-cover years for the 25-yr record and the five lowest JJA snow-cover years for the same period were also constructed using 500-mb heights and 850-mb specific humidity. For all variables, average values were mapped over the study period 1973–97. The diagnostic maps created from these analyses provide insight into the physical associations that anomalously high or low snow cover has with the North American monsoon.

Finally, encouraged by some of the results in the previous analyses, an initial attempt was made to determine the potential utility of snow cover as a lead indicator of upcoming monsoon intensity. Stepwise multiple regressions were run using average precipitation totals and frequency values for the “monsoon” component and severe weather for the entire monsoon region as the dependent variables. For all models, the independent variables were January–May snow-cover extents for western, eastern, and all of North America (15 predictors).

Finally, the snow cover, precipitation, and severe weather data were each stratified based on February total North American snow cover. Data for the years that corresponded to the top and bottom 25% (50% total) of February snow-cover extents were used in multiple regressions with the same independent and dependent variables as described above. The purpose of this technique was to better determine the potential association of snow cover with monsoon intensity by simply examining instances in which snow-cover extent was largely anomalous. For all stepwise regression models the alpha value for both entrance and removal from the model was 0.15.

3. Results

Isoline maps were created using correlation values between monthly and seasonal snow cover and individual stations' precipitation totals and precipitation frequency. Clearly indicated is the idea that as snow cover increased for JAS both precipitation totals and frequencies decreased. Correlation values were as high as −0.55 (99% confidence) with precipitation totals (Fig. 3a) and −0.65 (99%) with precipitation frequency (Fig. 3b). These represent the correlations at individual stations. These same maps were created for correlations between the two precipitation variables and monthly and seasonal snow-cover extent dating back to JFM [14 maps total (2 variables × 7 seasons), 12 not shown]. As the snow-cover extent values preceding the JAS period were employed, the pattern and relationships decreased in strength and were usually nonexistent by March–April–May (MAM). Results were very similar using monthly rather than seasonal snow-cover extent (not shown).

More important than the negative relationships shown are the spatial patterns demonstrated by the maps. Both maps show that the strongest relationships occur in a north–south area covering eastern Arizona and western New Mexico, extending into southern Colorado. The band is most likely a function of the tongue of moisture that moves into the area during the summer months. This technique of mapping correlation coefficients is identical to that used by Ellis and Hawkins (2001). However, the results can now be applied to the entire monsoon season rather than just August. Also it is now known that the snow cover–precipitation relationship weakens when using snow-cover extents from earlier in the year.

For the three components generated from the PCA, isoline maps were created for the monsoon region based on the component scores for each station (Fig. 4). Areas with positive component scores represent areas of positive precipitation z scores and can be considered moister. Areas with negative component scores represent drier areas.

Component 1 (13.5% variance) is termed “east–west” (EW) due to the east–west gradient of precipitation. Positive component scores are found in the eastern and southeastern parts of the study region. Drier regions are found in the western and northwestern regions. The zero line runs from the southern part of the Arizona–New Mexico border through central Colorado.

Component 2 (9.6% variance) is termed “north–south” (NS) due to the north–south gradient of precipitation. Positive component scores, where more precipitation occurred, are found in the northeastern section of the study region. Drier areas are found in the southern and eastern parts of the region. The zero line runs from the middle of the New Mexico–Texas border to the middle of the Utah–Nevada border.

Finally, component 3 (6.4% variance) is termed “monsoon” due to the tongue that stretches into the monsoon region from Mexico indicative of monsoon moisture advection. The eastern two thirds of Arizona, western two thirds of New Mexico, and part of southern Colorado are encompassed by positive (moist) component scores that delineate a dry interior. This shape is similar to the pattern in the isoline maps of correlations between snow cover extent and individual station variables (Fig. 3). The dry center for the monsoon component is located on the United States–Mexico border. This center is probably shifted northward due to the lack of Mexican data used in this study. While the pattern is represented as a dry monsoon area, these results may be misleading. A year that had a positive loading on this component would have a dry monsoon component while a year with a negative loading would have a wet monsoon region. It is the case with all components of this nature that what is truly important is not the areas of positive or negative component scores but rather the pattern of scores.

For each component a yearly loading was generated which is a representation of how well a particular year was represented by a component. These loadings were linearly regressed with monthly and seasonal snow-cover extents to ascertain how well snow-cover extent covaried with precipitation patterns. Results from correlations with monthly snow-cover extent (Fig. 5) were a bit stronger than results associated with seasonal snow-cover extent. The EW component had no significant relationship with snow cover. The NS component had a high correlation coefficient of −0.39 (95%) between August snow cover and NS loadings. Years with high snow cover were well represented by the opposite of the NS component map, which is to say, the southern part of the monsoon study region would be wet and the northern region would be dry.

The highest correlation coefficient of 0.66 (99%) occurred between August snow-cover extent and the monsoon component loadings. This positive relationship means that high snow cover extent correlates with the pattern shown in the component map and that high snow cover correlates to a dry monsoon region. This finding agrees with the previous results. From winter to summer, there is a general increase in the correlation coefficients between snow cover and the monsoon component loadings, with the maximum correlation occurring between August snow-cover extent and the monsoon component.

Based on the monsoon component (Fig. 4), precipitation totals and frequencies for stations contained within the zero line of the component were averaged and linearly regressed with monthly and seasonal snow cover (Fig. 6). By taking an areal average of the precipitation variables, and thereby minimizing the problem of spatial inhomogeneity associated with convective precipitation, as opposed to using individual stations, relationships between summer snow cover extent and precipitation totals and frequencies were slightly improved. In both cases the association between summer snow-cover extent and monsoon precipitation was expanded to a large area outward from a point location. Using JJA snow-cover extent, correlation coefficients were −0.55 (99%) and −0.39 (95%) with precipitation totals and frequency. More important is the fact that relationships were strengthened between the precipitation variables and winter and spring snow-cover extent (Fig. 6). This provides hope for using winter or spring snow cover as a lead indicator of monsoon intensity.

Severe weather for the monsoon study region (Fig. 1) was correlated with monthly and seasonal snow cover (Fig. 7). All severe weather except tornadoes (not shown) show strong relationships with snow cover extent. Correlations between JJA snow cover and severe wind (−0.74, 99%), total weather (−0.64, 99%), and hail (−0.50, 99%) were significant. More important from a potential forecasting standpoint is the −0.50 (99%) correlation between February–March–April (FMA) snow-cover extent and the frequency of damaging monsoon season winds. This provides hope for the development of snow cover as a lead indicator of monsoon intensity.

To gain a physical understanding of the numerous statistical relationships demonstrated in this study, composite JJA sea level pressure, 500-mb geopotential height, and 850-mb specific humidity maps were created (Figs. 8 and 9). The general JJA 500-mb flow shows ridging over the central United States (Fig. 8a). This ridging is associated with what is often referred to as the “Four Corners high,” a surface high pressure center located near the Arizona, Utah, Colorado, and New Mexico borders. The center of this high can be seen as the more darkly shaded area located over Colorado in Fig. 8a. The clockwise rotation around this surface high helps to advect moist air into the monsoon region from the Gulf of California on southerly winds. The Bermuda high off the southeastern United States coastline also serves to advect higher-level moisture from the south (Fig. 8a). The Bermuda high is not as important as the Four Corners high. The sea level pressure map (Fig. 8a) also indicates the typical thermal low present over the Colorado River valley. This feature also allows for moist air advection from the cooler Gulf of California. The difference map (Fig. 8b) shows the mean JJA 500-mb heights for the five highest snow-cover extent years subtracted from that of the five lowest snow-cover extent years. There exist positive differences over most of the United States with negative differences over the Pacific coast of the United States. The differences suggest that during low snow-cover extent years there exist higher 500-mb heights across most of the United States than during low snow-cover extent years. This study has indicated that during low snow-cover extent years, the monsoon intensity is increased. These findings correspond to the idea outlined in previous studies that an intensified monsoon is related to a northward displacement of the subtropical ridge (Carleton et al. 1990). This displacement will also affect the location of the Four Corners high, altering the southerly flow into the region.

Similar to the 500-mb height differences, the 850-mb specific humidity data (Fig. 9) correspond to the earlier findings in this study, as well as the results of previous studies outlined in the literature. The mean 850-mb specific humidity field (Fig. 9a) shows a maximum centered in Mexico associated with the monsoon circulation. As with the 500-mb heights, a differenced map was created that depicts the difference in mean humidity during high and low snow-cover extent years (Fig. 9b). Differences show that during low snow-cover extent years there tends to be an increase in the 850-mb level moisture across the monsoon region. The increase is as great as 1.8 g kg⁻¹. This corresponds to this study's findings of increased monsoon strength during low snow-cover extent years. The general conclusion that can be drawn from this analysis along with the statistical results is that decreased snow-cover extent corresponds to a northward-displaced subtropical ridge, which seems to allow for greater advection of moisture into the monsoon region possibly by affecting the location of the Four Corners high.

In an effort to explore the potential utility of the snow cover–monsoon relationship as a forecast tool for the monsoon season as a whole, stepwise multiple regressions were run using precipitation totals and frequency for the monsoon component (Fig. 4) and severe weather for the entire monsoon region (Fig. 1) as the dependent variables and 15 snow-cover variables (western, eastern, and total North American snow-cover extent for January–May) as the independent variables. Statistics were generated for precipitation totals, frequency, hail, damaging wind, and total severe weather using data for all years (Table 1). The 15 snow-cover extent variables do a relatively poor job explaining the variance in the five monsoon variables. The largest coefficient of determination, a measure of explained variance, is 34.35 for severe wind occurrences using western North American snow cover in March as a predictor (Table 1). Certainly there are intricacies in the regional components of the monsoon circulation (e.g., thermal circulation) that weaken the overall large-scale snow cover–monsoon relationship.

In order to better isolate the snow cover–monsoon relationship, statistics were rerun for only those years of the highest and lowest 25% (50% total) of February snow-cover extents for the 25-yr record. Data were stratified by February snow-cover extent because of the majority of correlations between monthly snow cover and numerous monsoon variables, including many not discussed here. February snow-cover extent often exhibited a stronger relationship with the monsoon variables than that of the surrounding months. This is most likely due to the greater variability of snow cover in spring as ablation occurs. Snowpacks are more consistent in the winter than in spring. Spring, as a transition season, exhibits wide variation in the timing and amount of ablation. This lack of consistency makes the more consistent February snow cover extent a better predictor.

Using extreme February snow-cover extent years dramatically improved the regression statistics. Combinations of snow cover extent variables explained significant amounts of variance in precipitation totals (83%), precipitation frequency (95%), hail (81%), wind (82%), and total weather (98%). Improved relationships are naturally expected through the reduction of the sample size of the data (12 yr instead of 25 yr). However, the overwhelming strength of the relationships suggests that anomalously high or low February snow-cover extents are associated with the strength of the monsoon. This technique was used not as a statistically sound, final forecast model but to provide insight into where in the temporal record the relationship was strongest. When snow-cover extent is neither anomalously large nor small (middle 50%), the relationship deteriorates, suggesting that an alternative mechanism(s) needs to be accounted for. It is possible that during anomalously high or low snow-cover extent years, snow cover is associated with the large-scale atmospheric platform for the monsoon flow (subtropical ridge) that either strongly encourages or inhibits the monsoon circulation. When snow cover is near normal, it is possible that smaller-scale influences on the strength of the monsoon are of greater importance.

4. Summary and conclusions

The research that was presented here provides a more in-depth look than previous research (Ellis and Hawkins 2001) at the apparent teleconnection between synoptic-scale snow-cover extent and the North American monsoon. More importantly, the research has provided evidence that further work may lead to utilization of snow cover as a tool for early monsoon season forecasting. An accurate assessment of the expected general monsoon intensity several months in advance would be invaluable to various aspects of society in the Southwest. The results of this research can be summarized as follows.

1. Correlations between monthly and seasonal snow-cover extent and monsoon season precipitation totals and frequencies for individual stations revealed that snow-cover extent covaries most strongly with precipitation values for stations in eastern Arizona, western New Mexico, and southern Colorado. These relationships were strongest with summer snow-cover extent and deteriorated as snow-cover extents from earlier in the year were used.

2. A PCA using monsoon season precipitation data produced three components that explained 29.5% of the variance in the precipitation data. The third component was termed the monsoon component. This component captured eastern Arizona, western New Mexico, and southern Colorado. The loadings for this component covaried with monthly and seasonal snow-cover extent much better than the other two component loadings.

3. Averaging precipitation values from the stations within the monsoon component region allowed results to be applied to a larger geographical area. The strength of the relationships between winter snow-cover extent and monsoon season precipitation were strengthened.

4. Correlations between monthly and seasonal snow cover and severe weather showed that summer snow cover covaried strongly with damaging winds as well as total severe weather. These relationships also decreased as snow-cover extent from earlier in the year was used.

5. Synoptic atmospheric composites revealed that during low JJA snow-cover extent years, 500-mb geopotential heights were higher over much of the United States and 850-mb specific humidity values were increased over southwestern North America compared with high snow-cover extent years. Changes in snow cover correlate with a displacement of the subtropical ridge, which in turn displaces the Four Corners high, which plays a part in low-level advection of moisture.

6. Using winter and spring snow-cover extent data from extreme February snow-cover extent years produced stepwise regression models, which explained a large amount of the variance in precipitation totals, precipitation frequencies, hail, wind, and total severe weather. Regressions using moderate snow-cover extent years were much less robust.

The research presented here has illuminated the fact that creating forecast algorithms for the North American monsoon using snow cover as a central predictor variable may be a strong possibility. As such, numerous small projects are under way with the goal of creating a set of forecast algorithms that integrate the results of these projects. Large-scale snow cover is being examined relative to the monsoon to determine if there are specific areas within North America for which snow cover exhibits a more intimate association with the monsoon. Local snow cover is also being examined to account for, on a smaller scale, the thermal portion of the monsoon circulation. This portion of the project is more in line with much of the Indian monsoon–snow cover work as well as Gutzler and Preston's (1997) and Gutzler's (2000) work on the North American monsoon.