Intercomparison and Evaluation of Optical Array Probe Data Analysis of Cloud Microphysical Measurments

Airborne cloud optical array probes (OAPs), widely used in airborne and ground measurements, capture binary digital images of cloud hydrometeors from which their microphysical properties can be derived. However, deriving cloud microphysical properties from these images is not straightforward. Corrections are needed for sizing of out-of-focus particles, for fogged and stuck pixels, for corrupt images, and for electronic noise. Additionally, removal of shattered particles and resizing of particles not entirely within the photodiode array are required. Furthermore, substantial uncertainties exist regarding how the probe sample volume depends on particle size and the appropriate metric to describe particle size. Various algorithms have been developed by different international research groups to process and analyse these probe images, and to derive various microphysical properties Prior intercomparisons of cloud properties derived from identical data files processed by different research groups have revealed persistent inconsistencies that make it challenging to derive coherent conclusions, intercompare results, or combine datasets for parameterization and modelling.

To determine the source of discrepancies, synthetic data for four optical array probes—the Stratton Park Engineering Company (SPEC) 2D-S, SPEC High Volume Precipitation Spectrometer ver.3 (HVPS-3), Droplet Measurement Technologies (DMT) Cloud Aerosol and Precipitation Spectrometer (CAPS) – Cloud Imaging Probe (CIP), and DMT Precipitation Imaging Probe (PIP)—were generated. The synthetic data use an optical model to simulate the signal that the OAPs would detect, including allowances for diffraction, discretization, and other optical phenomena. These simulations were generated for three habits: round drops, columns with a 3:1 aspect ratio, and a mix of irregular ice shapes. The simulations all assumed an exponential size distribution. Nine international research groups processed these data using their probe data processing software, generating both particle-by-particle information and bulk properties (size distributions). This presentation compares the variation between algorithms of the following quantities for round, columnar, and irregular particles: counts per second and timing of individual particles; length along the time array; width along the photodiode array; projected area; perimeter; and end diode shadowing.

To make the comparisons as simple as possible, comparisons were made in terms of counts rather than concentration to avoid dependence on assumed sample volume, and separate comparisons were made for complete and partial images. In the comparisons, the identity of the different codes was masked to preserve anonymity. Although two codes had exact agreement in the number of particles detected over the 1 minute of simulated time, there were variations in the counts between all other codes. For example, the average (maximum) discrepancy from the actual number of particles in the synthetic data set and those generated by the algorithms was 5% (8%) for the 2DS, 4.5% (5%) for the CIP, 4.5% (5%) for the PIP, and 4.5% (6%) for the HVPS-3. As there should be, there was no change in the number of particles detected after correcting the sizing of out-of-focus particles. The sizing corrections due to out of focus particles had the most impact on irregularly shaped particles, particularly those obtained from the HVPS-3 and PIP probes, which also showed changes in their distributions after applying corrections for removal of artifacts and shattering. Conversely, round particles, especially those detected by the 2DS probe, exhibited minimal change between different algorithms, as these particles were less affected by the artifacts being corrected. The reasons for differences between algorithms are discussed, and the implications for comparing microphysical properties from different field campaigns led by various research groups are addressed.