A. In-flight Icing

[Background] [An Inferred Icing Climatology]
[Analysis of Problematic Icing Diagnoses in the Pacific Northwest]
[Use of MM5] [Use of Radar Polarization Information]
[Effects of Ice Crystals on Brightness Temperature Measurements]
[An Initial Assessment of Radiometer Retrievals of Atmospheric Profiles]

---

1. Background

The goal of RAP inflight icing research is to develop more accurate and timely diagnoses and forecasts of conditions leading to ice accretion on aircraft during flight.  Since in-flight icing is the accretion of supercooled liquid water (SLW) during flight, the process is synonymous with finding regions favorable to SLW formation.  Thus, inflight icing research provides both interesting and challenging studies in cloud physics, remote sensing, and mesoscale meteorology.

The integrated icing diagnostic algorithm that combines sensor data with model output has been quite successful and the product should gain “operational” status with the National Weather Service and Federal Aviation Administration during Winter 2001/2003. 

The forecasting research has been guided toward further developments of the MM5 and WRF models, including improved microphysical parameterizations that can more accurately forecast cloud liquid, freezing drizzle and rain. These improvements are transferred to operational use via the NOAA Forecasting Systems Laboratory and the National Centers for Environmental Prediction. RAP has developed good working relationships with these organizations, which provides an effective means of technology transfer.

Remote sensing systems that utilize data from radars, radiometers, and satellites are being designed for icing detection in a variety of weather conditions.  These methods can be developed using knowledge of hydrometeor types and size distributions typically found in icing environments, combined with radiative transfer models that simulate the instrument response at various wavelengths and polarization states. Collaborations with the NOAA Environmental Technology Laboratory the U.S. Army Cold Regions Research and Engineering Laboratory, NASA Glenn Research Center and Radiometrics, Inc. have helped further this research.

2. An Inferred Icing Climatology from Sounding and Surface Dataset

A question commonly asked about in-flight icing is how frequently it occurs, at what altitudes and how this varies geographically and by season.  Although a database containing several years of direct observations using icing pilot reports (PIREPs) is available, it contains several important biases that limit its usefulness.  For example, PIREP databases are biased by the frequency of air traffic, both geographically and by time of day, and “no-icing” is grossly underreported. Thus, use of PIREPs would lead to an icing climatology that is biased toward high-traffic areas, while underestimating areas lacking icing conditions.

In an effort to eliminate some of these irregularities, B. Bernstein is developing an inferred icing climatology using regularly-observed data from 14 years of coincident, 12-hourly US and Canadian surface weather reports and balloon-borne soundings. Although these datasets do not provide direct observations of icing conditions aloft, they have been shown to give strong indications of its presence and absence when properly combined.  The observations are used to infer the characteristics of clouds, including their heights, depths, temperatures, and expected microphysical phase.  Cloud base and top height are determined from ceiling observations, and estimated using relative humidity with respect to both water and ice from the sounding, respectively.  Likely locations of cloud layers are determined from the relative humidity profile.  For each cloud layer, the phase is inferred from a combination of temperature, cloud-top temperature and precipitation and thunder reports from the coincident surface observation.  The result is an icing “potential” ranging from 0.0 (no icing) to 1.0 (icing extremely likely) for each level in every sounding. 

Soundings were examined at 121 sites for 14 years.  Horizontal coverage is fairly uniform across North America, eliminating air-traffic bias from the climatology. Interpolation of results between sites should be reasonable for all but the Intermountain West and, to some extent, the Appalachians and Great Lakes.

The climatology can be performed at any threshold of icing potential. For this discussion, a value of 0.75 is chosen, as it indicates locations where icing is likely to be present, while minimizing false alarms.  Icing is most common along the Pacific Coast, north of the California/Oregon border, with peak frequency in western Alaska (Figure A1). The Great Lakes and eastern Canadian provinces also have a significant amount of icing.  Lower values of icing potential occurs along and just east of the Rocky Mountains, as well as in the Southwestern U.S. and along the Gulf Coast.  Month-by-month results show large deviations from this pattern, as icing-prone areas migrate latitudinally with storm tracks, moisture plumes, and of course, common icing temperatures (-15 to 0^oC). 

This is quite evident in single-station time-height cross-sections (Figure A2). Icing is most frequent <9000 ft (all heights MSL) and is present ~23% of the time during the winter at Green Bay.  The pattern shifts upward to 12,000 - 21,000 ft and icing is present ~11% of the time during the summer.  While winter-summer altitude shifts still occur, the frequency changes are reversed at Resolute, Northwest Territories, with icing ~37% of the time during summer, nearly all <12,000 ft and ~1% of the time during winter, when even surface temperatures are frequently very cold.

This inferred icing climatology should provide us with a new tool to assess icing frequency. This will be a valuable addition to our in-flight icing diagnosis and prediction algorithms, to match observed or forecast conditions to expected icing-prone areas.


3. Analysis of Problematic Icing Diagnoses in the Pacific Northwest

Several recent formal and informal studies have been conducted to verify the NCAR Integrated Icing Diagnosis Algorithm (IIDA) using observations of icing from pilot reports (PIREPs) and research aircraft. Results have shown that IIDA sometimes performs differently depending on the meteorological situation and region of the U.S. In particular, IIDA occasionally has problems identifying, or underestimates, the potential for icing in the Pacific Northwest when vigorous synoptic-scale storms affect that area. These events and others where IIDA does not perform well appear to have common characteristics, suggesting that an analysis of the IIDA input datasets used and the synoptic-scale meteorology for these cases may provide insight and potential solutions.
M. Chapman and B. Bernstein began this analysis to matching available PIREPs with icing potential values and created a time-height cross-section for each day in question. The IIDA values used in these plots are the maximum values at the four grid points (on the 40-km RUC-2 model grid) surrounding either Portland or Seattle, and range from 0.0 (no icing) to 1.0 (icing very likely). Also, similar plots were made of the RUC model-based relative humidity and temperature values that IIDA ingested. Relative humidity and temperature are used in conjunction with observations from satellite, radar, and surface stations to assess where clouds are present and estimate their phase. Together these plots should help to construct an initial hypothesis of why IIDA did not perform well for these events. After the initial work is complete, a look into the synoptic-scale meteorology of these separate circumstances is needed to assess the type of weather pattern that was present during these events. Some basic tools that will be used during this aspect of the analysis are constant pressure charts from required levels, soundings from the closest upper-air station, surface charts, satellite and radar data, and surface observations. Preliminary results from one example case are presented below.

A fairly typical, deep, strong synoptic system impacted the Seattle area between 1200 - 2400 UTC on 1 February 2001, causing an outbreak of light and moderate icing between 5000 and 22,000 feet MSL (Figure A3a). While IIDA captured 30 of the 39 icing PIREPs near Seattle, it completely missed 9 PIREPs and had relatively low icing potentials (<0.5) for many others. Examination of RUC temperature output (Figure A3b) and the Quillayute, WA (UIL) sounding reveals that the missed PIREPs occurred at altitudes with T<-25oC. This is below the minimum temperature at which IIDA will indicate icing, which was chosen based on the rarity of icing conditions at such temperatures. Most of the other PIREPs occurred in places with temperatures between -15 and -25oC, and where RUC RH was <60% (e.g. 10-15 kft at 1800-2000 UTC, Figure A3c). Strong lift, such as that associated with vigorous storms and their interaction with the steep topography in the area, can cause supercooled liquid to form and be maintained even in the presence of ice crystals at relatively cold temperatures. Glaciated conditions are usually expected at such low temperatures. IIDA developers are considering the use of model vertical velocity fields to extend icing diagnoses to colder temperatures in such cases.

[top of page]

4. Use of MM5 to Simulate a Weakly-Forced Stratiform Cloud with High Liquid Water Content

Correct prediction of cloud liquid water content is critical for quantifying inflight icing hazards. G. Thompson, R. Rasmussen and W. Hall have been working on improvements to the microphysics parameterization in MM5, with a view toward future implementation in both the Rapid Update Cycle and Weather Research Forecast models.  

A recent simulation was performed on a case that included a shallow stratocumulus cloud sampled by the NASA Glenn Twin Otter research aircraft near Cleveland.  The conditions encountered were relatively small cloud droplets with a concentration of ~250 cm^-3, very light freezing drizzle and also some mixed-phase clouds downwind of Lake Erie.  During the initial ascent after takeoff from Cleveland the liquid water content increased with height to 0.8 g m^-3 near cloud top (see Figure A4). The cloud liquid water content profile was very similar during a missed approach to Canton-Akron airport later in the flight.

The Penn State/NCAR MM5 model was configured with 20-km grid spacing covering much of North America. The model was initialized at 1200 UTC 29 Jan 1998 and run for 36 simulated hours.  Thus, the model is compared to the observations approximately 24 - 27 h into the simulation.  The microphysics output is in good agreement with the aircraft observations.  Note the small amounts of snow, graupel and rain (freezing rain since the temperature is < 0^oC) just below cloud base.  

Thus far all MM5 simulations have used a threshold value of 0.35 g kg^-1 before converting cloud water to the rain category (no drizzle-specific category at this time). To test the sensitivity of the maximum liquid in the cloud to this threshold, four values (0.20, 0.35, 0.50 and 0.70 g kg^-1) were used. A CCN concentration of 250 cm^-3 (similar to the droplet concentration measured from the aircraft, and typical for continental air masses) was used for all these threshold values. As expected, the simulation with the lowest threshold for converting cloud water to rain produced the most rain and maintained the least liquid water content since cloud water transitioned to rain more rapidly.  Conversely, the simulation with the highest threshold produced the least rain and maintained nearly 0.7 g m^-3 liquid water content, which is near the value measured from the aircraft.  This confirms that the model can accurately reproduce the observed cloud liquid water, but only if an appropriate auto-conversion threshold is used.

Thus, in order to correctly predict such high liquid cases, it’s important to know whether the air mass had continental or maritime origins.  In the continental case, the auto-conversion threshold needs to be set near 0.75 g m^-3, which was also the result in another detailed modeling study of freezing drizzle conducted in RAP.  Field and modeling studies show that in maritime regions (or air with a source of maritime origin) liquid water content rarely exceeds 0.3 g m^-3 before drizzle or rain begins to form. For this reason, the plan is to modify the microphysics parameterization to accept a variable number of CCN and different thresholds for the conversion of cloud water to rain water, based on knowledge of the source region of the model air.

[top of page]

5. Use of Radar Polarization Information to Detect InFlight Icing Conditions

Icing situations encountered during several recent field programs involving the S-Pol radar and storm-penetrating aircraft are being investigated by E. Brandes and S. Ellis (ATD).  Hazard designation is facilitated in part by the capability of polarimetric radars to determine the 0^oC (melting) level within precipitation and thereby designate potential icing layers. The hydrometeor classification algorithm (HCA) that runs in real time on NCAR’s S-Pol radar is used to determine hydrometeor type. 

Figure A5 shows a vertical cross-section through two small convective showers observed in Florida on 14 September 1998 during the PRECIP98 field program.  The University of North Dakota’s Citation research aircraft penetrated the convective towers at the -10^oC level and encountered supercooled large drops (SLD).  Individual panels in the figure show radar reflectivity (Z), differential reflectivity (Z_DR), linear depolarization ratio (LDR), the correlation coefficient between reflectivities at horizontal and vertical polarization (RHOHV), and particle designations made with the hydrometeor classification algorithm.  Also shown are the melting level (as determined from the polarimetric measurements), some labeled hydrometeor classifications, and sample images of particles observed by the aircraft.  The aircraft location is indicated by a small black square.  At the time of data collection both liquid drops and frozen particles were observed.  The Rosemount icing probe indicated icing.  From the suite of radar measurements, supercooled liquid water (SLW), light rain, wet snow, and a graupel­rain mixture were all inferred at and slightly above the aircraft’s altitude.

An example of a light icing event that occurred on 20 September 1999 during the Mesoscale Alpine Experiment conducted in Italy is shown in Figure A6.  The NCAR Electra encountered icing conditions at a temperature of -15^oC in the uppermost portions of a precipitation layer.  The figure shows a matched time­space series of aircraft observations and radar measurements.  There were two periods of icing.  In Region I the reflectivity is low (averaging ~0 dBZ) suggesting that the median size of the particles is small.  The differential reflectivity (Z_DR) varies from 0­1 dB an indication that ice crystals are present. [Drizzle would have a Z_DR of ~0 dB.]  The correlation coefficient is significantly less than 1.0 suggesting the presence of mixed-phase hydrometeors (riming and potential icing conditions).  Region II is characterized by reflectivity generally > 10 dBZ, Z_DR is near 0 dB, and RHO_HV is close to 1.0.  In Region II the radar measurements are dominated by fairly-large irregular ice crystals with near spherical shapes in the mean which obscure the icing signatures.