A. In-flight Icing [Background] [Remote
sensing] [A
Physically-based statistical retrieval] [Icing detection
using AVHRR] 1. Background
Remote sensing systems are being designed to utilize data from radars, radiometers, and satellites. Simulations of the response of such systems to realistic cloud conditions are used to design the systems. Initial feasibility studies have been done using existing systems during the Winter Icing and Storms Projects and the Mt. Washington Icing Sensors Program (WISP and MWISP). Results of this work are being used to guide the construction of a second wavelength (0.86 cm) addition to NSF's S-Pol research radar, and for two new frequency receivers for the Radiometrics, Inc. TP/WVP-3000 profiling radiometer. 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. We also continue to work in the characterization of cloud and weather conditions associated with icing environments. This work has had a significant impact on international aviation regulation harmonization groups and on guiding the location and timing of future field projects such as the upcoming Alliance Icing Research Study slated for Montreal, Quebec, in winter 2003-2004. 2. Remote Sensing of Icing Conditions by Dual-wavelength Radar and Radiometer The pursuit of reliable methods for the remote sensing of super-cooled liquid water in clouds continued in FY02 with an examination of the use of dual-wavelength radar and microwave radiometer measurements. Because radar signals having distinct wavelengths are attenuated differently by small water droplets, the difference in returned signal from two co-located radars may be used to determine the amount of water present along their common direction. In addition, radar-measured reflectivity is approximately proportional to the square of liquid water content, and this proportionality constant may be estimated if an independent measurement of the total water along the radar beam, such as that provided by a radiometer, is known. These observations are the basis for several techniques for retrieving liquid water content from dual-wavelength radar and radiometer data. A primary goal of this year's work was to implement, compare, and refine these techniques with an eye towards developing a combined, robust algorithm. This research is sponsored by the FAA's Aviation Weather Research Program and is being conducted by J. Williams, G. Zhang, J. Vivekanandan and M. Politovich. John Williams used data from two recent field programs, the Mount Washington Icing Sensor Project (MWISP) and the Alliance Icing Research Study (AIRS) were used to evaluate three candidate detection methods. Both field programs provided total-path vapor and liquid water measurements from radiometers along with X, Ka, and W-band polarized Doppler radar data. In MWISP, a NOAA/ETL radiometer, the NOAA/ETL X- and Ka-band radars, and the University of Massachusetts CPRS Ka- and W-band radar were used; AIRS employed a Radiometrics WVR-1100 radiometer, McMaster University's IPIX X-band radar, and the UMASS CPRS radar. To test the three detection techniques Williams pre-processed radar data from dual-wavelength pairs by interpolating them onto a common range-time grid, and the data were smoothed by applying a weighted median filter. Ka-band linear depolarization ratio (LDR) and reflectivity were used to identify good test cases in which predominately liquid clouds were present. In order to isolate the effect of the cloud liquid water, filtered reflectivity values were corrected for the attenuation due to atmospheric vapor, oxygen, and nitrogen as estimated using measurements from a proximate sounding. Figures 1 and 2 show results produced by two candidate techniques on a sample case. For both methods, there appears to be reasonably good qualitative agreement between the liquid water content from an in-situ sounding and the remote retrievals within the radars' domain of sensitivity; however, the inhomogeneity of the atmosphere and the absence of true co-location makes a precise comparison difficult.
Figure 1. Results of the dual-wavelength LWC retrieval using MWISP CPRS Ka- and W-band data from 15 April 1999. The left-hand plot shows the in-situ LWC measured by an ATEK sounding released at 16:22 GMT, and the plot at right depicts the retrieved LWC for slant ranges 0-6 km (vertical axis) and times 16:20-16:40 (horizontal axis), with the blue-to-red colorscale representing values from 0 to 1 g/m^3.
Figure 2. Results from the Z/LWC-squared proportionality retrieval technique using CPRS W-band and radiometer data for the same case shown in Figure 1. The three techniques were observed to have different, somewhat complementary strengths and weaknesses. For instance, the dual-wavelength method does not make any assumptions about droplet size distributions and does not require radiometer measurements, but it does assume small droplet conditions, is quite sensitive to instrument noise and radar beam mismatch, and is only effective where valid data from both radars is available. On the other hand, the Z/LWC-squared proportionality technique requires data from only one radar and is not affected by a bias in measured reflectivity, but it does assume a fixed relationship between measured reflectivity and LWC for each profile. These observations suggest that a combined or hybrid algorithm, constructed using fuzzy logic principles, might well be more accurate and robust under varied conditions than any single technique. [Top]
Temperature profiles retrieved from a profiling radiometer such as the Radiometrics TP/WVP-3000 could be biased warm by scattering from ice particles. J. Vivekanandan, G. Zhang and M. Politovich have been working with Radiometrics to detect when such conditions exist, and to seek correction factors. The instrument senses 12 frequencies ranging from 22 - 58 GHz. The frequencies at V-band, from ~ 50-60 MHz, are used for temperature profiling, but are sensitive to some degree to scattering by ice particles. Lower frequencies do not have this sensitivity. Higher frequencies have greater sensitivity and could be used in combination with the lower frequencies to flag those soundings with significant ice. The optical thickness (t) at any frequency can be estimated using the measured brightness temperatures (TB) from the instrument. If there is no scattering in the sampled medium, such as at lower frequencies, this is a true optical depth. However, with scattering, this becomes an apparent optical depth. At ~ 20 and 30 GHz, total integrated vapor and liquid can be accurately retrieved. These values can be fed back into emission models for higher frequencies, and those results compared with the apparent optical depths obtained from the measured TB. With large differences in TB, ice can be assumed present. Thus a combination of TB observations at 20, 30, and 90 or 150 GHz could be used to detect a mixed-phase cloud and estimate the total ice water path. G. Zhang ran the RAP radiation transfer model to simulate the effect of total ice water path (IWP) and particle size (expressed as RES, the cube root of the sixth moment divided by the third moment of the size distribution). The model calculation assumed specific values of ice density, a Gamma particle size distribution, and mixed-phase layer configuration. Figure 3 shows a positive bias in V-band brightness temperature in a mixed-phase cloud, as a function of both IWP and RES. Modeled bias values are comparable to the bias in retrieved temperature profiles during the several field programs when mixed-phase clouds were observed. Figure 4 shows the bias in 90, 150 and 220 GHz TB due to ice scattering, again sensitive to both IWP and RES. At 90 GHz, the bias increases with both IWP and RES. However, at 150 and 220 GHz, the increased forward scattering reduces the bias. Hence, for particle sizes larger than 1 mm, the 90 GHz TB may be more appropriate for inferring RES and IWP. Figure 5 shows DTB at V-band channels as a function of DTB at 90 GHz. Thus, a small but significant DTB can be quantified by measuring DTB values at scattering-sensitive millimeter-wave channels. In response to these results, Radiometrics, Inc. is constructing receivers at 90 and 150 GHz which will be tested in winter 2002-2003. More detailed calculations, using measured ranges of IWP and RES, and realistic ice particle shapes and densities, should make it possible to bound the magnitude of the expected DTB and devise a means for real-time correction.
Figure 5. Bias in V-band brightness temperatures shown as a function of bias in 90 GHz brightness temperature.
J. Hagerty of ATD has been working with RAP scientists J. Vivekanandan, G. Zhang and B. Bernstein to investigate new techniques for inflight icing detection using satellite data. The AVHRR/3 sensor on the NOAA polar orbiting platforms has been operational since February, 2001. This sensor includes a daytime channel at 1.62 mm, as well as visible (0.63 mm) and IR (3.9, 10.8, and 12 mm) channels. The reflectivity of a cloud depends on the refractive index of the hydrometeors (liquid or ice) and the mean size of the hydrometeors. Water clouds are highly reflective at 0.63 and 1.62 mm, while ice clouds have a high reflectivity at 0.63 mm and lower reflectivity at 1.62 microns. The utility of the reflectivity ratio (R1.6/0.6, the ratio of reflectivities at 1.6 and 0.6 microns) method for cloud top phase discrimination has been demonstrated by Hutchison (1999). His work showed that thresholds can be applied to R1.6/0.6 to distinguish liquid and ice clouds under certain conditions. The value of the threshold will vary depending on solar illumination conditions and scattering geometry between the sun, satellite, and pixel. The presence of overlying thin cirrus clouds and/or a highly reflective surface may produce ambiguous results. Aircraft observations based in Cleveland during the winter of 2002 provide in situ data in icing conditions for comparison with the reflectivity ratio phase detection method. Supercooled liquid droplets were observed during a flight on February 21, 2002; coincident AVHRR data were obtained for this case and Figure 6 is an image of R1.6/0.6. Julie Hagerty (ATD) worked with RAP scientists to apply the Hutchinson Thresholds to distinguish areas of liquid and ice phase cloud tops. Areas within the red polygons have R1.6/0.6 < 0.7, and are designated as ice clouds. Areas within the green boundary have R1.6/0.6 between 0.7 and 1.0, and are designated as liquid clouds. Coupled with the information that 10.8-micron brightness temperature is around -9°C within the area designated as liquid, these areas could be assumed to include icing conditions. Aircraft measurements on this date confirm the presence of supercooled liquid at the time of the satellite overpass. Although the aircraft did not penetrate cloud top during this flight, observed temperatures suggest that it was near the top. Finally, note that the clear sky areas to the north and south of the cloud system have extreme (high and low) values of R1.6/0.6. Reflectivity in these areas depends on surface conditions, and should be screened out with an a priori cloud masking technique. Refinements to the R1.6/0.6 threshold technique are in progress. Modeling studies on the effect of solar illumination and scattering geometry are being conducted with M. Deeter of ACD and G. Zhang of RAP to adjust the thresholds to appropriate values for specific conditions. A cloud/land masking technique will also be incorporated. Methods for determining cloud phase at night using the AVHRR IR channels (3.7, 10.8, and 12 mm) are also being explored. Reference: Hutchison, K., 1999: Application of AVHRR/3 imagery for the improved detection of thin cirrus clouds and specification of cloud-top phase. J. Atm. Oceanic Tech., 16, 1885-1899.
Figure 6. Reflectivity ratio (R1.6/0.6) derived from 1.6 and 0.6-micron wavelength AVHRR images over the Great Lakes area. Areas with R1.6/0.6 < 0.7 are designated as ice phase cloud tops; areas with R1.6/0.6 between 0.7 and 1.0 are classified as liquid phase cloud tops. The flight track from the NASA Twin Otter is plotted in blue. [Top]
5. Improvements in Numerical Modeling for Icing This study evaluated the sensitivity of winter precipitation to numerous aspects of a bulk, mixed-phase, microphysical parameterization found in three widely-used mesoscale models (the National Center for Atmospheric Research/Pennsylvania State University Mesoscale Model version 5, MM5, the Rapid Update Cycle, RUC, and the Weather Research and Forecast model, WRF). Sensitivities of the microphysics to primary ice initiation, autoconversion, CCN spectra, treatment of graupel, and parameters controlling the snow and rain size distributions were tested. The sensitivity tests were performed by simulating various cloud depths (with different cloud-top temperatures) using flow over an idealized two-dimensional mountain. The height and width of the two-dimensional barrier are designed to reproduce an updraft pattern with extent and magnitude consistent with documented freezing-drizzle cases. Freezing drizzle can pose a significant threat to aircraft so particular emphasis is placed on model predicted freezing-drizzle formation. Upon testing the primary sensitivities of the microphysics scheme in two dimensions, MM5 was used in multiple case studies and results were compared to observations.
The major results of this study are: 1) the choice of ice initiation schemes is relatively unimportant for deep precipitating snow clouds but more important for shallow warm clouds having cloud-top temperature greater than -12°C; 2) autoconversion of cloud water to rain water is too simplistic and improperly represents fundamental differences in drizzle formation between cloud systems with maritime versus continental CCN spectra; 3) the assumed graupel size distribution and method of transforming rimed snow into graupel have major impacts on the mass of cloud water and formation of freezing drizzle; and 4) the assumption of the snow size distribution, particularly a shift from small to large diameters with increasing temperature is a key factor controlling formation of freezing drizzle in the model.
Forecasting icing severity is one of the more difficult jobs in aviation weather. One has to forecast a poorly-defined condition (current severity definitions are based on an airplane and pilot's response to the icing environment), using inadequate information (no direct measurement of cloud liquid water content). However, aircraft operating requirements force the issue and we must find a way to do this satisfactorily. This is an ideal case for application of fuzzy logic techniques. It has everything those methods could ask for: a vague concept of the outcome, and uncertain and occasionally ambiguous data sets. The current sources of icing severity information are: the Current Icing Product algorithm, CIP (see report in Technology Transfer section); the NCEP Rapid Update Cycle (RUC) weather prediction model; and pilot reports of icing (PIREPs). M. Politovich, F. McDonough and B. Bernstein are evaluating an initial severity algorithm as part of this inflight icing research. Icing severity is divided into four categories, trace, light, moderate and severe or heavy which are reported approximately 9, 42, 45 and 4% of the time, respectively (as in Politovich and Omeron-Bernstein, 2002). The various inputs to the severity algorithm are translated to severity on a 0 (none) to 1 (severe or heavy) scale, and calculations are performed on the 40-km RUC grid (soon to be upgraded to the new, 20-km resolution). Wherever CIP diagnoses an icing likelihood >0.05 (on a 0 to 1 scale) a value of 0.3 (light) is assigned. If the supercooled large drop potential is >0.05 or if convection is present, the value is 0.5 (moderate). In these areas, that base value is nudged upward or downward based on other clues. Where there is RUC model cloud and rain water, severity is assigned and mapped based on an accretion-based definition (Politovich, 2002). Icing severity reported by pilots (PIREPs) are gridded onto its own interest map, applying the worst reported severity in any grid box. Finally, RUC vertical velocity is mapped, where strongly upward (downward) motions are assigned higher (lower) severity values, the assumption being that greater vertical velocities are associated with more liquid water production. These inputs are combined to create a final value, which is illustrated in the figure. So far, our visual inspection confirms that in general, higher reported severities are associated with higher numbers, but the scale is as yet uncalibrated (see Figures 1 and 2 above). Calibration will be done after sufficient data are collected (PIREPS and NASA Glenn Research Aircraft measurements) to confidently do so. This product should enable the FAA and NWS to approve the CIP for fully operational use, rather than its current requirement to be augmented by the AIRMET icing severity content. References Politovich, M.K. and T.A.O. Bernstein, 2002: Aircraft icing conditions in northeast Colorado. J. Appl. Meteor., 41, 118 - 132. Politovich, M.K., 2002: Predicting aircraft icing intensity. Submitted, J. Aircraft.
Figure 7. Composite of icing severity for the CONUS. Scale is uncalibrated and as shown at the top, where 0=no severity, 1=severe or heavy icing. Letters are pilot reports of icing: R=rime, U=unknown type, and size is related to severity (larger are more severe).
Figure 8. Vertical cross section of icing severity for the route IAD-MSP. Waypoints are indicated at the bottom. Fields are as indicated with scales noted to the right. [Top]
An in-flight icing diagnostic algorithm (AK-CIP) was developed for the Alaska air space. F. McDonough is taking the lead on this project, working with B. Bernstein and M. Politovich of RAP along with forecasters at the Alaska Aviation Weather Unit. The algorithm combines data from the Alaska ETA (AK ETA) numerical weather prediction model, the GOES 10 satellite imager, and surface observations from the Alaska METAR sites. The observed data (satellite imagery and METARs) are mapped to the AK ETA model space and used in combination with the model soundings to create an hourly three-dimensional depiction of the icing potential. The GOES 10 mapping was done such that the higher resolution satellite imagery data was retained to some degree. The METARs are mapped to the model grid using a watershed approach. The METARs are only mapped to model grid points within a fixed distance and within the same watershed as the METAR site.
The algorithm initially defines a two-dimensional cloud field, and
the cloud top temperatures (CTT) for the cloud field. This is done
using the multi-channel satellite imager data and METAR cloud cover
information. Next the cloud top and cloud/precipitation base heights
are determined for the cloud field using the CTT, AK ETA model soundings,
and METAR cloud base heights or precipitation reports. Multiple cloud
layers within the cloud field are inferred using the model relative
humidity field. This results in a 3D cloud/precipitation field in
which the heights and CTT of all clouds and precipitation is known.
Finally the potential for supercooled liquid water (icing conditions)
to exist in the cloud/precipitation regions is determined. This is
done by first calculating supercooled liquid water interest maps of
temperature, relative humidity over water, pressure vertical velocity,
and cloud top temperature. Once the interest maps are determined the
icing potential equation using the interest maps can be calculated.
The final icing equation used by each model grid point is a situationally
dependent combination of the interest maps. For example if The algorithm updates hourly and the final icing and SLD potentials are plotted in both flight levels (Fig. 9) and vertical cross sections (Fig. 10). The algorithm was developed during the winter and spring of 2002 and was running in real time by mid-June 2002. The Alaska Aviation Weather Unit (AAWU) forecasters, who generate the operational icing AIRMETs for the Alaska airspace, were interviewed and many of the ideas they suggested are part of the algorithm. Once the algorithm was running, Peninsula Airlines (PenAir) was contacted about supplying supplemental icing pilot reports (PIREPs). They supplied PIREPs from mid July through September. These PIREPs are currently being compared to the AK-CIP icing and SLD potentials.
8.
An Inferred Climatology Of The Size of Supercooled Large Droplet Icing
Events Over North America Two
primary datasets were used to examine these issues. Fourteen years
of coincident soundings and surface observations of cloud cover and
precipitation taken across the U.S. and Canada were run through aspecial
version of the Integrated Icing Diagnosis Algorithm (IIDA) to infer
the presence of SLD. Gridded IIDA output which uses a combination
of model output and observations from satellite, radar and surface
stations to produce hourly, 3-D diagnoses of SLD over the lower-48
andsouthern was also examined. The sounding dataset was used to produce
the best estimate of the depth of SLD events, while the gridded dataset
was used to estimate their a real coverage and persistence. IIDA produces
a floating point SLD "potential" value, ranging from 0 (no
SLD is expectedto be present) to 1 (SLD is very likely to be present).
For these analyses, a threshold of 0.4 was chosen. Sensitivity tests
have shown that the choice of lower (higher) thresholds resulted in
somehwat larger (smaller) depths, areas and event logevity. To estimate the size and persistence of SLD conditions, gridded analyses were broken down into SLD "events" based on their footprint. The average 40km grid box has a footprint of ~1600km2. The number of contiguous pixels with SLD potential of 0.4 or greater determined the areal coverage of the SLD "event". Breaks of up to 1 pixel in width were allowed. Among the ~12,000 events, most were quite small, but there was a long tail to the distribution. About half of the events were <3200km2 (2 pixels, roughly the size of a metropolitan area), 10% were > 25,600km2 (16 pixels) and ~1% were > 128,000km2 (80 pixels, roughly the size of South Carolina) in size. The long axis of these events was typically ~200km long, but one event was found to be ~3400km long,extending from Texas to Massachussetts. Using consecutive 3-hourly diagnoses, an event was considered to persistfrom one hour to the next if SLD>0.4 was found within ~100km of the event's location at the previous time. This picks up on slow moving, continuous events, but will underrepresent the longevity of those thatare moving quickly. Most events (~65%) were found to last for <3hr, meaning that they were only present in one gridded analysis. Given the3-hr granularity of the data, such an event acutally may have been present for up to 7 hr. About 1% of all events persisted for more than 18hr, and the longest event lasted for ~70hr. Results are somewhat sensitive to the choice of a 100km-search radius. [Top] |
