1. Introduction

Freezing rainstorms are among the most severe winter weather events. Surface ice accumulation can halt air and ground transportation, weigh down and snap power lines, severely damage trees, and cause numerous traffic and pedestrian accidents. Forecasters closely monitor storms producing freezing rain; however, the mesoscale nature of freezing rain events makes their location and severity inherently difficult to forecast. A particularly difficult problem of weather forecasting is distinguishing regions that will receive freezing rain from regions that will receive ice pellets. Czys et al. (1996, hereafter CZ) reported a method to diagnose the boundary between freezing rain and ice pellets using a simple nondimensional parameter. An advantage of the CZ method is its simple application in the forecasting environment;easily measured quantities from either actual or forecast model-generated soundings can be used along with values of a nondimensional parameter obtained from a nomogram (Fig. 1) to map the boundary between ice pellets and freezing rain. Czys et al. (1996) suggested that after further testing their method could be combined with mesoscale model data. Forecast maps delineating the freezing rain–ice pellet boundary could then be generated. To derive the parameter, CZ assumed the atmosphere had a three-layer structure similar the sounding shown in Fig. 2, with 1) a cold (T < 0°C) cloud layer aloft that is a source of ice particles, 2) a midlevel layer where the temperature exceeds 0°C and ice particles melt, and 3) a surface layer where T < 0°C. If ice particles from aloft completely melt into raindrops within the warm layer, the drops will supercool in the surface layer and reach the ground as freezing rain. If ice particles from aloft partially melt in the warm layer, they will refreeze in the surface layer and reach the ground as ice pellets.

Czys et al. (1996) suggested that the parameter τ, defined as the ratio of the residence time of an ice particle in the warm layer (t_res) to the time required for complete melting (t_melt), might be useful to discriminate regions of freezing rain and ice pellets in forecasts. (The terminology used here follows CZ.) Czys et al. estimated t_res by considering the depth of the melting zone (ΔZ_w) and the velocity at which a particle falls through the layer [U(a) − V]. Here U(a) is the particle terminal velocity, a function of drop radius a, and V is the ambient air vertical velocity. The time t_melt was estimated by considering a balance between the rate of energy required to transform the solid to liquid and the rate at which the environment can supply the needed energy. The net result for τ was

View Expanded

where k_w is the thermal conductivity of water, L_sl is the latent heat of melting, ρ_i is the density of ice, r is the radius of the ice core in the drop, T_a the temperature of the particle–environment interface, and T₀ the temperature of the ice–water interface, assumed to be 273 K.

Since particles occupy a range of sizes in clouds, CZ considered the critical radius, a_c, to be the radius of the largest particle falling through the warm layer. If τ = 1 for this particle, τ > 1 for all other particles and the precipitation will fall as freezing rain. Otherwise, the precipitation will be a mixture of freezing rain and ice pellets, or ice pellets alone. To determine a representative value for a_c, CZ considered radar reflectivity measurements from the Valentine’s Day ice storm of 1990 (Martner et al. 1992, 1993; Rauber et al. 1994; Rasmussen et al. 1995). They further assumed an exponential size distribution for the particles with the parameters for the distribution taken from Marshall and Palmer (1948) and Gunn and Marshall (1958). For the 1990 ice storm, a value of critical radius a_c was found to be 400 μm. Czys et al. (1996) used this value with soundings from across the geographic area affected by freezing rain to diagnose the location of the boundary between freezing rain and ice pellets. Their results closely matched the observations.

Subsequent cursory evaluations of τ were conducted in operational environments at the St. Louis and Indianapolis National Weather Service Forecast Offices during the 1995/96 winter. Qualitatively, the method tended to identify where freezing rain and ice pellets occurred when a_c was assumed to be 400 μm. A cautionary situation mentioned by CZ are cases where freezing drizzle occurs and the sounding is everywhere colder than 0°C. In these cases, freezing precipitation forms through the warm rain process (Bocchieri 1980; Huffman and Norman 1988; Rauber et al. 2000). Czys et al. (1996) note that their 1995/96 experience suggests that this situation is “a highly localized phenomena.”

Based on initial successes, CZ developed a simple isonomogram that could be used with soundings as a diagnostic tool to distinguish freezing rain and ice pellet occurrences. The isonomogram, reproduced in Fig. 1, shows isopleths of τ = 1 computed over a range of critical particle radii, warm-layer depths, and warm-layer mean temperatures. Czys et al. (1996) suggested that the value a_c = 400 μm be used to delineate the freezing rain–ice pellet boundary. The technique proposed by CZ can be applied to model forecast soundings and can be used to produce forecast products such as maps delineating freezing rain–ice pellet (ZR–IP) boundaries. It is important, therefore, to test the CZ isonomogram with independent data. The purpose of this note is to evaluate the general applicability of the CZ isonomogram as a forecasting tool by testing it with a 25-yr sounding database.

2. Database

Soundings from all United States stations east of the Rocky Mountain states during the 25-yr period from 1 January 1970 through 31 December 1994 were considered for this analysis. The upper-air data were obtained from the National Climatic Data Center’s Radiosonde Data of North America. Freezing rain, freezing drizzle (ZL), and IP events were first identified using the National Climatic Data Center’s Storm Data reports (NCDC 1970–94). Each state report was carefully screened for any occurrence of ZL, ZR, or IP (sleet or snow grains). The 3-hourly surface charts (obtained from National Climatic Data Center microfilms) for the entire storm associated with each event identified in Storm Data were then analyzed to determine the regions where these events occurred. With this procedure, we identified 1091 12-h periods, an average of 43 per winter or about 8 per month (Nov–Mar). We then examined the 0000 and 1200 UTC charts to identify those sounding sites within the area of ZR, ZL, or IP that reported one of these conditions at the time of launch. In the case where there was not a surface report at the site itself, nearest-neighbor stations surrounding the site had to be reporting the same condition for the sounding to be included in the database. These soundings were retained for analysis. A total of 1051 soundings were identified, 581 reporting freezing drizzle, 391 reporting freezing rain, and 79 reporting ice pellets.

This method of choosing soundings has two obvious drawbacks. The first is that the surface observations and sounding data are not coincident in time and space. Surface observations are normally taken 10 min prior to the synoptic time, while soundings are launched 50 min prior to the synoptic time, a 40-min time difference. In addition, the rawinsondes drift downwind of the station, and the surface and sounding stations are not always collocated. This problem cannot be avoided using conventional data. We do not consider this issue to be significant to the results of this study, based on individual examination of each of the sounding profiles and the fact that similar conditions were, in most cases, reported at several nearby stations simultaneously.

The second drawback is that minor events, covering a small area over a short time, would be excluded from our analyses because they never would be reported in Storm Data. Recall that Storm Data was used to initially identify IP, ZL, and ZR events. Subsequent analysis of the surface charts nearly always identified a much broader area of IP, ZL, or ZR than indicated in the state reports in Storm Data. Much of the coverage area in any event received only minor amounts of these precipitation types. Most of the soundings in our database were associated with ZL, which, aside from air traffic hazards and automobile accidents, is generally associated with minor property damage. Furthermore, the conditions had to occur at the sounding site at the synoptic time. We believe that the odds are low that minor, short-lived events would have occurred at sounding stations at specific times at a high enough frequency to bias our results significantly.

This is the second paper reporting analyses from this database. The first paper (Rauber et al. 2000, hereafter RB), discussed the importance of the warm rain process in freezing events. Rauber et al. expanded upon earlier work on this same topic by Huffman and Norman (1988) and other researchers. The sounding database used by RB did not contain the 79 soundings from stations reporting ice pellets, nor did RB consider processes concerned with ice pellet formation. In this evaluation of the CZ method to diagnose the boundary between freezing precipitation and ice pellets, we use RB’s sounding categorization to identify freezing rain and drizzle soundings that conform to CZ criteria. We also introduce the database of 79 soundings from stations reporting ice pellets, and compare data from soundings in each of these databases to diagnostic predictions of the CZ nomogram.

3. Reduction of the database

Although 1051 stations reported ZL, ZR, or IP at the time of the soundings, only 306 clearly conformed to the environmental vertical temperature and moisture profile assumed by CZ. This profile consisted of a three-layer structure, with a cold cloud layer aloft that is a source of ice particles, a midlevel layer where the temperature exceeds 0°C and ice particles melt, and a surface layer where T < 0°C.

4. Results

Figure 3 shows the average warm layer temperature (°C) plotted as a function of warm layer depth (m) for (a) each of the 190 category 6 soundings where ZR was reported, (b) each of the 52 category 6 soundings where ZL was reported, and (c) each of the 64 IP soundings. The method for estimating the average warm layer temperature is described in the appendix. An obvious correlation between the depth and mean temperature of the warm layer appears in the data on all three diagrams; however, none of the diagrams supports a relationship between a specific critical radius and a freezing rain–ice pellet boundary. For example, if we consider the value of the critical radius suggested by CZ (400 μm), then 48.4% of the ZR soundings and 51.9% of the ZL soundings would lie in the IP regime (Figs. 3a,b). Similarly, 32.8% of the IP soundings would lie in the ZR and ZL regime (Fig. 3c). Other values of the critical radius have poor diagnostic skill as well.

Reasons for the lack of diagnostic skill most probably relate to the variability in ice particle shapes and size distributions in clouds that produce ZR and IP. Czys et al. (1996) calculations were based on the assumption that spherical ice particles were falling into the warm layer. In situ observations of particles entering the melting layer by Ohtake (1969), Stewart et al. (1984), Fujiyoshi (1986), Barthazy et al. (1996), and others show that aggregates, rather than graupel, are the common ice particle type falling into the melting layer in winter stratiform clouds. In fact, microphysical models of winter stratiform clouds (e.g., Passarelli 1978; Szyrmer and Zawadski 1999) typically do not even consider graupel. Spherical particles can be expected to form when elevated convection occurs above the warm layer leading to heavy riming. The soundings represented in Fig. 3 generally do not correspond to conditions associated with elevated convection. Figures 4a–c shows the geographic distribution of soundings appearing in Figs. 3a–c, respectively. Although the soundings occur throughout the United States east of the Rockies, most occur on the east side of the Appalachian Mountains and just north of the Ohio River Valley. These locations commonly experience freezing precipitation or ice pellets during warm frontal overrunning and Appalachian cold air damming and trapping events (Kocin and Uccellini 1990). To determine the likelihood that convection could lead to spherical particles in conditions represented by these soundings, each sounding in Fig. 3 was examined to determine if elevated convection was possible above the low-level stable layer. Of the 242 ZR–ZL soundings in Figs. 3a and 3b, 93% were absolutely stable. Of the IP soundings in Fig. 3c, 97% were absolutely stable. These findings conform to surface reports of freezing precipitation analyzed by Robbins (1998). Robbins analyzed a 15-yr database of surface observations and found that 97% of the freezing precipitation reports were of light precipitation, 3% of moderate precipitation, and none of heavy precipitation characteristic of convection. These data suggest that elevated convection is infrequent and generally not strong during freezing precipitation and IP events, and are consistent with the idea that ice particles falling into the melting level are typically aggregates.

Aggregates and ice crystals both have a large surface to volume ratio and undoubtedly melt at a different rate than spherical particles. This in itself might account for many of the ZR and ZL cases in the IP regime in Fig. 3a. It is also possible that particles that are almost completely melted do not have sufficient time to refreeze in their fall through the low-elevation cold layer if the layer is neither too cold nor deep. Many IP points occur in the ZR regime in Fig. 3b. Apparently, complete melting of these falling particles did not occur at the predicted rate. One possible reason for this would be that the air through which melting particles were falling was unsaturated. In this case, evaporation would cool the particles to the wet-bulb temperature, reducing the rate at which melting proceeds.

Another issue complicating the application of the CZ nomogram, even if it showed diagnostic skill, would be the real-time choice of a critical radius. Although CZ found that a critical radius of 400 μm “applied well to the 17 cases examined during the 1995/96 forecast experiment,” it is not clear what general criteria one could use to assign a critical radius in more general application of the CZ technique. Czys et al. (1996) recommended using rain-rate or reflectivity data to establish a critical radius, but these parameters have large temporal and spatial variability, making assignment of a unique value of a critical radius difficult.

5. Summary

The general applicability of the isonomgram developed by Czys et al. (1996) for diagnosing the position of the boundary between freezing precipitation (freezing rain or drizzle) and ice pellets (sleet or snow grains) was tested using a 25-yr sounding database consisting of 1051 soundings, 581 reporting freezing drizzle, 391 reporting freezing rain, and 79 reporting ice pellets. It was found that only 306 of the 1051 soundings clearly conformed to the Czys et al. (1996) environmental temperature and moisture profile, which consisted of a three-layer atmosphere with a cold cloud layer aloft that is a source of ice particles, a midlevel layer where the temperature exceeds 0°C and ice particles melt, and a surface layer where T < 0°C. The remaining soundings did not conform either because freezing precipitation formed due to the warm rain process or ice pellets formed due to riming rather than melting and refreezing. For the soundings conforming to the CZ profile, the isonomogram showed little skill in delineating between ZR–ZL and IP. Freezing rain or drizzle occurred about 50% of the time that ice pellets were diagnosed. Ice pellets occurred in nearly a third of the cases where freezing precipitation was expected. These results can probably be attributed to the fact that the physics of the melting process are more complicated than those used to develop the simple isonomogram.

Although these results are discouraging, it may be possible in the future to develop a forecasting technique that incorporates more of the complex physics of ice particle melting. This information may then be used to improve weather-type prediction algorithms in numerical forecast models (e.g., Baldwin and Contorno 1993). Key to the development of any technique are observations of the habits of ice particles falling into the melting layer, and observations that can delineate warm rain from ice processes. Polarization radar techniques hold promise for addressing these issues (e.g., Reinking et al. 1997; Vivekanandan et al. 1999; Zrnic and Ryzhkov 1999). Hopefully, when polarization capabilities are incorporated into the national Doppler network, the identification of precipitation types in the zone of freezing rain and ice pellets will become possible.