Regional Snowfall Distributions Associated with ENSO:
Implications for Seasonal Forecasting



Shawn R. Smith and James J. O'Brien




Center for Ocean-Atmospheric Prediction Studies
The Florida State University
Tallahassee, FL 32306-2840

contact: smith@coaps.fsu.edu



Submitted to the Bulletin of the American Meteorological Society

15 June 2000
Revised: 15 December 2000
In press: 31 January 2001


Abstract

   Regional changes in early, middle, and late winter total snowfall distributions are identified over the continental United States in association with warm and cold phases of the El Niño/Southern Oscillation (ENSO). The analysis is primarily motivated by a desire to improve winter season climate forecasts. Original interest in snowfall associated with ENSO was provided by requests for skiing forecasts during the 1997 ENSO warm phase. Geographic regions with internally similar ENSO warm, cold, and neutral phase snowfall distributions are identified using a composite technique. The composites reveal three early-winter, five mid-winter, and three late-winter regions with shifts in the upper, middle, and lower quartile seasonal snowfall. The quartile shifts revealed by the composite technique are important for forecasting applications; however, snowfall impact studies rely more on the absolute magnitude of the change in snowfall at individual stations. Potential impacts of the shifts snowfall distributions associated with ENSO are discussed using the quartile snowfall magnitudes for the stations in the composites. Shifts in regional snowfall distributions are compared to published ENSO winter climate studies, and hypotheses are presented to relate physical processes to the warm, cold, and neutral phase snowfall distributions.

   Principal findings include increased snowfall during an ENSO cold phase relative to warm and neutral phases in the northwestern states from early through mid-winter, less (more) snowfall during a cold (warm) phase relative to neutral years in the Northeast, and less snowfall (relative to neutral winters) in both warm and cold phases in the Ohio Valley (early winter) and Midwest (mid-winter). Combining these snowfall regions with an ever improving ability to forecast ENSO warm and cold phases will improve seasonal snowfall forecasts. The results should improve mitigation strategies for agencies adversely impacted by ENSO induced snowfall anomalies.

1. Introduction

   The El Niño/Southern Oscillation (ENSO) cycle has been shown to have significant impacts on winter precipitation over the continental United States (e.g., Smith et al. 1999, Smith et al. 1998, Ropelewski and Halpert 1996, Yarnal and Diaz 1986). Typically, past studies evaluated total precipitation without differentiating between the frozen and liquid forms of winter precipitation. Frozen precipitation impacts human activity in ways that present unique challenges to residents of the United States. The authors undertake a study focused on a common form of frozen precipitation, snow, to identify regional snowfall patterns associated with the ENSO cycle. The study was originally motivated by numerous requests for seasonal snowfall forecasts at ski resorts during the strong ENSO warm phase of 1997. Our primary focus is the identification of regions with significant changes in seasonal snowfall that can be used to improve seasonal snow forecasts. Improved forecasts are valuable not only to winter recreation industries, but also to urban planners, airport managers, and other activities impacted by snow. Combining new knowledge of regional snowfall patterns associated with ENSO and an ever improving ability to forecast ENSO warm and cold phases should result in improved seasonal snowfall forecasts. Improved seasonal forecasts may then be used by planners to mitigate costs or increase profits related to the upcoming season's snowfall.

   Snow is a hydrometeor with wide ranging impacts on human activities. For winter recreation industries, increased snowfall typically has a positive impact, increasing tourist revenues at resorts by improving conditions for winter sports. On the negative side, too much snow may increase the avalanche risk in the mountains. During winters when natural snowfall is limited, reduced tourist revenues may occur and the need to make snow may further increase costs for winter resorts. The transportation industry and government agencies charged with snow removal are adversely affected by unexpected increases in snowfall. Each year these organizations must budget for salt, equipment, and the personnel needed to keep roads and airports clear of snow and operational. Foreknowledge of the snow totals for an upcoming winter season has the potential to save these agencies hundreds of thousands or even millions of dollars.

   Recent studies of the impact of ENSO on snowfall are limited in the literature. Sweeny (1996) used statistical methods to show an increase or decrease in the number of hours with snowfall being reported at continental United States military bases during ENSO warm and cold phases. In general, Sweeny (1996) found fewer hours of reported snowfall during both warm and cold phases than in neutral years, with the exception of the Pacific Northwest, where increased hours of snowfall were reported during cold phases. More recently, Janowiak and Bell (1998) and Kunkel and Angel (1999) investigated changes in winter mean snowfall over the contiguous United States associated with moderate and strong ENSO cold and warm phases. They both showed decreases in mean snowfall over the Great Lakes, Ohio Valley, and Northern Rockies and increased snowfall over the Four Corners states during a warm phase. Interestingly, Kunkel and Angel (1999) revealed few areas with statistically significant snowfall anomalies during an ENSO cold phase, although their figures suggest increased snowfall in the Northwest and Northern Plains. Janowiak and Bell (1998) confirmed the increases in snow over the Pacific Northwest, North Dakota, and Minnesota. We find changes in the snowfall distributions during the cold phase that are generally in agreement with Sweeny (1996) and Janowiak and Bell (1998). In addition, our limited statistical testing shows more significant changes in seasonal snowfall during ENSO cold phases.

   Our method differs from these previous studies because we evaluate the quartiles of the seasonal snowfall distributions as opposed to the seasonal means. The seasonal mean is a poor measure of the central tendency of non-Gaussian precipitation distributions, and it provides no information about changes in seasonal snowfall toward the tails of the distribution. Furthermore, we analyze snowfall distributions during the early, middle, and late winter to show how snowfall distributions evolve through warm, cold, and neutral phase ENSO winters. Janowiak and Bell (1998) and Kunkel and Angel (1999) only evaluate the mean snowfall for an entire winter.

   Daily snowfall observations from first order weather stations are used to construct seasonal distributions when ENSO is in warm, neutral, and cold phases. The quartile snowfall for early, middle, and late winter distributions at each station are visually compared to neighboring stations. Stations with similar distributions (i.e., similar quartile shifts) for each of the three ENSO phases are grouped into geographic regions and composite seasonal snowfall distributions are created. One limitation of this approach is that we may exclude some (but not all, e.g., Northern Texas, Section 4b) ENSO signals where only one extreme phase shows a shift from neutral winters. The quartile shifts revealed by the composites are important for forecasting applications. Snowfall impact studies rely more on the absolute magnitude of the change in snowfall at individual stations; therefore, potential impacts of the shifts in snowfall distributions are discussed using quartile snowfall magnitudes from the climatologies for stations in the composites.

   Note that we only evaluate composite snowfall distributions associated with the phase of ENSO. We acknowledge that other climate factors, including but not limited to the Pacific Decadal Oscillation (Gershunov and Barnett, 1998) and the North Atlantic Oscillation (Hurrell 1996), might play a role in seasonal snowfall. In fact, these decadal oscillations may be partially responsible for the intraphase variability we note in ENSO snowfall distributions.

   We identify three early, five middle, and three late winter regions, each with multiple stations exhibiting similar snowfall distributions in all three ENSO phases (Section 4). ENSO cold phases are associated with increased snowfall relative to warm and neutral phase winters in the northwestern states from early through mid-winter. This pattern extends eastward to the northern Great Lakes in late winter. In the Northeast less (more) snowfall occurs in a cold (warm) phase relative to neutral years. In addition, less snowfall (relative to neutral winters) is associated with both warm and cold phases in the Ohio Valley (early winter), Midwest (mid-winter), and southwest Montana (late winter), while snowfall increases in Wyoming (late winter) in both warm and cold phases. Further results include increased snowfall during a warm phase in north Texas and notable changes in snowfall variability (measured by the breadth of the probability distributions) in some ENSO phases. Shifts in these regional snowfall distributions during the ENSO phases are compared to published snowfall studies. Large-scale jet stream and temperature anomaly patterns are discussed in Section 5 to develop hypotheses on the role atmospheric circulations play in the evolution of warm, cold, and neutral phases snowfall distributions.

2. Seasonal totals

   Daily snowfall data are extracted from the first order summary of the day (FSOD) dataset provided by the National Climatic Data Center. Stations in the FSOD dataset are located worldwide and the FSOD observations are collected by the National Weather Service, United States Air Force, United States Navy, and the Federal Aviation Administration. We focus on continental United States. The number of active stations has varied through the years. Although there were 428 active stations reporting FSOD data in 1985, only 143 stations meet the criterion for this snowfall study.

   The period of record 1950-1994 is chosen to maximize the number of FSOD stations with complete snowfall records. The analysis is limited to October through April as these are the months when snowfall typically occurs in the continental United States. The 143 stations chosen all have complete records of daily observations for the 45-year study period. The stations are distributed fairly evenly across the continental United States, with the largest gaps occurring in North Dakota, eastern Montana, Kansas, and Oklahoma. We note that these stations are not ideally located for assessment of mountain snowfall; however, they will capture snowfall patterns that are the result of synoptic-scale systems. Some stations in the Deep South are included to identify whether the rare occurrences of snow have any relationship to the phase of ENSO.

   The authors note a limitation of the FSOD snowfall data is the inclusion of hail in the daily snow totals from 1950-1955 and from April 1988 to present. Limiting the analysis to the cool season months should reduce the impact of hail on the results. In the remaining months, there is no way to differentiate the hail reports from actual snowfall.

   Seasonal snow totals for each station are created by summing the daily snowfall at each station. Snowfall seasons are defined as early winter (October, November, December; OND), mid-winter (December, January, February; DJF), and late winter (February, March, April; FMA). The early and late winter seasons are included to show both the evolution of snowfall distributions and to capture early and late winter snows in the western mountains and near the Great Lakes.

3. Regional Composites

   Early, middle, and late winter snowfall totals at each station are classified as occurring during a warm, cold, or neutral phase of ENSO using the Japanese Meteorological Agency (JMA) sea surface temperature (SST) index (Green 1996). The JMA SST index is a five-month running mean of SST anomalies over the equatorial Pacific Ocean from 4°N to 4°S and 150°W to 90°W. An ENSO warm (cold) phase is defined when the JMA SST index is greater than 0.5°C (less than -0.5°C) for at least six consecutive months. The consecutive months must begin prior to October and include October, November, and December.
   Extremes in the ENSO cycle typically develop during summer, peak in late fall, and decay in the following spring. Therefore an ENSO year is defined as the period from October of the onset year through September of the following year. As an example, the 1982 warm phase is defined as the year from October 1982 through September 1983. In this case, we would classify the early (OND'82), middle (D'82 JF'83), and late (FMA'83) winter snow totals as part of the warm phase distributions for each station. All years not classified as warm or cold according to the half-degree Celsius criterion listed above are considered neutral years (Table 1.)
Table 1: ENSO warm, neutral, and cold phase years based on the JMA SST index for the period 1950-1994. Each year marks the beginning of the ENSO year (e.g., 1951 = October 1951 through September 1952).
ENSO Phase ENSO Year

Warm 1951, 1957, 1963, 1965, 1969, 1972, 1976, 1982, 1986, 1987, 1991
Cold 1954, 1955, 1956, 1964, 1967, 1970, 1971, 1973, 1975, 1988
Neutral 1950, 1952, 1953, 1958, 1959, 1960, 1961, 1962, 1966, 1968, 1974, 1977, 1978, 1979, 1980, 1981, 1983, 1984, 1985, 1989, 1990, 1992, 1993, 1994

   There is an ongoing debate in the scientific community concerning which index (JMA, SOI, NINO3.4, etc.) best represents the warm and cold phases of ENSO. No single index of ENSO is definitive. These indices are continually undergoing refinement (Trenberth 1997), and both the SOI and JMA have been accepted as ENSO indices in numerous refereed articles. The patterns identified herein are expected to be refined as more warm and cold phase data become available.

   Box plots (Wilks 1995) are constructed at each station to display the distribution of total snowfall during warm, cold, and neutral ENSO phases. Quartile, maximum, and minimum values are calculated from the seasonal totals. An alternate method first fits gamma distributions to the warm, cold, and neutral phase snowfall distributions prior to calculating the quartiles; however, the resulting quartile values are not significantly different than those calculated directly from the data. For the present study gamma fitting is not essential to the results; however, future statistical tests using gamma distributions are planned.

   Stations with similar distributions during warm, cold, and neutral phases are identified through visual comparison of neighboring stations. As an example, six stations in the Pacific Northwest are compared (Fig. 1). Four of these stations, Bellingham, Seattle-Tacoma, and Olympia, WA, and Portland, OR, are grouped as having similar distributions in all ENSO phases. Each station's warm phase distribution has a minimum of zero snowfall and the snowfall increases from warm to neutral to cold phases at all quartiles. The distribution at Yakima, WA, is similar, but median snow totals decrease from the warm to neutral phase. At Pendleton, OR, the pattern is clearly different with less than neutral snowfall at most quartiles in both warm and cold phases. Another consideration for grouping these four stations is that they are all geographically located near the Pacific coast.


Fig. 1. Snowfall distributions for ENSO warm, neutral, and cold phase mid-winter seasons at six stations in the Pacific Northwest. The box plot displays the 75th, median, and 25th percentile snow totals with the top, middle, and bottom lines of the box, while the hairs mark the maximum and minimum snow totals. Units are cm.


   Composite warm, cold, and neutral phase snowfall distributions are created for each region that displays coherent distributions amongst individual stations. One complication is total snowfall often varies from station to station (Fig. 1) depending upon local factors (e.g., topography and lake effects). Therefore, seasonal totals for stations in a region cannot simply be averaged to create a reliable composite. Instead, the warm, cold, and neutral phase quartile values are divided by the quartile values for the 45-year snowfall climatology at each station. These scaled total snowfall quartiles are then averaged over the number of stations in each geographic region to create seasonal composites. The resulting composite distributions are displayed as a percentage relative to the individual station's 45-year snowfall climatology. The compositing technique allows one to estimate the warm, cold, and neutral phase quartile snow totals for each station in a region using the 45-year climatological quartiles for the desired station.

   Regions with significant shifts in median snowfall during ENSO warm and cold phases are identified using a non-parametric, two-sample median test (Bhattacharyya and Johnson, 1977). Regional significance (at the 0.95 level) is identified within each season (OND, DJF, and FMA) for warm versus neutral (WN), cold versus neutral (CN), and warm versus cold (WC) phase snowfall distributions. Statistical tests are only conducted within each season (no cross season comparisons) to limit concerns about the independence of the overlapping seasons chosen by the authors. The median test employed is a simple measure of significant shifts in snowfall distributions. The test does not reveal significant distribution changes at any percentile other than the median.

4. Results

   Within this section we, 1) present regional composite snowfall quartiles associated with warm, cold, and neutral ENSO phases for early, middle, and late winter, 2) identify significant median shift in the composites, and 3) show examples of how to estimate snowfall quartiles for an individual station in a region. Potential impacts to residents of these regions will also be discussed. Hypotheses relating surface temperature anomalies and tropospheric jet stream patterns associated with ENSO phases to changes in snowfall distributions are presented in section 5.

a. Early winter

   Three regions, the Northwest, Great Basin, and Ohio Valley, are identified as having coherent snowfall distributions during the early winter (Fig. 2). In the Northwest and Great Basin, a general increase in snowfall occurs during cold phases. A reduction in snowfall, relative to neutral years, occurs during both warm and cold phases in the Ohio Valley.

Fig. 2. The map displays three regions during the early winter (OND) that have similar total snowfall distributions during ENSO warm, neutral, and cold phases. Diamonds on the map mark the 143 FSOD stations. Composite quartiles for the (a) Pacific Northwest, (b) Great Basin, and (c) Ohio Valley are graphed for the warm (triangle), cold (diamond), and neutral (square) phase snowfall distributions. The warm, cold, and neutral phase symbols are offset slightly at each quartile to improve clarity of figure. The uncertainty in the mean for each composite value is displayed using error bars. The Great Basin and Ohio Valley had statistically significant differences (0.95 level) in median snowfalls for the cold versus neutral phases. Significance is noted by the CN in the lower right corner of (b) and (c).


   The Northwest composite distributions (Fig. 2a) show an increase in snowfall atall quartiles during ENSO cold phases relative to neutral years. Decreased snowfall occurs during a warm phase at the median and 75th percentile. The shift in median snowfall equates to a cold phase snowfall increase as small as 2 cm at Portland, OR, up to a 49 cm at Kalispell, MT (see Table 2). The percent increase in cold phase snowfall is slightly smaller for the 25th and 75th percentiles. During a warm phase, median snowfall is reduced by 1 cm and 23 cm, respectively, for Portland, OR, and Kalispell, MT. Although the shifts in median snowfall are not significant at the 0.95 level, the increase in cold phase snowfall may adversely affect the transportation sector and have a generally positive impact (increased avalanche threat would also be possible) on the winter recreation industry.

Table 2: Early winter (OND) 45-year climatological snowfall totals for stations in three regions with internally similar distributions.
Region Station Quartile snowfall (cm)
  25th 50th 75th
Ohio Valley South Bend Michiana, IN 49.6 62.7 88.5
Fort Wayne Baer Fld., IN 16.3 24.9 37.5
Indianapolis Intl. Arpt., IN 8.4 17.0 24.4
Columbus Intl. Arpt., OH 10.4 19.8 25.6
Akron-Canton Reg. Arpt., OH 24.5 34.8 47.0
Dayton Intl. Arpt., OH 9.6 19.0 28.6
Cinci-Northern Ky Arpt., KY 6.2 14.0 25.6
Lexington Bluegrass, KY 2.7 6.9 12.3
Charleston Knwa Arpt., WV 9.8 15.7 25.0
 
Northwest Bellingham Intl. Arpt., WA 0.0 5.1 20.6
Seattle-Tacoma Arpt., WA 0.0 3.0 11.8
Olympia Arpt., WA 0.0 5.6 20.6
Yakima Air Terminal, WA 9.3 18.3 39.6
Spokane Intl. Arpt., WA 26.0 48.0 77.1
Kalispell, MT 43.6 65.0 78.0
Great Falls Intl. Arpt., MT 38.0 52.3 70.7
Missoula Johnsn-Bell, MT 26.5 40.9 62.1
Butte Bert Mooney Arpt., MT 31.7 40.9 55.2
Bozeman Gallatin Fld. , MT 27.3 36.8 46.2
Boise Air Terminal, ID 8.6 17.3 32.1
Sheridan County Arpt., WY 45.6 56.4 76.2
 
Great Basin Reno, NV 6.0 15.5 27.9
Elko Municipal Arpt., NV 16.1 34.0 48.5
Ely Yelland Field, NV 20.7 35.3 57.8
Grand Junction Wlkr., CO 14.2 18.3 28.8
Salt Lake City Intl. Arpt., UT 31.1 50.0 77.5


   The total snowfall distributions in the Northwest also show marked changes in variability depending upon the phase of ENSO. For our purpose, we measure the distribution spread with the standard deviation (s.d.) of the seasonal snowfalls used to construct the composite. Cold phases s.d. are 1.7, 5.0, and 7.0 times larger than the neutral phase s.d. at the 25th, 50th, and 75th percentiles, respectively. For comparison, warm phase s.d. are 1.0, 2.2, and 3.6 times larger than the neutral phase s.d. As a result, we expect to see more variability in OND snow totals in the Northwest during a cold phase than occurs in a neutral or warm phase.

   The early winter snowfall quartiles for the Great Basin (Fig. 2b) show somesimilarity to the distributions in the Northwest. The cold phase is associated with more snowfall at all quartiles than occurs during neutral and warm phase years. One main difference is that the median snowfall for the warm phase is greater than the neutral phase median. In addition, the difference between median snowfall for the cold and neutral phases is significant (0.95 level). The variability in the median snowfall is nearly identical for all three phases (s.d. all near 11%); however, s.d. for the cold phase are greater than both the warm and neutral phases at the 25th and 75th percentiles.

   Total snowfall distributions in the Ohio Valley during the early winter (Fig. 2c) are evidence of the nonlinear response to changing ENSO phase (Hoerling et al. 1997). At all quartiles, both the warm and cold phase snowfalls are less than neutral phase snow totals. At the 75th percentile, neutral years are associated with 31% (24%) more snow than occurs in cold (warm) phases, which equates to 8 (6) cm and 28 (21) cm more snow during neutral phases in Indianapolis and South Bend, IN, respectively (see Table 2). Percent changes at the early winter median result in 24 (30) cm more snow during neutral phases at South Bend, IN than occurs in median cold (warm) phase winters. Depending upon whether this additional snow falls in one or more storms, there is the potential for increased snow removal costs in neutral ENSO years. When a warm or cold phase of ENSO is forecast, the Ohio Valley has an increased probability of receiving less early winter snowfall. Interestingly, only the shift from neutral to cold phase median snowfall is statistically significant (0.95 level). The authors expect this is due to the lower variability in cold phase median snowfall (s.d.=17%) as compared to the warm phase(s.d.=23%).

b. Mid-winter

   Changes in snowfall associated with ENSO are identified for five regions during the mid-winter season. The regions include the coastal Pacific Northwest, Northern Rockies, Midwest, Northeast, and Northern Texas (Fig. 3). Similar to the early winter, snowfall is heaviest during the cold phase in the Northwest. The Midwest follows the pattern of the Ohio Valley during the early winter with decreased snow in both warm and cold phases. Unlike the early winter, increased snowfall occurs in the Northeast and Northern Texas during mid-winter of warm phases. The median test only identifies statistically significant shifts in the Pacific Northwest and Northern Rockies; however, coherent distribution changes clearly occur in other regions.

Fig. 3. The map displays five regions during the mid-winter (DJF) that have similar total snowfall distributions during ENSO warm, neutral, and cold phases. Diamonds on the map mark the 143 FSOD stations. Composite quartiles for the (a) coastal Pacific Northwest, (b) Northern Rockies, (c) Midwest, (d) Northeast, and (e) Northern Texas are graphed for the warm (triangle), cold (diamond), and neutral (square) phase snowfall distributions. The warm, cold, and neutral phase symbols are offset slightly at each quartile to improve clarity of figure. The uncertainty in the mean for each composite value is displayed using error bars. The coastal Pacific Northwest and Northern Rockies had statistically significant differences (0.95 level) in median snowfalls for the warm versus cold phases (denoted by WC in (a) and (b)). In addition, the warm versus neutral medians are significantly different in the Northern Rockies (marked with WN in (b)).


   The coastal Pacific Northwest and Northern Rockies both show a pattern ofincreased (decreased) snowfall in the cold (warm) phase relative to neutral years (Figs.3a,b). Snowfall is reduced during a warm phase (relative to neutral years) to a point that no measurable snow occurs at the 25th percentile. The 98% (60%) reduction in snowfall at the 50th (75th) percentile of the warm phase distribution equates to 10 (16) cm decrease at Seattle, WA when compared to the neutral phase snowfall. In contrast, Seattle, WA, receives 7, 11, and 12 cm more snow in a cold phase (relative to neutral phase) at the 25th, 50th, and 75th percentiles, respectively. The warm and cold phase percentage shifts in the Northern Rockies are not as large as the coastal Pacific Northwest (averaging ±20% from the neutral phase), but the difference between the warm phase median and the neutral and cold phases median snowfalls are statistically significant (0.95 level). Although their snowfall patterns are similar, the coastal Pacific Northwest and Northern Rockies are separated by a region with no identifiable shifts in snowfall during the ENSO cycle. This gap is likely related to local topography. The authors note that both Janowiak and Bell (1998) and Kunkel and Angel (1999) revealed similar snowfall patterns, though neither showed them to have statistical significance.

   The pattern of ENSO phase snowfall distributions in the Midwest (Fig. 3c) appears to be a continuation and westward expansion of the early winter pattern over the Ohio Valley (Fig. 2c). Both the warm and cold phases are characterized by less snowfall than occurs in neutral winters. On average, a 25% reduction in snowfall occurs at all quartiles during cold and warm ENSO winters, relative to neutral phase winters. Due to the wide range of snowfall in the Midwest, reductions in quartile snowfall can be as small as 6, 10, and 13 cm in Cincinnati, OH, or as large as 37, 50, and 64 cm in the lake snow belt at Muskegon, MI (Table 3). The decrease in snowfall identified in the Midwest during the warm phase is consistent with both Janowiak and Bell (1998) and Kunkel and Angel (1999); however, neither showed a coherent signal of decreased snowfall during a cold phase. In this heartland of commerce and industry, the ability to forecast a 25% decrease in snowfall has the potential to bring substantial savings to the travel and transportation industries.

Table 3: Mid-winter (DJF) 45-year climatological snowfall totals for stations in five regions with internally similar distributions.
Region Station Quartile snowfall (cm)
25th
 50th
 75th
Pacific Northwest Bellingham Intl. Arpt., WA 4.3 18.3 42.2
Seattle-Tacoma Arpt., WA 2.7 10.7 26.8
Olympia Arpt., WA 5.6 22.1 58.0
Portland, OR 0.0 8.9 20.9
Northern Rockies Boise Air Terminal, ID 23.2 39.6 54.2
Missoula Johnsn-Bell, MT 58.3 72.9 104.8
Great Falls Intl. Arpt., MT 52.2 67.3 81.9
Helena Arpt., MT 34.2 55.9 75.2
Butte Bert Mooney Arpt., MT 39.1 53.1 66.3
Bozeman Gallatin Fld., MT 35.8 48.0 67.6
Billings Logan Arpt., MT 50.4 60.7 79.8
Sheridan County Arpt., WY 64.6 83.1 101.0
Midwest Sioux Falls Foss Fld., SD 29.7 49.8 67.3
Norfolk Stefan Arpt., NE 28.4 37.6 66.2
Sioux City Muni. Arpt., IA 30.2 43.7 59.2
Des Moines Intl. Arpt., IA 38.0 55.4 65.4
Mason City Arpt., IA 36.7 58.4 78.6
Rochester, MN 42.8 73.1 92.7
Minneapolis Intl. Arpt., MN 42.3 67.3 101.6
Eau Claire County Arpt., WI 42.7 62.7 89.7
Green Bay Austin Str., WI 49.4 75.7 99.6
Madison Dane Cnty. Arpt., WI 48.1 67.3 100.3
Milwaukee Mtchll. Fld., WI 58.2 71.4 121.7
Muskegon Co. Arpt., MI 150.0 202.2 257.3
Flint Bishop Arpt., MI 64.0 83.8 100.7
Moline Quad City Arpt., IL 39.4 56.9 77.3
Peoria Gtr. Peoria Arpt., IL 31.6 45.5 60.3
Springfield Captl. Arpt., IL 22.3 41.9 55.0
Indianapolis Intl. Arpt., IN 29.5 44.2 62.6
Fort Wayne Baer Fld, IN 41.8 59.2 69.2
South Bend Michiana, IN 112.0 125.5 174.0
Columbus Intl. Arpt., OH 39.0 47.5 63.6
Akron-Canton Reg. Arpt., OH 62.5 73.9 98.0
Cleveland Hopkins Arpt., OH 71.2 97.0 117.3
Dayton Intl. Arpt., OH 33.1 44.7 66.7
Cinci-Northern Ky Arpt., KY 24.9 39.4 53.6
Northeast Burlington Intl. Arpt., VT 112.6 141.0 166.1
Albany County Arpt., NY 86.1 115.8 135.8
New York Laguardia, NY 25.6 41.4 56.0
Newark Intl. Arpt., NJ 26.0 43.7 58.9
Philadelphia Intl. Arpt., PA 23.7 34.3 51.6
Williamsprt-Lycoming, PA 43.9 67.1 91.9
Allentown A-B-E Intl., PA 36.7 46.7 82.8
Washingtn DC Natl. Arpt., VA 16.6 26.2 40.8
Roanoke Woodrum Arpt., VA 18.7 34.5 64.9
Lynchburg Muni. Arpt., VA 13.7 30.2 48.5
Northern Texas Amarillo Intl. Arpt., TX 12.9 22.6 38.6
Lubbock Regional Arpt., TX 4.7 13.2 26.7
Midland Regional Ter, TX 3.3 6.9 12.4
Wichita Falls Muni. Arpt., TX 2.9 10.7 19.7
Abilene Muni. Arpt., TX 1.8 9.4 17.0

   In the Northeast, another core of commerce and population, the warm phase snowfall exceeds the neutral phase at all quartiles (Fig. 3d). In New York, NY, the warm phase yields 14 (18) cm more snow at the 50th (75th) percentile than occurs in neutral winters (Table 3). During a cold phase, snowfall is generally reduced in the Northeast. The largest reductions occur at the 75th percentile of the cold phase distribution, resulting in a 15 cm reduction in snow at New York, NY (relative to neutral winters). The authors note that the increase in snowfall during the warm phase extends southward to some stations in the southern Appalachians; however, the stations do not form a geographically continuous extension of the Northeast region. In addition, the s.d. for the warm phase quartiles averages four times the s.d. during neutral phases, indicating an increased variability in mid-winter snow totals during ENSO warm phases. These changes in snowfall in the Northeast are not identified by Janowiak and Bell (1998). Kunkel and Angel (1999) have a weak signal of less snow over the Northeast in an ENSO cold phase; however, they also show the Northeast to have less snow during warm phase winters.

   The final region showing coherent mid-winter changes in snowfall distributions is Northern Texas. The unique feature here is that little or no change in snowfall occurs between neutral and cold phase winters; however, snowfall increases at all quartiles during ENSO warm phases (Fig. 3e). A similar warm phase increase is shown in the figures from Kunkel and Angel (1999) and Janowiak and Bell (1998). Warm phase median snowfall increases by 69% and 78% over neutral and cold phase mid-winters, respectively. Although snowfall is generally light in Northern Texas (Table 3), median increases range from ~5 cm in Midland, TX, to ~17 cm in Amarillo, TX, and may have some negative impact on travel and transportation.

c. Late winter

   Three regions with coherent ENSO snowfall distributions are identified during late winter: the Northern Lakes, Southwest Montana, and Wyoming (Fig. 4). Overall, the Northern Lakes experience an increase in snowfall during cold phase winters and a decrease during warm phases. The two western regions have an inverse snow relationship relative to one another. The three identified regions each have small populations and snow is a normal part of winter life, so travel impacts would likely be minimal. However, these regions are known for winter recreation, so changes in late winter snow forecasts have the potential to impact tourist revenues.

Fig. 4. The map displays three regions during the late winter (FMA) that have similar total snowfall distributions during ENSO warm, neutral, and cold phases. Diamonds on the map mark the 143 FSOD stations. Composite quartiles for (a) Southwest Montana, (b) Wyoming, and (c) the Northern Lakes are graphed for the warm (triangle), cold (diamond), and neutral (square) phase snowfall distributions. The warm, cold, and neutral phase symbols are offset slightly at each quartile to improve clarity of figure. The uncertainty in the mean for each composite value is displayed using error bars. Wyoming and Northern Lakes had statistically significant differences (0.95 level) in median snowfalls for the cold versus neutral phases (denoted by CN in (b) and (c)). In addition, the warm versus neutral medians in Wyoming (marked with WN in (b)) and the warm versus cold medians in the Northern Lakes (marked with WC in (c)) are significantly different.


   The Northern Lakes region during a cold (warm) phase exhibits an increase (decrease) in snowfall at all quartiles relative to neutral years ( Fig. 4c). The increase in snowfall during a cold phase extends to some stations immediately downwind of lakes Erie and Ontario; however, the lack of Canadian data in our analysis does not allow us to connect these stations to the Northern Lakes region. Cold phase increases in quartile snowfall range from 11, 11, and 8 cm at Fargo, ND up to 29, 27, and 19 cm at Sault Ste. Marie, MI for the 25th, 50th, and 75th percentiles, respectively (see Table 4). Statistical testing shows the cold phase median snowfall to be significantly greater than both warm and neutral phase late winter snow totals. Although Kunkel and Angel (1999) and Janowiak and Bell (1998) showed cold phase snowfall increases in the Northern Lakes, neither showed the increases to be significant. Overall a positive impact on winter recreation is probable during ENSO cold phases.

Table 4: Late winter (FMA) 45-year climatological snowfall totals for stations in three regions with internally similar distributions.
Region Station Quartile snowfall (cm)
25th 50th 75th
Southwest Montana Helena Arpt., MT 29.6 45.2 61.8
Butte Bert Mooney Arpt., MT 43.1 59.4 84.7
Bozeman Gallatin Fld., MT 37.6 53.8 60.1
Pocatello Municipal, ID 24.5 37.6 55.0
Wyoming Sheridan County Arpt., WY 60.6 79.5 114.5
Lander Hunt Field, WY 83.8 119.9 181.4
Casper Natrona Co. Arpt., WY 69.6 87.6 106.7
Salt Lake City Intl. Arpt., UT 42.0 61.0 93.0
Northern Lakes Fargo Hector Field, ND 27.2 38.6 56.3
International Falls Arpt., MN 48.3 62.5 78.7
Duluth Intl. Arpt., MN 50.5 78.5 101.5
Minneapolis Intl. Arpt., MN 33.8 51.0 76.5
Eau Claire County Arpt., WI 31.9 41.1 63.5
Green Bay Austin Str., WI 29.8 48.8 69.3
Sault Ste. Marie Arpt., MI 71.5 91.7 131.3

   An interesting relationship occurs in the late winter distributions over Southwest Montana (Fig 4a) and Wyoming (Fig. 4b). Both warm and cold phases are associated with increased snow, relative to neutral years, at all quartiles in Wyoming; whereas, snow decreases in warm and cold phases nearby in Southwest Montana. The exception is the composite 75th percentile in Southwest Montana, which has the same percentage relative to the local 75th percentiles for the cold and neutral phases (Fig. 4a). Interestingly, the differences between neutral phase and both warm and cold phase median snowfall is statistically significant (0.95 level) in Wyoming, but not Southwest Montana.

5. Discussion

   Identifying shifts in the snowfall distributions associated with ENSO over regions of the continental United States is only the first step in improving our ability to forecast these anomalies. The other essential step is to be able to relate the changes in snowfall distributions to physical mechanisms in the atmosphere. In this section, hypotheses are presented to show how modifications in snowfall distributions can be related to changes in upper tropospheric jet streams and surface temperature anomalies that occur with most ENSO warm and cold phases. Clearly the jet streams play a role by providing dynamic uplift to produce snowfall while lower tropospheric temperatures, in part, determine the type of frozen precipitation reaching the surface. No attempts are made within this document to prove these relationships as this is an area of future research. The authors simply present plausible hypotheses, based upon previous findings, to show that regions with coherent snowfall distributions can be related to large-scale, ENSO-related, atmospheric anomalies.

   The majority of the previous work on ENSO related changes in troposphere circulations are limited to the mid-winter period. Therefore, an overview of the mid winter jet stream patterns (Smith et al. 1998) and surface temperature anomalies (Smith et al. 1999, Sittel 1994) is presented first and then related to changes in DJF snowfall distributions. Composite jet stream patterns for warm and cold ENSO phases are deduced using NCAR analyses on a 5-degree grid for the period 1947-1993 (for further details see Smith et al. 1998). The mean position of the composite jets are schematically shown in Fig. 5. Also displayed in the schematic are composite temperature anomalies derived from the GHCN results of Smith et al. (1999) and the USHCN results of Sittel (1994). Subsequent extrapolation of the mid-winter hypotheses is made for the early and late winter periods.


Fig. 5. Schematic representation of mid-winter surface temperature anomalies (1°C contours with respect to neutral years) and mean 300 hPa jet stream positions during (a) cold and (b) warm ENSO phases. Dominant jets are noted by thick arrows and weaker jets with thin arrows (adapted from Smith et al. 1998). Temperature anomaly patterns are from Sittel (1994).


   During the mid-winter, distinctly different jet stream and surface temperature anomaly patterns are present during ENSO cold and warm phases. The cold phase mid winter (Fig.5a) has a single dominant jet flowing from the Pacific Northwest, dipping slightly southward over the northern Plains, and exiting the East Coast near Delaware. Surface temperatures are slightly cooler in the western third of the nation due to clearer skies under the western ridge, while the Southeast is on average 2°C warmer than neutral phase winters primarily due to increased cloud cover associated with precipitation increases over the Tennessee Valley (Smith et al. 1998). In contrast, the mid-winter during a warm phase (Fig. 5b) has a split jet stream pattern with a dominant sub-tropical jet over the Gulf Coast and a weaker, northwest flowing polar jet sweeping south of the Great Lakes. ENSO warm phase winter temperatures are typically warmer than temperatures during a neutral phase winter over the northern states. The warming is a response to ridging in the Northwest and air masses of Pacific (not Arctic) origin entering the Northern Plains. Slightly cooler temperatures occur in west Texas, New Mexico, and Florida during an ENSO warm phase.

   Mid-winter shifts in the snowfall distributions from cold to warm ENSO phases can be related to these changes in jet positions and temperature anomalies. The Northwest and Northern Rockies snowfall regions both show increased (decreased) snowfall during cold (warm) phases (Figs. 3a,b). Hypothetically, the increased dynamics and more frequent/stronger Pacific storms associated with the jet over these regions during a cold phase will combine with slightly cooler temperatures and result in increased snow. The opposite occurs during a warm phase when the Northwest and Northern Rockies tend to be under a persistent ridge leaving these regions far from the dynamics of the jet streams (Smith et al. 1998).

   Our hypotheses for the Northeast involve temperature anomalies and modifications to the main cyclone tracks. The warmer than neutral temperatures during ENSO cold phases should play a role in reducing snowfall. In addition, the shift from a single strong jet in a cold phase to the strong sub-tropical jet during a warm phase (Fig.5) alters the dominant mid-winter storm tracks. An increase in cyclones tracking from the southwest to northeast along the East Coast has been identified in ENSO warm phases (Hirsch et al. 1999). These cyclones likely form in association with the stronger STJ at a position to the south and east of where East Coast cyclones form in a cold phase. In fact, Hirsch et al. (1999) showed that East Coast cyclones primarily occur north of Cape Hatteras during ENSO cold phases. A shift of East Coast cyclones to the south during a warm phase would place the Northeast on the cold side of the cyclone where easterly, upslope winds would increase the snowfall potential over the region. The authors also expect that the high degree of variability in warm phase winter snowfall over the Northeast is due to the fact that subtle changes in East Coast storm tracks can change Northeast precipitation from snow to rain. For example, during the ENSO warm phase of 1997, many strong East Coast cyclones occurred, but most travelled near the coast resulting in rain along the coast and heavy snowfall over the Appalachians (Kocin et al. 1999). In 1997, most of the heavy snowfall was to the west of our mid-winter Northeast snowfall region. Snowfall variability over the Northeast is low during a cold phase simply because the warmer than neutral temperatures and a more northern storm track preclude snowfall in most cases.

   The role of the jet streams and temperature anomalies in the decrease of both warm and cold phase snowfall over the Midwest (Fig. 3c) is not as clear for the mid winter season. Assuming that the majority of winter snowfall in the Midwest falls during storms that have access to Gulf of Mexico moisture, a plausible hypothesis can be developed. The warm phase decrease in snowfall can be associated with a decrease in dynamics (only a weak polar jet over the region) and limited access to Gulf of Mexico moisture. Most precipitation during a warm phase falls well to the south along the Gulf Coast (Ropelewski and Halpert 1996, Smith et al. 1998). During a cold phase, cyclones should have more access to Gulf moisture, but they also tend to track further north over Minnesota, Wisconsin, and Michigan (Kunkel and Angel 1999). The more northerly storm track draws the warm, moist air further north leading to warmer than neutral surface temperatures (Fig. 5a). As a result, the majority of the cold phase precipitation would fall as rain or freezing rain. From this simple hypothesis, a general decrease in snowfall in the Midwest is expected for both warm and cold phases.

   A mid-winter increase in snowfall over Northern Texas occurs only during ENSO warm phases (Fig. 3e). No hypothesis for this snowfall increase can be made at this time. Some atmospheric features that may play a role in the warm phase increase include enhanced moisture from the STJ, a more southerly cyclone track, and slightly cooler surface temperatures over the region (Fig. 5b). In addition, both increased upslope easterly flow and southward penetration of cold air from the Northern Plains would enhance snowfall in the region; however, the prevalence of these features during ENSO warm phases awaits further investigation.

   Early winter increases in snowfall during the cold phase over the Pacific Northwest and the Great Basin (Figs. 2a,b) are likely associated with the early development of a strong jet stream over the region. A pattern of negative temperature anomalies is present over these regions during the early winter (Sittel 1994), which would aid in the development of increased snow. The large variability in early winter snowfalls over the Pacific Northwest may result from changes in the onset time of the strong Pacific jet stream during ENSO cold phase winters. When the strong jet develops early, heavy OND snows occur over the region, while a late developing jet may lead to more typical early winter snow totals during cold phase years.

   The snowfall pattern over the Ohio Valley during the early winter (Fig. 2c) is very similar to the pattern in the Midwest during the mid-winter (Fig. 3c). The authors expect that atmospheric conditions similar to the mid-winter are responsible for the decreased early winter snows in both warm and cold phases. Warm surface temperature anomalies are already established in the early winter during a cold phase (Sittel 1994) while no appreciable warming or cooling occurs in a warm phases. One complication to deciphering the early winter snowfall in this region is the impact of lake effect snows. The role of the lakes in reducing snowfall in both the early and mid-winter is unclear at present.

   The impact of jet stream shifts and surface temperature anomalies on late winter snowfalls is not clear for the two western regions; however, a hypothesis is developed for the Northern Lakes region. During warm phases when the Northern Lakes region experiences less snow than neutral years (Fig. 4c), the region is far removed from the strong jet dynamics to the south (assuming the mid-winter pattern holds into the late winter) and temperatures are on average 1-1.5°C higher than neutral years (Smith et al. 1999; Sittel 1994). In contrast, surface temperatures in extreme northern Minnesota are 0.5-1°C cooler than neutral years during the late winter of a cold phase. In addition, cyclones following an "Alberta Clipper" storm track along the strong polar jet are more frequent during ENSO cold phases (Kunkel and Angel 1999) and likely enhance cold phase snowfall along the border between Canada and the United States.

6. Conclusions

   Regional changes in early, middle, and late winter total snowfall distributions are identified in association with warm and cold phases of ENSO. Snowfall is found to increase (decrease) during ENSO cold (warm) phases over the Pacific Northwest in early and mid-winter and over the Northern Rockies in mid-winter. These changes in the Northwest can be associated with a stronger (weaker) polar jet and colder (warmer) near surface air temperatures. Decreased snowfall relative to neutral years is noted over the Ohio Valley in the early winter and the Midwest in mid-winter. Causes for the decreased snow in both extreme ENSO phases are not clear but may be related to the availability of Gulf of Mexico moisture. Over the Northeast, more snowfall occurs in warm than in cold phases during the mid-winter, and the increase during warm phases may be associated with cyclones tracking further south and east of their typical winter track. In addition, Northern Texas receives more snowfall during ENSO warm phases than occurs in neutral or cold phase winters; however, the mechanism for this pattern is not clear at present. In the late winter, coherent snowfall patterns are found in southwest Montana, Wyoming, and the Northern Lakes. The role of jet streams and surface temperature anomalies is not clear in the western mountain regions; however, decreased snowfall in the Northern Lakes during warm phases can be associated with higher surface temperatures and the region being far removed from strong jet dynamics to the south.

   Limited non-parametric statistical testing identified significant shifts in median snowfall in the Great Basin (early winter), the Pacific Northwest and the Northern Rockies (mid-winter), and the Northern Lakes and Wyoming (late winter). Although many regions do not show statistically significant median shifts, a scarcity of long records of snowfall totals limits the number of ENSO phases studied and sampling errors are likely. The authors expect the identified regions to be modified when more seasonal samples become available. With these limitations in mind, the authors plan to identify long term snowfall records and compare them to the identified regions.

   The authors wish to add another cautionary note concerning these results. We specifically focused our analysis on differentiating between warm, cold, and neutral ENSO years. ENSO is only one oscillation that plays a role in altering tropospheric flow pattern on seasonal time scales. The North Atlantic Oscillation (Hurrell 1996) and the Pacific Decadal Oscillation (Gershunov and Barnett 1988) have been shown to influence circulation patterns. The authors expect that some of the intraphase snowfall variability is a result of these and other climate oscillations.

   Given the new knowledge of the regional changes in seasonal snowfall distributions associated with the phase of ENSO and an ever improving ability to forecast upcoming ENSO warm and cold phases, forecasters should be able to provide more accurate seasonal outlooks of total snowfall. When accepted by stakeholders, early predictions of ENSO enhanced or suppressed snowfall can aid planners to mitigate any negative impacts. On the other hand, for those where a forecast of increased or decreased snow is a benefit, plans can be made to take full advantage of the upcoming winter snowfall.

7. Acknowledgements

   The authors gratefully acknowledge the efforts of Mr. Deverraux Jones and Mr. Phil Hagen, students with the Young Scholars Program at the Florida State University, for their initial analyses of the snowfall data. The authors further thank Mr. Robert Outlaw, whose programming skills were invaluable during the initial processing of the summary of the day snowfall data. This work was funded by the NOAA Office of Global Programs under the Applied Research Center at Florida State University, which supports NCEP and the IRI with improved knowledge of seasonal to interannual climate variability. COAPS receives its base funding from the Secretary of Navy Grant to J. J. O'Brien through the Physical Oceanography Section of the Office of Naval Research.

8. References

Bhattacharyya, G. K., and R. A. Johnson, 1977: Statistical Concepts and Methods. John Wiley and Sons, 639 pp.

Gershunov, A., and T. P. Barnett, 1998: Interdecadal modulation of ENSO teleconnections. Bull. Amer. Meteor. Soc., 79, 2715-2725.

Green, P. M., 1996: Regional analysis of Canadian, Alaskan, and Mexican precipitation and temperature anomalies for ENSO impact. Tech. Rep. 96-6, Center for Ocean Atmospheric Prediction Studies, Florida State University, 104pp. [Available from COAPS, Florida State University, Tallahassee, FL, 32306-2840, USA.]

Hirsch, M. E., A. T. DeGaetano, and S. J. Colucci, 1999: The development and analysis of an East Coast winter storm climatology. Proceedings of the Eighth Conference on Climate Variations, Denver, CO, American Meteorological Society, 10-15.

Hoerling, M. P., A. Kumar, and M. Zhong, 1997: El Niño, La Niña, and the nonlinearity of their teleconnections. J. Climate, 10, 1769-1786.

Hurrell, J. W., 1996: Influence of variations in extratropical wintertime teleconnections on Northern Hemisphere temperature. Geophys. Res. Lett., 23, 665-668.

Janowiak, J. E., and G. D. Bell, 1998: Changes in the seasonal distribution of daily U. S. temperature, precipitation, and snowfall associated with the ENSO cycle. Proceedings of the 23rd Annual Climate Diagnostics and Prediction Workshop, Miami, FL, NOAA/CPC, 319-322.

Kocin, P. J., D. H. Graf, and W. E. Gartner, 1999: The 1997-1998 snow season. Weatherwise, 52(2), 62-67.

Kunkel, K. E., and J. R. Angel, 1999: Relationship of ENSO to snowfall and related cyclone activity in the contiguous United States to ENSO. J. Geophys. Res., 104 (D16), 19,425-19,434.

Ropelewski, C. F., and M. S. Halpert, 1996: Quantifying Southern Oscillation - precipitation relationships. J. Climate, 9, 1043-1059.

Sittel, M. C., 1994: Differences in the means of ENSO extremes for maximum temperature and precipitation in the United States. Tech. Rep. 94-2, Center for Ocean-Atmospheric Prediction Studies, Florida State University, 45pp. [Available from COAPS, Florida State University, Tallahassee, FL, 32306-2840, USA.]

Smith, S. R., D. M. Legler, M. J. Remigio, and J. J. O'Brien, 1999: Comparison of 1997-98 U. S. temperature and precipitation anomalies to historical ENSO warm phases. J. Climate, 12, 3507-3515.

Smith, S. R., P. M. Green, A. P. Leonardi, and J. J. O'Brien, 1998: Role of multiple-level tropospheric circulations in forcing ENSO winter precipitation anomalies. Mon. Wea. Rev., 126, 3102-3116.

Sweeny, S. R., 1996: Impact of ENSO on weather conditions at continental United States military bases. Tech. Rep. 96-2, Center for Ocean-Atmospheric Prediction Studies, Florida State University, 163pp. [Available from COAPS, Florida State University, Tallahassee, FL, 32306-2840, USA.]

Trenberth, K. E., 1997: The definition of El Niño. Bull. Amer. Meteor. Soc., 78, 2771-2777.

Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences, Academic Press, 467pp.

Yarnal, B., and H. Diaz, 1986: Relationships between extremes of the Southern Oscillation and the winter climate of the Anglo-American Pacific Coast. J. Climatol., 6, 197-219.