2.1 Optical array probes

This paper uses data from two types of commercially available OAP: a CIP-15 (cloud imaging probe, DMT Inc., USA; Baumgardner et al., 2001) and a 2D-S (2D stereo, SPEC Inc., USA; Lawson et al., 2006). The CIP-15 has a 64-element photodiode array and effective pixel size of 15 µm. The laboratory experiments were conducted with a CIP-15 with an arm separation of 70 mm (Sect. 3.1) and the field measurements with a second CIP-15 with an arm separation of 40 mm (Sect. 4.1). Images are recorded at three greyscale intensity thresholds. For this work, they were set to the manufacturer default settings of 25 %, 50 %, and 75 %. The 2D-S consists of two optical arrays and lasers orientated at right angles to each other and the direction of motion of the particles and aircraft. The laser beams overlap at the centre of the probe's arms, and each pair of transmit and receive arms are separated by 63 mm. Each optical array has 128 elements and 10 µm pixel resolution. The 2D-S is a monoscale probe with a single 50 % intensity detection threshold. Both probes are fitted with anti-shatter tips to minimise ice shattering on the leading edge of the probe during field measurements. This was further minimised by removing particles with inter-arrival times less than 1 × 10⁻⁵ s when calculating PSDs from field measurements (Field et al., 2006).

Baumgardner and Korolev (1997) show that the electronic time response of older probes can significantly reduce the DoF of small particles. This effect has been minimised in more modern probes such as the 2D-S and CIP-15, which have an order of magnitude faster time response.

A range of definitions have been used to define the diameter of ice crystals from OAP images. Here we test three metrics that have been widely used by the community. First, the mean of the particle extent along the axes parallel and perpendicular to the optical array (mean X–Y diameter). Second, the diameter calculated using $D=(\mathrm{4}A/\mathit{\pi }{)}^{\mathrm{1}/\mathrm{2}}$, where A is the particle area calculated as the sum of the pixels (circle equivalent diameter). Third, the major axis length of the ellipse that has the same normalised second central moments as the region (maximum diameter).

An image frame from the OAP may contain more than one object, where individual objects are defined as collections of pixels with eight-neighbour connectivity. This can be due to diffraction, with a single particle appearing as more than one object as the structure and intensity of the transmitted light degrades due to poor focus. However, it may also be due to shattering causing multiple particles to have sufficiently small separations that they are captured in the same image frame or occasionally when there are very high concentrations of ambient particles. A particle sizing metric can either relate to the largest object in an image frame or use the bounding box encompassing all objects. Some previous studies have filled any internal voids within objects in an image frame. For this work, unless otherwise stated, the mean X–Y, maximum, and circle equivalent diameters are calculated using the bounding box encompassing all objects in an image frame, and any internal voids are not filled.

2.2 Ice crystal analogues

Three-dimensional ice crystal analogues were grown from a sodium fluorosilicate solution (Ulanowski et al., 2003). These analogues have similar crystal habits to ice and a refractive index of 1.31, virtually identical to that for ice at visible wavelengths. Three rosette shapes were used in these experiments with approximate diameters 118 µm (ROS118), 250 µm (ROS250), and 300 µm (ROS300) (Fig. 1). The CIP-15 was mounted as shown in Fig. 2 so that the laser beam was vertically aligned. Each analogue was in turn placed on an anti-reflective optical window that was attached to a three-axis translation system that allowed the analogue's 3D position to be controlled. The stages that moved along the axes parallel (x axis) to the diode array and laser beam (z axis) each had a unidirectional position accuracy of 15 µm and travel ranges of 100 and 150 mm, respectively (X-LRM050A and Z-LRM150A, Zaber Technologies Inc., Canada). Movements along the axis that air flows through the probe under normal operation (y axis) were made using a belt-driven stage with a maximum speed of 1.1 m s⁻¹, positional accuracy of 200 µm, and maximum travel range of 70 mm (X-BLQ0070-EO1, Zaber Technologies Inc., Canada).

Figure 1Microscope images of sodium fluorosilicate crystals that were used as analogues for ice crystals. These are referred to as ROS118 (a), ROS250 (b), and ROS300 (c).

Download

Figure 2(a) Image of the experimental set-up for the ice crystal analogue tests of the CIP-15. Panel (b) shows a schematic of the experimental set-up. The CIP-15 is horizontally mounted on the left of the image. The translation stages used to move ice crystal analogues through the CIP-15 sample volume are shown on the right of the image. The x axis is perpendicular to the plane of drawing.

Download

CIP-15 images of the ice crystal analogues were collected by moving them through the laser beam along the axis of airflow. For each analogue, this was repeated five times before its position was stepped in 0.5 mm increments between the probe's vertical arms (along the z axis). This allows images of the analogues to be compared at different distances from the object plane.

Images were post-processed to take account of any difference in velocity between the stage and the CIP-15 data acquisition rate by resampling the images along the axis perpendicular to the optical array. This was performed to match the aspect ratio at Z = 0 of the CIP-15 image and a microscope image of each analogue. This typically corresponded to a particle stage velocity of ∼ 0.1 m s⁻¹.

2.3 Synthetic data (angular spectrum theory)

Theoretical shadow images of 2D non-spherical shapes were calculated using a diffraction model based on angular spectrum theory (referred to as the AST model). Several previous studies describe this model in detail (Vaillant de Guélis et al., 2019a, b). We initialised the model using a 2D binary image of an opaque shape at the object plane (Z = 0) and calculated the wave field for different positions between the probe arms in the z axis. This model has been shown to give good agreement with OAP images of several types of 2D rectangular columns using images printed on a rotating disc (Vaillant de Guélis et al., 2019a).

In this study, we use a variety of different shapes to initialise the model. In Sect. 3.1, the diffraction model is compared to CIP15 images of 3D ice crystal analogues. To initialise the model for the comparisons with ROS250 and ROS300, the CIP-15 image of them at Z = 0 is used. Due to the smaller size of ROS118 and coarse pixel size of the CIP-15, a microscope image of the analogue is used to initialise the model. This image was converted to a binary image.

In Sect. 3.2 the quality of OAP images of commonly occurring ice crystal habits is explored. This is done by initialising the model with a variety of different ice crystal images. The ice crystal dataset contains 1060 images that were collected using a cloud particle imager (CPI, SPEC Inc., USA) and has previously been used to train habit recognition algorithms (Lindqvist et al., 2012; O'Shea et al., 2016). It includes images of ice crystals from arctic, mid-latitude, and tropical clouds. These images have been manually classified into seven habits (rosette, column/bullet, plate, quasi-spherical, column aggregate, rosette aggregate, and plate aggregate). To initialise the model, each CPI image was converted to a binary image. Shadow images were calculated every 2 mm for the range Z = 0 to 100 mm. These images were averaged to 10 µm pixel resolution, which is typical of modern OAPs. All simulations were performed using a light wavelength of 0.658 µm.

Figure 3Diffraction simulations from an image of a rosette crystal collected in cirrus cloud using a CPI (see text for details). Panel (a) shows the image at Z = 0 that is used to initialise the model. The other panels show images at different distances from the object plane (Z = 5, 10, and 20 mm). Green, blue, and black pixels correspond to decreases in detector intensity of 25 % to 50 %, 50 % to 75 %, and greater than 75 %, respectively.

Download

Figure 4Same as Fig. 3 but for a column.

Download

An example simulation for a rosette crystal is shown in Fig. 3 and a column in Fig. 4; the top left panels show the images at Z = 0 that are used to initialise the model. The other panels show images of the crystals at different distances from the object plane. Green, blue, and black pixels correspond to decreases in detector intensity of 25 % to 50 %, 50 % to 75 %, and > 75 %, respectively. Figures 3 and 4 show the rapid deterioration in image quality within a few millimetres of the object plane, which will impact derived properties such as particle size, number, and habit. This compares to many tens of millimetres for the typical arm separation of modern OAPs.

2.4 Aircraft measurements

This paper uses measurements from three flights by the Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 research aircraft sampling frontal cirrus in the UK on the 11 March 2015 (nominal flight number B895), 7 February 2018 (C078), and 23 April 2018 (C097). The first two flights have previously been described in detail by O'Shea et al. (2016) and O19. For all three flights, the aircraft performed straight and level runs of approximately 10 min at different temperatures within the cloud. Ice crystals were dominated by rosettes, columns, and aggregates. Data from a 2D-S is available for the 11 March 2015 and CIP-15 for 7 February and 23 April 2018. On all flights the FAAM BAe-146 was fitted with a holographic imaging probe (HALOHolo). HALOHolo has a 6576 × 4384 pixel CCD (charged-coupled device) detector with an effective pixel size of 2.95 µm and arm separation of 155 mm. The probe acquires six frames per second, which equates to a volume sample rate of ∼ 230 cm³ s⁻¹. The detection of small particles is limited by noise in the background image. Therefore, a minimum size threshold of 35 µm is applied, above which it is estimated that the probe's detection rate is greater than 90 % (Schlenczek, 2017). Shattered particles were minimised by removing all particles with interparticle distances less than 10 mm (Fugal and Shaw, 2009; O'Shea et al., 2016).

Section 4.2 shows a comparison between the 2D-S and a cloud droplet probe (CDP, DMT Inc.) during a flight in liquid stratus on 17 August 2018 (C031). The CDP sizes particles (3 to 50 µm) using the scattered light intensity assuming Mie scattering theory and spherical particles (Lance et al., 2010). The probe was calibrated during the campaign using glass spheres.

3.1 OAP and AST model comparison using ice crystal analogues

This section compares CIP-15 images of ice crystal analogues with diffraction simulations using the AST model. Figures 5–7 show the image size of the ice crystal analogues ROS118, ROS250, and ROS300 at different distances (Z) from the object plane measured by the CIP-15 (black markers) and modelled using angular spectrum theory (red lines). Top left panels show the image diameter (mean X–Y), while the particle area is shown in the top right; both use a 50 % drop in light intensity for the detection threshold. The other panels show different combinations of simple greyscale ratios. The abbreviations A_25–50, A_50–75, and A_75–100 are used to denote the number of pixels associated with a decrease in detector signal of 25 %–50 %, 50 %–75 %, and 75 %–100 %, respectively. Example CIP-15 images of the ice crystal analogue ROS300 at three distance from the object plane are shown in Fig. 8.

Figure 5A comparison between CIP-15 images and diffraction simulations (red lines) of the ice crystal analogue ROS118. Grey dots show data from individual CIP-15 images, and black dots show the median for each 1 mm Z bin. Panel (a) shows the mean X–Y image diameter. Panel (b) shows the number of pixels using 50 % detection thresholds. Other panels show the ratio of the number of pixels (area) at different greyscale thresholds.

Download

Figure 6Same as Fig. 5 but for the ice crystal analogue ROS250.

Download

Figure 7Same as Fig. 5 but for the ice crystal analogue ROS300.

Download

Figure 8CIP-15 images of the ice crystal analogue ROS300 at three distances from the object plane.

Download

All three analogues have a general trend of diameter initially increasing with Z. The full DoF was sampled for ROS118 and shows the images fragmenting and diameter decreasing near the edge of the DoF. In addition to these general trends, there is a significant amount of fine-scale structure that is specific to each sample. There is a general trend of the greyscale ratio A_75–100 decreasing with Z, while both A_25–50 and A_50–75 initially increase for all three analogues. Like the diameter vs. Z plots, there is a significant amount of fine-scale structure overlaying these general trends.

In general, the AST model can capture the large-scale structure in these measured parameters, although some discrepancies are present in the finer detail. For ROS118, the DoF from the experiments and the model agree to within ±1 mm (Fig. 5). The size and greyscale parameters calculated from CIP-15 images are not completely symmetrical about Z = 0. The reason for this is unclear; potential causes are if the CIP-15 laser beam is not perfectly collimated, additional refraction caused by the optical window used to mount the sample, or changes to the CIP-15 background or dark current calculation due to attenuation by the optical window.

3.2 OAP ice crystal sizing

Having investigated the performance of the AST model using 3D analogues of complex ice, we will now use the AST model to examine the ability of OAPs to correctly determine the size of commonly occurring ice crystals. Figure 9 left panels show the ratio of the measured diameter (D) to the true diameter (D₀) vs. Z_d for diffraction simulations of 1060 ice crystals. The data for each individual ice crystal are shown as grey lines, while the coloured lines are the median for each habit. Top panels show plots using the circle equivalent diameter, while the middle panels use the mean X–Y diameter and maximum diameter. Right panels show histograms of $D/D_{\mathrm{0}}$ for each habit calculated for the Z_d range from 0 to 10.

Figure 9Left panels show the ratio of the measured diameter (D) to the true diameter (D₀) vs. Z_d for diffraction simulations of 1060 ice crystals. The data for each individual ice crystal are shown as grey lines, while the coloured lines are the median for each habit. Right panels show histograms of $D/D_{\mathrm{0}}$ for each habit calculated for the Z_d range 0 to 10. Top panels show plots using the circle equivalent diameter, while the middle panels use the mean X–Y and maximum diameters. Bottom panels show the diameter corrected using K07.

Download

Figure 9 shows large differences in these relationships depending on whether the mean X–Y, maximum, or circle equivalent diameters are used to define the particle size. For the 1060 ice crystal images used in this study, the median $D/D_{\mathrm{0}}$ over the Z_d range from 0 to 8 is 1.1 using circle equivalent diameter, 1.0 using the mean X–Y diameter, and 1.0 using the maximum diameter. However, there is significantly less variability between crystals using circle equivalent diameter, which has an interquartile range $D/D_{\mathrm{0}}$ of 0.2 compared to 1.1 and 1.3 using the mean X–Y and maximum diameters, respectively. This is also shown in Tables S1–S3 in the Supplement, which gives the median and interquartile range $D/D_{\mathrm{0}}$ at selected Z_d for each habit using the three different size metrics.

There is a general trend of increasing size with distance from the object plane. Oversized estimations are up to approximately 100 %, 200 %, and 50 % using mean X–Y, maximum, and circle equivalent diameters, respectively. However, the degree of oversizing is dependent on habit, with quasi-spherical and plate aggregates most significantly oversized using all D definitions. In agreement with O19, once D reaches a maximum, further increases in Z cause the images to fragment and their size to decrease until they are no longer visible.

K07 use the size of the internal voids within images of droplets to determine their Z_d and correct their size. O19 show that this algorithm is effective using modern OAPs for droplets with Z_d < ∼ 6. For Z_d > 6, the images are too fragmented for their size to be corrected. The K07 approach was derived by considering Fresnel diffraction from opaque discs and has only been tested for images of spherical droplets. However, in the absence of an alternative, previous studies have applied K07 to images of ice crystals (e.g. Davis et al., 2010). To examine the efficacy of this approach, Fig. 9 bottom panels show the mean X–Y diameter of the simulated images of ice crystals once the K07 approach has been applied. The ratio of their K07 corrected diameter to their true particle diameter is shown as a function of Z_d (left panel), while probability density functions of $D/D_{\mathrm{0}}$ for each habit are shown in the right panel. The median $D/D_{\mathrm{0}}$ for the Z_d range 0 to 8 is 0.9, and the interquartile range is 1.1. For a number of habits (rosette, plate, quasi-spherical, rosette aggregate, and plate aggregate), K07 reduce the number of oversized particles across most of the DoF. For bullets, columns, and column aggregates, the K07 approach has minimal impact on the probe sizing. For all habits, the K07 approach is not able to remove the small image fragments that occur when a particle is near the edge of the DoF.

3.3 Depth of field dependence on particle habit

Uncertainty of derived physical quantities (e.g. number concentration) from OAPs is dependent on the sample volume and therefore uncertainty in the DoF (see Eq. 2). The DoF of an OAP is commonly calculated using Eq. (1) with a single c value. The variable c in this equation is the Z_d where a particle is no longer detected by the OAP. If a single c value is used this would need to be independent of particle shape. Table 1 shows the median and interquartile range Z_d where particles are no longer visible for each habit using the maximum, mean X–Y and circle equivalent diameters. Using mean X–Y, the habit median DoF varies between Z_d = 5.0 and 9.9 for rosettes and quasi-spherical particles, respectively. Using the maximum as the particle sizing metric, the median DoF varies by a similar amount ranging between Z_d = 3.4 and 7.8 for bullets and quasi-spherical crystals. In addition, particles have significant intra-habit variability using both maximum and mean X–Y, with most habits DoF interquartile ranges greater than 2 Z_d. The variability is lower using circle equivalent diameter, with median DoFs ranging between 8.2 and 10.2 for plates and bullets, respectively, with habit interquartile ranges near 1 Z_d. As a result, derived physical quantities such as number concentration will have lower uncertainty if circle equivalent diameter is used to define the particle size compared to maximum and mean X–Y diameters.

Table 1Median and interquartile range (IQR) normalised dimensionless distance from the object plane (Z_d) where particles are no longer visible for different habits; this is equivalent to c in Eq. (1).

Download Print Version | Download XLSX

3.4 Greyscale information

Greyscale information in OAP imagery has previously been used to filter severely mis-sized images and enforce a DoF threshold that improves data quality (O19). Figure 10 shows combinations of simple greyscale ratios as a function of Z_d for the simulation of 1060 ice crystal images described in the previous section. Left panels use the size metric mean X–Y diameter in the Z_d calculation, whereas the right panels use circle equivalent diameter in the Z_d calculation. Like the ratio $D/D_{\mathrm{0}}$ (Fig. 9), the greyscale ratios also show significant variability between habits as a function of Z_d. Figure 10 shows this variability is greater if mean X–Y diameter is used to calculate Z_d, although it is still significant using circle equivalent diameter. The variability is larger still using maximum diameter (not shown).

Figure 10Combinations of number of pixels at different greyscale ratios as a function of Z_d for the simulated ice crystal images. Panels (a), (c), and (e) show plots where mean X–Y is used as the sizing metric, while panels (b), (d), and (f) use circle equivalent diameter.

Download

The O19 approach uses simple greyscale ratios to determine Z_d for spherical liquid droplets near the edge of the DoF (3.5 < Z_d < 8.5). This allows a new DoF to be defined that excludes fragmented images, removing significant biases in the PSD. This is possible since all spherical droplets independent of size have the same greyscale ratios at a given Z_d. Figure 10 shows that this is not true for ice crystals where the initial shape of the ice crystal has an impact on the greyscale ratios at a given Z_d. As a result, the O19 approach cannot be used to determine Z_d in the same way.

3.5 Habit recognition

The shape of ice crystals is a key microphysical parameter impacting cloud radiative properties in several ways. A variety of automatic image recognition algorithms have been applied to OAP datasets to classify particles into different habits (Korolev and Sussman, 2000; Crosier et al., 2011; Praz et al., 2018). These algorithms typically rely on geometrical features extracted from OAP images that have characteristic values for specific habits. These characteristic values are usually determined by manually classifying images into habits. These images are then used to set thresholds or train machine learning algorithms to automatically classify new images. For example, Crosier et al. (2011) used the following ratio to discriminate between ice crystals and liquid droplets:

where P is the particle perimeter, and A is the particle area including any internal void. Crosier et al. (2011) used a threshold of 1.25 to discriminate between these two categories. When images are manually selected to train habit recognition algorithms, only images that can be identified “by eye” as a specific habit will be included. For OAPs this is likely to be images that are “in focus”. However, the shape of an OAP image and therefore the geometrical features that are used in habit recognition algorithms depend on where in the probe's sample volume a particle is detected. For example, Fig. 3 shows a simulated 190 µm rosette at different distances from the object plane. It is only in the top left panel (Z = 0) that it can be identified as a rosette from its image alone. Figure 11 shows how this particle's circularity changes with Z and Z_d. At Z = 0 its circularity is near 4, while at Z = 20 mm it is near 1 and may be confused with a spherical droplet. Figure 11 demonstrates that the measured particle shape is highly dependent on the position in the sample volume Z_d (and Z) with the circularity decreasing by a factor 2 by Z_d = 1; in comparison the particle size has only changed by 15 %.

Figure 11The circularity (Eq. 4) of the rosette shown in Fig. 3 as a function of distance from the object plane Z and Z_d.

Download

The variance in geometrical features for each habit will not only be due to natural variability in the shape of ice crystals but also due to their position in the sample volume when measured. To date, this second effect has not been accounted for by habit recognition algorithms. Therefore, currently the results of habit classification algorithms on OAP datasets cannot be considered quantitative.

4.1 Greyscale filtering

Unlike for liquid droplets, the O19 approach does not accurately determine Z_d for non-spherical ice crystals. We now describe a new technique to use greyscale information to remove the most severely mis-sized ice crystals and constrain the sample volume with a reasonable uncertainty using circle equivalent diameter as the particle sizing metric. For example, if the diffraction simulations are filtered to only include images that have at least one pixel with a greater than a 75 % drop in light intensity (Fig. 7), then the median position where particles are no longer visible (using a 50 % intensity threshold) is Z_d = 4.6 (interquartile range 1.1 in Z_d). This removes the fragmented images that begin to occur at approximately $\left|Z_{\mathrm{d}}\right|>$ 6. The median ratio $D/D_{\mathrm{0}}$ for Z_d < 4.6 is 1.2 (interquartile range = 0.1); however, particles may still be oversized by approximately 40 % even with this filter applied (Fig. 7). Other greyscale thresholds may be used to provide a more or less restrictive DoF constraint. Table 2 shows the median (interquartile range) c values for various greyscale thresholds between 65 % and 85 %. Using a 65 % threshold the median c value is 6.2 (interquartile range = 1.3), while for 85 % it is 3.2 (interquartile range = 0.9). It should be noted that the lower the greyscale threshold, the higher the probability of a fragmented image being observed and the small particle concentration being biased.

Table 2Median (interquartile range) depth of field c value (Eq. 1) for 1060 ice crystal images using various greyscale intensity thresholds and circle equivalent diameter. The median (interquartile range) ratio $D/D_{\mathrm{0}}$ for Z_d < c is also given.

Download Print Version | Download XLSX

When determining the effective array width (Eq. 2), the image size along the direction of the photodiode array should be used. However, this size is a function of the particle's Z position, which is the reason why the effective array width needs to be integrated over the depth of field to determine the sample volume (Eq. 2). This can be calculated using the AST model if the true particle shape can be assumed (e.g. spherical particles in liquid cloud). However, if the true particle shape is not known, as is often the case for ice clouds, then it remains a source of uncertainty in the calculated sample volume.

Figures 12 and 13 apply this new methodology to ambient measurements collected during research flights in cirrus on 7 February and 23 April 2018. Figure 12 shows PSDs from the CIP-15 and HALOHolo for a run at −42 ^∘C on 7 February 2018 (16:02:00 to 16:10:00 GMT). This flight has previously been discussed by O19. Figure 13 shows equivalent PSDs for temperatures between −47 and −40 ^∘C collected on 23 April 2018. For both probes, the particle diameter given is the circle equivalent diameter, and particles in contact with the edge of the CIP-15 optical array have not been included in the PSD calculation. The black lines show the CIP-15 size distribution when images are filtered to only include those with at least one pixel at the 75 % intensity threshold. This threshold significantly reduces the concentration of small particles (< 200 µm) compared to when this filtering is not applied (grey lines) and generally is in much better agreement with HALOHolo (a holographic imaging probe) (blue markers). This suggests that for these cases using current data-processing techniques, a significant fraction of the ice crystal number concentration at sizes < 200 µm is an artefact due to optical effects.

Figure 12Size distributions from the CIP-15 and HALOHolo for a run at −42 ^∘C on 7 February 2018. The black line shows the CIP-15 size distribution when images are filtered to only include those with at least one pixel at the 75 % intensity threshold.

Download

Figure 13Size distributions from the CIP-15 and HALOHolo for runs between −47 and −40 ^∘C on 23 April 2018. The black line shows the CIP-15 size distribution when images are filtered to only include those with at least one pixel at the 75 % intensity threshold.

Download

HALOHolo's sample volume is not as strongly dependent on particle size as it is for OAPs. However, as described earlier, measurements of small particles from HALOHolo are limited by noise in the background image. For a complete description of the HALOHolo data-processing and quality-control procedures, see Schlenczek (2017). HALOHolo uses supervised machine learning to discriminate real particles from artefacts due to noise in the background image. However, it is possible that small particles could be misclassified as artefacts or vice versa, and as a result HALOHolo could either underestimate or overestimate the small ice concentration. For particles > 35 µm, it is estimated that the probe's detection rate is > 90 %, and previous work has shown excellent agreement with a CDP in liquid clouds (Schlenczek, 2017). However, HALOHolo PSDs should not be considered the true PSD but rather another piece of evidence that suggests for these cases OAPs overestimate small ice concentrations using current data-processing techniques.

4.2 Stereoscopic imaging

A second method that could be used to constrain the DoF of an OAP is to use the stereoscopic imaging that is possible with the 2D-S. The 2D-S in effect consists of two OAPs (known as channels) orientated perpendicular to each other and the direction of motion of the particle and instrument. Under normal operation the probe is oriented so that one laser beam is horizontal and the other is vertical. The two lasers overlap at the centre of each channel's arms. As well as increasing sampling statistics by having two channels which can be merged or averaged, this design also allows some ice crystals to be viewed from two orientations to study their aspect ratios. In this study we use this feature to constrain the probe's DoF, which greatly limits the magnitude of diffraction artefacts and represents the first implementation of stereoscopic analysis on an ambient OAP dataset. The 2D-S was designed so that Z = 0 on both channels is in the region where the two lasers overlap. We refer to particles observed by both channels as co-located particles. Co-located particles have tightly constrained Z position and should not be subject to significant mis-sizing due to diffraction. For the 2D-S, this is likely to be true for D₀ > 20 µm. For a hypothetical stereoscopic probe with larger optical arrays, it may be necessary to restrict the distance a particle can be from the centre of the optical array.

For the case where channel 0 is used for particle sizing and channel 1 is used to constrain the particle Z position, the sample volume of co-located particles is given by

where TAS is the true air speed, E is the number of array elements, R is the resolution of the probe, D is the measured particle diameter, and D_CH0 is the particle diameter measured along the axes of the channel 0 optical array. This requires that particles in contact with the edge of the channel 0 optical array have been removed. If channel 1 is used for particle sizing instead of channel 0, then particles in contact with the edge of the channel 1 optical array are removed instead of channel 0, and D_CH0 is replaced by D_CH1 in Eq. (5).

For this method to be applicable, it is important to validate that Z = 0 on both channels is in the laser overlap region. If it is significantly offset this would prevent small co-located particles from being observed, since the DoF from one channel would not overlap with the optical array of the other channel. Increasingly large offsets between the channels prevent increasingly large co-located particles from being observed. It is therefore important to check that this offset is not significant by regularly sampling in environments where small particles are present (i.e. in liquid cloud or using a droplet generator in a laboratory as in O19).

Co-located particles could be confused with shattered particles since they are also associated with short inter-arrival times. Figure 14 (top panel) shows a histogram of inter-arrival time for particles on the same channel for measurements in cirrus on 7 February 2018. To minimise shattering events, each channel was independently filtered for particles using an inter-arrival threshold of 1 × 10⁻⁵ s. It may still be possible to mistakenly detect shattered particles as co-located particles if one shattering fragment splits into two particles, triggering each channel simultaneously but in spatially independent parts of the sample volume. However, examination of co-located images suggest that this is rare.

Figure 14(a) Histograms of inter-arrival times for particles on the same 2D-S channel for measurements in cirrus on 7 February 2018. (b) A histogram of the difference in arrival time between a particle on one channel and their closest neighbour on the other channel.

Download

To identify co-located particles, we use the difference in arrival time between a particle on one channel and their closest neighbour on the other channel. Figure 14 shows a histogram of co-location times for measurements in cirrus on 7 February 2018. This distribution is bimodal with a larger mode centred at approximately 1 × 10⁻³ s and a smaller mode at 1 × 10⁻⁷ s. The larger mode is associated with the typical spatial separation between ambient particles, with its position dependent on the particle concentration. Examining pairs of images from the smaller mode suggests that these images are the same ice crystal viewed from different orientations. Figure 15 shows example pairs of co-located images, with channel 0 images shown in yellow and channel 1 images shown in blue. In addition to overall consistency in the geometrical shapes between channel 0 and channel 1 images, there is also excellent consistency in the particle size along the airspeed direction (x axis in Fig. 15) between these two channels.

Figure 15Example ice crystals observed by both channels of the 2D-S. Images from channel 0 are shown in yellow and images from channel 1 are shown in blue.

Download

Figure 14 shows that most co-located particles do not trigger both channels simultaneously within the time resolution of the data acquisition system but are offset by a few hundred nanoseconds. At 100 m s⁻¹ data slices from the detectors are acquired every 1 × 10⁻⁷ s, which corresponds to a spatial separation of 10 µm. Using the laboratory droplet generator system described in O19, we were able to generate a continuous stream of droplets of known size, velocity, rate, and with precise control over the position within the sample volume. These experiments with particle velocities of 1 m s⁻¹ resulted in a 1 × 10⁻⁵ s mode time delay in detection events between the two channels of the 2D-S. This also corresponds to an offset of 10 µm in the sample volume in the axis of airflow through the probe (y axis). These two sets of analysis provide a robust independent verification of the spatial offset between the two channels of the 2D-S. Therefore, when considering ambient data, we classify co-located particles as those with time separations less than 5 × 10⁻⁷ s.

Figure 16 shows a comparison between PSDs collected in liquid stratus cloud at 13 ^∘C on 17 August 2018. The grey lines show the 2D-S data for each channel using conventional data-processing protocols without constraining the DoF, while the green and red lines show PSDs for just the co-located particles. The CDP is shown in blue. For this case, no particles larger than approximately 200 µm are present. All data-processing methods are in good agreement up to 100 µm. For larger sizes, the measurements using the co-located particles are limited by counting statistics due to the low concentration of these particles. This illustrates the ability of the 2D-S to detect small co-located particles.

Figure 16Size distributions from the 2D-S and CDP for different temperatures during a research flight in liquid stratus on 17 August 2018 at 13 ^∘C. The grey lines show the 2D-S data using conventional data-processing protocols without constraining the DoF, while the green and red lines show size distributions for just the co-located particles. CDP size distributions are shown in blue.

Download

Figure 17Size distributions from the 2D-S and HALOHolo for different temperatures during a research flight in cirrus on 11 March 2015. The grey lines show the 2D-S data using conventional data-processing protocols without constraining the DoF, while the green and red lines show size distributions for just the co-located particles. The dashed black line shows a 2D-S processed using a hybrid of conventional and co-location data processing (see text for details). HALOHolo size distributions are shown in blue.

Download

Figure 17 shows size distributions from the 2D-S and HALOHolo for different temperatures (averaged over ∼ 10 min) during a research flight in cirrus on 11 March 2015 (see O'Shea et al., 2016). The grey lines show the 2D-S data for each channel using conventional data-processing protocols without constraining the DoF, while the red lines show size distributions for just the co-located particles. HALOHolo size distributions are shown in blue. For all temperatures, the conventional 2D-S data processing shows an ice crystal mode at small sizes (< 200 µm). At warmer temperatures (> −39 ^∘C) there is also a clear second mode at larger sizes. However, these high concentrations of small ice particles are not present in the co-located and the HALOHolo size distributions. This suggests that using only co-located particles on the dual channel 2D-S probe is effective at removing significant biases at small particle sizes. At larger sizes (> 300 µm) the 2D-S data processing using conventional and stereoscopic methods are in good agreement; however, the latter method is limited by sampling statistics.

Stereoscopic data processing has the advantage of removing out-of-focus artefacts that bias the PSD at small sizes, while at larger sizes traditional processing methods offer significantly improved sampling statistics. Therefore, a hybrid approach using stereoscopic processing for small sizes and traditional processing methods for larger sizes is advantageous. The choice of size threshold to switch between the two methods is dependent on the arm width of the probe and the level of mis-sizing that is deemed acceptable. To give an idea of a suitable threshold, we will choose a size limit that prevents all particles with Z_d > 2 from being included in the PSD. The maximum Z that the 2D-S can observe a particle is Z = 31.5 mm (2D-S arm_width/2). This corresponds to a 222 µm particle at Z_d = 2. However, since particles can be mis-sized by a factor 1.4, a size threshold of 300 µm is needed to ensure that no particle with Z_d > 2 is included. Figure 17 (dashed lines) shows 2D-S PSDs processed using this hybrid approach.

4.3 Other potential methods

There are several other potential methods that could be used to improve OAP PSD measurements. First, reducing a probe's arm width to physically limit a distance a particle can be from the object plane would reduce out-of-focus particles. The amount the arm width would need to be decreased depends on the level of mis-sizing that is deemed acceptable for a given particle size, with more accurate sizing and smaller particles requiring smaller arm widths. However, as well as decreasing the sample volume, reducing the probe's arm width is likely to increase the proportion of shattered artefact particles compared to ambient particles that the probe measures, since shattered artefacts are thought to cluster near the probe's arms.

Second, statistical retrievals have been applied to particle size distribution measurements where the instrument response is a distorted version of the true ambient distribution. These methods are reliant on knowing or empirically approximating the instrument function that distorts the ambient distribution. These methods have been applied to OAP measurements of spherical droplets (Korolev et al., 1998; Jensen and Granek, 2002). For non-spherical particles, the distortion function is dependent on the ice crystal habits present; therefore, the derived size distributions would have greater uncertainty, unless the particle shape is known a priori. However, this methodology may still result in an acceptable level of uncertainty if circle equivalent diameter is used, since its intra- and inter-habit $D/D_{\mathrm{0}}$(Z_d) variance is smaller than for the mean X–Y and maximum diameters.