Because of a lack of regular, direct measurements, limited information is available about the frequency and the 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. Surface observations of freezing fog and freezing precipitation provide additional insight, but cannot be used alone to assess the presence of icing aloft. Climatologies based solely on such observations can underestimate their presence in areas where subfreezing temperatures are uncommon. Other techniques can be used in an effort to reduce some of these biases and limitations. Expanding upon results for North America reported in Part I, the frequencies of icing and SLD over Europe and Asia are inferred here using 1) surface weather observations in conjunction with vertical profiles of temperature and moisture and 2) model reanalyses of temperature and moisture. Icing maxima and minima are found to migrate seasonally, both geographically and in the vertical, and to vary in their intensity. They are linked to the location of common storm tracks and other forcing, such as that associated with sloped terrain. After establishing reasonable consistency between the methods over data rich regions, the model analyses are used to examine icing frequencies over the remainder of the globe.
In-flight icing can pose a significant hazard to aircraft and has been implicated as a contributing factor to many accidents. Drop sizes associated with icing cover a broad range and include supercooled large drops (SLD; including freezing drizzle and freezing rain), which are not covered by aircraft certification envelopes (Federal Aviation Administration 1999). SLD has been shown to cause ice to form on unprotected surfaces, sometimes causing significant degradation in aircraft performance (e.g., Marwitz et al. 1997). Several icing-related accidents that occurred over North America were cited in Bernstein et al. (2007, hereafter Part I) and such accidents have also occurred over Europe, Asia, South America, Australia, and even Africa (Aviation Safety Network 2008; Pike 1995). Despite advances in forecast techniques and ice protection systems, this litany of incidents and accidents continues to grow (Petty and Floyd 2004; Green 2006).
Several observational, icing-related datasets exist, but limited information is available about the frequency, spatial, and temporal distribution of icing aloft. Direct observations such as those from pilot reports (PIREPs) have spatial and temporal biases, while those from research aircraft can be biased by the purpose of the flight program. However, indirect measures can be used in an effort to reduce some of these biases. Using a special version of the Current Icing Product (CIP), the frequency of icing and SLD over North America was inferred using coincident surface weather observations and vertical profiles of temperature and moisture from balloonborne instruments (Bernstein et al. 2005; Part I). Similarly, Le Bot and Lassegues (2004) completed a preliminary study of the frequency of icing over Europe and the globe by applying the icing index from Météo France’s System of Icing Geographic Identification in Meteorology for Aviation (SIGMA) algorithm (Le Bot 2004) to model reanalyses of temperature and moisture. In this paper, these two methods are used to examine climatological icing features over Europe, Asia, and the globe. Though they may serve to reduce some of the biases described above, they have the potential to introduce biases of their own.
2. Past icing climatologies
Estimates of icing and SLD frequencies have been made for Europe using a variety of datasets, including in situ observations from reconnaissance aircraft (e.g., Roach et al. 1984), ice detectors on mountains (e.g., Dobesch et al. 2005), surface observations (e.g., Carriere et al. 2000; Tammelin and Säntti 1998), balloonborne soundings (e.g., U.S. Air Force 1986), and model-based algorithms (e.g., Le Bot and Lassegues 2004). Among these, only the model- and sounding-based studies extended beyond Europe to cover the Northern Hemisphere and even the entire globe, while none of them have focused on Asia, a continent with significant air traffic that is expected to increase in the coming decades.
Among the data sources used, research aircraft have provided the best quality direct observations of icing aloft, but the geographic coverage of such datasets is limited—the sample size is small in climatological terms and can be biased by the purpose of the flight program (i.e., to find icing; e.g., Amendola et al. 1998). Thus, they do not provide a full assessment of the climatology of icing conditions aloft.
Surface-based studies primarily relied on reports of freezing drizzle (FZDZ), freezing rain (FZRA), and ice pellets (PL) to assess the locations of SLD icing. Carriere et al. (2000) found that these precipitation types were rare over most of Europe, except at high elevations and in the far north. As in the North American portion of this study (Part I), it will be demonstrated here that SLD is likely to exist over portions of Europe and Asia, including those where freezing precipitation is rare at the surface. Sounding-based studies estimated the frequency of the presence of icing by testing 10+ yr of data for specific combinations of dewpoint depression and temperature (U.S. Air Force 1986). While such studies provided estimates of icing frequency over land masses, little information was derived for data sparse regions, including over the oceans. Also, the presence of clouds was inferred, without confirmation from satellite or surface observations, while important factors in the microphysical makeup of clouds, such as their cloud-top temperature (CTT), were not taken into account.
3. Datasets and analysis methods
Two methods are used here to assess the potential for icing conditions aloft. The first uses a version of the Current Icing Product that was tailored to determine the potential for icing and SLD using coincident observations from balloonborne soundings and surface stations (CIP-sonde). This method was described in Part I of this paper, but a brief description is provided here for completeness. The second uses Météo France’s SIGMA algorithm to assess the potential presence of icing from global-scale model reanalyses of temperature (T) and relative humidity (RH). This method is described in detail below.
The Current Icing Product (Bernstein et al. 2005) was developed as a multiple data source, hybrid approach to the diagnosis of icing. CIP became an official Federal Aviation Administration and National Weather Service icing product in 2002. It combines satellite, radar, surface, and lightning observations with numerical model output and PIREPs to create an hourly, 3D diagnosis of icing and SLD. CIP uses these datasets to estimate the locations of clouds and precipitation, and then combines those using physically based decision trees and fuzzy logic.
CIP-sonde similarly examines the vertical structure of the atmosphere using coincident observations from soundings and surface stations. Based on this information, each profile is described by one of several icing scenarios (e.g., classical freezing rain). Within each scenario, fuzzy logic membership functions are applied to the data and then the potential for icing and SLD is determined at each level. The icing and SLD potentials are essentially the confidence or likelihood that those conditions were present, on a scale of 0 to 1. At high potentials, the presence of icing or SLD is considered to be very likely.
In Part I, it was shown that CIP-sonde was efficient at capturing icing events while warning for icing over a relatively small volume of airspace. The threshold of 0.15 resulted in the detection of 87.3% (PODy) of good-quality positive icing PIREPs matched to soundings over North America, while only indicating icing at 1.6% of all altitudes for all soundings at all sites combined. The 0.15 threshold represented a reasonable balance between detecting the majority of icing and minimizing overwarning. Regardless of threshold choice, the geographic patterns and vertical distributions of icing and SLD were similar.
2) Input datasets
Vertical profiles of temperature and moisture were derived from a National Climatic Data Center database of balloonborne soundings from 1980 to 1994. Soundings were only included in this study if they 1) had matching surface observations that included the presence or absence of cloud cover, precipitation, and its type, 2) reached the −35°C and 400-hPa levels, 3) had at least 10 data points at or below those levels, 4) did not contain any superadiabatic layers, and 5) did not have any data that were misplaced vertically (e.g., pressure increasing with height). This limited the database to profiles that appeared to be of good quality, had information about the presence or absence of clouds and precipitation, were deep enough to reach temperatures where ice-phase cloud tops could dominate, and had adequate resolution at levels where icing typically exists. The 15-yr dataset resulted in up to ∼10 500 soundings per site. Horizontal coverage was fairly uniform across Europe and most of Asia (Fig. 1).
Surface observations were derived from a National Oceanic and Atmospheric Administration (NOAA) Techniques Development Laboratory database. At each sounding site, all observations made within a 100-km radius were considered in the assessment of cloud cover and precipitation type. The number of surface stations available varied by site, with several sites having fewer than 5 surface stations, while others had 20 or more. The presence or absence of clouds was determined using the maximum cloud cover reported at the valid time of the sounding (0000 and 1200 UTC). Only soundings matched to surface observations of at least “broken” sky cover were considered to have any chance for icing, since nearly all icing occurs when such cloud cover is present (Bernstein et al. 1997). Ceiling height was set to the height of the lowest deck that met this criterion. Surface observations were also checked for the presence of FZDZ, FZRA, PL, rain (RA), drizzle (DZ), and snow (SN), all of which provide important clues as to the presence or absence of icing, including SLD (see Part I).
b. SIGMA index
To assess the location and severity of in-flight icing over western Europe, scientists at Météo France have developed SIGMA. The goal of SIGMA is to identify the areas for which the factors supporting icing conditions are present, based on observations and experience from international icing flight campaigns (e.g., Carriere et al. 2000; Amendola et al. 1998; Cober et al. 2001), which indicated that certain observations and model-based fields were strong indicators of the presence of icing. The operational version of SIGMA combines model forecasts of T, RH, and vertical velocity with real-time observations of cloud and precipitation characteristics from satellite and radar. Real-time runs of SIGMA first examine model forecasts of T and RH to produce an icing index that ranges from 0 (no icing) to 10 (icing very likely). The SIGMA index only exceeds 0 where −15° ≤ T ≤ 0°C and RH ≥ 80%. While icing can certainly occur when the model indicates lower T and/or RH, most icing occurrences fall within the ranges described above. Output from the icing index is then overlaid with the observational data to produce a more complex analysis of icing in three dimensions and includes information on the physical icing scenario and the expected icing severity (Le Bot 2004).
As was the case for CIP, the full suite of ingredients needed for the operational SIGMA was not readily available for climatological analysis. However, because the SIGMA index only required grids of T and RH to produce its initial analysis, Le Bot and Lassegues (2004) were able to estimate the frequency of icing over Europe and the globe, using 13 yr (1989–2001) of 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) model reanalyses of the state of the atmosphere. ERA-40 analyses were created using observations from soundings, surface stations, and satellites (among others) and were placed onto a 1° × 1° grid covering the globe. Available fields included T and RH, and these fields were interpolated from constant pressure levels to constant height levels from 500 to 16 000 m (Kållberg et al. 2004). Values of the SIGMA index were calculated at each point in the ERA-40 grids valid at 0000, 0600, 1200, and 1800 UTC each day. Using this output, icing frequencies were found by calculating the percentage of time that SIGMA index values were greater than or equal to 4 anywhere in the column.
This threshold was chosen based upon experience from forecasters and researchers who examined output from SIGMA on a day-to-day basis, as well as statistical verification. Because PIREPS are difficult to obtain over Europe, SIGMA was adapted to run over the United States and southern Canada on the datasets used by the operational CIP. A large database of SIGMA runs was compared with PIREPs and results compared favorably to those of CIP (Chapman et al. 2006). They also indicated that a value of 4 provided a good trade-off between PODy and overwarning.
c. Methodology differences
Though both methods make use of T and RH data, they apply different ranges and interest maps to them. They also directly or indirectly make use of other complementary, icing-relevant data. In addition, the periods of the databases overlap, but are not the same. Because of such differences, it is expected that the results derived from these systems and databases will be similar, but will not match perfectly. Some occurrences of this are noted in the results section below. Because each method has its strengths and limitations (which are discussed in section 5), there is great value in providing results from each and examining their similarities and differences. Rather than having only one perspective on the climatology of icing, two are presented here, as well as corroborating evidence from other sources.
1) Geographic distributions of icing
Over Europe, the primary full-year icing maximum occurred in a swath from Iceland and Scandinavia southeastward into Russia and southward to the Alps and Balkans (Fig. 2a). Since much of the western portion of this feature was over the ocean, it was primarily identified by the SIGMA index, but this was consistent with CIP-sonde results for Iceland, Norway, and several North Atlantic islands. High values in this region are not surprising, given the predominance of synoptic-scale storms and cloudy skies here (Joly et al. 1999; Rossow and Dueñas 2004).
Icing frequencies decreased gradually toward the southeast, but still exceeded 35% across western Russia, and 25% as far south as Turkey. They dropped off more sharply across France and through the Alps, toward Italy, reaching their minima over relatively cloud-free southern Spain and Greece. They also dropped off quickly between Iceland and the coast of Greenland, to the north of the main storm track. A strong gradient was present from Norway southeast to the Baltic Sea and the Gulf of Bothnia (between Sweden and Finland). This was likely associated with flow across the elevated terrain of Norway and northwestern Sweden. The gradient was reversed toward the northeast, across Finland. Dobesch et al. (2005) documented similar features in their surface-based assessments of icing frequency at 100 m AGL over this region.
The primary SLD maximum found using CIP-sonde was somewhat more confined to central and northwestern Europe (Fig. 2b). The area with SLD potentials exceeding 0.15 at least 8% of the time was focused around Germany, western Scandinavia, the United Kingdom, and Iceland. Many of these areas are located downstream of the North Atlantic and the North Sea. Relatively clean, maritime air masses with low drop concentrations can move onshore in these areas, enhancing the opportunity for SLD to form there. SLD has also been observed on numerous occasions within relatively clean source air over North America (e.g., Cober and Isaac 2006; Curry et al. 2000; Ikeda et al. 2007; Bernstein et al. 2004). During an icing research program over southern Germany, severe icing from nonclassical SLD was observed on two of the eight total flight days over a 2-week period. Both events occurred in northwesterly flow off the North Sea. Icing was found on 5 days and nonclassical DZ was observed on another (Hauf and Schröder 2006).
As was found for icing, SLD frequencies decreased sharply toward Spain and the Mediterranean Sea. They also decreased toward the east from Germany, falling to <3% across most of Russia, and from the Ukraine to the Baltic States. Values also appeared to decrease toward the Arctic and Greenland, but there were relatively few sounding sites in these remote areas and the trend may be partially attributable to a relative lack of surface stations there (see below). Still, some maritime sites with relatively few surface observations (e.g., Stornoway, Scotland, and Keflavik, Iceland) had some of the highest SLD frequencies found in this study.
While maximum icing and SLD frequencies on the order of 40% and 10% may seem large, recall that they were calculated using low-to-moderate thresholds and that they indicated the frequency of occurrence of conditions that were conducive to icing and SLD at any altitude up to ∼10 km. In CIP-sonde, all locations within 100 km of a sounding site were considered, resulting in a 314 000 km3 volume of airspace above each station. In SIGMA index calculations, analyses from 1° × 1° grid boxes were checked, each of which represented on the order of 106 km3 of airspace, depending on the longitude. Thus, the icing frequencies presented here should not be interpreted as being representative of point or instantaneous values, which are expected to be much lower. Later in this section, it will be demonstrated that frequencies are lower within specific altitude bands and months. Overall, icing frequencies for Europe were similar to those found over North America in Part I, while SLD frequencies for Europe may be slightly larger.
Icing results found by the two methods generally agreed, though there were some notable exceptions. For example, both methods indicated that the European maximum extended across Germany and that values decreased toward the west and south from there, but CIP-sonde values over Germany tended to be larger than those from the SIGMA index. One possible cause for the larger CIP-sonde values was the number of surface observations matched to German soundings, relative to those in neighboring countries. To examine the possible effects of this factor, stations within the box shown in Fig. 2a were tested. Within it there were 54 sounding sites with a wide range of 1) matched numbers of surface stations and 2) icing and SLD frequencies.
The correlation coefficient between the number of stations matched to each site and the annual icing frequencies found at those sites was 0.02, indicating no correlation whatsoever. For SLD, the correlation coefficient was somewhat higher (0.30), but still fairly weak. These results do not explain what appear to be elevated CIP-sonde icing frequencies over Germany, but they do hint that SLD frequencies over Germany and other countries where the surface network was relatively dense (see Fig. 3) may be somewhat elevated. In contrast, SLD frequencies in portions of the Arctic may be underestimated because of the relative lack of surface stations there.
Frequencies of surface observations of freezing precipitation and subfreezing fog were relatively large over Germany and bordering areas of France and Benelux compared to surrounding areas, especially to the west (Fig. 3; Carriere et al. 2000). As will be discussed later, frequencies from German mountain stations roughly matched CIP-sonde icing estimates at those elevations from nearby sounding sites, including seasonal trends. These results suggest that the elevated CIP-sonde frequencies around Germany may be reasonable.
2) Surface indications
As expected, locations with high icing frequencies also had a high frequency of significant cloud cover in historical surface observations from 0000 and 1200 UTC (Fig. 3). Strong gradients in the frequency of clouds were evident across topographic features such as the Alps and Pyrenees. SLD-prone portions of Europe were quite cloudy and had a high frequency of DZ and RA, but most experienced little or no FZDZ, FZRA, or PL. Approximately 97% of the SLD that CIP-sonde identified in Europe occurred when DZ and/or RA was coincident with warm cloud tops (>−12°C), a combination that is commonly associated with the presence of SLD aloft. This value is somewhat larger than the 92% value found for North America in Part I and suggests that the “classical” SLD mechanism is even less common over Europe.
Because surface FZDZ and FZRA were rarely observed across most of Europe, very little of the SLD that CIP-sonde identified appears to have reached the surface. With the exception of northeastern Europe, FZDZ and FZRA were only observed with significant frequency at stations located at elevations above ∼1 km MSL (as in Carriere et al. 2000). This matches the lack of SLD below the 1-km level across most of Europe, discussed in the next section.
Like freezing precipitation, freezing fog (FZFG) tended to occur at elevations >1 km (Carriere et al. 2000). However, Makkonen and Ahti (1995) showed that the relief of the local terrain is more critical than elevation in determining the frequency and severity of icing at surface sites. In either case, an observation of FZFG is a good indicator of icing at the ground. But close examination of surface reports indicated that FG at T < 0°C was often not indicated as FZFG. Though a portion of such “cold fog” (hereafter CFG) observations was actually “ice fog,” most was likely to have been at least partially composed of supercooled liquid water (SLW). Overall, 85% of the 116 763 CFG observations used to make Fig. 3d occurred at T > −10°C, where SLW is expected. Thus, it is reasonable to use frequencies of CFG as surrogates for icing frequencies at the ground, though they likely represent somewhat of an overestimate.
It is useful to compare the frequencies of CFG and freezing precipitation from surface stations in elevated terrain with CIP-sonde icing and SLD frequencies at nearby sites for the same times (0000 and 1200 UTC). To eliminate the possibility that the data from the surface stations were used by CIP-sonde, surface stations used were always more than 100 km from all sounding sites. This distance complicates matters somewhat, since regional- and local-scale phenomena can affect the occurrence of icing and SLD. One-to-one correlations were not expected, since surface stations that provide point observations were being compared with sounding estimates from a 100-km radius cylinder. Also, FZDZ and FZRA only represent a subset of the SLD drop size spectrum, so their frequency at the surface should underestimate the frequency of SLD at the same altitude.
Still, comparisons proved to be quite favorable in many parts of Europe, especially between CFG and icing. This was true both in terms of overall annual values and seasonal variations both in areas with relatively high (e.g., Germany, Bulgaria, Romania, and the United Kingdom) and low (e.g., Italy, Spain, and Portugal) icing frequencies. For SLD, reasonably good matches with freezing precipitation were found in several regions with high frequencies (Germany, Ukraine, Poland, Yugoslavia), as well as relatively SLD-free areas (Spain, southern Italy). For example, CFG and freezing precipitation frequencies from Kleiner Feldberg (elevation of 805 m) were compared to the average of CIP-sonde icing and SLD frequencies surrounding that altitude at Meiningen, Germany. Results showed that surface frequencies for icing and SLD were 64% and 34% of those from the soundings. The favorability of such comparisons varied by location.
3) Seasonal movement of icing and SLD aloft
Monthly maps show that the icing and SLD maxima and minima moved during the year (Fig. 4). In January, icing frequencies exceeding 35% were found across much of Europe, especially its northern and eastern portions, and extended southeast through the Balkans to northern Turkey. Some places with notably high frequencies included Iceland, the northern United Kingdom, and areas extending eastward from Norway, eastern France, and the Balkans. Icing was indicated >25% of the time across the southern United Kingdom, western France, and northern Spain. Northern Spain was the site of a research flight campaign that featured numerous cases of both stratiform and cumuliform icing, with nearly all of it occurring within the temperature ranges covered by the SIGMA index and CIP-sonde (Bain and Gayet 1982).
A small amount of icing was also indicated over Italy, southern France, and Spain, as well as portions of northern Africa. Baddour and Rasmussen (1989) documented icing events over the mountains of Morocco during this time of the year. Note that January results from the two methods matched reasonably well, including many details such as the strong gradients found near the Adriatic Sea as well as across France and the United Kingdom. As shown earlier for annual results, there were some disparities in the monthly results, especially around Germany, where CIP-sonde values generally exceeded those of the SIGMA index.
In the spring, the icing moved northward with the storm track and by July most of it had shifted to the Arctic, from northern Russia and Norway to Iceland. Still, both indices found that values exceeding 15% remained over the United Kingdom and from Germany eastward. By October, the icing maxima have begun to move south and west toward their cool season centers. Again, CIP-sonde values were relatively high over Germany and its vicinity, but frequencies matched quite well across most of the continent.
January SLD frequencies were maximized across western Europe, including the United Kingdom, southern Scandinavia, Benelux, and Germany, into Poland. SLD frequencies exceeding 5% extended down to the north coast of Spain, southern France, the Alps, and Balkans. Following the icing pattern, the SLD maximum moved northward during the summer, and returned southward in the fall. Maxima found in places like the United Kingdom and Germany remained somewhat intact throughout most of the year, though frequencies waxed and waned. Maximum values occurred during the peak of the winter, when particularly cloudy, drizzly conditions tended to exist in these areas. It is interesting to note that SLD was somewhat common along the west coast of France and north coast of Spain from October to April, and then became very uncommon from June to September. In contrast, Arctic stations had most of their SLD between late spring and early fall, with relatively little during the winter, when clouds were more likely to be glaciated. Frequencies over much of Scandinavia tended to peak in the autumn, except in its southern sections, where the peak tended to last into January. Vedin (1998) found similar patterns in a surface-based climatology of freezing precipitation compiled for Sweden.
4) Vertical distributions and examination of individual stations
The altitudes at which icing and SLD occurred most frequently also changed during the year, as demonstrated in plots of their frequencies versus height and time of year for all European stations combined (Figs. 5a,b). Both techniques indicated a similar time–height pattern, though CIP-sonde values were generally somewhat larger than those from the SIGMA index. Icing altitudes generally rose in the spring and dropped in the fall, following the freezing level. The peaks in icing (20%–40%) and SLD (4%–7%) frequency occurred during winter and late fall, when they were most often found between 1 and 2 km. This pattern matches the relative lack of freezing precipitation and CFG at surface stations below ∼1 km. Icing extended upward to slightly higher altitudes than SLD because CIP-sonde only diagnosed nonclassical SLD when CTTs ≥ −12°C were present and only attributed SLD to the lowest cloud layer when multiple layers were identified. Icing potentials exceeding 0.15 were possible with colder CTTs and thus, colder temperatures, which were found at higher altitudes. Seasonal transitions were quite evident, as values decreased and moved to higher altitudes during the spring, then moved downward in the autumn, and then values increased to reach their maximum at 1–2 km from November to March.
Time–height distributions for individual stations sometimes had patterns that were very similar to the European average, while others were noticeably different. A reasonably good match was present at Trappes, France (near Paris; Fig. 6a), near the western edge of the continental icing maximum. To the south, Rome, Italy, had less frequent icing, with most of it during the cool season (Fig. 6b). Toward the southeast, patterns at Sofia, Bulgaria, (Fig. 6c) were fairly similar to those found over Paris and icing was found to climb from 1–3 km in winter to above 4 km in the summer. Nikolov and Moraliiski (2000) found similar seasonal, time–height patterns in icing at Bulgarian mountain stations between 0.8- and 3.0-km elevation. Point frequencies at these sites typically peaked at 12%–22%, which are somewhat smaller than the 20%–30% values inferred by CIP-sonde and the SIGMA index at 1–3 km. This serves as a good example of the difference between point observations and the larger-scale estimations made using the two icing algorithms.
To the northwest of the continental maximum, icing frequencies were generally maximized in autumn, winter, and spring, with a general increase across the United Kingdom toward Iceland. Moderate and high values were fairly persistent from autumn through spring over Belfast, United Kingdom, and Keflavik, Iceland (near Reykjavik; Figs. 6d,e). Icing at Keflavik was typically maximized in the 1–2-km altitude range throughout most of the year, only rising into the 2–4-km range during summer. Similar patterns were found over northern Scotland, and the nearby Shetland and Faeroe Islands (not shown). Oslo, Norway, to the southeast, had a pattern that reflected aspects of both the northern sites and the continental maximum (Fig. 6f). Hirvonen et al. (2007) reported that a Rosemount ice detector at the Luosto Fell site (well to the north of Oslo; elevation of 0.52 km) was prone to heavy icing between October and April. This observation matches the low-altitude, seasonal pattern at Oslo and other Norwegian sites (not shown) quite well.
b. Asia and neighboring portions of Russia
1) Geographic distribution
This portion of the study focuses on eastern Asia and neighboring portions of Russia. Unfortunately, very little data were available for large portions of southwestern Asia, so those areas will mostly be discussed in terms of global patterns (next section). The primary Asian icing maxima were found in three swaths. The northernmost of these is an eastward extension of the European maximum, continuing over much of northwestern Russia, with values decreasing gradually toward the east to 20%–35% over Siberia (Fig. 7). The central maximum runs eastward from the Himalayan Mountains, across central China and South Korea to portions of Japan, then continues northeast into the Pacific. Full-year frequencies within this maximum were mostly on the order of 25%–45%, with higher frequencies along the Himalayas. Only the SIGMA index had sufficient data to detect the terrain-driven maximum there, because of a lack of sounding data. It was in this area that several icing accidents occurred during World War II (Aviation Safety Network 2008). A third maximum was found to the south, over Indonesia and Malaysia, and was associated with persistent cloudiness and showers in the intertropical convergence zone (ITCZ). There was only one SLD maximum and it was collocated with the central Asian icing maximum, from southeastern China to western Japan, and had frequencies of 5%–8%. Most stations in Asia had SLD (exceeding 0.15) less than 3% of the time. Overall, peak annual icing and SLD frequencies over Asia were lower than those found over Europe and North America.
As noted earlier, icing was found most often in places where at least broken skies were common and the clouds have a combination of ideal temperatures within them and at their tops. It is not surprising to note that the locations of the three icing maxima roughly match the areas with peak frequencies of cloudy skies (Fig. 8a). Though not observed as frequently as at some European sites, CFG was commonly reported at mountain stations within the central icing maximum, including Japan, China, and Taiwan. Several stations had CFG > 5% of the time (Fig. 8d) and had seasonal patterns that closely matched those from nearby sounding sites.
SLD maxima tended to occur where this combination was collocated with clean source air and/or the classical freezing rain process. Across Asia, the primary maritime air sources are the western Pacific Ocean and the large seas on its western edge. FZDZ at the surface was rare across Asia, except for at a few stations in southeastern China, where it was found in 0.2%–0.5% of all observations. Note that FZDZ frequencies decreased toward the east in Russia and were near zero over most of eastern Russia, matching the results of Berzukova et al. (2006).
FZRA was also rare across Asia, except in parts of southeastern China, where it was present in 1%–5% of all 0000 and 1200 UTC observations and has been reported to cause heavy icing on power lines there (Fig. 8c; Li et al. 2008). Most of the stations with significant FZRA were at elevations exceeding 1 km, though a few were in the lower terrain to the east. It was in this lower terrain, toward the east side of the southeast China FZRA maximum, that PL occurred most often, but with much lower peak frequencies of 0.2%–0.4%. These precipitation types and the classical FZRA process appear to have been important contributors to the presence of a pronounced SLD maximum over southeastern China (Fig. 7b). Overall, the classical structure was present in 10%–25% of all SLD cases in this region, compared to 5% for all of the Asian sites combined. Comparisons between frequencies of freezing precipitation at surface sites and CIP-sonde SLD frequencies at the same heights were reasonably good throughout southeastern China.
As was found for Europe, monthly maps indicate that there are seasonal movements to Asia’s icing and SLD maxima (Fig. 9). In January, April, July, and October, all three of the primary icing maxima were present, but their locations and intensities varied by month. During January, the northern icing area had a strong gradient from western Russia to Siberia. Siberia appeared to have almost no icing at this time of the year, because of bitter cold temperatures (<−30°C at the surface; Hydrometeorological Center of Russia 2008). Note that SIGMA index values tended to be larger than CIP-sonde values over western Russia, but similar toward Siberia.
The central maximum was fairly intense from the Himalayas across southeastern China and Korea to western Japan. Regional features embedded within this included north–south gradients across China, a secondary gradient across Korea, and a strong west-to-east gradient across Japan that was probably associated with the mountains there. The southern icing maximum was also quite active, especially between 10°S and 5°N, while there was a pronounced minimum between 10° and 25°N, where relatively few clouds tend to exist in winter (Rossow and Dueñas 2004).
With the warming of spring and summer, the central icing band weakened (except over the Himalayas), while the northern band shifted toward the north and northeast, and reached the Arctic by July. Arctic values peaked at more than 50% in midsummer, then decreased rapidly as cold air settled in during late fall and the main icing maximum moved southward. Monsoonal clouds and their associated icing also became quite evident in a swath from India and Burma to southwestern China in July. This appears to have been associated with a northward movement of the equatorial band of high-altitude icing. Summer was the only season when an adequate number of soundings was available for CIP-sonde to estimate frequencies in this region and comparisons between results from the two systems were favorable.
Seasonal patterns in SLD frequencies were similar to those found for icing. During January, the largest SLD frequencies (>7%) were present over the same region as the eastern portion of the central icing maximum and the locations where FZRA and PL were most common at the surface. The SLD maximum also extended from southeast China across Taiwan to western Japan. As part of their climatology of surface observations, Matsushita and Nishio (2004) found FZRA and PL in the ravines of western Japan during December and January. However, the classical FZRA structure was present in <5% of all Japanese SLD soundings. Based on this, and the fact that both FZDZ and DZ are rarely observed in this region, it seems that most of the SLD identified over Japan was associated with warm cloud tops and light rain at the surface.
Another strong gradient in SLD frequency ran from south to north across China. Both SLD maxima moved to the north and northeast during the late spring to briefly impact southeastern Russia, continued north to the Arctic by July, and returned southward in October.
2) Vertical distributions
As noted for Europe and North America (Part I), the altitudes at which icing and SLD occurred most frequently changed during the year over Asia, and the patterns could be different from region to region and station to station. Plots of their frequencies with both height and time of year for all Asian stations combined did not demonstrate this well, so the combined plots have been broken down into latitudes to the north and south of 25°N latitude (Figs. 5c–f).
Typical seasonal transitions in the northern latitudes were fairly evident, though weaker than those found over Europe and North America. Maximum icing frequencies were mostly found below 4 km from October to April, and in the 3–7-km range from June to August, with transition periods in May and September. Values and patterns from the two techniques proved to be quite similar. The southern icing pattern was quite different, due to tropical temperatures and persistent clouds found within the ITCZ, as well as monsoonal moisture over southern mainland Asia. Icing was mostly evident above 5 km, with a persistent maximum in the 5–7-km altitude band. While patterns from the two techniques matched reasonably well, the SIGMA index primarily indicated the presence of icing below 7 km, while CIP-sonde had icing up to 9 km. This is mostly due to the difference in the temperature ranges for the thresholds used by the two techniques. Some of the icing extended down to slightly lower altitudes during winter.
SLD frequencies were found to be quite low (<3%) in both latitude bands, but some SLD was still indicated at relatively low altitudes during the cool season and higher altitudes during the warm season. It is important to note that confidence in high-altitude SLD assessments is relatively weak because it is difficult to relate surface precipitation to such high-altitude clouds. While there is certainly potential for SLD to exist within these clouds, the methodology used here is probably inadequate to address a good portion of it.
Time–height distributions for individual stations revealed a fairly typical seasonal pattern at Taipei, Taiwan, and the more inland site of Guiyang, China (Figs. 6g,h). The results from these sites represent a transition from a southern latitude pattern to a more central Asian pattern, where cold season icing was common at lower altitudes. Frequencies at Taipei City were highest from January to June, as they climbed from 3 to near 7 km. On 21 December 2002, an ATR-72 lost control and crashed just to the west of Taiwan after encountering severe icing that included SLD at 5–6 km (Aviation Safety Council 2005). These altitudes were within the normal range for icing at this site for late December, which appears to be near the typical start to the main icing season over Taiwan. Guiyang, located within the southeastern China icing and SLD maxima, had peaks in icing and SLD frequencies from December to February, especially in the CIP-sonde results. SLD frequencies (not shown) exceeded 10% at altitudes below 2 km in January and February, the same months in which FZRA, PL, and RA occurred most often in this region. The seasonal change in altitude of icing was quite large at Guiyang and this feature was evident at numerous sites in east-central Asia, including over Japan.
Over southwestern and western Japan, icing and SLD appeared to be relatively common. Along the Sea of Japan, Wajima had a distinctive maximum below 3 km from November to March, when icing and SLD frequencies exceeded 25% and 6% (Figs. 6i, 10a). To the east, Tateno (near Tokyo; Figs. 6j, 10b) had similar seasonal patterns, but the wintertime frequencies were roughly one-quarter of the magnitude of those found at Wajima and they were slightly more elevated. This was likely due to an upslope–downslope couplet across the mountains of central Japan, with more frequent cloudiness and precipitation on the upwind, west side. Murakami et al. (2007) noted the frequent presence of clouds dominated by supercooled liquid water on the upslope side of west-central Japan in November and December, and a decrease in SLW occurrences to the east. Stratocumulus clouds sometimes form over the Sea of Japan as cold air blows across it (e.g., Inoue et al. 2005). When formed at the right temperatures, such “sea effect” clouds are likely candidates for icing. Heavy icing has been documented in similarly formed lake effect clouds over North America (Bernstein 2000).
c. Global icing patterns
An examination of global patterns provides additional perspective on the results already described for the Northern Hemisphere here and in Part I, as well as a look at the patterns of the Southern Hemisphere. For example, the full-year, global icing chart (Fig. 11) indicates that the maxima found over Alaska and southeastern China were inland extensions of a broad icing maximum present over the North Pacific. Similarly, a broad swath of high icing frequencies connects the maxima found over the Canadian Maritimes to those over northwestern Europe. These areas tended to have active storm tracks and cloudy skies (Rossow and Dueñas 2004).
In general, Northern Hemisphere icing tended to be most common in a somewhat zonal band poleward of about 40°N latitude, with broad meridional variations superimposed. These variations appear to be tied to the large-scale storm tracks and their interaction with major landmasses. They were quite evident in January (Fig. 12a), when the northern jet stream is strong and there tends to be sharp delineation between polar and midlatitude air masses. There was also a fairly distinct southwest–northeast orientation to the icing maxima over both the Atlantic and Pacific Oceans, from the Great Lakes to northwestern Europe and from southeast China to southern Alaska. Large mountain ranges like the Rocky Mountains disrupted the pattern via the impact of their large-scale upslope and downslope forces on cloudiness. Other large terrain features like the Alps and Himalayas also tend to have large changes in icing frequency across them since they act as barriers to flow on a regional scale.
The meridional nature of the Northern Hemisphere pattern weakened slightly in April, as the storm track moved northward. It became almost nonexistent in July, when most of the icing tends to be confined to the Arctic (Figs. 12b,c). This matches the timing of icing found at far northern latitudes during field programs. As the Arctic quickly cools in the fall and the storm track pushes southward, meridional deviations are reestablished by October (Fig. 12d).
In the tropics, high-altitude icing was generally found within 20° of the equator and moved latitudinally throughout the year. This relatively weak icing belt occasionally made excursions toward the midlatitudes via summer monsoons, such as those seen over southern Asia and southwestern North America in July, as well as over northern parts of Chile and Brazil in January. A relatively weak icing maximum moved northward from southern Africa in January to central Africa in July.
In the Southern Hemisphere, icing was quite latitudinal in nature, with the primary seasonal changes mostly affecting frequency. Icing tended to be most common between 50° and 70°–80°S, with the southern limit found along the coast of Antarctica. In the Western Hemisphere, the 50°–80°S latitude band is dominated by ocean, with the exceptions of southern Chile and Argentina and the Antarctic Peninsula, just to the south. The southern part of Chile is on the windward side of the Andes Mountains, causing it to have upslope flow downwind of a long fetch over the Pacific Ocean. Its icing frequencies were quite high relative to those across much of southern Argentina, which is on the leeward side of the Andes. Southern Chile’s elevated icing frequencies were semipermanent, but maximized in July—the peak of winter there. This fact is well known within the aircraft icing community and several aircraft manufacturers have traveled to this region to perform natural icing tests for certification when schedules necessitated flight at this time of the year. While icing has certainly been found in this area, SLD has also been found on many occasions (e.g., Burick and Ryan 1999), and is likely associated with upslope forcing of highly maritime air.
Icing over most of Antarctica was only evident in January, the peak of their summer, when temperatures could warm to the point where supercooled liquid water was more likely to exist. Limited observations taken there have shown the existence of SLW during the brief summer (December–January) and into the fall (February–March), when it was observed at unusually cold temperatures (T < −30°C; Ellison et al. 2006) not covered by this icing climatology. In the eastern portion of the Southern Hemisphere, icing was limited to between 50° and 70°S, following the more northern extent of the Antarctic coastline there. It reached as far north as southern New Zealand, southeastern Australia, and Tasmania, with frequencies peaking near 35% during July, then receded southward by October. Such icing frequencies did not quite reach southern Africa, even in July.
5. Limitations and conclusions
The results described here were based on the inferred presence of icing and SLD aloft derived from 1) an icing algorithm run on model-based analyses of profiles of T and RH and 2) coincident balloonborne soundings and surface observations of cloud cover and precipitation. Though not compared in detail, the Northern Hemisphere results of this study are consistent with Part I of this paper, as well as sounding-based icing climatologies compiled by Katz (1967), Heath and Cantrell (1972), and the U.S. Air Force (1986). These studies all found similar overall patterns and that icing migrated latitudinally and vertically with season. To the authors’ knowledge, there was no comparable study in the literature for the Southern Hemisphere results presented here.
The techniques used in this study have important limitations that affect the results described above. The SIGMA index and the version of CIP-sonde applied here did not take deep convection into account. 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. This choice is likely to have caused an underestimate of actual icing frequencies in areas prone to thunderstorms, especially at relatively high altitudes.
Both systems used somewhat coarse spatial data, so they are likely to have missed fine details associated with topographic features and land–sea interfaces. This is especially true for CIP-sonde, which had very little data in mountainous regions and away from mainland portions of continents. The potential for such large variations in icing frequency was demonstrated by Byrkjedal (2008) using finescale model predictions over the complex terrain and coastline of western Norway.
Frequencies shown here were calculated using low-to-moderate thresholds. This allowed the majority of icing and SLD situations to be captured, but may have resulted in an overdiagnosis in cases when conditions for icing and/or SLD were somewhat marginal. Increases (decreases) in threshold choice would have resulted in decreases (increases) in frequencies. Guan et al. (2001) reported that the application of an icing algorithm based solely on T and RH (Appleman 1954) to model output resulted in significant overforecasting and a lack of skill. Though the SIGMA index is based on T and RH, because it was applied here to model reanalyses that employed an abundance of observational data sources, it is expected that the overestimation of icing frequencies would have been reduced. While no statistical verification of the SIGMA index was done for this paper, the seasonal frequencies and patterns that it generated over North America were quite similar to those from CIP-sonde (see Fig. 12). As noted in section 3a and Part I, CIP-sonde was shown to capture 87.3% (PODy) of “good-quality” positive icing PIREPs while warning for a small volume of air space.
In CIP-sonde, any SLD aloft that was not reflected as freezing or liquid precipitation at the surface was missed. Since the authors were not aware of a reliable method to determine the presence of such situations from historical data, this will remain a shortcoming of the study. Also, a 100-km-radius circle was used to find surface reports in the vicinity of balloon launches. Appropriate precipitation types were almost always observed by a subset of the stations within the circle. To a lesser extent, this was also true for cloud cover. Thus, only a portion of the 100-km-radius cylinder was likely to have contained icing or SLD. One must consider this when estimating the actual chances that an aircraft flying over a given location would be exposed to these conditions.
CIP-sonde and SIGMA index results were based on data from 0000 and 1200 UTC and from 0000, 0600, 1200, and 1800 UTC, respectively. Thus, SIGMA results covered the diurnal cycle more thoroughly. Brown et al. (1997) demonstrated a diurnal pattern in the frequency of icing PIREPs and surmised that much of this was driven by the daily cycle of air traffic. Diurnal patterns in the frequency of FZDZ and FZRA have been found at the surface (e.g., Cortinas et al. 2004), and may also be present aloft. The potential influence of the diurnal cycle on the results has not been addressed here. The same is true for the frequencies of FZDZ, CFG, and so on, which were limited to surface observations from 0000 and 1200 UTC for the purpose of comparison to sounding data taken at these times.
Regardless of their frequency, depth, and mechanism, when combined with the wrong set of circumstances, encounters with icing and especially SLD can result in significant performance effects and even disaster. It is imperative that pilots be aware of the presence or expectation of such conditions along their flight route and know the appropriate escape routes to allow for a quick exit from the conditions. Climatological information, such as that presented here, can provide a sense of where these threats are mostly likely to exist. Aircraft manufacturers, on the other hand, may find this information useful for the planning of icing flight tests. In both cases, the user of this information must bear in mind that it was based on large amounts of historical data and that results can vary widely from year to year and location to location, as noted in Part I.
This research is in response to requirements and funding by the Federal Aviation Administration through the FAA Technical Center in Atlantic City, New Jersey, and funding supplied by Météo France. The views expressed are those of the authors and do not necessarily represent the official policy or position of the FAA or Météo France. The authors thank Cory Wolff for his help with figures, Frank McDonough for his work on CIP-sonde code and data processing, Kyoko Ikeda for her help with access to Japanese literature, Pierre Lassegues for preparing the SIGMA index data, and Agathe Drouin, Patrick Josse, and Christian Page for their help with figures and their insights on the paper. Thanks also are given to Stewart Cober, George Isaac, James Riley, Richard Jeck, Eugene Hill, Thomas Ratvasky, Thomas Bond, Lasse Makkonen, and the anonymous reviewers for their valuable feedback on the results of this work.
* Current affiliation: Leading Edge Atmospherics, Longmont, Colorado.
Corresponding author address: Ben C. Bernstein, Leading Edge Atmospherics, 3711 Yale Way, Longmont, CO 80503. Email: firstname.lastname@example.org
# The National Center for Atmospheric Research is sponsored by the National Science Foundation.