1. Introduction
In situ microphysics measurements are a key source of storm data, but they are difficult to obtain. With potential applications to dual-polarization radar validation, storm dynamics, cloud modeling, and lightning research, it is important that measurements accurately represent a wide range of spatially and temporally varying conditions. For example, surface disdrometers such as the Particle Size and Velocity (Parsivel; Löffler-Mang and Joss 2000; Friedrich et al. 2013; Yuter et al. 2006; Löffler-Mang and Blahak 2001) and the 2D video disdrometer (2DVD) (Schuur et al. 2001; Cao et al. 2008, 2010) have been used to examine the drop size distribution (DSD) in precipitation. These observations provide valuable comparisons to radar data in particular, but they suffer from a separation of sampling volumes in which the radar provides a volumetric estimate at altitude, while the surface disdrometer provides a fixed-point measurement. Additionally, several assumptions about spatiotemporal homogeneity and evaporation must be made prior to using these data (Schuur et al. 2001).
Many previous projects with access to storm-penetrating aircraft have employed wing-mounted particle probes, such as the Particle Measuring Systems (PMS) optical array probe (2D-OAP) models 2D-C and 2D-P and the PMS forward scattering spectrometer probe (FSSP). These systems have been used extensively in a variety of projects (e.g., Jorgensen and Willis 1982; Heymsfield et al. 2013; Smith et al. 2009; McFarquhar et al. 2007) to collect microphysics measurements at altitude in various types of convection. This data collection strategy has the advantage of being able to take DSD measurements over large areas; however, a degree of spatial homogeneity must be assumed to interpret the large samples that are normally required to obtain statistically representative datasets. Particle shattering and flow trajectories caused by the aircraft itself can cause errors in the measured DSD (Norment 1988). Additionally, safety factors often prevent aircraft from sampling certain areas of storms (such as hail cores, strong supercells, near lightning or active icing areas).
To address these issues and provide a relatively inexpensive, lightweight instrument for collecting in situ particle data, balloonborne devices known as “videosondes” and other microphysics probes have been developed. Such balloonborne instruments can be flown in a variety of conditions, including those generally too hostile or unreachable for more conventional means (e.g., surface instruments or aircraft) of measurements. Miloshevich and Heymsfield (1997) used a Formvar replicator on a balloonborne device to measure ice crystal habits and structures of particles smaller than 100 μm, while others used video cameras to observe precipitation particles. Murakami et al. (1987, hereafter M87) and Takahashi (1990, hereafter T90) have pioneered the development of the videosonde using a camera-based approach to obtain video images of precipitation particles inside convection and retrieve DSD information from the video images.
The M87 videosonde utilizes a filmstrip to physically record the impressions of impacting particles between 7 μm and 2 cm. The filmstrip is then imaged with a non-high-definition (non-HD) camera, and the recording is transmitted to a ground station using a 1.6-GHz microwave antenna link. M87 have carefully documented the particle sampling efficiency of their videosondes, which varies from about 0.12 to 0.77 as a function of increasing particle diameter.
The T90 videosonde is similar to the M87 instrument in that it uses a filmstrip to capture the sizes of smaller particles whose imprints are then imaged by a camera. However, for particles larger than 0.5 mm diameter, a flash is triggered that illuminates the particle for direct video capture. These flashes occur at a rate less than 2 Hz due to the lag time required to recharge the strobe lamp. As with the M87 videosonde, the T90 videosonde is not recovered, since it likely lands in the nearby ocean. Hence, the non-HD images from the T90 videosonde are also transmitted to a ground receiving station for processing.
Boussaton et al. (2004, hereafter B04) have expanded videosonde capabilities using a camera-based system similar to M87 and T90 combined with particle charge measurements (which was also added to the Takahashi videosonde later). The B04 videosonde measures a particle diameter from 0.5 mm to 2 cm and includes an induction ring to measure particle charge in the range from ±1 to 400 pC. As an alternative to direct video imaging of particles, B04 employ a shadowing technique to determine particle size. Two lights near the camera illuminate each particle, which cast two shadows on the back plane toward either side of the object. Knowing the distance between the induced shadows on the back plane, the distance of each particle from the camera and thus its original size can be determined. The B04 videosonde design assumes that all images contain a single particle to uniquely relate particle charge to size, a valid approximation given their small sampling volume and balloon ascent rate. The shutter speed of the camera used for particle capture is not fixed and varies with illumination. As with the M87 and T90 videosondes, the image data are transmitted to the ground, since the instrument is lost after launch.
A different approach to measuring particle distributions on a balloonborne device was utilized by Mahlke et al. (2008) via their development of a “Flying Parsivel.” The authors modified the standard Parsivel disdrometer unit as created by Löffler-Mang and Joss (2000) to fit within a balloonborne package. The Parsivel device is capable of measuring particle size and velocity through the use of a laser diode. One advantage of the Parsivel system, aside from the size and velocity measurements, is the fast scanning rate of the laser. This allows the unit to sample a large number of particles in a short amount of time, thus possibly measuring nearly every particle that passes through the sensing area in typical precipitation rates.
Although the M87, T90, and B04 videosondes and the Mahlke et al. (2008) Flying Parsivel, have pioneered a novel approach to collecting unique in situ particle data at altitude, they all suffer from a number of operational drawbacks. The large cost associated with the fabrication of the videosondes can be prohibitive in more extensive field campaigns, since each instrument is lost after launch. Additionally, because the instrument is lost, the data must be transmitted to the ground via a radio link. This requirement reduces the quality of the data transmitted, as it is difficult to move large quantities of data quickly over a radio link. Thus, lower-resolution images are used and a slower frame rate is required to be able to complete the transmission between frames, both of which reduce the sensitivity of the instrument.
A drawback of the Flying Parsivel system is that the laser cannot determine either particle habit or its possible departures from the assumed spherical drop form (e.g., Battaglia et al. 2010). For example, highly elliptical ice particles that fall through the horizontal laser beam at an angle could potentially account for variable and possibly large biases of inferred particle size and velocity (Battaglia et al. 2010). Hence, Parsivel measurement errors may be introduced via the needed simplifying assumptions regarding the size, velocity, and habit. Mahlke et al. (2008) assumed that all particles were rain droplets, which led to large discrepancies between the radar reflectivity as calculated from the measured DSD and that measured by radar when ice was present. Furthermore, Mahlke et al. (2008) did not take into account the balloon rise rate with respect to still air in their figures, leading to some uncertainty in their reported results. The approach to using the Parsivel on balloonborne instruments can be improved.
Advances in technology during roughly the past decade have made it possible to develop an improved balloonborne videosonde. Lightweight HD video cameras now record their data on small flash drives and provide higher-resolution imagery. Furthermore, the availability of low-cost, low-power, lightweight GPS tracking technology makes the retrieval of instruments more feasible, at least in operations over land. Retrieval and reuse of instruments lowers the per-mission cost even if the individual instruments are somewhat more expensive, and this makes it easier to afford many flights in a field program. However, the greatest benefit of retrieval is that it allows data to be recorded on board, and much more data can be recorded on board than can be readily transmitted during balloon flights. Thus, video can be recorded with much improved temporal and pixel resolution.
The latter progression of balloonborne precipitation particle sensing technology has helped provide a basis to develop an updated videosonde that provides more detailed particle measurements at altitude in storm precipitation. The present paper reports the development and testing of a retrievable, multisensor HD-camera-Parsivel hybrid balloonborne instrument that is known as the National Severe Storms Laboratory (NSSL) Particle Size, Image, and Velocity (PASIV) probe. The primary objectives of developing the balloonborne PASIV probe are to obtain detailed storm observations of liquid and ice precipitation particle sizes and concentrations, to determine particle habits, and to estimate the particle size distribution (PSD) or “particle spectrum” in storms (i.e., as defined by the particle concentration per unit volume per unit size interval). Section 2 describes the design and performance aspects of the PASIV probe, while section 3 presents two brief, illustrative examples of PASIV measurements in storms.
2. The PASIV probe
a. Overview of PASIV design
To obtain in situ microphysics measurements in storms, a balloonborne instrument known as the PASIV probe has been developed at NSSL following the work of B04, Murakami and Matsuo (1990), M87, and Takahashi (2010). The U.S. Federal Aviation Administration (FAA) regulation Part 101 requires that a free-flying balloon package weigh no more than 2.72 kg. It also states that the entire instrument train cannot weigh more than 5.44 kg, with no single instrument weighing more than 1.81 kg with a weight-to-area ratio of 85.05 g (6.45 cm2)−1 on any side of the instrument.
To comply with these regulations, the main support structure of the instrument is composed of parts made from residential-grade R-4 extruded sheet Styrofoam and assembled by hand with a combination of Monokote film, packing tape, and glue. This lightweight, rigid, and durable structure allows room within the weight limit for sampling instruments, while providing a structure capable of being carried on a balloon through severe weather. The current version of PASIV is approximately 1.5 m long, 0.3 m wide, and 0.3 m tall (Fig. 1). Each piece within the structure is cut from varying thicknesses of sheet Styrofoam (ranging from 12.7 to 25.4 mm) using a computer numerical control (CNC) router. The diamond shape of the body has been chosen to provide a rigid structure that reduces drag as the instrument is pulled through the ambient environment. The opening on the top of the instrument where the particles fall through to be measured is surrounded by small angled fins. The angled fins are intended to suppress particle splashing and rebounds that are of significant concern when making PSD measurements (Grossklaus et al. 1998; Habib et al. 2001; Kruger and Krajewski 2002), as such events can artificially modify the sampled PSD by creating an abundance of small particles. The angled fins are used to deflect low trajectory particles away from the opening, minimizing the opportunity for these disturbed particles to enter the instrument. Bright colors were chosen for all materials to increase the visibility of the instrument for help in the retrieval process.
The instrument is flown on either a 105.7 m3 Aerostar Stratofilm balloon or a 1500-g latex balloon and is supported using waxed nylon line rated at 18.14 kg of strength. In operations, the entire instrument train consists of the PASIV, a radiosonde to provide location and thermodynamic data, and a parachute to slow the descent of the unit once the balloon bursts. To reduce the likelihood of a situation where the balloon actively alters or shadows the particles from the sampling instrument, a let-down reel is used with 30 m of line that unspools approximately 1 min after launch using a delayed timer. The let-down reel method eases the launch operation, especially during high winds, via a shortened instrument train (approximately 3 m) at launch (Fig. 2). To further aid in the launch process, a launch tube is used to hold the balloon during inflation and instrument preparation (Rust and Marshall 1989).
b. Particle sizing instruments
The PASIV is a hybrid instrument system that provides redundant in situ measurements of the particle distribution. The upper portion of the PASIV system consists of a largely modified and repackaged Parsivel laser disdrometer (Löffler-Mang and Joss 2000) that provides particle counts, size distributions, and velocity distributions of the particles that pass through the intake portion of the instrument. This is similar to the Mahlke et al. (2008) Flying Parsivel. Below the Parsivel is an imaging chamber through which the particles subsequently fall and are imaged by a standard handheld-style HD video camcorder. The images obtained are digitally analyzed to identify particles and to measure their properties. Together, these measurements provide information about the size, shape, orientation, and composition (e.g., habit) of sampled particles. It should be noted that these measurements are two-dimensional at best. The camera is only capable of viewing particles on a single two-dimensional plane, and the Parsivel gives only a maximum diameter (making it a one-dimensional measurement). Although each instrument in theory is capable of estimating the PSD (subject to certain measurement constraints to be discussed), their synthesized data can combine the strengths of the individual sensors to yield increased confidence in the resulting analyses.
1) Video camera
The video camera is a Panasonic model HDC-SD9P charge-coupled device (CCD) high-resolution camera capable of shooting 1920 × 1080 resolution (2.07 × 106 total pixels) at 24 noninterlaced frames per second (fps), with a user-selectable maximum shutter speed of 1/8000 s. The camera is mounted on the end of the PASIV that is opposite the viewing chamber (Figs. 1a,b) and is rotated so that the long axis of the image is in the vertical plane of the PASIV. The imaged portion of the viewing chamber (Fig. 1c) measures 108 mm × 183 mm × 150 mm (width × height × depth). The opening to the top of the viewing chamber is smaller in cross section and is located in the center of the upper-viewing chamber face, and measures 110 mm × 130 mm (depth × width). The area of this opening, combined with the depth of the imaged viewing chamber, results in an effective videosonde sampling volume of 0.00217 m3 per image. The resulting physical image dimensions are 108 mm × 183 mm, resulting in a pixel size of 0.01 mm2. The pixel size of 0.01 mm2 is assuming an HD image (1920 × 1080 pixels). If a video graphics array (VGA) resolution of 640 × 480 is assumed, then the pixel size increases to 0.064 mm2. The HD images also provide a much clearer picture than VGA, thus allowing the end user to better identify particle size, shape, and composition. However, pixel size does not equate to identifiable object size. A better quantification for the resolving capability of the camera is given by lines per inch. This tests the ability of the camera to identify increasingly smaller lines as distinct lines rather than blurring them together. Through testing it has been determined that the camera is capable of resolving 100 lines per inch (each line would be 0.01 in. or 0.254 mm thick). This provides a minimum size resolution for the camera. Given the balloon’s maximum relative ascent rate of about 5 m s−1, the 24 fps frame rate, and the vertical dimension of the viewing chamber, the successive sampling volumes during a typical flight are closely stacked, though nonoverlapping, in height. Hence, the videosonde particle samples obtained from successive images may be considered statistically independent.
An aspect of camera optics known as “forced perspective” causes an object close to the camera to appear larger than the same-sized object farther away. In the PASIV instrument, this would lead to errors in the apparent size of a particle in the viewing chamber if the camera is close to the particle. The apparent size of a particle depends on the angle of rays from the top and bottom of the particle relative to the lens. The larger the angle subtended by the particle, the larger its apparent size. Therefore, a particle in the front of the viewing chamber would appear larger than a same-sized particle at the back of the chamber. In general, there is an inverse relationship between a particle’s apparent size and its distance to the lens. Two objects of equal size with a 1:2 ratio of distance from a camera lens would have a 2:1 ratio in apparent size. To reduce this issue, the separation distance of 1.22 m between the center of the viewing chamber and the camera is employed in estimating particle size. This separation distance decreases the forced perspective effect of particles in the front or back of the viewing chamber by reducing the average subtended particle angle, thus minimizing the apparent size difference resulting from a particle’s distance from the camera lens. At this separation distance, a particle in the PASIV probe will have an apparent size difference in the range of 3%–10% (experimentally measured) between the front and the back of the viewing chamber. With no particle depth information available in actual particle samples and assuming a random particle position relative to the center of the viewing chamber, the expected value of uncertainty in the particle size measurement from this effect is about 7%.
Because of the camera settings, the viewing chamber must be well illuminated to avoid dark images wherein particles are hard to discriminate from the dark background. Six high-intensity light-emitting diodes (LEDs) are located on the sides of the viewing chamber to provide adequate particle illumination (Fig. 1c) and are powered by a set of eight CR123 batteries on each side of the viewing chamber (Figs. 1a,b). The video data are recorded on a standard 8- or 16-GB Secure Digital (SD) card. The LED lights and the rechargeable camera battery last approximately 1.5 h. The camera, when combined with an image analysis program, is capable of resolving the size, shape, and orientation of precipitation particles in the form of raindrops, graupel, small hail, and both pristine and aggregate snow particles (e.g., Fig. 3).
2) Parsivel
The Parsivel disdrometer is an optical sensor manufactured by OTT Hydromet (Loffler-Mang and Joss 2000). The system uses a 780-nm wavelength, 30-mm wide, approximately 1-mm-thick laser beam to detect particles as they pass through the sensing area. The amount of light blocked by the particle passing through the beam is proportional to its time-varying linear dimension in the plane of the beam, while the length of time the light is blocked provides information about the particle velocity (e.g., Battaglia et al. 2010). A proprietary algorithm by the manufacturer bins each detected particle according to its size and velocity, and the data output interval can be set to a minimum of 10 s. The 32 size bins are nonlinearly spaced between 0.062- and 24.5-mm diameter, with more bins located in the lower portion of the range (Table 1). The velocity bins are similarly structured between 0.05 and 20.8 m s−1. The Parsivel unit presently does not utilize the smallest two diameter or velocity bins. The factory-designed, ground-based Parsivel unit has been used in a number of studies to examine drop size distributions at ground level (Friedrich et al. 2013; Yuter et al. 2006; Löffler-Mang and Blahak 2001). The Parsivel measures the number of drops falling into an area in a set amount of time. To convert this size distribution to a volumetric size distribution, the particle velocity through the laser must be taken into account to determine sample volume and consequently particle concentrations for each particle size.
Bin diameter, size class spread, and error of the Parsivel disdrometer.
The factory-configured Parsivel unit is too heavy for the requirements of mobile ballooning (i.e., weighing nearly 6.35 kg in its original metal casing). To use this instrument on a balloonborne system, the optics and electronics have been removed, condensed, and repackaged into a small aluminum box dimensioned 50.8 cm × 17.8 cm × 3.8 cm (Fig. 4a). The completed assembly is capable of running for approximately 8 h using four CR123 batteries housed within the box. The Parsivel’s raw ASCII particle data stream is output as an RS-485 signal that is converted into a 3.3-V transistor–transistor logic (TTL) signal and recorded on a small microSD datalogger as ASCII text. This process bypasses the need for computer software to drive the Parsivel and turns the completed unit into a stand-alone instrument requiring no user interaction once initially reconfigured. The final weight of the covered Parsivel box assembly (Fig. 4b) is approximately 0.91 kg. The Parsivel box is mounted above the viewing chamber on the assembled PASIV (Figs. 1a,b). The aluminum housing effectively serves as a Faraday cage to shield the internal Parsivel electronics from high electric fields. Although Parsivel 1 was originally used, the current PASIV version has recently been upgraded to Parsivel 2 (i.e., as in Fig. 4, with all subsequent discussion referring to Parsivel 2).
c. Operational considerations
Certain caveats have been carefully considered during the PASIV design process to increase the accuracy and overall scientific utility of the PASIV measurements. Chief among these caveats is that the Parsivel is somewhat limited in providing only a 1D estimate of the equivalent volume diameter within the plane of the flattened laser beam. As will be discussed in section 2e, the result is that the equivalent volume diameter of nonspherical particles will likely be underestimated by the Parsivel. Hence, care must be taken when comparing the Parsivel observations to those of the videosonde–particle analyzer (PA) system. The Parsivel also does not allow for observations of single particles; rather, it obtains bulk observations over a time period while making no distinction between potentially liquid and solid particles of varying habits.
The PASIV is presently not capable of streaming data back to a ground receiving station. Transmitting data increases the cost, weight, and power requirements of an instrument, and exposes the system to potentially destructive electrical interference in thunderstorms. Line-of-sight requirements for signal strength and data compression/conversion issues are also limiting factors. Instead, the launched instrument must be retrieved to obtain the recorded data. This considerably reduces the overall cost of the instrument by allowing it to be reused with minimal refurbishment and reduces the design complexity. To recover the instruments, however, a GPS tracker is required to relay the landing coordinates to a retrieval team. A SPOT1 device reliably accomplishes this goal. The uniqueness of this device is its ability to relay its location every 10 min to the Globalstar satellite constellation, which in turn relayed those data through the Internet to a user in near–real time. With this device, the instrument may be located easily via the Internet-accessible location page, and typically driving and walking directly to the device with minimal searching. This retrieval is particularly feasible in the central United States with large areas of open land and relatively small bodies of water. To limit situations where the instrument is destroyed or otherwise separated from its SPOT device, two SPOT trackers are flown with one on each end of the PASIV. This procedure has been successfully employed during ballooning operations in Oklahoma, Texas, and Florida within the United States.
d. Data processing
After the PASIV is deployed via an in-storm sounding and the instrument retrieved, the data must be processed to obtain the particle concentration and size information. The ASCII Parsivel data are processed by a custom MATLAB script that retrieves the particle concentration and size information from the raw particle counts output by the Parsivel. The initial processing scans the data to obtain the raw particle bins and separate the particle counts into 10-s distributions. Data are trimmed to remove records before launch and after landing. During the launch process, the Parsivel’s laser is intentionally blocked in the viewing chamber to force its signal strength to zero. This clearly marks the launch in the data record, as the Parsivel’s signal strength returns to a normal operating level once this block is removed at launch.
A second program matches the trimmed Parsivel measurements with radiosonde observations to obtain a record of the Parsivel’s particle concentration and size data that is paired with temperature, pressure, humidity, and location information. Assuming that the first record of the trimmed Parsivel data is at launch, the subsequent radiosonde-derived atmospheric profile is determined from pressure, altitude, and vertical velocity data beginning at the known radiosonde launch time. The balloon burst time is determined by the beginning of the descent of the radiosonde (evident in altitude decrease, pressure increase, and direction of the vertical velocity), and is used to further trim the data, so that only the ascent portion is examined. Since the Parsivel outputs data every 10 s, the radiosonde data are averaged over a 10-s period to match the temporal resolution of the Parsivel. The final merged Parsivel and radiosonde dataset is output as a comma-delimited ASCII text string that can be input into other programs.
The measurement of particle concentration and size from the videosonde camera requires intensive and complex image processing following general digital processing techniques that are increasingly being employed in physical science studies (e.g., E. Rasmussen 2010, personal communication; Ogliore et al. 2012; Liu et al. 2014). The original video camera movie files are processed into individual portable network graphics (PNG) images (i.e., one image per frame). Each image is analyzed to identify particles using a commercially available Interactive Data Language (IDL) program known as “particle analyzer” (e.g., Fig. 5).2 The PA program first enhances the image contrast, then scans each pixel to see if its brightness is above a running background threshold. Insufficiently bright pixels are masked by resetting their color as black. Conversely, sufficiently bright pixels are reset as white. After the brightened pixels are identified and whitened, tightly grouped whitened pixels are merged together to identify “particle objects.” These particle objects are then checked to see if multiple occurrences in the same location over a series of images exist. Multiple particle occurrences may be a result of the particle object being imaged more than once without moving, typically from being stuck in the viewing volume due to accumulated water substance on the black background surface but also due to an occasional camera system aberration (the latter being described in later sections). In the event of multiple-particle detections, the redundant particles are objectively removed from the analysis.3 The identified particle objects are quantified regarding their size, shape, and major/minor axis length distributions, and a custom graphical user interface (GUI) is used to manage the processing and view results of the PA program (Fig. 5). If a sounding is available and input along with the image data, then the PA program will automatically align the images with the radiosonde data by matching the first image in the analysis with the launch record of the sounding that has also been input. However, utilizing a sounding is not necessary for executing the PA program and may optionally be deferred. Available outputs of the PA program and their descriptions are listed in Table 2.
Output parameters from the PA image analysis program.
A particle identification test of the Parsivel and camera has been performed by dropping spheres of known diameter through the PASIV instrument (e.g., processed PA images in Fig. 6). A comparison of the raw and PA-processed particle images shows that the identified whitened pixels and an elliptical model fit to the particle size and shape by the PA program are consistent with the raw particle image. Note that the PA-processed equivalent spherical volume diameters are slightly smaller than the known particle diameters, likely due to a combination of forced perspective uncertainty, edge blurring, and background noise thresholding effects involved in the identification of the particle objects (i.e., a systematic function of diameter to be elucidated in section 2e).
The PA program utilizes a number of user-defined input parameters that control main aspects of the image processing. These user-defined input parameters allow the PA analysis to be more or less sensitive to brightness contrast and frame-to-frame brightness increases. The operable size range of detected particles may also be shifted on the lower end. These parameters can be tested on objects of known size until robustly consistent and reasonable results are objectively obtained. A user-defined selection area allows a subsection of the dataset to be analyzed and can control which image is being shown in the PA algorithm’s particle image area. Sample statistics can be calculated from the PA-analyzed dataset, and histograms of the radius, eccentricity, irregularity, and tilt utilizing both raw counts (e.g., number per size bin) and concentrations (e.g., number per cubic meter per millimeter size) to display the synthesized results spanning some PA-processed data time interval. The resulting PA-output data table provides this information for each particle (one record per particle), facilitating follow-on analysis. To aid in comparison with the Parsivel particle data, the resulting PA-analyzed videosonde image data are binned using the same bin size employed by the Parsivel (Table 1).
e. Measurement validation
1) Sizing accuracy and error correction
Sizing tests have been performed following the method described in section 2d (e.g., Figs. 5, 6) to estimate the accuracy of the PASIV probe and characterize errors. To test the PASIV capability of identifying and sizing particles, a series of balls of known sizes have been dropped through the PASIV probe and the resulting output data analyzed (Fig. 7). Three types of spheres (acrylic, Delrin, and steel ball bearings) are used. An acrylic sphere closely resembles a visual image of a raindrop when viewed by the camera4 (i.e., including external reflection and internal refraction of incident light), while a Delrin sphere closely resembles a graupel particle in dry-growth mode. Steel bearings are used to achieve a larger range of test sizes, since the available size range of acrylic and Delrin spheres is limited.
The Parsivel also has a slight tendency to underestimate the size of the test spheres (Fig. 7). This is largely caused by the size of the bins used to separate the detected spherical objects (i.e., the object size is reported as the center of each bin) and results in a maximum size difference of approximately ±8%. It is noted that the PASIV flights in storms (e.g., as discussed in section 3), specifically the camera data, have revealed a tendency for many PA-analyzed precipitation particles to be significantly elliptical in shape (e.g., long, narrow, needle-shaped ice crystals). As the plane of the major axis is often not parallel to the flattened Parsivel laser beam, the Parsivel observations are thus often characterized by smaller measured particle diameters than the equivalent spherical PA-analyzed particle diameters. A regression relationship similar to Eq. (1) could be (though is not presently) fitted to the Parsivel data from the drop tests using the same power-law form as the PA data analysis. A regression line was not fit, as it is hypothesized that the ellipticity errors could locally dominate in storm observations and represent an uncertainty that cannot be corrected using the Parsivel data alone. Instead, the overall Parsivel error of each size bin is simply estimated according to the class spread of that particular bin (Table 2).
2) Detection accuracy and error correction
Given the frame rate of the camera and the viewing chamber size, it is possible that faster-moving particles may be missed between frames of the camera, whereas slower-moving particles could be counted multiple times. Each image of the viewing chamber is taken in 0.000 125 s (i.e., the shutter speed of the camera), and there is a period of 0.0417 s between successive images (i.e., the frame rate). If an average balloon ascent rate of 5 m s−1 is assumed, then a particle falling at 2 m s−1 relative to ground travels through the viewing chamber at 7 m s−1 and has a residence time of 0.026 s in the viewing chamber. A particle at this speed is capable of traveling through the viewing chamber entirely between frames without being imaged. Given the shutter speed of the camera and this particle motion, blurring of particles could be expected due to their motion during the frame capture. This behavior has not been experimentally observed. While the entire frame is captured in 0.000 125 s, the pixels containing the particle are likely scanned by the camera in a fraction of this time, thus reducing the overall particle motion during the exposure of a given particle.
3. Example observations
The PASIV instrument has been successfully deployed in more than 20 research flights in several projects spanning a range of scientific objectives. The Deep Convective Clouds and Chemistry (DC3) experiment has been the first major deployment of PASIV. The objectives of DC3 include the examination of PSDs in central plains severe storms and the analysis of their relationships with lightning, in situ electric fields, and cloud chemistry. The PASIV instrument has also been utilized in Florida during a Defense Advanced Research Projects Agency (DARPA) storm electricity project to investigate triggered lightning. More recently, a small winter ballooning project has been conducted from Norman, Oklahoma, to examine the microphysics of refreezing signatures associated with sleet events. The number of deployable Parsivel units was limited, so many of these flights did not have a Parsivel system; however, the camera system alone is capable of providing quantitative information about the particle distribution. While detailed descriptions of several of these cases will be presented in future work, a brief description of two PASIV cases follows to illustrate the ability of the PASIV instrument to provide critical in situ data in severe weather environments.
a. 21 June 2012—DC3
A PASIV instrument launched at 1654 UTC (all times are universal time) on 21 June 2012 near Binger, Oklahoma, has revealed the internal microphysical structure of individual cells within a complex of weak pulse-type storms (Fig. 9). The launch conditions are relatively benign, with light stratiform rain falling around the time of launch. The PASIV rises completely through the storm to a maximum altitude of 18.7 km at 72 mb and samples a total of over 316 000 particles via the Parsivel (e.g., Fig. 10), with several 10-s periods exceeding 4000 particles.
The derived histograms of particle count versus size from the Parsivel and the videosonde-PA system both reveal a sharp increase in both the number and size of particles starting at 600 mb, which coincides with the melting level (Fig. 10). The abrupt increase in total particle concentration near the melting level is also noticeable in videosonde-PA system measurements (Fig. 11). At altitudes below the 600-mb level, it is verified from individual videosonde images (not shown) that the rapidly decreasing particle sizes and counts associate with melting ice particles that collapse to form raindrops that can subsequently begin to evaporate in the dry subcloud layer. There is a relative minimum of particle sizes and concentrations around 500 mb below a much deeper layer containing larger particle counts and sizes exceeding 4 mm. Individual videosonde images suggest that several of the latter particles are large aggregates (e.g., Fig. 12) that lower-level measurements imply do not survive the subsequent descent to the melting layer.
The videosonde-PA distribution appears to cut off at the top of the sounding (Fig. 10b), albeit at nearly the top of the storm as indicated by the Parsivel (Fig. 10a). This cutoff feature is thought to be caused by sunlight reflecting on the viewing chamber’s black background, resulting in the PA program having greater difficulty resolving particles due to degraded background contrast at storm top. The Parsivel is immune to the latter background contrast problem, and thus it is able to resolve the few particles present at this altitude.
The particle sizes from the videosonde-PA system and the Parsivel, and concentrations from the videosonde-PA system, also contain one or more camera aberrations, false detections, or system noise (i.e., as indicated by statistically rare outlying concentrations at large diameters) that collectively add uncertainty to the large-particle upper tails of the size distributions. A sorting algorithm is currently being developed and tested to remove such spurious outlying large-particle concentrations from future analyses. The relatively rare outlier data points primarily affect the scattered detections on the edges of the measured distribution without significantly altering the primary microphysical structure.
The particle size profiles measured by the Parsivel and the videosonde-PA system are similar in structure. However, the mean diameters of the particle histograms appear to be shifted toward smaller sizes in the Parsivel data. The offset of mean particle diameters in the Parsivel is likely caused by the asymmetry of sampled particles less than 1 mm in diameter as previously discussed in section 2 (e.g., Battaglia et al. 2010) that videosonde images indicate are frequently highly elliptical and oriented at an angle to the horizontal (not shown). The videosonde-PA system is able to measure two spatial particle dimensions, and thus it is more likely to measure the major axis than the Parsivel, which measures only one horizontal dimension.
The Parsivel-measured particle velocities are compared to various theoretical fall speed relationships to qualitatively assess their information content (Fig. 13). The melting layer was avoided in this analysis due to difficulties separating mixed-phase particles for comparisons with theoretical relations. The assumed still-air rise rate of 4.5 m s−1 for the balloon has been added to the theoretical fall speed curve to compare with Parsivel velocities, which are the sum of the particle terminal fall speed with respect to still air and the balloon ascent rate. The Parsivel observations in the rain layer (Fig. 13a) show particles increasing in relative velocity from roughly 5 m s−1 at diameters around 0.5 mm to approximately 11 m s−1 at diameters around 3.5 mm. These observations appear to agree reasonably well with terminal velocity relations (Beard 1977; Gunn and Kinzer 1949), though the Parsivel observations are systematically slower velocities. This could be an indication of a slower balloon ascent rate. The velocity relation from Gunn and Kinzer (1949) at altitude has been adjusted using a density correction at the midpoint of the layer from the surface to the melting level. It is rather likely that some Parsivel-sampled ice particles in various stages of melting are included in the sample.
For the ice region (Fig. 13b), several diameter–velocity relations are shown due to the presence of various particle habits as imaged by the videosonde. As with the rain region, the velocity relations were calculated using a midpoint of the atmospheric conditions across the ice layer. Since rimed ice particles in this rather weak, decaying convective cloud likely encounter rather low supercooled liquid water contents at cold riming temperatures and small impaction velocities, two lump graupel calculations were performed using ice densities of 300 (black line with diamonds) and 100 kg m−3 (black line with circles) using the generalized ice particle fall speed expression of Böhm (1989). To represent columns, the “Cle” relation from Davis and Auer (1974) was used (blue line). Magono and Lee (1966) defined crystal classifications for nearly 80 crystal types that Davis and Auer and others have used; the Cle relation refers to solid bullets. Two relations for plates were used—“Pla” (green dashed line) following Davis and Auer (1974) for hexagonal plates and “P1e” (green line with “x” symbol) for ordinary dendritic crystals from Pruppacher and Klett (1997). The curve “R1d” (black line with “x” symbol) is a relation for rimed stellar crystals from Pruppacher and Klett (1997). Finally, “AggDu” (red line) is for aggregates of unrimed radiating assemblages of dendrites from Pruppacher and Klett (1997), while “AggDr” (red line dashed line with circles) is for rimed assemblages. The various PASIV-relative ice particle fall speed relationships are also broadly consistent with the Parsivel-measured particle velocities, though the mixture of imaged ice particle habits and range of theoretical fall speed magnitudes somewhat limits quantitative comparison.
It has been noted that the balloon is assumed to have had a constant ascent rate relative to still air through the observed 8-km-deep ice particle layer on 21 June. While this assumption is likely valid in an average sense, there may be occasional periods during ascent in which the balloon rise rate is slower than 4.5 m s−1 (e.g., due to possible variations of the form drag of the balloon that are not exactly compensated by decreasing air density). Any such deviation from the assumed still-air ascent rate would result in slower measured relative velocities, thus requiring a corresponding downward shift of the theoretical fall speed relations. One possible explanation for the lower tail of measured velocities below 4 m s−1 and 1-mm diameter (Fig. 13b) could be the local transient ascent rate. A possibly more likely explanation for the lower velocity tail are local occurrences of the aforementioned concentration of highly elliptical ice particles that are tilted relative to the Parsivel laser beam, thus overestimating fall time relative to underestimated size and reducing derived fall speed accordingly (Battaglia et al. 2010). These interesting questions regarding the appropriate PASIV reference frame and impact of particle shape each bear further study and will be explored in subsequent research.
b. 29 May 2012—DC3
A severe weather outbreak on 29–30 May 2012 in central and northern Oklahoma has provided the opportunity to obtain PASIV soundings within a supercell storm. Launched at 2323 UTC 29 May, roughly 14 km north of Kingfisher, Oklahoma, a PASIV sounding has revealed characteristics of the internal microphysical structure of the forward overhang precipitation core of the southernmost supercell (Fig. 14). After launch the instrument train is carried initially into the downstream supercell flank and subsequently into the left anvil region as revealed by a triple-Doppler radar analysis (Fig. 14). The balloon location at 2342 UTC and at approximately 6-km altitude corresponds to a plume of ~35–40-dBZ reflectivity, with implied horizontal precipitation particle advection that originates from the storm core to the west-southwest and proceeds through the balloon’s location.
The derived histogram profile of particle count versus size from the PASIV flight launched at 2323 UTC is restricted to the camera system (Fig. 15) due to the limited number of available Parsivel units on 29 May. In very broad similarity to the 21 June PASIV flight, high numbers of ice particles are noted above the base of the anvil layer around 500 mb and smaller raindrop counts are observed around 950 and 750 mb. Note that there are several hypothetically spurious particle detections in the portion of the lowest 400 mb characterized by otherwise negligibly small mean local counts in the height–diameter domain. A few excessively large (>10 mm) spurious particles in this region (not shown) have been traced in detailed postanalysis to vibrations of the unit that caused the camera to image a thin edge section of the viewing chamber structure (as verified by viewing the questioned image), resulting in the PA analysis program identifying these false objects. As mentioned previously, a sorting algorithm is currently being developed to remove the rather minor concentrations of erroneous particles from future analysis. The low total particle concentrations in the lowest 400 mb (including an absolute minimum concentration around the melting level near 600 mb) and the abrupt increase in concentration above the anvil base at 500 mb are also evident in the total particle concentration profile from the camera-PA system analysis (Fig. 16).
Several aspects of the 29 May PASIV flight illustrate both the hazardous supercell internal conditions and the overall robustness of the PASIV and associated balloonborne instrumentation. A radiosonde malfunction resulted in the loss of temperature and RH data above 500 mb. Pressure and location information are nevertheless still available, which allows the vertical placement of the PSD data in the storm. Furthermore, the sounding terminates prematurely at 350 mb due to a lightning strike to the PASIV. This strike severs the rigging supporting the instrument from the balloon and parachute and thus initiates rapid PASIV descent. The camera and GPS tracker both survive the high-speed descent from altitude and the corresponding impact undamaged.
4. Conclusions
A balloonborne hybrid in situ precipitation particle measuring system known as the Particle Size, Image, and Velocity (PASIV) probe has been designed and tested by NSSL. The PASIV consists of an HD video camera and a modified Parsivel (Löffler-Mang and Joss 2000) that have been combined to form a single-balloonborne instrument while staying within the operating conditions required by the FAA for unmanned balloons. Following its creation, the PASIV has been used in a variety of field campaigns spanning a wide range of conditions. Partly based on the work of M87 and T90, this dual-instrument, battery-powered, semiautonomous system is capable of high-spatiotemporal measurements of particle size, shape, velocity, orientation, eccentricity, and composition within a quasi-vertical profile through any type of precipitation.
The PASIV instrument has performed rather well across the wide range of conditions encountered. The structure of the instrument is able to withstand the punishing wind shears and heavy precipitation encountered inside severe weather, including a direct strike by lightning. Thus, the instrument is capable not only of collecting high-resolution particle data but operating in difficult observing environments critical to the understanding of storm phenomena. As with any instrument or project, a few minor problems have been encountered, including faulty camera settings (e.g., the user-settable shutter speed and the gain of the cameras used) that led to blurred particles. These errors are correctible via operator training and have become less significant as experience deploying the instrument has increased.
Preliminary results from two cases during the DC3 field campaign have highlighted the ability of the PASIV to provide critical vertical profiles of the PSD in severe weather. These cases demonstrate the feasibility of the PASIV probe to collect much-needed in situ data for use in studies of storm electrification and cloud microphysics, as well as dual-polarization radar validation. The processing steps for the PASIV data will subsequently be refined to eliminate minor particle distribution artifacts, diagnose particle habit information, and estimate radar measurands, with the resulting data analyses used to examine the microphysical structure of several severe convective storms.
Acknowledgments
We thank Dr. Erik Rasmussen of the Cooperative Institute for Mesoscale Meteorological Studies at the University of Oklahoma (OU) and the National Severe Storms Laboratory (NSSL) for his significant contribution to this work. His efforts in writing the IDL Particle Analyzer program that processes the PASIV image data have been invaluable. We also gratefully acknowledge Daniel Betten and Dr. Michael Biggerstaff of the OU School of Meteorology for providing their radar analysis for this study, as well as Dr. Ted Mansell, who provided considerable support regarding various fall speed calculations. Finally, we thank the four anonymous reviewers, whose comments and suggestions were instrumental in improving the quality of this manuscript. Support for this research was provided by NSF Grants AGS-10145102 and AGS-1063945 and by the NSSL, the latter including federal funding for Rasmussen Systems, LLC, to develop the Particle Analyzer program. Funding was also provided by the NOAA/Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA11OAR4320072, U.S. Department of Commerce. A number of students have contributed to this work through their help in assembly of PASIV units and assisting balloon launches. Without their dedication and willingness to help, this work would not have been possible.
REFERENCES
Battaglia, A., Rustemeier E. , Tokay A. , Blahak U. , and Simmer C. , 2010: PARSIVEL snow observations: A critical assessment. J. Atmos. Oceanic Technol., 27, 333–344, doi:10.1175/2009JTECHA1332.1.
Beard, K. V., 1977: Terminal velocity and shape of cloud and precipitation drops aloft. J. Atmos. Sci., 34, 1293–1298, doi:10.1175/1520-0469(1977)034<1293:TVAFCA>2.0.CO;2.
Böhm, H. P., 1989: A general equation for the terminal fall speed of solid hydrometeors. J. Atmos. Sci., 46, 2419–2427, doi:10.1175/1520-0469(1989)046<2419:AGEFTT>2.0.CO;2.
Boussaton, M. P., Coquillat S. , Chauzy S. , and Gangneron F. , 2004: A new videosonde with a particle charge measurement device for in situ observation of precipitation particles. J. Atmos. Oceanic Technol., 21, 1519–1531, doi:10.1175/1520-0426(2004)021<1519:ANVWAP>2.0.CO;2.
Cao, Q., Zhang G. , Brandes E. A. , Schuur T. J. , Ryzhkov A. , and Ikeda K. , 2008: Analysis of video disdrometer and polarimetric radar data to characterize rain microphysics in Oklahoma. J. Appl. Meteor. Climatol., 47, 2238–2255, doi:10.1175/2008JAMC1732.1.
Cao, Q., Zhang G. , Brandes E. A. , and Schuur T. J. , 2010: Polarimetric radar rain estimation through retrieval of drop size distribution using a Bayesian approach. J. Appl. Meteor. Climatol., 49, 973–990, doi:10.1175/2009JAMC2227.1.
Davis, C. I., and Auer A. H. Jr., 1974: Use of isolated orographic clouds to establish the accuracy of diffusional ice crystal growth equations. Preprints, Conf. on Cloud Physics, Tucson, AZ, Amer. Meteor. Soc., 141–147.
Friedrich, K., Kalina E. A. , Masters F. J. , and Lopez C. R. , 2013: Drop-size distributions in thunderstorms measured by optical disdrometers during VORTEX2. Mon. Wea. Rev., 141, 1182–1203, doi:10.1175/MWR-D-12-00116.1.
Grossklaus, M., Uhlig K. , and Hasse L. , 1998: An optical disdrometer for use in high wind speeds. J. Atmos. Oceanic Technol., 15, 1051–1059, doi:10.1175/1520-0426(1998)015<1051:AODFUI>2.0.CO;2.
Gunn, R., and Kinzer G. D. , 1949: The terminal velocity of fall for water droplets in stagnant air. J. Meteor., 6, 243–248, doi:10.1175/1520-0469(1949)006<0243:TTVOFF>2.0.CO;2.
Habib, E., Krajewski W. F. , and Kruger A. , 2001: Sampling errors of tipping-bucket rain gauge measurements. J. Hydrol. Eng., 6, 159–166, doi:10.1061/(ASCE)1084-0699(2001)6:2(159).
Heymsfield, A. J., Schmitt C. , and Bansemer A. , 2013: Ice cloud particle size distributions and pressure-dependent terminal velocities from in situ observations at temperatures from 0° to −86°C. J. Atmos. Sci., 70, 4123–4154, doi:10.1175/JAS-D-12-0124.1.
Jorgensen, D. P., and Willis P. T. , 1982: A Z–R relationship for hurricanes. J. Appl. Meteor., 21, 356–366, doi:10.1175/1520-0450(1982)021<0356:AZRRFH>2.0.CO;2.
Kruger, A., and Krajewski W. F. , 2002: Two-dimensional video disdrometer: A description. J. Atmos. Oceanic Technol., 19, 602–617, doi:10.1175/1520-0426(2002)019<0602:TDVDAD>2.0.CO;2.
Liu, X. C., Gao T. C. , and Liu L. , 2014: A video precipitation sensor for imaging and velocimetry of hydrometeors. Atmos. Meas. Tech., 7, 2037–2046, doi:10.5194/amt-7-2037-2014.
Löffler-Mang, M., and Joss J. , 2000: An optical disdrometer for measuring size and velocity of hydrometeors. J. Atmos. Oceanic Technol., 17, 130–139, doi:10.1175/1520-0426(2000)017<0130:AODFMS>2.0.CO;2.
Löffler-Mang, M., and Blahak U. , 2001: Estimation of the equivalent radar reflectivity factor from measured snow size spectra. J. Appl. Meteor., 40, 843–849, doi:10.1175/1520-0450(2001)040<0843:EOTERR>2.0.CO;2.
Magono, C., and Lee C. E. , 1966: Meteorological classification of natural snow crystals. J. Fac. Sci. Hokkaido Univ. Ser. 7, 2, 321–362.
Mahlke, H., Corsmeier U. , Kottmeier C. , and Löffler-Mang M. , 2008: The new balloon-borne disdrometer ‘Flying Parsivel.’ Seventh Convective and Orographically Induced Precipitation Study (COPS) Workshop, Strasbourg, France, Karlsruhe Institute of Technology, C8. [Available online at https://www.imk-tro.kit.edu/download/Poster_Flying_Parsivel.pdf.]
McFarquhar, G. M., Timlin M. S. , Rauber R. M. , Jewett B. F. , Grim J. A. , and Jorgensen D. P. , 2007: Vertical variability of cloud hydrometeors in the stratiform region of mesoscale convective systems and bow echoes. Mon. Wea. Rev., 135, 3405–3428, doi:10.1175/MWR3444.1.
Miloshevich, L. M., and Heymsfield A. J. , 1997: A balloon-borne continuous cloud particle replicator for measuring vertical profiles of cloud microphysical properties: Instrument design, performance, and collection efficiency analysis. J. Atmos. Oceanic Technol., 14, 753–768, doi:10.1175/1520-0426(1997)014<0753:ABBCCP>2.0.CO;2.
Murakami, M., and Matsuo T. , 1990: Development of the hydrometeor videosonde. J. Atmos. Oceanic Technol., 7, 613–620, doi:10.1175/1520-0426(1990)007<0613:DOTHV>2.0.CO;2.
Murakami, M., Matsuo T. , Nakayama T. , and Tanaka T. , 1987: Development of cloud particle video sonde. J. Meteor. Soc. Japan, 65, 803–809.
Norment, H. G., 1988: Three-dimensional trajectory analysis of two drop sizing instruments: PMS* OAP and PMS* FSSP. J. Atmos. Oceanic Technol., 5, 743–756, doi:10.1175/1520-0426(1988)005<0743:TDTAOT>2.0.CO;2.
Ogliore, R., Floss C. , Stadermann F. , Kearsley A. , Leitner J. , Stroud R. , and Westphal A. , 2012: Automated searching of stardust interstellar foils. Meteorit. Planet. Sci., 47, 729–736, doi:10.1111/j.1945-5100.2011.01325.x.
Pruppacher, H. R., and Klett J. D. , 1997: Microphysics of Clouds and Precipitation. 2nd ed. Kluwer Academic Press, 954 pp.
Rust, W. D., and Marshall T. C. , 1989: Mobile, high-wind, balloon-launching apparatus. J. Atmos. Oceanic Technol., 6, 215–217, doi:10.1175/1520-0426(1989)006<0215:MHWBLA>2.0.CO;2.
Schoenberg Ferrier, B., 1994: A double-moment multiple-phase four-class bulk ice scheme. Part I: Description. J. Atmos. Sci., 51, 249–280, doi:10.1175/1520-0469(1994)051<0249:ADMMPF>2.0.CO;2.
Schuur, T. J., Ryzhkov A. V. , Zrnić D. S. , and Schönhuber M. , 2001: Drop size distributions measured by a 2D video disdrometer: Comparison with dual-polarization radar data. J. Appl. Meteor., 40, 1019–1034, doi:10.1175/1520-0450(2001)040<1019:DSDMBA>2.0.CO;2.
Smith, A. M., McFarquhar G. M. , Rauber R. M. , Grim J. A. , Timlin M. S. , Jewett B. F. , and Jorgensen D. P. , 2009: Microphysical and thermodynamic structure and evolution of the trailing stratiform regions of mesoscale convective systems during BAMEX. Part I: Observations. Mon. Wea. Rev., 137, 1165–1185, doi:10.1175/2008MWR2504.1.
Takahashi, T., 1990: Near absence of lightning in torrential rainfall producing Micronesian thunderstorms. Geophys. Res. Lett., 17, 2381–2384, doi:10.1029/GL017i013p02381.
Takahashi, T., 2010: The videosonde system and its use in the study of East Asian monsoon rain. Bull. Amer. Meteor. Soc., 91, 1231–1246, doi:10.1175/2010BAMS2777.1.
Ulbrich, C. W., and Atlas D. , 1998: Rainfall microphysics and radar properties: Analysis methods for drop size spectra. J. Appl. Meteor., 37, 912–923, doi:10.1175/1520-0450(1998)037<0912:RMARPA>2.0.CO;2.
Yuter, S. E., Kingsmill D. E. , Nance L. B. , and Löffler-Mang M. , 2006: Observations of precipitation size and fall speed characteristics within coexisting rain and wet snow. J. Appl. Meteor. Climatol., 45, 1450–1464, doi:10.1175/JAM2406.1.
The SPOT device (www.findmespot.com) was first employed during a 2008 field experiment.
Particle Analyzer software and documentation are available at http://www.rasmsys.com/.
A similar precipitation particle image processing algorithm has been reported by Liu et al. (2014).
Alternatively, Liu et al. (2014) employed glass spheres to simulate raindrops in their drop tests.