a. Laboratory calibrations

The overlap of the trigger PMT sensing volumes has been measured in the laboratory using an oscillating ice-analog crystal. An ice-analog rosette with a maximum dimension of 25 μm was fixed on the end of a 7-μm-diameter carbon fiber. A piezoelectric transducer caused the ice analog to oscillate into and out of the trigger sensing volume. The transducer/ice-analog assembly is moved using a micropositioner, which allows the trigger area to be mapped out (giving 0.88 mm²).

The size calibration of SID-2 was carried out in the laboratory using 5, 10, and 20 μm polystyrene latex (PSL) spheres. Scattering theory indicates that, for larger spheres, the scattering cross section scales with the projected area of the sphere. The detector response therefore depends on the square of the particle size. The function D = aS^0.5, where S is the mean outer detector response, is fitted to the PSL data to give the constant a. To correct the size calibration for the different refractive index of 1.32 (water and ice average) and 1.59 (polystyrene), Lorentz–Mie theory calculations were carried out. The calculations showed that, over the forward-scattering angles 9°–20°, the PSL calibration must be divided by 1.35 to obtain the corresponding water/ice sphere diameter. For spherical particles, the corrected diameter calibration is therefore given by

View Expanded

The angular dependence of scattering from ice crystals is expected to be different from spheres. The phase functions of various ice crystal habits were calculated and compared with that from an ice sphere of equivalent projected area. For scattering angles corresponding to the outer azimuthal elements, the scattered flux from the sphere is larger by a factor of 3.0, with only a weak dependence on the crystal habit. These calculations were confirmed by measuring the phase function of a 72-μm ice-analog rosette and an ice sphere of equal average projected area held in a levitation trap using a laser diffractometer and CCD array. Assuming that the scattering cross section is proportional to the projected area of the particle, it is expected that the ice crystals are undersized by a factor of 3^0.5 = 1.7. This ignores the crystal habit details and any size dependence. Therefore, prior to particle size and concentration determination in the data processing, each particle’s phase (liquid or ice) must be classified.

Laboratory tests indicated that each photodiode element has a different background noise level (detector signal when there is no illumination) and a different gain. There is no correlation of background with gain; the background is from the dark current of the photodiode element or offsets in the detector electronics. The background noise level is temperature dependent and will therefore vary during a flight. The background noise variation must be taken into account if small particles, which give a low detector response compared to the noise, are to be accurately characterized.

Surrounding the small trigger volume is an extended sensing volume (60 times larger). The coincidence of particles in this sensing volume while another particle triggers the detector readout will add to the scattered light intensity of the triggered particle. This will lead to a nonuniform azimuthal detector element response and lead to skewing the particle size distribution toward larger particle sizes. The probability of coincidence can be calculated using Poisson statistics and depends on the particle concentration (there is a 5% probability when the concentration is 30 cm⁻³).

b. Data processing

The data processing involves the following steps:

• (i) Photodetector background noise: All photodiode elements are read every second without the need for a particle trigger. These are referred to as “forced triggers.” Only out-of-cloud values are used to calculate the background noise. When in cloud, there is a high probability of a cloud particle being in the sensing volume when this forced trigger occurs. The typical variation in background noise during a flight is equivalent to a change of 1.0 μm in the diameter of a 4.0-μm particle. The ability to determine the individual detector element’s noise level is a significant improvement over SID-1, giving improved potential for single-particle shape discrimination.

• (ii) Nonuniformity of photodetector gain: Each photodiode gain is corrected so that its mean response from a large number of particles is similar. By assuming that the nonspherical particles have random orientation through the laser beam, all particles measured during the entire flight are used. The gain ranges from 0.58 to 1.88, where 1.0 is the average response.

• (iii) Particle asphericity: The particle asphericity A_f is defined as the variation of the scattered light intensity around the 24 azimuthal detector elements, and it is given by

  

  

  View Expanded

  

  where S_i is the ith detector element response (after background noise subtraction and gain correction) out of n elements and k is a scale factor (k = 100/n² − n), so that A_f < 100.0. The simplest particle classification for liquid drop or ice crystal tests whether A_f is above some threshold value.

• (iv) Particle radius: For spherical particles, the corrected diameter calibration is used to calculate the diameter. If the particle asphericity is high and the classification gives a nonspherical ice crystal, the calculated diameter is increased by a factor of 1.7. For the aerosol comparisons shown here, the factor 1.7 was not applied.

• (v) Particle number concentration: For each second, the particle number concentration N is given by

  

  

  View Expanded

  

  where f is the count frequency; A is the area of the triggering zone (from the laboratory calibration); and υ is the average particle velocity, which is equated to the true airspeed in flight.

• (vi) “Missed particles”: The recording system can, because of electronics dead time, read a maximum of 8000 particles distributed uniformly over 1 s, which corresponds to a maximum concentration of around 150 cm⁻³ at typical airspeed. However, the SID-2 data acquisition can keep a sum of the number of particle triggers that occur while the detector element signals are read out. These particles are added to the recorded particle data by distributing over the particle size spectrum. However, the probability of multiple particles in the sensing volume is over 50% at concentrations of 150 cm⁻³, so this is an upper limit for reliable sensing.

a. Liquid water cloud

One flight in extensive stratocumulus (flight B256) is selected to characterize the response of SID-2 to liquid water drops. This flight was during the Winter Experiments (WINTEX) series of Met Office winter experiments flown from Cranfield from November 2006 to January 2007 (campaign details available online at http://www.faam.ac.uk/).

The comparisons with the liquid water cloud cases provide the best test of the particle number concentration because of the overlap with the FFSSP size range (from 2 to 47 μm in diameter). Figure 2b is the scatterplot of SID-2 and FFSSP number concentrations during one straight and level run. This was at 300 m above mean sea level (AMSL) with air temperature around 9°C. The correlation coefficient is 0.929 and the best-fit gradient is 1.048: the concentration measured by SID-2 is 1.048 higher than that from the FFSSP. This level of agreement is similar in all other runs throughout this stratocumulus flight, where only liquid water was present. This confirms the estimate of the laboratory-based trigger area that defines the sample volume.

The particle size spectra for two runs, one near the base and one near the top of the stratocumulus, are shown in Fig. 3. The agreement is good, but SID-2 is slightly undersizing the liquid water drops compared with the FFSSP: the average particle size measured during the 900-m run is 23.8 μm from SID-2 and 26.5 μm from the FFSSP.

The SID-2 liquid water content (LWC) obtained from integration of the particle size spectrum is compared to the FFSSP LWC (also obtained by integration of the size spectrum) and to the Nevzorov bulk-water probe is shown in Fig. 4. The density of the water drops is assumed to be 1.0 g cm⁻³. In Fig. 4b, the comparison of SID-2 and the Nevzorov probe, the correlation coefficient is 0.938 and the best-fit gradient 0.640: the liquid water measured by the SID-2 is around 0.64 of that measured by the Nevzorov probe. The underestimate of the LWC by SID-2 is also evident in the comparison with the FFSSP LWC shown in Fig. 4c, where the correlation coefficient is 0.923 and the best-fit gradient is 0.775. The LWC error can be explained if SID-2 number concentration is correct, but the particle size is underestimated. The particle size spectrum for the 900-m run implies that the SID-2 is on average 0.90 that of the FFSSP. This ratio leads to an underestimate of volume and hence LWC of around 0.73, which is similar to the FFSSP LWC discrepancy.

The SID-2 sizing error might be due to a decrease in the laser power because of the laboratory calibration or a different HPD voltage supplied during the flights.

b. Ice cloud

One flight in frontal cirrus (flight B257) is selected to characterize the response of SID-2 to small ice crystals. This flight was during the Cirrus and Anvils: European Satellite and Airborne Radiation (CAESAR) experiment.

The SID-2, 2D-C, and Nevzorov ice water contents (IWC) during one straight and level run near the cirrus top are shown in Fig. 5. The SID-2 ice water content is obtained from integration of the particle size spectrum similarly to the liquid water content but assuming a density of 0.9 g cm⁻³ for the small ice particles. The 2D-C estimates the ice crystal mass from the mean diameter of the measured particle image using

View Expanded

where M is the particle mass (g), D is the mean diameter (μm), a = 7.38 × 10⁻¹¹, and b = 1.9 (Brown and Francis 1995). For diameters smaller than 100 μm, ice crystals are assumed to be solid ice spheres with a density 0.9 g cm⁻³. This was at 9.75 km with air temperature around −57°C, and at this cold temperature no liquid water will be present. Near the cloud top, all ice particles are small enough to be measured by the SID-2 without detector saturation, and it may be expected that the incidence of any particle shattering on the probe will be low (Field et al. 2003).

The IWC shows a much larger spread than the LWC in Fig. 4. For the SID-2 and Nevzorov IWC, the correlation coefficient is 0.663 and the best-fit gradient is 3.55. This poor agreement could be due to the range of ice crystal shape (for the SID-2 IWC, all ice crystals are assumed spheres with a constant density) and problems with the Nevzorov probe. The collection efficiency of the Nevzorov probe can be less than unity, especially for small cloud particles (Korolev et al. 2003). In particular, in ice cloud, the Nevzorov probe is known to underread the ice water content, because some of the ice particles bounce out of the collection region (Korolev et al. 2008).

The particle size spectra, including the 2D-C data, for runs at two different altitudes are shown in Fig. 6. Data from the 2D-C for ice crystals smaller than 100 μm are discarded because of measurement uncertainties for small particles. The agreement in ice crystal diameter is good, and it is repeated for all other runs at different heights in the cirrus flight. The comparison of SID-2 with the 2D-C in cirrus shows the importance of measuring small ice particles. The total ice crystal number concentration using only the 2D-C will be significantly underestimated, leading to an error in effective radius (the error in ice water content is less, because most of the mass is in the larger particles).

c. Mineral dust aerosol

One flight in moderate loadings of mineral dust aerosol (flight B301) is selected. This flight was during the Geostationary Earth Radiation Budget Intercomparison of Longwave and Shortwave Radiation (GERBILS) experiment (Marsham et al. 2008). The size distribution for aerosols in the accumulation and coarse modes were measured using the PCASP and SID-2. Figure 7a shows two size spectra during a straight and level run along 18°N between 3.6° and 14°W. The airborne dust was generated by a combination of Saharan dust from the north and more localized dust uplifted in relatively moist, cool monsoon air from the south; the highest dust concentrations occurred during the latter events around mesoscale convective systems. Figure 7a shows a decrease in concentration of all particle sizes to the west. The comparison of SID-2 with the PCASP in desert dust aerosol shows a discontinuity at the overlap region. SID-2 has not been calibrated for dust aerosols, and there is no information regarding the mineral composition of the desert dust. The discontinuity and gradient change could be due to sizing error, which affects the number concentration via the particle size–dependent sample volume. The effect of particle shape on the ability of SID-1 to derive a meaningful radius was noted by Osborne et al. (2004), where it was shown that nonspherical biomass-burning aerosols tended to be undersized by a factor of 1.5–2.0. Even though the optics are different on SID-2, a similar undersizing is expected for nonspherical particles such as mineral dust; that is, the determined radius will be smaller than the true equivalent spherical radius. The size spectra for sea-salt aerosol, shown in Fig. 7b, are in better agreement.

d. Particle shattering on the probe inlet

Field et al. (2003) show observations of ice particle interarrival times measured with a FFSSP. The distribution of interarrival times was bimodal with two modes at ∼10⁻² and ∼10⁻⁴ s instead of the expected exponential distribution. The smaller interarrival time interval corresponds to ∼1-cm horizontal spacings, and these are suggested to originate from shattering of ice crystals on the probe inlet tube. If shattering is responsible for the interarrival time bimodality, then the ice number concentration will be overestimated by up to a factor of 5.

Shattering on the probe housing of optical array probes such as the 2D-C is also possible, leading to incorrect particle number concentration and size spectra. Field et al. (2006) again show that the distribution of interarrival times was bimodal and indicate that the particle number concentration can be overestimated by up to a factor of 4.

The SID-2 inlet design was designed as open path to reduce possible large particle shattering. The distributions of particle interarrival times for example sections of the flight data are shown in Fig. 8. All are monomodal and indicate that no shattering is occurring, particularly for the cirrus case, where you might expect it. Figure 8 also indicates the minimum resolvable interarrival time, which is 0.04 ms.

a. Cloud particle phase

The efficiency of using A_f to discriminate particle phase is shown in Fig. 9. There is not a clear separation; however, by inspection of Fig. 9, an A_f threshold of 6 best separates liquid drops from ice crystals. The fraction of cloud particles with high asphericity (A_f > 6) is 0.184 for the stratocumulus data, 0.031 for the supercooled altocumulus lenticularis data, and 0.961 for the cirrus data. The relatively high fraction of droplets in the stratocumulus data with high A_f is due to the high concentrations and presence of precipitation drops giving multiple particles in the SID-2 sensing volume, which leads to an asymmetric detector response. An A_f threshold of 9 best separates dust from deliquesced aerosols.

The plots of frequency of occurrence as a function of the particle asphericity A_f and radius (Fig. 10) show the ability of SID-2 to discriminate the phase of the cloud particles is dependent on the particle size. For the stratocumulus data (Fig. 10a), most particles have A_f < 6, but there are significant numbers with larger A_f . These large A_f particles exhibit irregular azimuthal scattering patterns. However, in a cloud that can be assumed on physical grounds to be unglaciated, such asymmetric scattering is assumed to indicate multiple particles in the SID-2 sensing volume. For the supercooled altocumulus lenticularis data (Fig. 10b), there are negligible numbers of particles with large A_f , confirming the absence of large ice observation. The lower concentrations (∼20 cm⁻³) and lack of large precipitation-sized drops reduces the chance of multiple particles in the SID-2 sensing volume compared to the stratocumulus case (∼40 cm⁻³). The asphericity tends to increase for smaller particles, because the detector noise becomes significant relative to the response to the scattered light. For the cirrus data (Fig. 10c), the cloud particles typically have larger A_f . This flight data were taken near the top of thin cirrus so that there was no aggregation and the ice particles are small and pristine (most have a diameter between 10 and 20 μm). This is a more difficult test for the SID-2 phase discrimination capability than using larger, possibly aggregated particles. The ability of SID-2 to discriminate small ice particles from supercooled liquid drops is summarized by the difference between Figs. 10b,c. The ice particles typically have A_f values higher than the supercooled liquid drops. Small liquid drops at the limit of the SID-2 detection threshold, however, might be misinterpreted as ice particles. Data from the supercooled altocumulus lenticularis flight show that the SID-2 detection limit for cloud particles (observed at the leading edge of the cloud) is around 3-μm diameter. The mixed-phase growing cumulus data (Fig. 10d) contain both high and low A_f values, which indicate that there are significant amounts of both liquid drops and ice particles present. An asphericity threshold, which varies with particle size rather than a single A_f threshold of 6, would better discriminate the liquid drops from ice crystals in this mixed-phase cloud.

b. Aerosol shape

The distinction between spherical sea-salt particles (deliquesced salt solution droplets in a humid marine boundary layer) and nonspherical mineral dust particles is clearly shown in Figs. 9 and 10. The mineral dust loadings relative to the sea-salt loadings are very low with concentrations within the SID-2 size range of 1–3 cm⁻³. The sea-salt data show some larger particles resulting from proximity to the aerosol source.

The dust data are biased to the smallest size bins of SID-2, smaller on average than the sea-salt data, and are at the limit of the probes resolution. Nonetheless, the differences in A_f between dust and sea salt are clear.

c. Spatial scattering diagrams

Figure 11 contains polar plots related to the azimuthal variation of scattered light intensity for a sample of cloud particles and aerosols from each of the flight datasets. The plot radius for each photodetector is drawn as , where S_i, the ith detector response (i = 1, 24), is proportional to the scattered light intensity. Hence, plot area is proportional to particle cross-sectional area; for spherical particles (with uniform azimuthal response), plot radius is proportional to particle radius. The scale is the same for each polar plot, with the circle representing the calculated equivalent spherical radius.

The scattering patterns for the stratocumulus data (Fig. 11a) show both large precipitation drops and small drizzle drops. The high particle concentrations lead to coincidences where there are multiple particles in the sensing volume, and the detector response is typified by the particle third from the left on the bottom row. The supercooled altocumulus lenticularis data (Fig. 11b) also show some coincidences. The cirrus data (Fig. 11c) show a wide variation of irregular scattering patterns. Surface roughness might be increasing the scattering variation from that expected from pristine ice crystals. Supercooled mixed-phase growing cumulus data (Fig. 11d) show a mixture of small cloud drops and larger ice particles. The two particles third and fourth from the top are interesting in that, over part of the azimuth, the scattering is uniform as from a spherical particle and that, over the rest, the scattering is varied. The dust data (Fig. 11e) include one large aerosol similar to the scattering from an ice particle in the mixed-phase cumulus.