Effects of ice particles shattering on the 2D-S probe

Recently, considerable attention has been focused on the issue of large ice particles shattering on the inlets and tips of cloud particle probes, which produces copious ice particles that can be mistakenly measured as real ice particles. Currently two approaches are being used to mitigate the problem: (1) Based on recent high-speed video in icing tunnels, probe tips have been designed that reduce the number of shattered particles that reach the probe sample volume, and (2) Post processing techniques such as image processing and using the arrival time of each individual particle. This paper focuses on exposing suspected errors in measurements of ice particle size distributions due to shattering, and evaluation of the two techniques used to reduce the errors. Data from 2D-S probes constitute the primary source of the investigation, however, when available comparisons with 2D-C and CIP measurements are also included. Korolev et al. (2010b) report results from a recent field campaign (AIIE) and conclude that modified probe tips are more effective than an arrival time algorithm when applied to 2D-C and CIP measurements. Analysis of 2D-S data from the AIIE and SPARTICUS field campaigns shows that modified probe tips significantly reduce the number of shattered particles, but that a particle arrival time algorithm is more effective than the probe tips designed to reduce shattering. A large dataset of 2D-S measurements with and without modified probe tips was not available from the AIEE and SPARTICUS field campaigns. Instead, measurements in regions with large ice particles are presented to show that shattering on the 2D-S with modified probe tips produces large quantities of small particles that are likely produced by shattering. Also, when an arrival time algorithm is applied to the 2D-S data, the results show that it is more effective than the modified probe tips in reducing the number of small (shattered) particles. Recent results Correspondence to: R. P. Lawson (plawson@specinc.com) from SPARTICUS and MACPEX show that 2D-S ice particle concentration measurements are more consistent with physical arguments and numerical simulations than measurements with older cloud probes from previous field campaigns. The analysis techniques in this paper can also be used to estimate an upper bound for the effects of shattering. For example, the additional spurious concentration of small ice particles can be measured as a function of the mass concentration of large ice particles. The analysis provides estimates of upper bounds on the concentration of natural ice, and on the remaining concentration of shattered ice particles after application of the post-processing techniques. However, a comprehensive investigation of shattering is required to quantify effects that arise from the multiple degrees of freedom associated with this process, including different cloud environments, probe geometries, airspeed, angle of attack, particle size and type.


Introduction
Ice particles shattering on the inlets and tips of cloud particle probes produce small ice artifacts that can be erroneously included in measurements of ice particle size distributions.Artifacts produced from shattering of ice particles on optical cloud particle probes were reported in the literature over three decades ago (e.g., Cooper, 1978;Gardiner and Hallett, 1985).Cooper (1978) recognized the problem and introduced a method for removing shattered ice in post processing.Based on examination of two-dimensional images from the Particle Measuring Systems (PMS) model 2D-C probe (Knollenberg, 1970), he suggested that a burst of closely spaced particles, which is the typical pattern resulting from shattering, could be removed by comparison of individual particle arrival times.In simplistic terms, if ice particles are assumed to be randomly distributed in a cloud with concentration <∼1 cm −3 , than particles with arrival times R. P. Lawson: Effects of ice particles shattering on the 2D-S probe equivalent to a spacing that is less than about 2 cm are considered artifacts, and are removed.Cooper (1978) introduced the arrival time approach, which was later refined by Field et al. (2003Field et al. ( , 2006)), Korolev and Isaac (2005) and Baker et al. (2009).The work presented by Baker et al. (2009) considers removal of splashing raindrops, which closely resembles shattered ice particles.
While the issue surrounding ice particles shattering on the inlets and tips of optical particle probes (hereafter referred to simply as "shattering") has been known since the 1970's, it has only been recently that the magnitude of the effect has been brought to the attention of the cloud physics community.Advances in high-speed digital videography and cloud particle probes have provided new insights into the shattering process.High-speed videography of ice particles shattering on probe tips in the Cox & Company Incoporated icing wind tunnel by the National Aeronautics and Space Administration (NASA) Glenn Research Center (GRC) and Environment Canada (EC) showed some remarkable results.Korolev et al. (2010a) shows digital videography of ice particles a few hundreds of micrometers in size shattering on probe tips, with small ice particles bouncing several mm upstream into the 100 m s −1 airflow, and then traversing up to 3 cm across the airflow into the probe sample volume.
Advances in the electro-optics of linear-array cloud particle probes over the past four decades have provided new insights into measurements of cloud particle size distributions.To summarize, the 2D-C probe (Knollenberg, 1970) has 25-µm pixels with 32 photodiodes that are strobed at 5 MHz.The cloud imaging probe (CIP), designed and built by Droplet Measuring Technology (DMT) in the late 1990's (Baumgardner et al., 2001), has 64 photodiodes, 25-µm pixels and is strobed at 8 MHz.The 2D-S probe has 128 photodiodes, 10-µm pixels and is strobed at 20 Mhz.Lawson et al. (2006a) show laboratory results that demonstrate the ability of the 2D-S to image an 8-µm pixel fiber at speeds exceeding 200 m s −1 .In comparison, limitations of the time response of the photodiode array and front end amplifier in CIP and 2D-C probes may result in under sizing of small (e.g., <∼100 µm) particles.Lawson et al. (2006a) showed measurements that suggest that the 2D-C does not image particles <∼125 µm at an airspeed of 103 m s −1 .Strapp et al. (2001) report on the efficiency of a 2D-C probe to detect 60-µm opaque circular dots on before a clear disk spinning at 100 m s −1 .The results show that the depth of field (DoF) of the 2D-C probe reduces from about 75 mm to 10 mm when the disk speed increases from 10 to 100 m s −1 , and that 80 % of the 60-µm spots are detected exactly at the center of the DoF, with the percentage of detected dots (including zero area images) decreasing to about 10 % within 3 mm of the center of DoF.Recent measurements show that a newer version of the CIP is capable of imaging 50-µm drops at 150 m s −1 , but that performance degrades at higher airspeeds (Lawson et al., 2010).These results suggest that the time response, and therefore the ability of the 2D-C and CIP probes to image small particles, have improved over the past two decades.However, there is no evidence to date that shows that the 2D-C and CIP probes can detect particles less than 50 µm in size at jet aircraft speeds.On the other hand, the 2D-S probe has demonstrated the ability to image 10-µm particles at jet aircraft speeds. 1  This paper is focused on exposing suspected errors in measurements of ice particle size distributions due to shattering, and evaluation of techniques used to reduce these errors.It is not intended to be a comparison of the relative performance of various imaging probes.However, Korolev et al. (2010b) recently evaluated shattering effects on 2D-C and CIP probes and reported that specially modified tips were more effective than an arrival time algorithm in reducing the effects of shattering.In this paper it is seen that, after evaluating limited data collected by two 2D-S probes, one with and one without modified tips, we find a different result; i.e., an arrival time algorithm is more effective in reducing the apparent effects of shattering than modified tips.The limited dataset with and without modified probe tips is not intended to provide guidance for quantitative assessment of shattering on the 2D-S probe.However, the data do show that, contrary to the results presented for 2D-C and CIP probes in Korolev et al. (2010b), the 2D-S probe with modified tips detects shattered particles in significant quantities.This may be due to the improved time response and size resolution of the 2D-S probe, or other factors, such as the probe tip design.Regardless, the results show that the arrival time algorithm is more effective than the modified probe tips in reducing the number of small (shattered) particles in these regions of large ice particles.
We concentrate on 2D-S measurements, which themselves contain uncertainties, some known and others that will likely be exposed over time.However, when available we have included comparable 2D-C and CIP measurements.A comparison of 2D-S and historical measurements also leads to implications regarding how uncertainties may impact cloud particle data in archives.

Comparison of measurements from a 2D-S and historical measurements
Historical aircraft measurements of ice particle size distributions using optical probes in deep stratus cloud systems, such as thick cirrus, have generally revealed a vertical profile In contrast to the vertical distribution of small ice particles seen in Fig. 1, the measurements in Fig. 2 were collected using a 2D-S probe in a deep cirrus cloud investigated from 193630 -195900 UTC on 10 February 2010 during the Small PARTIcles in CirrUS (SPARTICUS) project.This example was chosen from the SPARTICUS dataset because it shows high (2.7 cm -3 ) concentrations of small ice near cloud top (and the CPI images reveal nearly all small ice), but only 44 L -1 near cloud base, including bullet rosettes with sizes of hundreds of microns.A comparison of the size distributions in Figs. 1 and 2 shows that both the concentration and mass distributions have similar shapes near cloud top where there were few large ice particles.However, lower in the cloud where there are higher concentrations of large ice particles, the mode of the mass distribution in Fig. 1 peaks between 10 and 100 m, whereas the mass mode peaks between 100 and 500 m in Fig. 2. The much smaller mass where small ice particles exist and typically dominate the size distribution throughout the depth of cloud (e.g., Lawson et al., 2006b).This is contrary to conventional thinking, which suggests that smaller particles will nucleate in higher concentrations at cold temperatures near cloud top and subsequently sublimate and disappear, grow via vapor diffusion, or aggregate into larger ice particles as they fall toward cloud base.
To help visualize the effects of shattering on archival data, we show two examples of vertical profiles of ice particle size distributions collected in relatively deep cirrus clouds.The first example shows average ice particle size distributions using older cloud particle probes that are believed to be subject to errors from shattering.The second example shows data from the 2D-S probe, which used modified probe tips based on the Korolev design technique and particle arrival times (Baker et al. 2009) to remove shattered ice in post processing.Figure 1 shows an example from Lawson et al. (2006b) of particle size distributions and number concentrations based on multiple penetrations of cirrus clouds.Composite size distributions were put together using measurements from a forward scattering spectrometer probe (FSSP), a cloud particle imager (CPI) and a 2D-C probe.The FSSP was used to establish the small particle end of the size distribution (generally less than about 30 µm) and the 2D-C established the large end.CPI data were scaled to merge with the FSSP and 2D-C measurements (see Lawson et al., 2006b for details).
A combination of gravity waves and homogenous nucleation at these cold temperatures is a possible theoretical explanation for the relatively high (846 l −1 ) average ice concentration near cloud top in Fig. 1 (Kärcher and Ström, 2003;Jensen et al., 2009).Some investigators have reported even higher (>1 cm −3 ) ice concentrations in regions where the maximum particle size is about 100 µm (Gayet et al., 2002;Kärcher and Ström, 2003;Lawson et al., 2006b).Shattering is not thought to be a major contributor to ice concentration in this situation.However, the high (2.17 cm −3 ) average ice concentration near cloud base in Fig. 1 cannot be explained theoretically.
Using particle arrival times from a fast FSSP, Field et al. (2003) shed light on this issue when they showed that the FSSP was very sensitive to shattering, and that the shattered artifacts could significantly increase the small particle concentration.Most of the measurements that suggest high concentrations of small ice in regions with large ice (i.e., > a few hundreds of microns) have been reported using (or in the case of the CPI scaled by) a scattering probe such as the FSSP (e.g., Fig. 1).
In contrast to the vertical distribution of small ice particles seen in Fig. 1, the measurements in Fig. 2 were collected using a 2D-S probe in a deep cirrus cloud investigated from 19:36:30-19:59:00 UTC on 10 February 2010 during the Small PARTIcles in CirrUS (SPARTICUS) project.This example was chosen from the SPARTICUS dataset because it shows high (2.7 cm −3 ) concentrations of small ice near cloud top (and the CPI images reveal nearly all small ice), but only 44 l −1 near cloud base, including bullet rosettes with sizes of hundreds of microns.A comparison of the size distributions in Figs. 1 and 2 shows that both the concentration and mass distributions have similar shapes near cloud top where there were few large ice particles.However, lower in the cloud where there are higher concentrations of large ice particles, the mode of the mass distribution in Fig. 1 peaks between 10 and 100 µm, whereas the mass mode peaks between 100 and 500 µm in Fig. 2. The much smaller mass mode in Fig. 1 is most likely due to particle shattering on the inlet of the FSSP used in these studies.Jensen et al. (2009) show that the shattering on scattering probes with inlets like the FSSP has a significant effect on the second moment (i.e., extinction) of the cloud particle size distribution.Errors from shattering that have affected measurements of extinction coefficient suggest that significant errors could occur in radiative transfer models and remote retrievals.

SPARTICUS field campaign
During the SPARTICUS project the SPEC Learjet was flown on a special mission with two 2D-S probes; one probe with "unmodified" ("standard") tips and one probe with tips "modified" using the Korolev design technique.Korolev developed probe tips to reduce shattering based on theoretical considerations and high-speed video in the Cox & Company icing tunnel.The process was iterative and the design of (2D-C and CIP) probe tips evolved over time.Korolev eventually patented the probe tip design used on the 2D-C and CIP probes.Lessons learned from the high-speed video revealed some fundamental concepts regarding the shattering process that Korolev communicated to SPEC engineers.The 2D-S probe that is termed the "standard" or "unmodified" probe in this paper had tips that were designed by SPEC to minimize shattering, but without benefit of information later provided by Korolev.The 2D-S with "modified" probe tips was designed The images in Fig. 5 suggest, and the size distributions in Fig. 6 support the premise that the modified probe tips significantly reduce, but do not eliminate shattering.During the penetration of precipitating dendrites, post processing of the 2D-S probe with standard tips identified 153 out of 507 (30%) of the images > 1 mm as shattered images.In comparison, the modified probe tips eliminated more than half this amount, with 54 out of 450 (12%) of the images > 1 mm being identified as shattered images.
Figure 6 shows that the size distribution with the unmodified probe tips and without application of the arrival time removal algorithm (green trace) contains the most particles with sizes < 200 m.The size distribution with modified probe tips and without application of the arrival time removal algorithm (red trace) contains the second most particles with sizes < 200 m, which also shows that the probe with modified probe tips is effective in reducing shattering.However, the two size distributions that have been processed using the arrival time algorithm to remove shattered particles contain far fewer particles < 200 m than either paper, the two (H and V) channels of each probe were averaged, and only regions where the two channels were in good agreement were used, avoiding any regions where obvious instrumentation effects adversely influenced the measurements.and fabricated based on information that Korolev delivered in person to SPEC engineers.Appendix C shows photographs and drawings of 2D-S probe tips that are germane to this paper.As shown in Fig. 3, the "standard" and "modified" 2D-S probes were about 1.5 m apart and were identical except for the probe tips.However, it is always important to keep in mind that even though two particle imaging probes are constructed to be identical, their comparative performance is rarely identical (e.g., Gayet et al., 1993).
On 23 July 2010 the Learjet penetrated a cloud region with only small cloud drops, where no shattering is expected, and a region of precipitation with large ice aggregates, where shattering is expected.Figure 4 shows average 2D-S drop size distributions in the region with only small cloud drops, with and without application of the shattering algorithm, for both 2D-S probes shown in Fig. 3.There is reasonably good agreement between the two probes in the cloud with only small drops where no shattering is expected, and application of the shattering algorithm had a negligible effect on the drop size distributions.This demonstrates that on this mission both probes recorded similar concentrations of small particles when there were no shattering effects.Later in the mission, the Learjet penetrated near the base of a thunderstorm anvil that contained a wide range of ice particle sizes, extending out to a few millimeters.This 3-min segment was selected because it was the only period with millimeter-size precipitating ice that the Learjet encountered on this mission.Figure 5 shows examples of typical images from both probes, i.e., with unmodified and with modified probe tips.Figure 6 shows a comparison of particle size distributions from the two 2D-S probes, with and without modified tips; and with and without application of the arrival time shattering removal algorithm.Figure 6 shows that all of the size distributions standard tips, and from 2.5 km to 2.2 km with the modified tips.The arrival algorithm reduces IWC from 100 mg m -3 to 84 mg m -3 with the standard tips, and from mg m -3 to 90 mg m -3 with the modified tips.Thus, in this example application of the arr time algorithm makes a much more significant impact on the second and third moment the size distribution than does the modified tips.The light green trace is from the probe with standard tips and includes shattered particles.A dark green trace is from the probe with standard tips after applying the shattering algorithm, but is not visible behind the light green trace.The red trace is from the probe with modified tips and includes shattered particles.A blue trace from the probe with modified tips after applying the shattering algorithm is barely visible near the red trace.
tend to converge at particle sizes larger than about 200 µm, which suggests that (in this case) the erroneous effects of shattering are less apparent in the larger portion of the size distribution. 3 The images in Fig. 5 suggest, and the size distributions in Fig. 6 support the premise that the modified probe tips significantly reduce, but do not eliminate shattering.During the penetration of precipitating dendrites, post processing of the 2D-S probe with standard tips identified 153 out of 507 (30 %) of the images >1 mm as shattered images.In comparison, the modified probe tips eliminated more than half this amount, with 54 out of 450 (12 %) of the images >1 mm being identified as shattered images.
light green trace.The red trace is from the probe with modified tips and includes shattered particles.A blue trace is from the probe with modified tips after applying the shattering algorithm is barely visible near the red trace.Figure 6 shows that the size distribution with the unmodified probe tips and without application of the arrival time removal algorithm (green trace) contains the most particles with sizes <200 µm.The size distribution with modified probe tips and without application of the arrival time removal algorithm (red trace) contains the second most particles with sizes <200 µm, which also shows that the probe with modified probe tips is effective in reducing shattering.However, the two size distributions that have been processed using the arrival time algorithm to remove shattered particles contain far fewer particles <200 µm than either of the other size distributions, regardless whether the probe has modified tips, or not.Thus, in this case, the arrival time algorithm is more effective than the modified probe tips in reducing the effects of shattering on these 2D-S probes.
The average bulk parameters, total number concentration (N), extinction coefficient (β ext ) and ice water content (IWC) are also shown in Fig. 6.In this example, without application of the arrival time algorithm the modified tips make a significant difference in N (707 l −1 vs. 214 l −1 ).Application of the arrival time algorithm reduces N from 707 l −1 to 37 l −1 with the standard tips and from 214 l −1 to 50 l −1 with the modified tips.Thus, the modified tips reduce N by 493 l −1 and application of the arrival time algorithm reduces N by only an additional 177 l −1 .On the other hand, without application of the arrival time algorithm the modified tips make no difference in β ext and only 3 mg difference in IWC.When the arrival algorithm is used in the computation of β ext and IWC, the differences are much larger.Application of the arrival time algorithm reduces β ext from 2.5 km −1 to 2.0 km −1 with the standard tips, and from 2.5 km −1 to 2.2 km −1 with the modified tips.The arrival time algorithm reduces IWC from 100 mg m −3 to 84 mg m −3 with the standard tips, and from 103 mg m −3 to 90 mg m −3 with the modified tips.Thus, in this example application of the arrival time algorithm makes a much more significant impact on the second and third moments of the size distribution than does the modified tips.
Another method for examining the effects of shattering is to generate a scatter plot of the concentration of small (shattered) particles versus the mass of large (shattering) particles (Jensen et al., 2009).In this way, it is possible to see if the concentration of smaller particles that may (or may not) Another method for examining the effects of shattering is to generate a scatter plot of the concentration of small (shattered) particles versus the mass of large (shattering) particles (Jensen et al. 2009).In this way, it is possible to see if the concentration of smaller particles that may (or may not) have been generated from shattering, are correlated with increasing mass of large particles, which are responsible for generating shattered particles.Figure 7 shows 1-Hz scatter plots of 2D-S data with standard tips and with modified tips from the 213418 -213716 UTC 23 July 2010 SPARTICUS anvil penetration discussed above.As seen in Fig. 7a, the scatter plot with the standard tips shows a very strong correlation between the concentration of small particles and increasing ice water content without application of the arrival time algorithm.After application of the arrival time algorithm there is still a correlation, but the magnitude of the trend is considerably less.Figure 7b shows that the modified tips reduces the number of small (shattered) particles, but the correlation between the concentration of small particles and increasing ice water content is still strong.have been generated from shattering, are correlated with increasing mass of large particles, which are responsible for generating shattered particles.Figure 7 shows 1-Hz scatter plots of 2D-S data with standard tips and with modified tips from the 21:34:18-21:37:16 UTC, 23 July 2010 SPAR-TICUS anvil penetration discussed above.As seen in Fig. 7a, the scatter plot with the standard tips shows a very strong correlation between the concentration of small particles and increasing ice water content without application of the arrival time algorithm.After application of the arrival time algorithm there is still a correlation, but the magnitude of the trend is considerably less.Figure 7b shows that the modified tips reduces the number of small (shattered) particles, but the correlation between the concentration of small particles and increasing ice water content is still strong.Application of the arrival time algorithm reduces the shattering trends in Fig. 7a and b to about the same magnitude, regardless whether the probe has standard or modified tips.The data in Fig. 7 suggests that, in this case, the arrival time algorithm produces approximately the same result, regardless of whether the 2D-S probe has standard or modified tips.The data also sug-gest that using the 2D-S probe with modified tips alone is not sufficient to reduce shattering to the level achieved after the arrival time algorithm is applied.
The scatter plots in Fig. 7 can also be interpreted to provide an estimate of the effectiveness of removing small ice particles due to shattering, or alternatively, an estimate of the maximum number of natural small ice particles.For example, the data in Fig. 7 show that for this particular case, for the unmodified tips, approximately 8500 l −1 of (spurious) small ice particles are produced for each g m −3 of large ice.Similarly, measurements from the probe with the modified tips show that about 2000 l −1 small ice particles are produced for each g m −3 of large ice.Since both probes yield about the same result after application of the arrival time algorithm, it is tempting (and possible) that the remaining ∼350 l −1 per g m −3 are real ice particles, naturally correlated with large ice mass content.However, it is also possible that these ice particles are still spurious events not removed by the tips and algorithm, i.e., there are no small ice particles in this region of precipitating aggregates.If this is the situation, then these data provide an estimate of the remaining spurious effects of ure 7. Scatter plots of the concentration of ice particles < 100 m versus ice water tent.Data collected in large ice aggregates with two 2D-S probes installed side-by-side he SPEC Learjet (Figs. 3 -6).One probe (a) had standard probe tips and the other probe was equipped with probe tips modified to reduce shattering.Effect of the arrival time rithm to remove shattering is shown on each plot.shattering, i.e, in this case ∼350 l −1 of small ice particles per g m −3 of large ice remain after shattering prevention with modified tips and removal with the arrival time algorithm.Since it is not possible to know the actual concentration of small ice particles, in this case: (1) 350 l −1 per g m −3 is an estimate of the upper bound of the possible remaining effects of shattering, or (2) an upper bound of the natural ice concentration in the cloud.
The measurements shown in Figs. 6 and 7 suggest that the modified tips reduce the number of small (shattered) particles, but not as effectively as the arrival time algorithm.Also, there is still a trend for increasing small particles with increasing ice water content, even with modified tips and application of the arrival time algorithm.This can be explained either by a process that is actually generating small particles when there are more large ice particles (e.g., particle-particle collisions), or that not all of the shattered particles are being removed by the modified tips and arrival time algorithm.We would like to point out a scenario where a shattered particle can be counted as a natural ice particle.If only one shattered small particle passes through the sample volume (i.e., the remainder of the shattered small particles are out of the depth of field), the one small particle in the depth of field will be not be rejected by the arrival time algorithm and will be counted as a natural ice particle (Korolev et al., 2010a).Since the depth of field of imaging probes is very small for small particles, the effective particle concentration is increased dramatically.The probability of this occurring is unknown at this time and would require a dedicated investigation, perhaps requiring high-speed video of shattered particles in various airborne flight and cloud conditions.However, the methodology presented here (i.e., Fig. 7 and associated discussion) is a method for estimating the maximum contribution from shattering.
After examining the data in Fig. 7, it is tempting here to state that shattering may have artificially increased the concentration of small particles by an order of magnitude.However, it is important to keep in mind that the contribution from shattering is relative to the natural concentration of small particles.If the contribution from shattering of very large particles in the example in Fig. 7 was hypothetically added to the concentration of natural small particles at the top of the cirrus cloud example in Fig. 2, then the contribution from shattering would add < 10 % to the total particle concentration.For this reason, we recommend reporting quantified additive effects and to avoid reporting multiplicative values.

AIIE field campaign
The Airborne Icing Instrumentation Evaluation (AIIE) field campaign was conducted near Ottawa in March-April 2009 (Korolev et al., 2010b).Data were collected with the National Research Council (NRC) of Canada Convair 580 research aircraft in deep precipitating glaciated cloud systems associated with frontal clouds.There was only one 2D-S probe available, so the probe was flown on one research flight without the modified tips, and then on another flight with the modified tips.Figure 8 is reproduced from Fig. 5 in Korolev et al. (2010b), with the addition of 2D-S measurements from the same time period.The data in Fig. 8a show that, contrary to 2D-S data shown in Fig. 7, modified 2D-C tips are more effective in removing small (shattered) particles than is the arrival time algorithm (indicated by "corr." in the figure) 4 .The CIP data in Fig. 8b shows the same trend as the 2D-C in Fig. 8a.In Fig. 8a and b, 2D-S data without ("no corr.") and with ("corr.")arrival time corrections are shown with modified probe tips.A comparison of all data in Fig. 8 suggest that the 2D-S probe with application of arrival time correction removes the most small (shattered) particles.The 2D-C probe with arrival time correction is closest to the 2D-S PSD, with the CIP probe showing the most deviation from the 2D-S results.
Figure 9 shows 2D-S measurements from data collected during the AIIE field program on two different flights in similar cloud conditions; one flight when the probe was flown 4 The particle arrival time algorithm applied to the AIIE data was developed and applied by Alexei Korolev.without modified probe tips, and the second flight with modified tips.Each data point in Fig. 9 represents a 15 to 30 s average calculated in the following way.One-hertz data 2D-S are screened for periods when the large (>55 µm) particle concentration exceeds 1 m −3 for a minimum of 15 s.During this period the 1-Hz median volume diameter must remain between 0.8 and 1.2 of its mean and the large particle concentration must remain between 0.4 and 2.0 of its mean.If an accepted time period exceeds 30 s the first 15 s are cut off as a separate period and the algorithm continues.So all accepted periods are between 15 and 30 s in duration.2D-S data from the flights with and without the modified tips were processed with and without application of the arrival time algorithm, producing the four scatter plots seen in Fig. 9.The data in Fig. 9 show that without applying the arrival time algorithm, the concentration of small (<55 µm) ice particles is about 6000 l −1 per g m −3 with the unmodified tips, compared with about 1000 l −1 per g m −3 with the modified tips.However, there is still a strong correlation between increasing concentration of small particles and ice water content, until the application of the arrival time algorithm.Once the arrival time algorithm is applied the average concentration of small ice particles is about 20 l −1 per g m −3 with both the unmodified and modified tips, and there is no correlation between increasing small ice particles and ice water content.
The data in Fig. 9 suggest that, for the 2D-S probe in these cloud conditions, the modified tips reduce, but do not eliminate the trend of increasing small particles with increasing ice water content.On the other hand, the data in Fig. 9 do show that, in this case, the arrival time algorithm eliminates the correlation between large and small (shattered) particles.It should be pointed out, however, that because there is no way of knowing the actual concentration of small particles (i.e., there could be none), this does not imply that the arrival time algorithm eliminates all of the shattered particles.As in Fig. 7, though, an estimate of the upper bound on the amount of shattered ice particles and natural ice particles can be derived from these scatter plots.The results shown in Figs.7 and 9 are only two examples, and shattering is likely to depend on many factors, including ice crystal size, type, airspeed, angle of attack and temperature, to mention some of the more important factors.For example, the data in Fig. 7 still show a (weak) correlation between large and small particles in the large particle region of an anvil cloud, even with modified tips and application of the arrival time algorithm.

ISDAC field campaign
The 2D-S, 2D-C and CIP probes were also flown together in April 2008 on the NRC Convair 580 research aircraft during the Department of Energy (DOE) Atmospheric Radiation Measurement (ARM) Indirect and Semi-Direct Aerosol Campaign (ISDAC).Figure 10 shows typical particle images and size distributions from several particle probes that were flown together below the base of a precipitating Arctic www.atmos-meas-tech.net/4/1361/2011/The lower left sides of Figs. 10 and 11 show size distributions without removing shattered particles, while the right sides show the same time periods using an arrival time algorithm to remove shattered (i.e., closely spaced) particles on the SPEC 2D-S and fast FSSP, the EC 2D-C and the DMT CIP. 5 The size distributions without application of the arrival time algorithm are all in reasonably good agreement, both below cloud base in precipitating dendrites (where small particles are not thought to be abundant), and in the mixed-phase region where the CDP and FSSP probes show about 80 cm −3 .In the precipitating dendrite size distributions (Fig. 10), the SPEC fast FSSP and 2D-S probes show a significant reduction in the concentration of small particles with the arrival time algorithm applied.The particle concentration in the size range from 5 to 300 µm is reduced from about 20 to 2 l −1 .The 2D-C, which has a 25-µm pixel size, also shows a reduction in particle concentration of about 3 to 0.3 l −1 in the 25 to 300 µm size range.On the other hand, there is very lit-tle change in the CIP size distribution and the small particle concentration actually increases in the smallest bins (due to re-sizing of some of the larger donut-shaped particles).In this case the application of an arrival time algorithm has a result similar to the AIIE results shown in Fig. 8, where both the 2D-S and 2D-C size distributions show significant reductions in small particles, whereas the CIP shows much less of an effect.
In the mixed-phase region of the same cloud (Fig. 11), the natural small particle (i.e., cloud drop) concentration is about 80 cm −3 , which is much higher than below cloud base.Even though the concentration of large particles in the mixed-phase is about the same as in the precipitation below cloud base, the total particle concentration is not significantly affected when the arrival time algorithm is applied.The most significant difference when the arrival time algorithm is applied is seen in the region from about 50 to 150 µm, but the percentage change is still quite small.When shattered particles are removed particle concentration in the 50 to 150 µm size range changes from 1.4 to 0.4 l −1 , extinction coefficient goes from 0.08 to 0.05 km −1 and ice water content changes from 7 to 5 mg m −3 .The percentage change in small (cloud drop) particles when shattered particles are removed is negligible.Particle concentration changes from 66 304 to 66 295 l −1 , extinction coefficient goes from 10.78 to 10.76 km −1 and liquid water content changes from 40.   makes a significant contribution to total particle concentration, whereas this is not the case when the natural ice concentration is high.physical credibility to the very large differences in 2D-S particle concentrations observed in the two regimes.The high particle concentration in the anvil cirrus is the result of homogeneous freezing of drops in convective updrafts and subsequent outflow in the anvil.Fridlind et al. (2004) show that model simulations produce in excess of 10 cm −3 of ice particles in the convective outflow regions of anvils.On the other hand, Comstock et al. (2008) show numerical simulations of synoptic cirrus with typical concentrations of 1 to 100 l −1 .Consistent with simple physical arguments and numerical simulations, Fig. 12 shows relatively low (11 l −1 ) concentrations of small particles in cirrus that was generated synoptically, and much higher (2059 l −1 ) concentrations of small particles in anvil cirrus, located several km downwind of convection.Note that unlike Figs. 1 and 2, the ordinate in Fig. 3 is dN/dlogD, which tends to emphasize the reduction of small particles in the plot.

MACPEX field campaign
relatively low (11 L -1 ) concentrations of small particles in cirrus that was generated synoptically, and much higher (2,059 L -1 ) concentrations of small particles in anvil cirrus, located several km downwind of convection.Note that unlike Figs. 1 and 2, the ordinate in Fig. 3 is dN/dlogD, which tends to emphasize the reduction of small particles in the plot.The data in Fig. 12 are also a good example of how quoting multiplicative shattering factors can be very misleading.The shattering algorithm reduces the total number concentration in the synoptic cirrus by an order of magnitude (i.e., from 109 L -1 to 11 L -1 ), while the anvil particle concentration is only reduced by 32% (i.e., from 3,017 L -1 to 2,059 L - The data in Fig. 12 are also a good example of how quoting multiplicative shattering factors can be very misleading.The shattering algorithm reduces the total number concentration in the synoptic cirrus by an order of magnitude (i.e., from 109 l −1 to 11 l −1 ), while the anvil particle concentration is only reduced by 32 % (i.e., from 3017 l −1 to 2059 l −1 ).However, the total number concentration of anvil particles removed by the algorithm (958 l −1 ) is an order of magnitude greater than the synoptic cirrus value (98 l −1 ).Thus, reporting only the multiplicative factor in the synoptic cirrus case can lead readers to assume that the order of magnitude concentration enhancement due to shattering could also apply to other cases, leading to erroneous conclusions.
The data shown in Fig. 12 also beg a question: why, with roughly the same total mass in each size distribution, does shattering in the anvil regime appear to produce 10 times the concentration of shattered particles?Fig. 13, which shows the anvil and synoptic cirrus size distributions as a function of mass, provides some insight.While the total mass in particles >∼100 µm is roughly the same in both size distributions, there is significantly more mass in sizes >1 mm in the anvil compared to the synoptic cirrus.Also, as shown in several previous studies, and the 2D-S images in Fig. 12, the mass distribution in synoptic cirrus is dominated by bullet rosette shapes, while anvils are typically composed of plates, aggregates of plates and columns (Connolly et al., 2005;Lawson  Two 2D-S probes, one with and one without modified probe tips, were flown side-byside on the SPEC Learjet in the SPARTICUS field project.The modified probe tips were designed with the assistance of Dr. Korolev, and are based on theory and knowledge learned from the icing tunnel videography.Analysis of data collected in large aggregates shows that the modified 2D-S probe tips substantially reduce the number of small (shattered) particles; however, post-processing with the arrival time algorithm is more effective, whether applied to the probe with modified or unmodified tips.This is contrary to the results for the 2D-C and CIP presented by Korolev et al. (2010b).et al., 2006b, 2010;Protat et al., 2011).In addition, Connolly et al. (2005) show that mid-latitude anvils often contain aggregated chains of small plates that could easily break apart when shattered.The observations in Figs. 12 and 13 further emphasize the complexity of the shattering issue and the need for more detailed study, including systematic studies of the shattering characteristics of particles with different sizes and shapes.

Summary and discussion
The effects of ice particles shattering on the tips of the 2D-S optical array probe, referred to in this paper as "shattering", are investigated.While shattering has been known for over 35 yr, under certain cloud conditions the magnitude of the contribution of shattered particles can be significant.NASA GRC supplied high-speed video photography in the Cox & Company icing tunnel of ice particles shattering on the tips of cloud particle probes.Korolev et al. (2010a) show that ice particles a few hundreds of microns in size shatter into hundreds of small ice particles with sizes that range from about 10 to 100 µm.As suggested from the comparison of 2D-S particle images in Fig. 5 and size distributions in Figs. 6, 10, 12 and 13, the large majority of these particles are in the size range from 10 to 50 µm (or perhaps even smaller).
Two techniques are currently used to reduce the effects of shattering: (1) probe tips designed to reduce the number of shattered ice particles that enter the sample volume, and (2) algorithms that remove shattered particles in post processing, mainly based on the observation that the shattered www.atmos-meas-tech.net/4/1361/2011/Atmos.Meas.Tech., 4,2011 particles are much more closely spaced than natural ice distributions.Korolev et al. (2010b) report results from analysis of 2D-C and CIP data collected from a field campaign (AIIE) that was designed to evaluate the effects of shattering on cloud particle probes.2D-C and CIP probes, each with standard tips and tips modified by Korolev 6 were flown side-by-side on the Canadian CV-580.The results show that the modified probe tips were more effective reducing shatterers than postprocessing with an arrival time algorithm.
Two 2D-S probes, one with and one without modified probe tips, were flown side-by-side on the SPEC Learjet in the SPARTICUS field project.The modified probe tips were designed with the assistance of Korolev, and are based on theory and knowledge learned from the icing tunnel videography.Analysis of data collected in large aggregates shows that the modified 2D-S probe tips substantially reduce the number of small (shattered) particles; however, post-processing with the arrival time algorithm is more effective, whether applied to the probe with modified or unmodified tips.This is a different result than Korolev et al. (2010b) obtained for the 2D-C and CIP probes.Korolev et al. (2010a) discuss a possible explanation for the apparent ineffectiveness of the arrival time approach when applied to the 2D-C and CIP probes.Due to the (50 µm to 100 µm) effective size resolution of the CIP and 2D-C, the imagery may miss many of the small shattered particles, producing large gaps that defeat the arrival time algorithm, and/or groups of small particles that are blurred together and appear as one larger particle.The explanation offered by Korolev et al. (2010a) is that it may appear that one (shattered) particle passes through the sample volume and is counted as a real particle.Because depth of field and thus sample volume are inversely proportional to the square of particle size, the one event generates a much higher concentration than does a large particle.The faster time response and greater effective size resolution of the 2D-S probe may enable it to more accurately reproduce particle spacing in a burst of shattered particles, which may make its particle arrival time algorithm more effective.
Results from SPARTICUS, ISDAC, AIIE and MACPEX show that post-processing 2D-S data with the arrival time algorithm (see Appendices A and B) is very effective in minimizing the effects of shattered ice particles.For the first time, measurements from major field programs (e.g., SPARTI-CUS, TC4, MACPEX) show vertical profiles of ice particle concentration in deep ice clouds (anvils and deep cirrus) that are consistent with physical arguments and numerical models.That is, relatively high concentrations of small ice near cloud top with decreasing concentration of small ice toward cloud base.Data from previous field campaigns (e.g., Lawson et al., 2006b) show increasing concentrations of small ice toward cloud base due to shattering on the older probes.A 6 See footnote # 2.
comparison of 2D-S particle concentrations from MACPEX in synoptically generated cirrus and a convectively generated anvil also more closely represent results from physical arguments and numerical simulations.
The effect of shattering on the second (extinction) and third (ice water content) moments of a cloud particle size distribution can be either minimal, or significant, depending on the instrument, cloud conditions and application of the measurements.The worse situation appears to be when scattering probes, such as the FSSP and the cloud and aerosol spectrometer (CAS), with tubular inlets, are used to measure ice particles in environments with high mass concentrations of large ice.Jensen et al. (2009) suggests that measurements under these conditions significantly skew the second moment of the size distribution, which generates the potential for misleading radiative computations.Ice water content is not as strongly affected as extinction, but significant errors can occur, such as suggested when an FSSP is used to measure small ice particles near the bases of deep cirrus clouds (Lawson et al., 2006b), or when a CAS is used to measure small particle concentration in thunderstorm anvils (Garrett et al., 2005;Fridlind et al., 2004).
All of the shattering results presented here are based on analyses of a few cases, using only three research aircraft, and in a limited number of cloud conditions and flight configurations.Results of the effects of shattering from the SPAR-TICUS, AIIE, ISDAC and MACPEX projects cannot be considered comprehensive or statistical.A statistical analysis was not possible with the available dataset and was not the focus of this paper.The takeaway message is that the limited dataset indicates that a post-processing algorithm based mainly on particle arrival times is more effective than modified 2D-S probe tips in reducing the effects of shattering.This result differs from analysis of a shattering analysis reported by Korolev et al. (2011b), who found that modified probe tips were more effective than a particle arrival time algorithm in reducing shattering on the CIP and 2D-S probes.Korolev et al. (2011b) suggest that, based on high-speed video in an icing tunnel, several factors appear to influence the amount of shattering on the inlets and tips of optical cloud particle probes.We suggest that factors that could potentially influence shattering include the probe geometry, time response, size resolution, probe installation on the aircraft, aircraft speed and angle of attack, ice particle size, type, air density and temperature.A comprehensive documentation of the results of shattering will likely require a comprehensive set of flight programs that address these variables.It is likely that estimates of the probability that one shattered ice particle will pass through a probe sample volume, thereby defeating any arrival time algorithm, will require analysis of high-speed, in-flight video.
5 Major conclusions -The effects of ice particles shattering on the tips of 2D-S probes are reduced by modified tips based on the design of Korolev 7 , and by post-processing based on an interarrival time algorithm (Appendices A and B).However, the inter-arrival time algorithm appears to be more effective than the modified tips with applied to 2D-S data, although use of both techniques is synergistic and recommended.This is a different result than obtained by Korolev et al. (2010b) from the AIEE field campaign, where it is reported that modified tips on CIP and 2D-C probes are more effective than an arrival time algorithm in reducing the effects of shattering.
-When compared to measurements from previous field campaigns that used older cloud particle probes, 2D-S data (with shattering removed) collected in recent (i.e., TC4, SPARTICUS, MACPEX) field campaigns show results that are consistent with physical arguments and numerical simulations.For example, ice particle concentrations in synoptic cirrus are two orders of magnitude less than anvil cirrus, even though total mass concentrations are similar in the regions selected for this analysis.Also, the vertical distribution of small ice particles in deep ice clouds decreases with distance from cloud top while the total mass of ice increases.
-A rigorous quantification of the effects of ice particles shattering on the inlets and tips of optical cloud particle probes is a complicated undertaking, because shattering is contingent on many factors that are a function of the probe, aircraft and environmental properties.However, Figs. 7 and 9, and other scatter plots of this type found in the literature, provide a methodology that forms a crude quantification of the upper bound of the effects of shattering.A comprehensive experiment with perhaps high-speed, in-flight video and extensive measurements in varying cloud regimes and aircraft configurations could shed more light on the physical processes involved in shattering, and also form the basis for statistical analysis.

Software processing of 2D-S Data
Processing of 2D-S image data is a complex process that has evolved based on both theoretical and empirical approaches.Image analysis and derived products include the convolution of multiple algorithmic processes.The convolution of algorithms includes software techniques that make adjustments to particle concentration and size that are used in both the 7 See footnote #2.
basic processing and removal of shattered particles.Thus, there is crossover of some techniques from one process to another (and consequently from Appendix A to Appendix B, and vice versa).The processing can loosely be divided into three broad steps: -Various methods to determine "characteristic" lengths, L i , and areas, A i , of an image.
-Removal of what are called here "spurious" events (also referred to as artifact rejection), which can include electronic noise, optical contamination, particle shattering and splashing effects.
-Various methods, M i , of estimating the bulk physical parameters; concentration, extinction, and mass as functions of size.These include correction for diffraction effects based on the Korolev (2007) methodology and adjustments to sample volume as a function of particle size.
These algorithmic processes require the introduction of various parameters that are defined throughout this Appendix and in Appendix B.
There are several different ways that L i can be measured.Figure A1 is a schematic depicting four measures of image length used in 2D-S analysis.L 1 is the number of slices (pixels in the direction of travel) for which a particle event lasted.L 2 is the number of shaded photodiodes (pixels in the direction along the array) for the slice for which the same quantity is maximized.L 4 is the number of diodes between, and including the shaded end diodes, for the slice that maximizes the same quantity.L 5 is the distance between (and including) the shaded end diodes considering all of the slices together. 8he appropriate selection of L i depends on the size and type of particles that are being imaged.For example, if particle sizes exceed the (1.28 mm) viewing area of the 2D-S, use of L 2 would limit the maximum particle size to 1.28 mm, whereas use of L 1 would provide a one-dimensional measurement of particles of any length, providing any part of the particle remained in the viewing area.
Several other size parameters are used in processing an image.L 7 is the diffraction-corrected length for out-of-focus ("donut") images based on Korolev (2007).A s is the number of occulted pixels for the entire image (summed over all slices).A t is the total number of pixels (occulted or not) contained within an image, which was developed by Korolev and is used in the Korolev (2007) diffraction-correction method.
Next we present general equations for calculating bin particle concentration, bin particle area assuming spheres, bin particle area not assuming spheres, bin liquid water content (LWC) and bin ice water content (IWC).The adjustment factor SV adj and sample volume (SV) are defined in specific terms later in Appendix B and below, respectively.
The general equation for a bin's concentration is: where # counts is the number of particles counted in that size bin, BW is the width of the size bin, and SV is the sample volume, defined below according to the method used.
The general equation for a bin's particle projected area assuming spheres is: The final step is another loop through the data to calculate the inter-particle spacings and reject particles that are close together.This algorithm is only performed if the current precipitation status is 'yes'.The current precipitation status is 'yes' if there is one or more particle(s) with L7 > 100 µm in 10,000 particles accepted by the first two cleaning steps, centered approximately on the current particle.It is approximately centered because the The second step of this first cleaning loop is another noisydiode removal method, based on the statistics of image center locations calculated over 4000 images9 approximately centered on the image being evaluated.Figure B2 shows an example of data with noisy photodiodes and the distribution of image centers across the array.When a photodiode is noisy, there are more image centers located on that diode than are located on quiet diodes.A photodiode is labeled noisy when it has more particle center locations than the threshold value (TH= max(M + 5 × √ M,1.5)).Where M is the mean number of image centers per diode across the array, for diodes having more than a minimum number of image center counts.This minimum number is Mt − 3 ×

√
Mt, where Mt is the mean number of image centers per diode across the array, for all diodes.If less than 33 photodiodes satisfy the requirement of having more than Mt counts then M is calculated from all diodes (i.e.M = Mt).Mt and M are recalculated after noisy diodes are identified, ignoring those diodes, and the process 2D-S images.This does not imply that the physics of ice particles shattering on probe tips are identical to raindrops splashing on probe tips, but instead to show that the resulting 2D-S images are very similar, suggesting that the particle inter-arrival time algorithm can be effectively applied in both cases.

Figure B.3.
Example of (left) 2D-S images in ice with particles in blue identified as artifacts using the algorithms described in this Appendix, and (right) plot of particle events versus inter-arrival distance showing (in red) the inter-particle distances before removing shatterers, (in black-grey) the remaining ("Accepted") particles' inter-particle distances after "Removing" shatterers, and (in green) an exponential distribution with the same mean as the after-shattering removal distribution, for comparison (labeled "Theoretical Distribution").For the sake of simplicity in nomenclature, the difference between the black and green distributions is what is labeled "Put Back", but in reality the results are adjusted by the factor SV adj , as described in the text.Fig. B3.Example of (left) 2D-S images in ice with particles in blue identified as artifacts using the algorithms described in this Appendix, and (right) plot of particle events versus inter-arrival distance showing (in red) the inter-particle distances before removing shatterers, (in black-grey) the remaining ("Accepted") particles' inter-particle distances after "Removing" shatterers, and (in green) an exponential distribution with the same mean as the after-shattering removal distribution, for comparison (labeled "Theoretical Distribution").For the sake of simplicity in nomenclature, the difference between the black and green distributions is what is labeled "Put Back", but in reality the results are adjusted by the factor SV adj , as described in the text.
time and the following 10 particle's inter-particle time are less than C 2 11 .
An exception occurs when a particle has L 1 > 10 and the inter-particle time for that particle (or the next particle) is less than 10, and the previous particle (next particle) has area <10.In these cases, the previous particle (next particle) is considered a diffraction satellite and is ignored.i.e. the previous (next), non-satellite, particle's inter-particle time is used.
A variable, SV adj , is calculated for each image in the following manner.SV adj is 1 if precipitation status is "no".When precipitation status is "yes" a new true average interparticle time is calculated.TaW = the average inter-particle time of particles accepted by all the steps minus C 1 .Assuming a true waiting time distribution with a mean of TaW, then 1 − e −C 1/ TaW is the probability of, or fraction of events with, inter-arrival times less than C 1 . 1 − e −2C 1/ TaW is the fraction of events eliminated due to the first criterion, leaving e −2C 1/ TaW not eliminated by the first criterion.Of those, a fraction 1 − e −C 2/ TaW 2 are eliminated by the second criterion.Thus, the fraction of events eliminated by the criteria given a true waiting time distribution with mean of TaW is ) 11 Events eliminated by steps 1 and 2 are not considered as following events.If the next event has been eliminated by either step 1 or step 2 then the next event, not eliminated by step 1 or step 2 is used and its inter-particle time is the time from the last event not eliminated by step 1 or step 2. Similarly, the current particles wait time must be from the previous event not rejected by step 1 or step 2. i.e., the wait times for particles not rejected by step 1 or step 2 are recalculated at this point.

Appendix C: Description of 2D-S Probe Tips
As described in the Introduction to this paper, the probe tips on the 2D-S were modified based on information communicated from Dr. Alexei Korolev to SPEC engineers.The 2D-S probe that is termed the "standard" or "unmodified" probe in this paper had tips that were designed by SPEC to minimize shattering, but without benefit of information later provide by Korolev.The 2D-S with "modified" probe tips was designed and fabricated based on information that Korolev delivered in person to SPEC engineers.Further, the modified probe tips were first applied as a retrofit to the standard probe by fastening adapter tips onto the existing probe arms.SV adj = 1 1−k (but capped at 1.11) is used elsewhere in the algorithms for calculating concentrations, area, etc., to increase the weight of each accepted particle to account for those good particles rejected by the inter-particle time criteria.See Appendix A for details. Figure B3 shows an example of 2D-S images in ice, identifying those images that have been removed via the algorithms described in Appendix B. The panel on the right in Fig. B3 shows (in red) the inter-particle distances before removing shatterers, (in black-grey) the remaining particles' inter-particle distances after removing shatterers, and (in green) an exponential distribution with the same mean as the after-shattering removal distribution, for comparison (labeled Theoretical Distribution in the figure).Note, these exponential and near exponential curves have apparent maxima only because they are plotted dN/dlog(N).For the sake of simplicity in nomenclature, the difference between the black and green distributions is what is labeled "Put Back" in the figure, but in reality the results are adjusted by the factor SV adj , as described above.
For comparison purposes, Fig. B4 shows an example of small drops generated from raindrops splashing on the tips of the 2D-S.The "splashers" are shown to highlight the similarity between ice particle "shatterers" (see Fig. B3) and raindrop "splashers" seen in 2D-S images.This does not imply that the physics of ice particles shattering on probe tips are identical to raindrops splashing on probe tips, but instead to show that the resulting 2D-S images are very similar, suggesting that the particle inter-arrival time algorithm can be effectively applied in both cases.

Appendix C Description of 2D-S Probe Tips
As described in the Introduction to this paper, the probe tips on the 2D-S were modified based on information communicated from Alexei Korolev to SPEC engineers.The 2D-S probe that is termed the "standard" or "unmodified" probe in this paper had tips that were designed by SPEC to minimize shattering, but without benefit of information later provide by Korolev.The 2D-S with "modified" probe tips was designed and fabricated based on information that Korolev delivered in person to SPEC engineers.Further, the modified probe tips were first applied as a retrofit to the standard probe by fastening adapter tips onto the existing probe arms.Figure C1.shows photographs of the standard probe tips and the standard probe tips with the adapters.2D-S data from the ISDAC field campaign and from the "standard" probe during the SPARTICUS field campaign were collected with standard probe tips.Data collected using a 2D-S probe with tips retrofitted with adapters were collected in the AIIE field campaign and by the modified probe during SPARTICUS.Lastly, new probe arms and tips based on the design used in fabricating the adapter tips were fabricated and installed on a new 2D-S probe for the MACPEX field campaign.A photograph of the 2D-S probe tips used in the MACPEX project and an engineering drawing showing plan and profile views of the modified probe tips are also shown in Fig. C1.Edited by: M. Wendisch

Figure 1 .
Figure 1.Example of (left) vertical profile of CPI images in a deep orographically generated cirrus, and (middle and right panels) number and mass particle size distributions for three temperature ranges generated from 102 horizontal legs in cirrus, where average size distribution distributions are shown in red.Small end of the size distributions are based on measurements from FSSP and large end from 2D-C probe, with CPI data scaled to fit in between.Adapted from Lawson et al. (2006b).

Fig. 1 .
Fig. 1.Example of (left) vertical profile of CPI images in a deep orographically generated cirrus, and (middle and right panels) number and mass particle size distributions for three temperature ranges generated from 102 horizontal legs in cirrus, where average size distributions are shown in red.Small end of the size distributions are based on measurements from FSSP and large end from 2D-C probe, with CPI data scaled to fit in between.Adapted from Lawson et al. (2006b).

Figure 2 .
Figure 2. Example from the SPARTICUS project showing (left) concentration and m particle size distributions derived from the 2D-S and (right) images from the CPI.

Fig. 2 .
Fig. 2. Example from the SPARTICUS project showing (left) concentration and mass particle size distributions derived from the 2D-S and (right) images from the CPI.

Figure 3 .
Figure 3. Photograph showing the SPEC Learjet used in the SPARTICUS project with two 2D-S probes, one with tips modified to reduce shattering, and the other with standard tips.

Fig. 3 .
Fig. 3. Photograph showing the SPEC Learjet used in the SPARTI-CUS project with two 2D-S probes, one with tips modified to reduce shattering, and the other with standard tips.

Figure 4 .
Figure 4. 2D-S size distributions from Learjet penetration (205408 205421 UTC on 23 2010) of a small cumulus cloud containing only water drops.The light green trace is from probe with standard tips and includes shattered particles.A dark green trace is from

Fig. 4 .
Fig. 4. 2D-S size distributions from Learjet penetration (20:54:08 to 20:54:21 UTC, 23 July 2010) of a small cumulus cloud containing only water drops.The light green trace is from the probe with standard tips and includes shattered particles.A dark green trace is from the probe with standard tips after applying the shattering algorithm, but is not visible behind the light green trace.The red trace is from the probe with modified tips and includes shattered particles.A blue trace from the probe with modified tips after applying the shattering algorithm is barely visible near the red trace.

Figure 5 .
Figure 5. Examples of 2D-S images from two 2D-S probes flown side-by-side on the SPEC Learjet (Fig. 3) during the SPARTICUS project, one probe had standard 2D-S probe tips (left) and (right) the other with tips modified to reduce the effects of shattering.

Fig. 5 .
Fig. 5. Examples of 2D-S images from two 2D-S probes flown side-by-side on the SPEC Learjet (Fig. 3) during the SPARTICUS project, one probe had standard 2D-S probe tips (left) and (right) the other with tips modified to reduce the effects of shattering.

Figure 6 .
Figure 6.Average particle size distributions derived from 2D-S measurements collected in large ice aggregates from 213418 -213716 UTC on 23 July 2010.Data are from two 2D-S probes installed side-by-side on the SPEC Learjet (Figs. 3 -5).One probe had standard probe tips and the other probe was equipped with probe tips modified to reduce shattering.Size distributions are shown with and without the effects of an arrival time algorithm to remove shattering.Total particle concentration (N), extinction coefficient ( ext ) and ice water content (IWC) were derived from each average size distribution.

Fig. 6 .
Fig. 6.Average particle size distributions derived from 2D-S measurements collected in large ice aggregates from 21:34:18-21:37:16 UTC, 23 July 2010.Data are from two 2D-S probes installed side-by-side on the SPEC Learjet (Figs. 3-5).One probe had standard probe tips and the other probe was equipped with probe tips modified to reduce shattering.Size distributions are shown with and without the effects of an arrival time algorithm to remove shattering.Total particle concentration (N ), extinction coefficient (β ext ) and ice water content (IWC) were derived from each average size distribution.

Fig. 7 .
Fig. 7. Scatter plots of the concentration versus ice water content of ice particles <100 µm.Data collected in large ice aggregates with two 2D-S probes installed side-by-side on the SPEC Learjet (Figs. 3-6).One probe (a) had standard probe tips and the other probe (b) was equipped with probe tips modified to reduce shattering.Effect of the arrival time algorithm to remove shattering is shown on each plot.

Fig. 8 .
Fig. 8. Particle size distributions from 14:09:00-14:21:00 UTC, 8 April 2009, collected during the AIIE field project.Data are reproduced from Korolev et al. (2010b) with the addition of 2D-S data for the same time period."stand."Means standard probe tips; "modif."means modified probe tips, "corr."means data have been adjusted using an arrival time algorithm and "no corr."means that no arrival time algorithm has been applied.

Fig. 9 .
Fig. 9. Scatter plots of the concentration of ice particles (<55 µm) versus ice water content =>55 µm.Data were collected during the AIIE field project with a 2D-S probe installed on the NRC Convair 580.Data were collected with standard tips from 17:54:53-20:16:44 UTC, 1 April 2009.Data were collected with modified tips from 14:32:32-16:24:37 UTC, 4 April 2009.The 2D-S probe was flown with standard tips on one flight (top two panels), then with modified tips in similar conditions on a second flight (bottom two panels).Effect of the arrival time algorithm to remove shattering is shown on plots on the right side.
3 to 40.0 mg m −3 .A comparison of Figs. 10 and 11 emphasizes the reason why shattering should be reported as an additive effect and not a multiplicative effect.Jensen et al. (2009) show a result similar to Figs. 10 and 11 for low and high concentrations of natural small ice at the top of an aged tropical anvil cloud; i.e., shattering with low natural iceimages and size distributions from the same instruments flown in the mixed-phase region 300 m above the base of the same Arctic stratus cloud from 02:48:40 to 02:48:53 UTC.Only the 2D-C had probe tips with an aggressive design to reduce shattering.However, the 2D-S probe did have tips that were designed to reduce shattering, based on understanding of ice particle shattering at that time.

Figure 10 .
Figure 10.(top) 2D-S Images and (bottom) particle size distributions from several cloud particle probes flown on the Canadian Convair 580 in precipitating dendrites below cloud

Figure 12 Figure 11 .
Figure12shows another example of 2D-S measurements from a very recent field campaign, the Mid-latitude Airborne Cirrus Properties Experiment (MACPEX).The measurements are consistent with numerical models of cirrus and cirrus anvils, which are formulated on basic physics that lend

Figure 12 .
Figure 12.Particle size distributions as a function of concentration from 2D-S probe measurements in anvil and synoptic cirrus sampled on 14 April 2011 from 183600 -185100 UTC (synoptic cirrus), and 11 April 2011 from 173750 -184610 UTC (anvil cirrus) by the NASA WB-57 during the MACPEX field campaign.Example 2D-S images from each time period are shown below the size distributions.

Figure 13 .
Figure 13.As in Fig. 12 except the size distributions are shown as a function of particle mass.

Fig. 13 .
Fig. 13.As in Fig. 12 except the size distributions are shown as a function of particle mass.

Figure A. 1 .
Figure A.1.Two example particle images designed to demonstrate the four measures of image size described in this Appendix.

Fig. A1 .
Fig. A1.Two example particle images designed to demonstrate the four measures of image size described in this Appendix.
Figure B.2. Example of noisy photodiode data intermixed with good particle data.The images highlighted in grey are rejected.Horizontal lines indicate the location of photodiodes determined to be noisy by the criteria described in the text and exemplified in the particlecenter-location distribution shown above.

Fig. B2 .
Fig. B2.Example of noisy photodiode data intermixed with good particle data.The images highlighted in grey are rejected.Horizontal lines indicate the location of photodiodes determined to be noisy by the criteria described in the text and exemplified in the particlecenter-location distribution shown above.

Figure B. 4 .
Figure B.4. Example of 2D-S images in raindrops with particles in grey identified as artifacts using the algorithms described in this Appendix (from Baker et al. 2009)

Figure C. 1 .
shows photographs of the standard probe tips and the standard probe tips with the adapters.2D-S data from the ISDAC field campaign and from the "standard" probe during the SPARTICUS field campaign were collected with standard probe tips.Data collected using a 2D-S probe with tips retrofitted with adapters were collected in the AIIE field campaign and by the modified probe during SPARTICUS.Lastly, new probe arms and tips based on the design used in fabricating the adapter tips were fabricated and installed on a new 2D-S probe for the MACPEX field campaign.A photograph of the 2D-S probe tips used in the MACPEX project and an engineering drawing showing plan and profile views of the modified probe tips are also shown in Fig. C.1.

Fig. B4 .
Fig. B4.Example of 2D-S images in raindrops with particles in grey identified as artifacts using the algorithms described in this Appendix (from Baker et al. 2009).

Figure C. 1 .
Figure C.1.(top left) Photograph of "standard" (unmodified) 2D-S probe tips used in ISDAC and by the standard probe in SPARTICUS.(upper right) Photograph of 2D-S probe with adapter tips fitted onto the standard probe arms, which was used in the AIIE field campaign and by the modified probe during SPARTICUS (note that the outer surfaces of the adapter tips have rime ice on them).(lower left) Photograph of the 2D-S probe tips that are designed based on the adapter tips shown at the upper right, but are integrated with the 2D-S probe arms; these tips were used in the MACPEX project.(lower right) Engineering drawing showing shape and dimensions of the modified probe tips.

Fig. C1 .
Fig. C1.(top left) Photograph of "standard" (unmodified) 2D-S probe tips used in ISDAC and by the standard probe in SPARTI-CUS.(upper right) Photograph of 2D-S probe with adapter tips fitted onto the standard probe arms, which was used in the AIIE field campaign and by the modified probe during SPARTICUS (note that the outer surfaces of the adapter tips have rime ice on them).(lower left) Photograph of the 2D-S probe tips that are designed based on the adapter tips shown at the upper right, but are integrated with the 2D-S probe arms; these tips were used in the MACPEX project.(lower right) Engineering drawing showing shape and dimensions of the modified probe tips.