Because of a lack of regular, direct measurements, little information is available about the frequency and spatial and temporal distribution of icing conditions aloft, including supercooled large drops (SLD). Research aircraft provide in situ observations of these conditions, but the sample set is small and can be biased. Other techniques must be used to create a more unbiased climatology. The presence and absence of icing and SLD aloft can be inferred using surface weather observations in conjunction with vertical profiles of temperature and moisture. In this study, such a climatology was created using 14 yr of coincident, 12-hourly Canadian and continental U.S. surface weather reports and balloonborne soundings. The conditions were found to be most common along the Pacific Coast from Alaska to Oregon, and in a large swath from the Canadian Maritimes to the Midwest. Prime locations migrated seasonally. Most SLD events appeared to occur below 4 km, were less than 1 km deep, and were formed via the collision–coalescence process.
In-flight icing, including that from supercooled large drops (SLD; drops with diameter >50 μm), can pose a significant hazard to aircraft. Icing has been implicated as a contributing factor in many accidents, including that of an ATR-72 that crashed near Roselawn, Indiana, in 1994 after holding in SLD icing conditions, and an EMB-120 that crashed on final approach to Detroit, Michigan, in 1997 after descending through icing conditions (Marwitz et al. 1997; NTSB 1996, 1998). These accidents alone resulted in the deaths of all 68 and 29 people on board, respectively. Icing-related incidents and accidents continue to occur (Petty and Floyd 2004), but because of a lack of regular, unbiased, direct measurements, little information is available about the frequency, spatial, and temporal distribution of icing aloft, including that which is associated with SLD.
Research aircraft provide the most reliable information on the presence or absence of icing, and essentially the only in situ observations of SLD aloft. However, geographic coverage for these datasets is limited: the sample set is small, and it may be biased by the purpose of the flight program (i.e., to find icing; e.g., Cober et al. 1996; Miller et al. 1998; Isaac et al. 2001). Pilot reports (PIREPs) provide observations of icing, but they tend to be concentrated around major airports, where air traffic is greatest; are made most frequently during daylight hours; and do not adequately reflect the frequency of icing-free conditions (Brown et al. 1997; Kelsch and Wharton 1996; Young et al. 2002). Pilots also avoid areas of deep convection that often contain icing. Thus, PIREPs are inappropriate for developing a climatology of icing. Surface-based climatologies of freezing precipitation (a subcategory of SLD) can provide some clues, but this approach misses SLD events aloft that occur in the absence of surface freezing precipitation, and they provide no information about the vertical extent of SLD. Thus, other techniques must be used to create a more complete, unbiased climatology.
By properly combining coincident surface weather observations and vertical profiles of temperature and moisture from balloonborne instruments, the potential for icing and SLD to be present aloft can be inferred. These datasets provide more uniform spatial coverage and are available twice daily. In this paper, 14 yr of data from Canada and the continental United States will be used to create geographic and altitude distributions for icing and SLD. Month-to-month variability and the depth of SLD layers will also be discussed. Results will be compared with independent observations of icing and SLD to demonstrate the robustness of the method. They will also be placed in the context of commuter flights to describe a typical aircraft’s potential exposure to icing conditions in flight.
2. Past icing climatologies
Researchers have attempted to estimate icing and SLD frequencies aloft over parts of the globe using a variety of datasets. Icing climatologies have been constructed using balloonborne soundings (Katz 1967; Heath and Cantrell 1972; Grelson 1997), surface observations (e.g., Stuart and Isaac 1999; Carriere et al. 2000; Bernstein 2000a; Young et al. 2002), model-based icing algorithms (USAF 1986; Fowler et al. 2002; Le Bot and Lassegues 2004), ice detectors on mountains (Ryerson 1988), in situ reports from reconnaissance and research aircraft (Roach et al. 1984; Politovich and Bernstein 2002), and manually generated icing forecasts (Fowler et al. 2002). Some approaches were two-dimensional, while others included information on occurrence versus altitude or pressure. Most covered specific regions (e.g., Europe or the United States); others covered the entire hemispheres or even the globe. Results varied with the datasets and methodologies employed, location, time of year, and altitude, but the patterns were reasonably consistent among the reports. While it is beyond the scope of this document to intercompare the results of some past climatologies, comparisons between the results described here and past climatologies will be made. In general, the North American studies found the highest frequencies along the Pacific Coast from Alaska to Oregon and in a swath from the Canadian Maritimes to the Midwest.
SLD climatologies had generally similar results, except that almost no SLD occurrences were indicated along the Pacific Coast. Most of these studies relied solely on surface reports of freezing drizzle (FZDZ), freezing rain (FZRA), and/or ice pellets (PL), which strongly imply the presence of SLD aloft (Strapp et al. 1996; Bernstein et al. 1998, 2005; Stuart and Isaac 1999). While such analyses can provide a first cut at SLD frequency, location, time of year, and even time of day, SLD commonly occur aloft in the absence of FZDZ, FZRA, or PL at the surface (Bernstein et al. 2005; Cober et al. 2001). Such an approach results in an underestimate of SLD frequencies aloft, especially in areas where subfreezing surface temperatures are uncommon, as in the Pacific Northwest and southern United States. In these situations, FZDZ or FZRA aloft fall into above-freezing air, often resulting in drizzle (DZ) or rain (RA) at the surface.
One could try to include DZ and RA observations, which are common along the Pacific Coast, into a surface-based SLD climatology, but these observations do not necessarily reflect the presence of FZDZ and FZRA aloft. Most RA within the region of interest is likely to have resulted from snow melting before it reached the surface. A significant percentage of DZ at the surface is likely to have resulted from clouds that are at above-freezing temperatures throughout their depth, and a small percentage from the melting of very light snow. Thus, a surface-based analysis using FZDZ, FZRA, PL, DZ, and RA would result in an overestimate of SLD frequency aloft. Likewise, an estimation of the frequency of icing aloft that was based on surface observations of the presence of significant cloud cover would result in an overestimate, since many of these clouds would either have been glaciated or occurred entirely at above-freezing temperatures. To correctly determine whether such surface observations are associated with the likely presence of icing and SLD aloft, it is imperative to place them into the context of the thermodynamic structure in which they existed.
3. Datasets and analysis techniques
To assess the potential for icing and SLD aloft, observations from two primary data sources were used: balloonborne soundings of temperature and moisture and surface observations of cloud cover and precipitation. The combined data are examined using a version of the Current Icing Potential algorithm (CIP; now known as the “Current Icing Product”; Bernstein et al. 2005) that has been tailored to determine the potential for icing and SLD using these datasets.
Vertical profiles of temperature and moisture were derived from a National Climatic Data Center (NCDC) database of balloonborne soundings taken at 118 sites across North America between 1977 and 1990 (NOAA 1996). Quality control was applied to the data beyond NCDC standards to only include those soundings that had good temperature T and moisture data down to at least T = −35°C, reached 400 hPa, had at least 25 levels in the file, and had no superadiabatic layers. This limited the database to profiles that were of good quality, had adequate resolution, and were deep enough to reach temperatures where ice phase cloud tops could dominate. Soundings were launched at 1100 and 2300 (all times UTC). The 14-yr dataset included ∼10 000 soundings per site and ∼1 million, overall. Horizontal coverage was fairly uniform across North America (Fig. 1), eliminating most geographic biases. Station density was greatest over the contiguous United States, however, introducing some biases that will be discussed later.
b. Surface observations
Surface observations were derived from the National Oceanic and Atmospheric Administration (NOAA) Techniques Development Laboratory and NCDC Solar and Meteorological Surface Observation Network (SAMSON) archives (USDC 1994; NOAA 1993a, b,c). Both datasets spanned the period of interest and were combined to maximize the data available for analysis. At each sounding site, all observations made within a 100-km radius at the sounding launch time were considered in the diagnosis of cloud cover and precipitation type. Station elevations had to be within 609 m (2000 ft) of the sounding site to eliminate the use of potentially irrelevant observations in areas of steep terrain.
The presence or absence of clouds was determined from the maximum cloud cover reported at all stations within 100 km. If all of the stations reported either “clear” or “scattered,” then the sounding was considered to be in a “cloud-free” environment for icing purposes, since nearly all icing occurs in places where at least “broken” sky cover is reported (Bernstein et al. 1997). If any of the stations within the 100-km radius reported broken, “overcast,” or “obscured” sky cover, then the ceiling height was set to the height of the lowest deck that met these criteria (all heights MSL). Precipitation observations were checked for the presence of the following precipitation types: FZDZ, FZRA, PL, RA, DZ, and snow (SN). The number of surface stations available varied from site to site, with several sites having only one matching surface station (e.g., Inuvik, Northwest Territories), while others had 10 or more (e.g., Sterling, Virginia) within 100 km. Sites with more surface observations were more likely to capture the presence of clouds and precipitation and thus, their chances for diagnosing icing and SLD were somewhat enhanced.
To diagnose the potential for in-flight icing (ICEPOT) and SLD (SLDPOT) to exist, a special version of the CIP (Bernstein et al. 2005) was applied to the matched surface and sounding data. CIP is normally applied to real-time observations from satellites, radars, surface stations, lightning sensors, and pilot reports matched to numerical model output. The determination of ICEPOT and SLDPOT is complex, so the reader is referred to Bernstein et al. (2005) for complete details. In short, if a grid point is determined to be cloudy via satellite and surface observations, then the range of altitudes where clouds and/or precipitation exist are examined for their ICEPOT and SLDPOT. Between cloud top and cloud/precipitation base, a decision tree is used to place the column into one of five icing situations: single-layer cloud, multilayered cloud, cloud-top temperature gradient, classical freezing rain, and deep convection. The observations and model output are then run through fuzzy-logic membership functions and combined appropriately to determine the potential for supercooled liquid water to be present at sufficiently cold temperatures to form icing on a typical prop-aircraft (ICEPOT) and for a portion of those drops to fall into the SLD size range (SLDPOT). CIP calculates ICEPOT and SLDPOT at each altitude on a scale of 0.0 (no icing/SLD) to 1.0 (icing/SLD very likely). These fields are calculated independently, but share the use of numerous input parameters (e.g., temperature). A flowchart of CIP-sonde is shown in Fig. 2 and results for example cases are shown in Fig. 3.
CIP-sonde has some key differences from the model-based CIP. They are as follows:
Cloudiness is determined exclusively from surface observations of sky cover, rather than from both satellite and surface observations. At least broken sky cover has to be reported by at least one station for conditions to be considered cloudy.
If cloudy conditions are identified using the surface observations, then the cloud top is estimated to be at the highest altitude where the relative humidity (RH) criteria of Wang and Rossow (1995) are met. Their cloud layer scheme identifies cloud top as the first level in the top-down direction where either (i) RHi or RHw exceed 87% or (ii) RHi or RHw exceed 84% and the level above has RHi or RHw that is lower by at least 3%. Cloud-top temperature (CTT) is set to temperature T at cloud top. See the example in Fig. 3a.
No radar or lightning observations are used.
The deep convection scenario is not used in this study. Aircraft typically avoid areas of deep convection because of the myriad of threats that they present, including icing. Clouds in these situations are still identified, but are treated as nonconvective when determining their icing and SLD potentials. This results in a decrease in icing and SLD frequencies in areas prone to deep convection. The vertical extent of the icing and SLD is also decreased somewhat because CIP only considers a chance for icing in standard clouds down to −25°C, while icing in deep convection can occur at colder temperatures.
Identification of “dry layers” that separate cloud decks is made using dewpoint or frost point depression, depending upon temperature. A layer has to be adequately thick and dry (>2000 K m) to clearly separate two cloud layers (see the example in Fig. 3b). To assess the “dryness” of a layer, the average difference between the temperature and dewpoint at two levels in the sounding is multiplied by the difference in height between the two levels. This value is accumulated in the downward direction from the top of a given cloud layer. Once the threshold is met or exceeded, an adequately deep, dry layer is considered to be present such that condensate falling from the cloud layer will not fall into a cloud layer below. After a dry layer has been identified, the search for another, lower cloud layer is begun, using the Wang and Rossow (1995) technique. The model-based CIP finds dry layers using the presence of RHw < 50% over at least 75-hPa depth.
There is no cloud-top temperature gradient branch to the decision tree.
The RH membership function is more stringent than that applied to the Rapid Update Cycle (RUC) model profiles, since the sounding provides actual measurements rather than forecasts. This function and its resultant interest map (Fig. 4) are intended to provide an indication of the likelihood of icing based solely on RH, given that all other parameters (e.g., T and CTT) are ideal for icing.
Boosting factors based on nearby PIREPs, and model forecasts of vertical velocity and explicit supercooled liquid content were not applied because they were not available.
d. SLD icing scenarios
Two primary mechanisms are responsible for the formation of nonconvective SLD: classical and nonclassical (Huffman and Norman 1988). By determining the mechanism and examining the temperature and moisture profiles in conjunction with the observed precipitation type, CIP-sonde can diagnose the SLD potential. “Classical” SLD is diagnosed when a layer of T > 0°C (“warm nose”) is located between two layers with T < 0°C, freezing or liquid precipitation is observed at the surface, and the CTT is less than −12°C (see example in Fig. 3c). The relatively cold CTTs indicate that snow was likely to have been present above the warm nose, melted to form liquid precipitation within the warm nose, and then fell into the lower subfreezing layer to form classical SLD. The precipitation typically reaches the surface in the form of FZRA, RA, or PL, depending on the strength of the warm nose and the T and RH within and beneath the lower subfreezing layer (Hanesiak and Stewart 1995; Zerr 1997). It occasionally reaches the surface in the form of FZDZ or DZ when very light snow starts the process above the melting zone or the process is entirely nonclassical (see below).
“Nonclassical” (collision–coalescence) SLD is diagnosed using one of three scenarios. 1) Freezing or liquid precipitation is observed when a classical warm nose is present, but the CTT is greater than −12°C, indicating a good chance that an all-liquid process is responsible for the precipitation formation (Geresdi et al. 2004). 2) No warm nose is present, only RA and/or DZ are observed at the surface, and CTT is greater than −12°C. 3) No warm nose is present and freezing precipitation is observed at the surface (see example in Fig. 3d). CTT is not a factor in case 3, since the precipitation must form via collision–coalescence. SLD associated with deep convection is not explicitly considered in this study.
4. Comparison of CIP-sonde with PIREPs
A 5-yr independent dataset of matched soundings and surface observations for 1997–2001 was used to test the ability of CIP-sonde to detect icing PIREPs. In-flight icing PIREPs made within 40 km of the sounding sites were compared with CIP-sonde icing diagnoses for the period of the balloon ascent (1100–1159 and 2300–2359). Verification was done by matching all sounding altitudes within one reporting level of the PIREPs icing altitude(s) and was performed both on positive and negative (no icing) reports. All icing severities were included in the positive PIREP group.
Overall, CIP-sonde correctly identified the presence of icing for 71.3% of the 5619 positive icing PIREPs tested (PODy) and a lack of icing for 75% of the 2802 explicit negative icing PIREPs tested (PODn), using an ICEPOT threshold of 0.01. Higher thresholds had lower PODy and higher PODn and were more efficient predictors of icing conditions (had a higher ratio of PODy to the portion of all sounding levels where icing was indicated; see Fig. 5). When examining the icing PIREPs themselves, 98.3% of all positive reports occurred where at least broken sky cover was reported, and 88.7% occurred between the highest indicated cloud top and cloud/precipitation base (“in cloud”). The remaining positive PIREPs were reported at altitudes either above the highest cloud top (7.2%), below cloud base when no precipitation was reported (2.4%), or where only clear skies or few or scattered clouds were observed (1.7%). Among the 4982 in-cloud, positive PIREPs, 2.6% and 0.5% were reported to be at altitudes with T > 0°C and T < −38°C, respectively, while 4.9% occurred with RHw < 50% at temperatures between −38° and 0°C. Positive in-cloud icing PIREPs were typically associated within a narrow range of T and RHw, while positive out-of-cloud and negative icing PIREPs were more randomly distributed (Fig. 6). Note that many negative PIREPs occur at T–RH combinations where positive PIREPs are concentrated because pilots tend to report a lack of icing when they observe it in the vicinity of locations where other pilots recently reported that icing was present.
In all, 18.4% of all of the positive icing PIREPs occurred either outside of clouds and/or precipitation identified by CIP-sonde or at a T or RHw (<50%), where icing is unlikely to be present. Of course, local variations in T, RH, and cloud altitudes may account for some of these cases, but such errors are commonly due to misreported locations and mistakes in the encoding and/or decoding of the reports. A typical example of this is when a pilot calls in a report after climbing through an icing layer, and the altitude of the icing gets miscoded as at the altitude where the report was called in rather than where the icing actually occurred. CIP-sonde always indicates ICEPOT = 0 in the cloud-free, above-freezing, very cold, and/or dry situations described above. This result explains most of the difference between the CIP-sonde’s PODy of 71.3% and unity. When such questionable PIREPs are removed, CIP-sonde captured 87.3% of all positive icing PIREPs that appeared to be of good quality using an ICEPOT threshold of 0.01 (Fig. 5). A PODy of 71.4% is achieved using a threshold of 0.15. ICEPOT exceeding these thresholds was indicated at only 2.8% and 1.6% of all altitudes for all soundings at all sites combined over the 5-yr test period.
Overall, CIP-sonde achieved high PODy and PODn values, while only indicating icing over a small amount of the airspace. Comparing the PODy with the percentage of the altitudes covered by all soundings (Fig. 5) provides an indication of the efficiency of the CIP-sonde technique at each threshold. From this chart, it is clear that there are tradeoffs in PODy, PODn, and efficiency with the choice of threshold. For this study, thresholds of 0.15 and 0.40 will be used. The threshold choices are somewhat arbitrary, but the 0.15 threshold captures nearly all icing conditions, while the 0.40 threshold captures slightly more than one-half of all icing PIREPs while warning much less frequently (0.9% vs 1.6%; Fig. 5). Thus, the results using 0.15 can be interpreted as the frequency of conditions with “at least some chance” for icing or SLD to be present, while results using 0.40 can be interpreted as the frequency of conditions with a “good chance.” Though absolute frequencies change with threshold choice, patterns in the results presented in the following sections are very similar.
a. Geographic distributions for icing
Figure 7 shows the icing frequencies calculated for all altitudes combined for the entire year, using thresholds of 0.15 and 0.40. These frequencies represent the percentage of all soundings tested where ICEPOT met or exceeded the thresholds at any level in the sounding. There are two primary icing maxima over North America. The first is essentially constrained to west of the continental divide, over Alaska, western British Columbia, and the Pacific Northwest. The second maximum extends southwestward and westward from the Canadian Maritimes to the Great Lakes, Ohio Valley, and Hudson Bay. Both maxima are located where clouds and precipitation are frequently observed (Rossow and Dueñas 2004; Adler et al. 2003; Cortinas et al. 2004). Common storm tracks bring moisture into these areas. Low pressure centers are often present in the northeastern Pacific Ocean, off the coasts of Alaska, British Columbia, Washington, and Oregon, while others commonly track across the central and eastern parts of the United States and Canada (Zishka and Smith 1980). The western maximum is generally associated with topographic and frontal lift, as maritime air is advected inland, while the eastern maximum is generally associated with synoptically forced clouds, including overrunning ahead of warm fronts and widespread clouds in the wake of cold fronts. Lake effect–induced clouds as well as upslope clouds along the Appalachians also play a role in the production of icing within the eastern maximum. Storms in both regions can produce copious amounts of icing and precipitation, though the two are not necessarily coincident (Bernstein et al. 1997).
While peak frequencies in Fig. 7 may seem large, recall that they are calculated using low-to-moderate thresholds and that they indicate the frequency of occurrence of conditions that are conducive to icing (i.e., those with clouds that have a relatively ideal combination of temperature, cloud-top temperature, and relative humidity) at any altitude up to ∼10 km (30 000 ft), anywhere within 100 km of the sounding site. This covers a 314 000-km3 volume of airspace above a given station. Likewise, values for individual 1-km- (∼3000 ft) thick altitude ranges (to be discussed later) cover a 31 400-km3 volume of airspace. Thus, the percentages do not represent point or instantaneous frequencies of icing (i.e., that icing would be encountered during 60% of the time during random flight within the area of peak frequencies), which are expected to be much lower. Later in this section, it will be demonstrated that frequencies are much lower within specific altitude bands and at different times of the year, and that the frequency and pattern of icing pilot reports in the vicinity of major airports are similar.
Icing is least common over the Southwest, portions of the Deep South and along the east slope of the Rocky Mountains. While clouds and icing do impact these regions on occasion, their frequency is relatively low. For example, upslope conditions on the east slope of the Rocky Mountains can cause significant icing events (e.g., Rasmussen et al. 1995), but cloudy skies are less frequent in this area (Rossow and Dueñas 2004) as the synoptic pattern often results in westerly, downslope winds, especially during the cool season. Common storm tracks tend to lie to the north of the Gulf Coast region, so areas in the Deep South commonly lie in the warm sector of storm systems, where icing is less prevalent (Bernstein et al. 1997). Note the significant gradient in icing frequency running south and west from the Great Lakes. This feature is associated with variability in the storm track in those areas. While common storm tracks frequently bring clouds to the Great Lakes and areas to their north and east, they less frequently bring them to areas to the south and west (Zishka and Smith 1980; Rossow and Dueñas 2004).
Frequencies for select months (Fig. 8) show significant deviations from the overall full-year pattern. In general, icing conditions are at their southernmost point during January and February, with the highest frequencies found across the Great Lakes, across the Canadian Maritimes, and from southern Alaska to Oregon and northern Utah. Note that the western maximum extends southward into California in midwinter, matching the wintertime peak in storm activity there. Icing frequencies are briefly maximized in the Southeast as well, while conditions in the Arctic are too cold to support icing on most days. The eastern icing maximum migrates northward as winter transitions to spring, reaching the Arctic Coast in late spring and persisting there through early autumn. The migration pattern reverses itself during autumn, moving quickly southward into the midlatitudes. The western maximum exhibits more subtle latitudinal movement with season, but transitions are still evident, especially over Alaska and California. These migrations reflect the annual movement of storm tracks. Grelson (1997), Katz (1967), Heath and Cantrell (1972), USAF (1986), Young et al. (2002), and Fowler et al. (2002) found similar patterns and icing frequencies using independent methods.
During summer, the far north often has clouds at icing temperatures, while widespread cloudiness and strong frontal systems become a rarity in the south. Summertime icing over most of the contiguous United States is associated with deep convection. This is particularly evident over the Southwest and southern Rocky Mountains, where monsoonal moisture breeds convective activity on a regular basis (as noted in Grelson 1997). It is during this time of year that the annual maximum in icing frequency is found at sites across this region (e.g., Tucson, Arizona, and Denver, Colorado; see section 5d). Florida also has a summertime icing maximum that is associated with nearly daily convection. Of course, such convection is often spotty and is only located within close proximity to sounding sites on some days. It is important to recall that icing from deep convection is not explicitly considered in this study, and that these situations are treated the same as nonconvective situations. Because convection commonly results in the presence of clouds at temperatures conducive to nonconvective icing, relatively high icing frequencies are found within areas with high frequencies of convection, such as within the western monsoon.
b. Geographic distributions for SLD
Figure 9 shows the full-year geographic distribution of SLD frequency, using thresholds of 0.15 and 0.40. Not surprising is that the primary SLD maxima were found in similar locations to the icing maxima. The Pacific maximum was more confined to coastal regions and had frequencies on the order of 8% and 6% (thresholds = 0.15, 0.40; values that follow use the 0.15 threshold). Frequencies dropped off much more quickly with distance from the coast than they did for icing (Fig. 7), dropping to less than 4% at neighboring inland sites. Within the eastern icing maximum, SLD frequencies exceeded 4% in a swath from the Ohio River Valley, through the eastern Great Lakes, Appalachian Mountains, and northern New England to the Canadian Maritimes. Peak values exceeded 6% over eastern Newfoundland and around the eastern Great Lakes, where research aircraft have observed SLD on many occasions (Miller et al. 1998; Cober et al. 2001). Isaac et al. (2001) reported that SLD were observed during 8.6% and 13% of their flight time near Ottawa, Ontario, and St. John’s, Newfoundland, Canada, respectively, though many of these flights were made with the intent of sampling icing and even SLD environments. Minimum SLD frequencies were less than 2% across the South, Southwest, Intermountain West, much of western Canada, and the Arctic.
Monthly charts (Fig. 10) show that SLD maxima move similarly to icing maxima, with the eastern wintertime maximum roughly located on the northern side of the typical storm track and the surface 0°C line, where surface freezing precipitation is most common (Bennett 1959; Cortinas et al. 2004). SLD was relatively common across the south-central states, Appalachian Mountains, Great Lakes, Northeast, and Canadian Maritimes during winter. This maximum moved northward to the Arctic during the period between late spring and early autumn, after which the SLD maximum moved back to the south, toward the Great Lakes. Along the North Atlantic coast, near St. John’s, there was a semipermanent maximum that simply varied in intensity. Frequencies were highest there during the spring. Such results nicely match those from climatologies of surface observations of freezing precipitation (e.g., Stuart and Isaac 1999). The western maximum also had varying intensity, and was strongest during the autumn and winter. It moved latitudinally during the year, but less dramatically than the eastern maximum.
The locations of the primary SLD maxima are ones where subfreezing clouds with relatively clean source air and/or relatively large supercooled liquid water contents are common. Such conditions have been found to be conducive to the formation of FZDZ and have been observed in research flights made along the coasts of Newfoundland, the Arctic, southeast Alaska, the Pacific Northwest, and over inland areas of the Midwest and near Montreal, Canada (Cober et al. 1996, 2001; Rasmussen et al. 2002; Ikeda et al. 2007; Bernstein et al. 2004). Areas with significant maritime influence are particularly likely to have clean source air, and this nicely matches the strong SLD maximum along the Pacific Coast, as well as the relatively high frequencies over the Canadian Maritimes. The Great Lakes is not a region of significant maritime influence, but warm frontal lifting in relatively stable environments may serve to isolate the primary lifting zone from the relatively dirty boundary layer air typically associated with continental air masses. Such clean air has been associated with several continental FZDZ cases over this region (Bernstein et al. 2004) and has been observed over eastern Colorado (Rasmussen et al. 1995). Areas with relatively little SLD tend to be cut off from maritime air sources by steep orography (e.g., Intermountain West) and/or distance (e.g., northern Plains), have nearby maritime sources that are often covered in ice (Arctic), and/or have relatively infrequent or short-lived warm frontal activity (e.g., High Plains, Deep South, and Intermountain West).
c. Mechanism frequencies
Examination of thermodynamic structures found in SLD soundings showed that the nonclassical mechanism accounted for ∼92% of all SLD icing events, using the definitions outlined in section 3d. Classical SLD was most common along the east slope of the Appalachian Mountains, across New England and eastern Canada, in a swath across the south-central states, and in isolated portions of Alaska and northwestern Canada (Fig. 11). The central and eastern classical maxima are considered to be in the “freezing rain belt” where classical FZRA and PL were relatively common at the surface (Bennett 1959; Strapp et al. 1996; Stuart and Isaac 1999; Bernstein 2000a). Note the relative minimum in classical activity along the west slope of the Appalachians. The nonclassical mechanism was particularly active along the West Coast and along the east slope of the Rocky Mountains, where it accounted for nearly 100% of all SLD events. Rauber et al. (2000) found similar geographic features in their study of sounding structures associated with freezing precipitation but found that the nonclassical mechanism was present only 75% of the time. The fact that their study was limited to portions of the United States east of the Rocky Mountain states, where the classical mechanism is most common, and their more conservative definition of nonclassical events (CTT > −10°C) explain much of this discrepancy. Isaac et al. (2001) found that the nonclassical mechanism was responsible for 80% of the SLD encountered in their research flights, which were made over the eastern quarter of the continent.
d. Altitude and time–height distributions
The altitudes at which icing and SLD were most common also changed seasonally. Plots of the frequency of icing and SLD with both height and time of year for all stations combined (Fig. 12; threshold = 0.15) demonstrate this well. These frequencies represent the percentage of all soundings tested where ICEPOT met or exceeded the thresholds within a given height range (e.g., 2–3 km) in the sounding. The peak in icing (20%) and SLD (1.9%) frequency occurred during the winter months, when these conditions were most frequently found between the surface and 3–4 km. Seasonal transitions were evident for both fields. During the summer months, peak icing frequencies of 14% were found between 4 and 5 km and peak SLD frequencies of 1.2% were found between 3 and 4 km. Because of the exclusion of the “deep convection” scenario from CIP-sonde, icing and SLD frequencies at such altitudes may be underestimated, especially during the spring and summer.
It is important to note that the results for all sites combined were somewhat weighted toward lower latitudes because of the relatively low density of stations in Canada (Fig. 1). Low-altitude summertime icing and SLD tended to occur over this more sparsely sampled area, so its frequency was underrepresented in Fig. 12. A denser network of soundings over Canada would have likely broadened the low-altitude maxima into the warm season and smoothed out the seasonal transition somewhat.
The time–height distributions for many individual stations were very different from those in Fig. 12. For example, the annual pattern at Resolute Bay, Nunavut, (CYRB; Fig. 13a; see Fig. 1 for locations) north of the Arctic Circle, is the antithesis of the all-sites annual pattern. Nearly all icing and SLD appeared to occur between May and October at altitudes below 5 km. A double peak in near-surface icing was evident in June and September, matching evidence from field programs such as the Mixed-Phase Arctic Cloud Experiment (MPACE; J. Pinto 2005, personal communication). It was during these times of the year that adequately “warm” temperatures were present to support nonglaciated clouds.
The patterns at Flint, Michigan, and Caribou, Maine (KFNT and KCAR; Figs. 13b,c), were very similar to the pattern for all stations combined, as the icing maximum was located near these stations during much of the year. Notice that the prime icing season lasts until late spring at KCAR. Similar patterns were found at Quillayute, Washington (KUIL; Fig. 13d), but icing altitudes did not vary as widely. Icing and SLD rarely occurred below 1 km at KUIL, as the freezing level only occasionally drops to levels near the surface, even in winter. A great deal of drizzle and light rain is observed in this area in conjunction with warm cloud-top temperatures, indicative of a collision–coalescence process forming SLD aloft that falls into above-freezing surface air (Fig. 14a). This result demonstrates why SLD frequencies aloft are grossly underestimated by climatologies based solely on surface observations of freezing precipitation, which is rare in the Pacific Northwest (Stuart and Isaac 1999; Cortinas et al. 2004). Though Denver (KDEN) is located within the icing minimum during most of the year, its time–height chart (Fig. 13e) shows a peak in icing frequency at relatively high altitudes (5–7 km) during the summer months, when monsoonal moisture from the south is often present and brings clouds and afternoon thunderstorms to the area. Similar features are evident at Albuquerque, New Mexico, and Tucson (not shown). Grelson (1997) noted similar time–height patterns in icing at stations across North America.
Sounding-derived time–height cross sections can be compared with independent observations of icing at altitude in some cases. Around busy airports where icing would commonly be reported if it was present, time–height distributions of PIREPs can provide an alternative indication of icing frequencies. For this test, the percentage of days with positive icing reports within 100 km of several major airports was calculated for each month in the same altitude bands where the sounding frequencies were generated. Such results for Detroit (near KFNT; Figs. 13b, 15a), and Seattle, Washington (near KUIL; Figs. 13d, 15b), indicate that the sounding results are comparable to those from PIREPs. PIREP-based values are somewhat higher because these observations can be made at any time of the day (though most occur during the daytime and evening hours), rather than only at 0000 and 1200 UTC. Similarly reasonable comparisons were found at other major cities near sounding sites.
Another comparison is between observations made near the top of Mount Washington, New Hampshire (KMWN; 1917 m), and CIP-sonde results for nearby Albany, New York, and Caribou and Portland, Maine (263, 385, and 104 km from KMWN, respectively; see Fig. 1). Ryerson (1988) found that ice detectors indicated the presence of icing 38.8% of the time at mountaintop, with peaks in icing frequency during the spring and autumn and a relative minimum during winter. This roughly matches the frequencies and pattern found near 2 km at both Caribou (KCAR; Fig. 13c) and Portland (not shown), though the midwinter minimum was not as prevalent at Albany (not shown). Albany was located closer to the midwinter icing maximum that shifts away from northern New England in December (Fig. 7a). Manual surface observations taken at KMWN between 1977 and 1990 indicated that FZDZ or FZRA occurred ∼1% of the time between October and May, with peak frequencies of ∼2% during autumn and spring and a relative minimum of ∼0.5% in January and February. This nicely matches the SLD frequencies around 2 km over KCAR (Fig. 14b) and also roughly matches those for Portland and Albany (not shown).
e. Year-to-year variability
As with many climatological calculations of meteorological parameters, the overall mean can represent an average of extremes. The same is certainly true for in-flight icing, including SLD, as frequencies at given sites and over regions can vary greatly, thanks to variations in the storm track that change the location of prime combinations of temperature, moisture, and source air. Examination of monthly icing averages calculated over Denver for each year of the dataset indicates that monthly frequencies can deviate dramatically from the 14-yr average (Fig. 16a). Denver’s February icing average was 18%, for example, while frequencies for individual Februaries ranged from 4% in 1977 to 39% in 1989. During February and March of 1990, a major field program took place over the Denver area, with numerous icing events, including several with SLD and one where a plane crashed after encountering icing (Politovich and Bernstein 1995; Rasmussen et al. 1995). This period proved to be above average in terms of icing and SLD diagnosed at Denver (30% and 2.5% in 1990, as compared with 14-yr averages of 21% and 1.3%).
Large variations around the mean are common in locations where icing is relatively infrequent, while the presence of icing is more consistent from year to year at more icing-prone locations, where cloudiness and ideal icing temperatures occur with a high frequency in the cool season, year after year. The February average of 55% for Flint includes values that range from 44% in 1982 to 70% the following year (Fig. 16b). Relatively icing-free San Diego, California, had a particularly large amount of variability around its mean values, and occasionally very active winter months when the storm track impinged strongly on southern California (Fig. 16c). A good example of this is the period of nearly all-time-high monthly frequencies of 35% and 34% for San Diego during February and March 1983, as compared with the overall 19% average for these months. A strong El Niño brought well above average moisture and rainfall across that region during this period. In contrast, February and March of 1984 featured extremely dry weather and almost no icing there (3%).
f. SLD layer depths
SLD layer depths were calculated by finding the lowest (highest) altitude with values meeting or exceeding the threshold and then checking each level above (below) this until a value less than that threshold was found (e.g., Fig. 3a, where the icing layer base and top were at 603 and 959 m, respectively). Linear interpolation was used to find the level where the 0.15 threshold was met. Layer depth was the difference between the heights of the highest and lowest altitudes that met these criteria.
Using this approach, the top of the layer was often at the top of the lowest cloud deck. While nonclassical SLD often forms near cloud top, and proliferates through all levels below (as in Pobanz et al. 1994; Korolev and Isaac 2005), this may not always be the case. SLD may have only existed through a portion of the cloud deck that produced the precipitation (Korolev and Isaac 2000). The height of the top of the SLD layer was impossible to determine with absolute certainty using sounding data for nonclassical layers. Thus, the nonclassical layer depths represented the greatest possible vertical extent of the SLD and may be an overestimate. For SLD formed via the classical FZRA mechanism, the depth of the layer was easily determined from the height of the base of the warm nose and either 1) the ground or 2) the height of the next 0°C level below, if one existed. Classical SLD layer depth estimates are expected to be very accurate.
SLD layers occurred over a wide range of depths, and their distribution was dependent on the formation mechanism. The vast majority of classical SLD was based at or near the ground, and the peak in the distribution was in the 0–150-m range (Fig. 17). About 23% of all classical SLD events fell into this range, while about half were more than 400 m deep and ∼10% were more than 1000 m deep. About 2% of all classical events were >1500 m deep and the deepest classical SLD layer was 2300 m deep, using the definitions described here. Surprisingly, that event occurred at Del Rio, Texas (see Fig. 1 for location), when a deep cold pool was overrun by above-freezing air near 700 hPa.
Nonclassical SLD events occurred over a greater variety of altitudes, were generally of greater depth, and had a much broader peak in the distribution that was centered on the 600–1500-m range. About half of all nonclassical SLD events fell into this range. About 50% had depths exceeding 1000 m, ∼20% had depths exceeding 1500 m, and ∼1% were more than 3000 m deep. The deepest event was found over Frobisher Bay, Nunavut, (CYFB; see Fig. 1) and may have been as much as 5000 m deep. Recall, however, that it was assumed that SLD would have formed at the top of the deck forming the precipitation, so the depth of this and other very deep nonclassical events may be overestimated. SLD layers documented by research aircraft over Canada, the Great Lakes, and Colorado had depths that reasonably matched the distributions found in the climatology (Cober et al. 2001; Bernstein 2000b; Politovich and Bernstein 1995; Rasmussen et al. 1995).
6. Implications for commuter flights
Many short-range commuter aircraft are very likely to encounter icing conditions and are fairly likely to encounter SLD icing over the course of a given year. Such aircraft complete many short flights per day and spend a relatively large percentage of their flight time at altitudes where icing and SLD are most common. The frequency of exposure is certainly impacted by flight location. Given random flights over the United States and Canada at all times of the year, there is at least some chance of icing (threshold = 0.15) within 100 km of an aircraft during 34.6% of descents or climbs between the surface and 10 km. This value is 3% for SLD icing. These drop to 22.8% and 2.2% when the threshold of 0.40 is used. Recall that these percentages are calculated over 314 000 km3 volume of airspace over each site and that indications of the presence of the conditions could have occurred at any location within that airspace. When 1-km-thick layers are considered (31 400 km3 of airspace), the percentages are 2.2% and 0.3%, using the 0.40 threshold.
Aircraft spending most of their time flying in the Southwest, Rocky Mountains, and Deep South will encounter icing and SLD much less frequently than those flying over the Pacific Northwest, Alaska, eastern Canada or the Great Lakes. Pilots flying within the latter regions should be particularly aware of the hazard, since they are more likely to encounter it on a given flight. Still, pilots flying in relatively icing- and SLD-free areas will not avoid these phenomena altogether and should remain vigilant, since most areas are not immune to icing and can occasionally have active periods. Of course, the vast majority of icing and SLD encounters do not result in crashes; otherwise, icing-related crashes would happen frequently. A unique combination of meteorological conditions and aviation parameters must coincide for icing to contribute to an accident. The results here only represent one aspect of the meteorological part of the equation.
Given the typical altitude range for icing and SLD, most commuter aircraft encounters are likely to occur during hold, descent, or climb. Summertime encounters in the contiguous United States may occur during cruise, especially in the vicinity of deep convection and elevated layer clouds. Since the vast majority of SLD events (98% of classical and 90% of nonclassical) appeared to be less than 1.5 km deep, escape from many SLD encounters may only require a change in altitude of a few hundred meters. In the deepest cases, however, such an altitude change may not result in an exit from the conditions. Knowledge of the altitudes of the freezing level (if it is above ground), cloud top, and cloud base may prove to be critical in these situations.
7. Limitations and conclusions
The results described here were based on the inferred presence of icing and SLD aloft, derived from coincident balloonborne soundings and surface observations of cloud cover and precipitation. The techniques applied have important limitations. First, icing and SLD from deep convection was not considered explicitly. This was done because aircraft specifically avoid flight into thunderstorms because of the many hazards they present, such as hail and lightning. Thus, their inclusion would have inflated the likely potential exposure rate. Sensitivity tests for the inclusion of thunderstorms showed additional icing and SLD during the warm season, mostly over the eastern two-thirds of the United States and the monsoonal region of the Southwest above 3 km. Second, any SLD aloft that was not reflected as freezing or liquid precipitation at the surface was missed. Since the authors are not aware of a reliable method to determine the presence of such situations from climatological data, this will likely remain a shortcoming of the study.
Regardless of their depth, size, shape, longevity, and formation mechanism, when combined with the wrong set of circumstances, encounters with icing and SLD, in particular, can result in significant performance effects and even disaster. It is imperative that pilots be aware of the presence or expectation for such conditions along their flight route and to know the appropriate escape routes to allow for a quick exit from the conditions.
This research is in response to requirements and funding by the Federal Aviation Administration (FAA). The views expressed are those of the authors and do not necessarily represent the official policy of position of the FAA. Special thanks are given to Stewart Cober, George Isaac, James Riley, Richard Jeck, Eugene Hill, and other members of the Ice Protection Harmonization Working Group for their valuable feedback. Thanks also are given to Marcia Politovich and Roy Rasmussen for their reviews of the manuscript and to Scott Landolt and Jamie Wolff for help with quality control of the sounding dataset.
* The National Center for Atmospheric Research is sponsored by the National Science Foundation
Corresponding author address: Ben C. Bernstein, National Center for Atmospheric Research, P.O. Box 3000, Boulder, CO 80307-3000. Email: firstname.lastname@example.org