Datasets:

---

LLMDH

/

OpenScience

like 0

Follow

LLM - Digital Humanities 16

ERROR: type should be string, got "https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. The open source Video In Situ Snowfall Sensor (VISSS) is introduced as a novel instrument for the characterization\nof particle shape and size in snowfall. The VISSS consists of two cameras with LED backlights and telecentric lenses that\nallow accurate sizing and combine a large observation volume with relatively high resolution and a design that limits wind\ndisturbance. VISSS data products include per-particle properties and integrated particle size distribution properties such as\nparticle maximum extent, cross-sectional area, perimeter, complexity, and—in the future—sedimentation velocity. Initial anal-\n5\nysis shows that the VISSS provides robust statistics based on up to 100,000 particles observed per minute. Comparison of the particle maximum extent, cross-sectional area, perimeter, complexity, and—in the future—sedimentation velocity. Initial anal-\n5\nysis shows that the VISSS provides robust statistics based on up to 100,000 particles observed per minute. Comparison of the\nVISSS with collocated PIP and Parsivel instruments at Hyytiälä, Finland, shows excellent agreement with Parsivel, but reveals\nsome differences for the PIP (Precipitation Imaging Package) that are likely related to PIP data processing and limitations of\nthe PIP with respect to observing smaller particles. The open source nature of the VISSS hardware plans, data acquisition\nsoftware and data processing libraries invites the community to contribute to the development of the instrument which has\n10 particle maximum extent, cross-sectional area, perimeter, complexity, and—in the future—sedimentation velocity. Initial anal-\n5\nysis shows that the VISSS provides robust statistics based on up to 100,000 particles observed per minute. Comparison of the\nVISSS with collocated PIP and Parsivel instruments at Hyytiälä, Finland, shows excellent agreement with Parsivel, but reveals\nsome differences for the PIP (Precipitation Imaging Package) that are likely related to PIP data processing and limitations of\nthe PIP with respect to observing smaller particles. The open source nature of the VISSS hardware plans, data acquisition\nsoftware, and data processing libraries invites the community to contribute to the development of the instrument, which has\n10\nmany potential applications in atmospheric science and beyond. 5 software, and data processing libraries invites the community to contribute to the development of the instrument, which has\n10\nmany potential applications in atmospheric science and beyond. 10 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Introducing the Video In Situ Snowfall Sensor (VISSS)\nMaximilian Maahn1, Dmitri Moisseev2,3, Isabelle Steinke1,*, Nina Maherndl1, and Matthew D. Shupe4,5\n1Leipzig University, Leipzig Institute of Meteorology (LIM), Germany\n2Institute for Atmospheric and Earth System Research/Physics, Faculty of Science, University of Helsinki, Finland\n3Finnish Meteorological Institute, Helsinki, Finland\n4University of Colorado, Cooperative Institute for Research in Environmental Sciences, Boulder, Colorado, USA\n5National Oceanographic and Atmospheric Administration, Physical Sciences Laboratory, Boulder, Colorado, USA\n*Now at Geoscience and Remote Sensing Department, Delft University of Technology, Delft, Netherlands\nCorrespondence: Maximilian Maahn (maximilian.maahn@uni-leipzig.de) Introducing the Video In Situ Snowfall Sensor (VISSS)\nMaximilian Maahn1, Dmitri Moisseev2,3, Isabelle Steinke1,*, Nina Maherndl1, and Matthew D. Shupe4,5\n1Leipzig University, Leipzig Institute of Meteorology (LIM), Germany\n2Institute for Atmospheric and Earth System Research/Physics, Faculty of Science, University of Helsinki, Finland\n3Finnish Meteorological Institute, Helsinki, Finland\n4University of Colorado, Cooperative Institute for Research in Environmental Sciences, Boulder, Colorado, USA\n5National Oceanographic and Atmospheric Administration, Physical Sciences Laboratory, Boulder, Colorado, USA\n*Now at Geoscience and Remote Sensing Department, Delft University of Technology, Delft, Netherlands\nCorrespondence: Maximilian Maahn (maximilian.maahn@uni-leipzig.de) Abstract. The open source Video In Situ Snowfall Sensor (VISSS) is introduced as a novel instrument for the characterization\nof particle shape and size in snowfall. The VISSS consists of two cameras with LED backlights and telecentric lenses that\nallow accurate sizing and combine a large observation volume with relatively high resolution and a design that limits wind\ndisturbance. VISSS data products include per-particle properties and integrated particle size distribution properties such as\nparticle maximum extent, cross-sectional area, perimeter, complexity, and—in the future—sedimentation velocity. Initial anal-\n5\nysis shows that the VISSS provides robust statistics based on up to 100,000 particles observed per minute. Comparison of the\nVISSS with collocated PIP and Parsivel instruments at Hyytiälä, Finland, shows excellent agreement with Parsivel, but reveals\nsome differences for the PIP (Precipitation Imaging Package) that are likely related to PIP data processing and limitations of\nthe PIP with respect to observing smaller particles. The open source nature of the VISSS hardware plans, data acquisition\nsoftware, and data processing libraries invites the community to contribute to the development of the instrument, which has\n10\nmany potential applications in atmospheric science and beyond. Abstract. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Remote sensing observations of snowfall are indirect, which limits their ability to identify snow particle shape by design. Ground-based in situ observations of ice and snow particles can identify the fingerprints of the snowfall formation processes and\n25\nprovide detailed information on particle size, shape, and fall velocity. Using assumptions about fall velocity or an aggregation\nand riming model as a reference, the particle mass-size and/or density relationship can also be inferred from in situ observations. (Tiira et al., 2016; von Lerber et al., 2017; Pettersen et al., 2020; Tokay et al., 2021; Leinonen et al., 2021; Vázquez-Martín\net al., 2021a). Various attempts have been made to classify particle types and identify active snowfall formation processes using\nvarious machine learning techniques (Nurzy´nska et al., 2013; Grazioli et al., 2014; Praz et al., 2017; Hicks and Notaroš, 2019;\n30\nLeinonen and Berne, 2020; Del Guasta, 2022); these classifications are needed to support quantification of snowfall formation\nprocesses (Grazioli et al., 2017; Moisseev et al., 2017; Dunnavan et al., 2019; Pasquier et al., 2023). In situ observations have\nalso been used to characterize particle size distributions (Kulie et al., 2021; Fitch and Garrett, 2022), investigate sedimentation\nvelocity and turbulence of hydrometeors (Garrett et al., 2012; Garrett and Yuter, 2014; Li et al., 2021; Vázquez-Martín et al.,\n2021b; Takami et al., 2022), and for model evaluation (Vignon et al., 2019). In combination with ground-based remote sensing,\n35\nin situ snowfall data have been used to validate or better understand remote sensing observations (Gergely and Garrett, 2016;\nLi et al., 2018; Matrosov et al., 2020; Luke et al., 2021), to develop joint radar in situ retrievals (Cooper et al., 2017, 2022),\nand to train remote sensing retrievals (Huang et al., 2015; Vogl et al., 2022). Remote sensing observations of snowfall are indirect, which limits their ability to identify snow particle shape by design. Remote sensing observations of snowfall are indirect, which limits their ability to identify snow particle shape by design. Ground-based in situ observations of ice and snow particles can identify the fingerprints of the snowfall formation processes and\n25\nprovide detailed information on particle size, shape, and fall velocity. Using assumptions about fall velocity or an aggregation\nand riming model as a reference, the particle mass-size and/or density relationship can also be inferred from in situ observations. 1\nIntroduction It is well known that \"every snowflake is unique\". The shape of a snow crystal is very sensitive to the processes that were active\nduring its formation and growth. Vapor depositional growth leads to a myriad of crystal shapes depending on temperature,\nhumidity, and their turbulent fluctuations. Aggregation combines individual crystals into complex snowflakes. Riming describes\n15\nthe freezing of small droplets onto ice crystals, causing them to rapidly gain mass and form a more rounded shape. In other\nwords, the shape of snow particles is a fingerprint of the dominant processes during the lifecycle of snowfall.i It is well known that \"every snowflake is unique\". The shape of a snow crystal is very sensitive to the processes that were active\nduring its formation and growth. Vapor depositional growth leads to a myriad of crystal shapes depending on temperature, yl\nq\np\ny\ny\np\nduring its formation and growth. Vapor depositional growth leads to a myriad of crystal shapes depending on temperature,\nhumidity, and their turbulent fluctuations. Aggregation combines individual crystals into complex snowflakes. Riming describes\n15\nthe freezing of small droplets onto ice crystals, causing them to rapidly gain mass and form a more rounded shape. In other\nwords, the shape of snow particles is a fingerprint of the dominant processes during the lifecycle of snowfall. 15 humidity, and their turbulent fluctuations. Aggregation combines individual crystals into complex snowflakes. Riming describes\n15\nthe freezing of small droplets onto ice crystals, causing them to rapidly gain mass and form a more rounded shape. In other\nwords, the shape of snow particles is a fingerprint of the dominant processes during the lifecycle of snowfall. Better observations of the fingerprints of snowfall formation processes are needed to advance our understanding of ice and Better observations of the fingerprints of snowfall formation processes are needed to advance our understanding of ice and\nmixed-phase clouds and precipitation formation processes (Morrison et al., 2020). Given the importance of snowfall formation processes for global precipitation (Mülmenstädt et al., 2015; Field and Heymsfield, 2015), the lack of process understanding\n20\nleads to gaps in the representation of these processes in numerical models. In a warming climate, precipitation amounts and\nextreme events, including heavy snowfall, are expected to increase (Quante et al., 2021), but the exact magnitudes are associated\nwith large uncertainties (Lopez-Cantu et al., 2020). 1 1 This limitation can be overcome when adding a second line camera as for the 2DVD (2-dimensional video disdrometer,\nSchönhuber et al., 2007), but particle shape estimates can still be biased by horizontal winds (Huang et al., 2015; Helms et al., 2022). The 2DVD’s resolution of approx. 190 µm per pixel (px) and the lack of grayscale information prohibits resolving\n45\nfine-scale details of snow particles. To get high resolution images, a group of instruments uses various approaches to obtain particle images with microscopic\nresolution at the expense of the measurement volume size. For example, the MASC (Multi-Angle Snowfall Camera, Garrett\net al., 2012) takes three high resolution (30 µm px−1) images of the same particle from different angles. This allows for resolv- 2022). The 2DVD’s resolution of approx. 190 µm per pixel (px) and the lack of grayscale information prohibits resolving\n45\nfine-scale details of snow particles. To get high resolution images, a group of instruments uses various approaches to obtain particle images with microscopic\nresolution at the expense of the measurement volume size. For example, the MASC (Multi-Angle Snowfall Camera, Garrett\net al., 2012) takes three high resolution (30 µm px−1) images of the same particle from different angles. This allows for resolv- ing very fine particle structures, but during a snowfall event Gergely and Garrett (2016) observed only 102 - 104 particles which\n50\nis not sufficient to reliably estimate a PSD on minute temporal scales needed to capture changes in precipitation properties. (Del Guasta, 2022) have developed a flatbed scanner (ICE-CAMERA) that has a resolution of 7 µm px−1 and can provide\nmass estimates by melting the particles, but this approach only works at low snowfall rates. The images of the D-ICI (Dual Ice\nCrystal Imager, Kuhn and Vázquez-Martín, 2020) have even a resolution of 4 µm px−1 and show particles from two perspec- tives, but similar to the MASC, the small sampling volume does not allow for the measurement of PSDs with a sufficiently\n55\nhigh accuracy. The SVI (Snowfall Video Imager, Newman et al., 2009) and its successor the PIP (Precipitation Imaging Package, Pettersen\net al., 2020) use a camera pointed to a light source to image snow particles in free fall. The open design limits wind field tives, but similar to the MASC, the small sampling volume does not allow for the measurement of PSDs with a sufficiently\n55\nhigh accuracy. (Tiira et al., 2016; von Lerber et al., 2017; Pettersen et al., 2020; Tokay et al., 2021; Leinonen et al., 2021; Vázquez-Martín\net al., 2021a). Various attempts have been made to classify particle types and identify active snowfall formation processes using Ground-based in situ observations of ice and snow particles can identify the fingerprints of the snowfall formation processes and\n25\nprovide detailed information on particle size, shape, and fall velocity. Using assumptions about fall velocity or an aggregation\nand riming model as a reference, the particle mass-size and/or density relationship can also be inferred from in situ observations. (Tiira et al., 2016; von Lerber et al., 2017; Pettersen et al., 2020; Tokay et al., 2021; Leinonen et al., 2021; Vázquez-Martín\net al., 2021a). Various attempts have been made to classify particle types and identify active snowfall formation processes using various machine learning techniques (Nurzy´nska et al., 2013; Grazioli et al., 2014; Praz et al., 2017; Hicks and Notaroš, 2019;\n30\nLeinonen and Berne, 2020; Del Guasta, 2022); these classifications are needed to support quantification of snowfall formation\nprocesses (Grazioli et al., 2017; Moisseev et al., 2017; Dunnavan et al., 2019; Pasquier et al., 2023). In situ observations have\nalso been used to characterize particle size distributions (Kulie et al., 2021; Fitch and Garrett, 2022), investigate sedimentation\nvelocity and turbulence of hydrometeors (Garrett et al., 2012; Garrett and Yuter, 2014; Li et al., 2021; Vázquez-Martín et al., 2021b; Takami et al., 2022), and for model evaluation (Vignon et al., 2019). In combination with ground-based remote sensing,\n35\nin situ snowfall data have been used to validate or better understand remote sensing observations (Gergely and Garrett, 2016;\nLi et al., 2018; Matrosov et al., 2020; Luke et al., 2021), to develop joint radar in situ retrievals (Cooper et al., 2017, 2022),\nand to train remote sensing retrievals (Huang et al., 2015; Vogl et al., 2022). Different design concepts have been used for in situ snowfall instruments. Line scan cameras are commonly used by optical 2021b; Takami et al., 2022), and for model evaluation (Vignon et al., 2019). In combination with ground-based remote sensing,\n35\nin situ snowfall data have been used to validate or better understand remote sensing observations (Gergely and Garrett, 2016;\nLi et al., 2018; Matrosov et al., 2020; Luke et al., 2021), to develop joint radar in situ retrievals (Cooper et al., 2017, 2022),\nand to train remote sensing retrievals (Huang et al., 2015; Vogl et al., 2022). Different design concepts have been used for in situ snowfall instruments. Line scan cameras are commonly used by optical Different design concepts have been used for in situ snowfall instruments. Line scan cameras are commonly used by optical\ndisdrometers such as the OTT Parsivel (Löffler-Mang and Joss, 2000) and their relatively large observation volume reduces the\n40\nstatistical uncertainty for estimating the particle size distribution (PSD). However, additional assumptions are required to size\nirregularly shaped particles such as snow particles correctly due to the one-dimensional measurement concept (Battaglia et al.,\n2010). This limitation can be overcome when adding a second line camera as for the 2DVD (2-dimensional video disdrometer,\nSchönhuber et al., 2007), but particle shape estimates can still be biased by horizontal winds (Huang et al., 2015; Helms et al., Different design concepts have been used for in situ snowfall instruments. Line scan cameras are commonly used by optical\ndisdrometers such as the OTT Parsivel (Löffler-Mang and Joss, 2000) and their relatively large observation volume reduces the\n40\nstatistical uncertainty for estimating the particle size distribution (PSD). However, additional assumptions are required to size\nirregularly shaped particles such as snow particles correctly due to the one-dimensional measurement concept (Battaglia et al.,\n2010). This limitation can be overcome when adding a second line camera as for the 2DVD (2-dimensional video disdrometer,\nSchönhuber et al., 2007), but particle shape estimates can still be biased by horizontal winds (Huang et al., 2015; Helms et al., disdrometers such as the OTT Parsivel (Löffler-Mang and Joss, 2000) and their relatively large observation volume reduces the\n40\nstatistical uncertainty for estimating the particle size distribution (PSD). However, additional assumptions are required to size\nirregularly shaped particles such as snow particles correctly due to the one-dimensional measurement concept (Battaglia et al.,\n2010). After MOSAiC, the original VISSS was deployed at Hyytiälä, Finland (Petäjä et al.,\n2016) in 2021/22 and at Gothic, Colorado as part of the SAIL campaign (Surface Atmosphere Integrated Field Laboratory, Feldman et al., 2021). During a test setup in Leipzig, Germany, the VISSS was used to evaluate a radar-based riming retrieval\n80\n(Vogl et al., 2022). An improved second generation of VISSS was installed at the French-German Arctic research base AW-\nIPEV (the Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research - AWI - and the French Polar Institute\nPaul Emile Victor - PEV) in Ny-Ålesund, Svalbard (Nomokonova et al., 2019) in 2021. A further improved third generation\nVISSS is currently being built at the Leipzig University. The VISSS hardware plans and software libraries have been released Feldman et al., 2021). During a test setup in Leipzig, Germany, the VISSS was used to evaluate a radar-based riming retrieval\n80\n(Vogl et al., 2022). An improved second generation of VISSS was installed at the French-German Arctic research base AW-\nIPEV (the Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research - AWI - and the French Polar Institute\nPaul Emile Victor - PEV) in Ny-Ålesund, Svalbard (Nomokonova et al., 2019) in 2021. A further improved third generation\nVISSS is currently being built at the Leipzig University. The VISSS hardware plans and software libraries have been released under an open source license (Maahn et al., 2023; Maahn, 2023a, b) so that the community can replicate and further develop\n85\nVISSS. The VISSS hardware design and data processing are described in Sects. 2 and 3, respectively. Example cases including\na comparison with the PIP are given in Sect. 4 and concluding remarks are given in Sect. 5. under an open source license (Maahn et al., 2023; Maahn, 2023a, b) so that the community can replicate and further develop\n85\nVISSS. The VISSS hardware design and data processing are described in Sects. 2 and 3, respectively. Example cases including\na comparison with the PIP are given in Sect. 4 and concluding remarks are given in Sect. 5. The SVI (Snowfall Video Imager, Newman et al., 2009) and its successor the PIP (Precipitation Imaging Package, Pettersen\net al., 2020) use a camera pointed to a light source to image snow particles in free fall. The open design limits wind field 2 2 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. perturbations and the large measurement volume (4.8 x 6.4 x 5.5 cm for a 1 mm snow particle) limits statistical errors in\nderiving the PSD. However, the resolution of 100 µm px−1 is not sufficient to study fine details. Further, the open design\n60\nrequires that the depth of the observation volume is not constrained by the instrument itself. As a consequence, particle blur\nneeds to be used to determine whether a particle is in the observation volume or not which is potentially more error prone\nthan a closed instrument design. A similar design was used by Testik and Rahman (2016) to study the sphericity oscillations\nof raindrops. Kennedy et al. (2022) developed the low-cost OSCRE (Open Snowflake Camera for Research and Education)\nsystem that uses a strobe light to illuminate particles from the side allowing for the observation of particle type of blowing and\n65\nprecipitating snow but the observation volume is not fully constrained. 60 This study presents the Video In Situ Snowfall Sensor (VISSS). The goal was to develop a sensor with an open instrument\ndesign without sacrificing the quality of measurement volume definition or optical resolution. It uses the same general principle\nas the PIP (Fig. 1): grayscale images of particles in free fall illuminated by a background light. Unlike the PIP, this setup is gi\ng\nq\nyi\np\ng\np\np\nas the PIP (Fig. 1): grayscale images of particles in free fall illuminated by a background light. Unlike the PIP, this setup is\nduplicated with overlapping measurement volumes so that particles are observed simultaneously from two perspectives at a\n70\n90° angle. This robustly constrains the observation volume without the need for further assumptions. In addition, having two\nperspectives of the same particle increases the likelihood that the observed maximum dimension (Dmax) and aspect ratio are\nrepresentative of the particle. While the VISSS does not reach the microscopic resolution of the D-ICI or ICE-CAMERA, its\nresolution of 43 to 59 µm px−1 is significantly better than the PIP, and the use of telecentric lenses eliminates sizing errors duplicated with overlapping measurement volumes so that particles are observed simultaneously from two perspectives at a\n70\n90° angle. This robustly constrains the observation volume without the need for further assumptions. In addition, having two\nperspectives of the same particle increases the likelihood that the observed maximum dimension (Dmax) and aspect ratio are\nrepresentative of the particle. While the VISSS does not reach the microscopic resolution of the D-ICI or ICE-CAMERA, its\nresolution of 43 to 59 µm px−1 is significantly better than the PIP, and the use of telecentric lenses eliminates sizing errors duplicated with overlapping measurement volumes so that particles are observed simultaneously from two perspectives at a\n70\n90° angle. This robustly constrains the observation volume without the need for further assumptions. In addition, having two\nperspectives of the same particle increases the likelihood that the observed maximum dimension (Dmax) and aspect ratio are\nrepresentative of the particle. While the VISSS does not reach the microscopic resolution of the D-ICI or ICE-CAMERA, its\nresolution of 43 to 59 µm px−1 is significantly better than the PIP, and the use of telecentric lenses eliminates sizing errors caused by the variable distance of snow particles to the cameras. 75\nThe VISSS was originally developed for the MOSAiC (Multidisciplinary drifting Observatory for the Study of Arctic Cli-\nmate) experiment (Shupe et al., 2022) and deployed at MetCity and, after the sea ice became too unstable in April 2020, on the\nP-deck of the research vessel Polarstern. After MOSAiC, the original VISSS was deployed at Hyytiälä, Finland (Petäjä et al.,\n2016) in 2021/22 and at Gothic, Colorado as part of the SAIL campaign (Surface Atmosphere Integrated Field Laboratory, caused by the variable distance of snow particles to the cameras. 75\nThe VISSS was originally developed for the MOSAiC (Multidisciplinary drifting Observatory for the Study of Arctic Cli-\nmate) experiment (Shupe et al., 2022) and deployed at MetCity and, after the sea ice became too unstable in April 2020, on the\nP-deck of the research vessel Polarstern. After MOSAiC, the original VISSS was deployed at Hyytiälä, Finland (Petäjä et al.,\n2016) in 2021/22 and at Gothic, Colorado as part of the SAIL campaign (Surface Atmosphere Integrated Field Laboratory, caused by the variable distance of snow particles to the cameras. 75\nThe VISSS was originally developed for the MOSAiC (Multidisciplinary drifting Observatory for the Study of Arctic Cli-\nmate) experiment (Shupe et al., 2022) and deployed at MetCity and, after the sea ice became too unstable in April 2020, on the\nP-deck of the research vessel Polarstern. It\nalso does not allow for sufficiently large roofs over the camera windows to protect against snow accumulation in all weather conditions. This problem was partially solved by the increased budget (22kEUR) for the second generation VISSS2, which\n110\nused a 600 mm working distance lens and a camera with an increased frame rate of 250 Hz, resulting in a resolution of 43.125\nµm px−1. However, the optical quality of the lens proved to be borderline for the applications, resulting in slightly blurred\nparticle images, so the lens was changed again for the third generation VISSS3 (currently under construction), which also has\na working distance of 1300 mm. This was motivated by the result of (Newman et al., 2009) that the air flow is undisturbed at a distance of 1 m from the instrument. The lens-camera combinations and backlights are housed in waterproof enclosures that\n115\nare heated to -5°C and 10°C, respectively. The low temperature in the camera housing is to prevent melting and refreezing of\nparticles on the camera window. The cameras of VISSS1 and VISSS2 are connected to the data acquisition systems via separate 1 Gbit and 5 Gbit Ethernet\nconnections, respectively. Due to the increased frame rate, two separate systems are required to record data in rea- time for\nVISSS2. 120 The cameras of VISSS1 and VISSS2 are connected to the data acquisition systems via separate 1 Gbit and 5 Gbit Ethernet\nconnections, respectively. Due to the increased frame rate, two separate systems are required to record data in rea- time for\nVISSS2. 120 q\ny\np\nconnections, respectively. Due to the increased frame rate, two separate systems are required to record data in rea- time for\nVISSS2. 120 VISSS2. 120 VISSS2. 120 2\nInstrument design The VISSS consists of two camera systems oriented at a 90° angle to the same measurement volume (Fig. 1). Both cameras\nhave 1280x1024 grayscale pixels and operate at a frame rate of 140 Hz (250 Hz since the 2nd generation). One camera acts\n90\nas the leader, sending trigger signals to both the follower camera and the two LED backlights that illuminate the scenes from The VISSS consists of two camera systems oriented at a 90° angle to the same measurement volume (Fig. 1). Both cameras have 1280x1024 grayscale pixels and operate at a frame rate of 140 Hz (250 Hz since the 2nd generation). One camera acts\n90\nas the leader, sending trigger signals to both the follower camera and the two LED backlights that illuminate the scenes from 3 3 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. behind with 350,000 lux. Green backlights (530 nm) were chosen because the camera and lenses are optimized for visual light. The leader-follower setup results in a slight delay in the start of exposure between the two cameras. To compensate for this,\nthe background LEDs are turned on for a duration of 60 µs only when the exposure of both cameras is active. Thus, the 60\nµs flash of the backlights determines the effective exposure time of the camera as long as there is no bright sunlight, which\n95\nis a rare condition during precipitation. The two camera-lens-backlight combinations are at a 90° angle so that particles are\nobserved from two perspectives, reducing sizing errors. Leinonen et al. (2021) found that using only a single perspective for\nsizing snow particles can lead to a normalized root mean square error of 6% for Dmax and Wood et al. (2013) estimated the\nresulting bias in simulated radar reflectivity to be 3.2 dB. For the VISSS, the accuracy of the measurements can potentially be behind with 350,000 lux. Green backlights (530 nm) were chosen because the camera and lenses are optimized for visual light. The leader-follower setup results in a slight delay in the start of exposure between the two cameras. To compensate for this,\nthe background LEDs are turned on for a duration of 60 µs only when the exposure of both cameras is active. Thus, the 60\nµs flash of the backlights determines the effective exposure time of the camera as long as there is no bright sunlight, which\n95\nis a rare condition during precipitation. The two camera-lens-backlight combinations are at a 90° angle so that particles are\nobserved from two perspectives, reducing sizing errors. Leinonen et al. (2021) found that using only a single perspective for\nsizing snow particles can lead to a normalized root mean square error of 6% for Dmax and Wood et al. (2013) estimated the\nresulting bias in simulated radar reflectivity to be 3.2 dB. For the VISSS, the accuracy of the measurements can potentially be further improved by taking advantage of the fact that the VISSS typically observes 8 to 11 frames of each particle (assuming\n100\na fall velocity of 1 m s−1 and a frame rate of 140 to 250 Hz), and additional perspectives can be obtained from the natural\ntumbling of the particle. Telecentric lenses have a constant magnification within the usable depth of field, eliminating sizing errors. They also typically\nhave a greater depth of field than standard lenses. The disadvantage is that the lens aperture must be as large as the observation further improved by taking advantage of the fact that the VISSS typically observes 8 to 11 frames of each particle (assuming\n100\na fall velocity of 1 m s−1 and a frame rate of 140 to 250 Hz), and additional perspectives can be obtained from the natural\ntumbling of the particle. Telecentric lenses have a constant magnification within the usable depth of field, eliminating sizing errors. They also typically\nhave a greater depth of field than standard lenses. The disadvantage is that the lens aperture must be as large as the observation area, making the lens bulky, heavy and expensive. For the first VISSS (VISSS1), a lens with a magnification of 0.08 was\n105\nchosen, resulting in a pixel resolution of 58.75 µm px−1 (Table 1). The working distance, i.e. the distance from the edge of\nthe lens to the center of the observation volume, is 227 mm. This partly undermines the goal of having an instrument with an\nobservation volume that is not obstructed by turbulence induced by nearby structures, but was caused by budget limitations. It\nalso does not allow for sufficiently large roofs over the camera windows to protect against snow accumulation in all weather area, making the lens bulky, heavy and expensive. For the first VISSS (VISSS1), a lens with a magnification of 0.08 was\n105\nchosen, resulting in a pixel resolution of 58.75 µm px−1 (Table 1). The working distance, i.e. the distance from the edge of\nthe lens to the center of the observation volume, is 227 mm. This partly undermines the goal of having an instrument with an\nobservation volume that is not obstructed by turbulence induced by nearby structures, but was caused by budget limitations. 3\nData processing The cameras transmit every captured image to the data acquisition systems which are standard desktop computers running\nLinux. Based on simple brightness changes, the computers save only moving images and discard all other data (this was not\nimplemented for MOSAiC yet). The raw data of the VISSS consists of the video files (mov or mkv video files with h264 4 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 10 mm\na)\nb)\nc)\nLeader\nFollower\nLED backlight\nLED backlight\nx\ny\nz\nOfz\nθ\nφ\nfollower \ntransformation\ncoordinate \nsystem\nobservation\nvolume 10 mm\na)\nb)\nc)\nLeader\nFollower\nLED backlight\nLED backlight\nx\ny\nz\nOfz\nθ\nφ\nfollower \ntransformation\ncoordinate \nsystem\nobservation\nvolume\nFigure 1. a) Concept drawing of the VISSS (not to scale with enlarged observation volume). See Sections 3.2 and 3.3 for a discussion of\nthe joint coordinate system and the transformation of the follower’s coordinate system, respectively. b) VISSS deployed at Gothic, Colorado\nduring the SAIL campaign (Photo by Benn Schmatz), c) Randomly selected particles observed during MOSAiC on 15 November 2019\nbetween 6:53 and 11:13 UTC. a)\nLeader\nFollower\nLED backlight\nLED backlight\nx\ny\nz\nOfz\nθ\nφ\nfollower \ntransformation\ncoordinate \nsystem\nobservation\nvolume a) 10 mm\nb)\nc)\ng c) Figure 1. a) Concept drawing of the VISSS (not to scale with enlarged observation volume). See Sections 3.2 and 3.3 for a discussion of\nthe joint coordinate system and the transformation of the follower’s coordinate system, respectively. b) VISSS deployed at Gothic, Colorado\nduring the SAIL campaign (Photo by Benn Schmatz), c) Randomly selected particles observed during MOSAiC on 15 November 2019\nbetween 6:53 and 11:13 UTC. compression), the first recorded frame as an image (jpg format) for quick evaluation of camera blocking, and a csv file with the\n125\ntimestamps of the camera (capture_time) as well as the computer (record_time) and other meta information for each frame. The\ncameras run continuously and new files are created every 10 minutes (5 minutes for MOSAiC). In addition, a daily status csv\nfile is maintained that contains information about software start and stop times and when new files were created. Both cameras\nrecord completely separately which requires an accurate synchronization of the camera and computer clocks for matching the observations of a single particle. 3\nData processing 130\nObtaining particle properties from the individual VISSS video images requires (1) detecting the particles, (2) matching the\nobservations of the two cameras, and (3) tracking the particles over multiple frames to estimate the fall velocities. These three\nprocessing steps comprise the level1 products, which contain uncalibrated properties for each observed particle. For the level2\nproduct, the level1 observations are calibrated and distributions of particle size, aspect ratio, and other properties are estimated observations of a single particle. 130\nObtaining particle properties from the individual VISSS video images requires (1) detecting the particles, (2) matching the\nobservations of the two cameras, and (3) tracking the particles over multiple frames to estimate the fall velocities. These three\nprocessing steps comprise the level1 products, which contain uncalibrated properties for each observed particle. For the level2\nproduct, the level1 observations are calibrated and distributions of particle size, aspect ratio, and other properties are estimated based on the per-particle properties. In addition to the level1 and level2 products, there are metadata products: metaEvents is\n135\na netcdf version of the status files along with a camera blocking estimate based on the jpg images. metaFrames is a netcdf based on the per-particle properties. In addition to the level1 and level2 products, there are metadata products: metaEvents is\n135\na netcdf version of the status files along with a camera blocking estimate based on the jpg images. metaFrames is a netcdf 5 5 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Table 1. Technical specifications of the three VISSS instruments. Table 1. Technical specifications of the three VISSS instruments. VISSS1\nVISSS2\nVISSS3\nResolution [µm px−1]\n58.75\n43.125\n46,0\nObs. volume (w x d x h)\n[mm]\n75.2 x 60.1 x 60.1\n55.2 x 44.2 x 44.2\n58.9 x 47.1 x 47.1\nUsed frame size [px]\n1280 x 1024\n1280 x 1024\n1280 x 1024\nFrame rate [Hz]\n140\n250\n250\nExposure time [µs]\n60\n60\n60\nWorking distance [mm]\n227 mm\n600 mm\n1300 mm\nCamera\nTeledyne\nGenie\nNano\nM1280\nMono\nTeledyne Genie Nano 5G M2050\nMono\nTeledyne Genie Nano 5G M2050\nMono\nLens\nOpto Engineering TC12080\nSill S5LPJ1235 (modified working\ndistance)\nSill S5LPJ1725 (modified working\ndistance)\nMaker\nUniversity of Colorado Boulder\nUniversity of Cologne\nLeipzig University\nDeployments\nMOSAiC\n2019/20,\nHyytiälä\n2021/22, SAIL 2022/23\nNy-Ålesund since 2021\nHyytiälä (planned)\nVISSS data \nacquisition\nStatus Files\nCSV Files\nImage Files\nMovie Files\nmetaEvents\nmetaFrames\nmetaRotation\nlevel1detect\nlevel1match\nlevel1track\nlevel2match\nlevel2track\nmeta data\nlevel 1: per \nparticle properties\nlevel 2: time \naveraged properties\nimagesL1detect\nstatus:\nfile frequency:\nfinished\nin development\n10 min\ndaily\nlegend\nlevel 0 raw data\nboth cameras processed separately\nboth cameras\nprocessed jointly\nFigure 2. Flowchart of VISSS data processing. Daily products have rounded corners, 10-minute resolution products have square corners. Completed and under development products are indicated by solid and dashed boxes, respectively. VISSS data \nacquisition both cameras processed separately Image Files Figure 2. Flowchart of VISSS data processing. Daily products have rounded corners, 10-minute resolution products have square corners. Completed and under development products are indicated by solid and dashed boxes, respectively. version of the csv file. metaRotation keeps track of the camera alignment as detailed below. The imagesL1detect product\ncontains images of the detected particles which is required for creating quicklooks like Fig. 1.c. version of the csv file. metaRotation keeps track of the camera alignment as detailed below. The imagesL1detect product\ncontains images of the detected particles which is required for creating quicklooks like Fig. 1.c. version of the csv file. metaRotation keeps track of the camera alignment as detailed below. T\ncontains images of the detected particles which is required for creating quicklooks like Fig. 1.c. rsion of the csv file. metaRotation keeps track of the camera alignment as detailed below. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Figure 3. Estimation of particle contour (cyan), maximum dimension Dmax (via smallest enclosing circle, magenta), smallest rectangle\n(red), region of interest (green), and elliptical fits using openCV’s fitEllipseDirect (white) and fitEllipse functions (blue). The particles were\nobserved during MOSAiC on 15 November 2019 05:25 UTC except the particle on the right (Hyytiälä 23 January 2022 04:10 UTC). Figure 3. Estimation of particle contour (cyan), maximum dimension Dmax (via smallest enclosing circle, magenta), smallest rectangle\n(red), region of interest (green), and elliptical fits using openCV’s fitEllipseDirect (white) and fitEllipse functions (blue). The particles were\nobserved during MOSAiC on 15 November 2019 05:25 UTC except the particle on the right (Hyytiälä 23 January 2022 04:10 UTC). 3.1\nParticle Detection\n140 AR and α are estimated in three different ways, from the smallest rectangle fitted around the contour\n(minAreaRect) or from an ellipse fitted to the contour (fitEllipse and the more stable fitEllipseDirect). Particle complexity c\n(Garrett et al., 2012; Gergely et al., 2017) is derived from the ratio between particle perimeter p to the perimeter of a sphere\nwith same area A\n160 (Zeevi, 2016) in the image coordinate system (horizontal dimension X, vertical dimension Y pointing to the ground). This\n145\nroutine is faster than commonly used background detection algorithms, but still works well with the—relatively simple—\ndetection problem of VISSS. The ROI identified by the background subtraction methods cannot be used directly for particle\nsizing because it contains a few blurred pixels around the particle that would introduce a bias. Therefore, we select a 10\npixel padded box around the ROI and use openCV’s Canny filter (after applying a Gaussian blur with a standard deviation (Zeevi, 2016) in the image coordinate system (horizontal dimension X, vertical dimension Y pointing to the ground). This\n145\nroutine is faster than commonly used background detection algorithms, but still works well with the—relatively simple—\ndetection problem of VISSS. The ROI identified by the background subtraction methods cannot be used directly for particle\nsizing because it contains a few blurred pixels around the particle that would introduce a bias. Therefore, we select a 10\npixel padded box around the ROI and use openCV’s Canny filter (after applying a Gaussian blur with a standard deviation of 1.5 pixels) to identify the edges of the particles. To fill in small gaps in the contour, we use dilate contour by 1 pixel,\n150\nfill the contour, erode by 1 pixel, and identify the new contour. Since filling the contour also closes potential holes in the\nparticles, the background detection and Canny filter masks are combined. As a result, VISSS can detect even relatively small\nparticle structures, as shown in Fig. 3. The use of only 1 pixel (i.e., 43 to 59 µm) for dilation was found to be sufficient and\nallows to potentially resolve more details of the particles than MASC and PIP, which dilate by 200 µm (Garrett et al., 2012) of 1.5 pixels) to identify the edges of the particles. The\nntains images of the detected particles which is required for creating quicklooks like Fig. 1.c. images of the detected particles which is required for creating quicklooks like Fig. 1.c. In the following, the processing of the level1 and level2 products is described in detail (Fig. 2). In the following, the processing of the level1 and level2 products is described in detail (Fig. 2). 6 Figure 3. Estimation of particle contour (cyan), maximum dimension Dmax (via smallest enclosing circle, magenta), smallest rectangle\n(red), region of interest (green), and elliptical fits using openCV’s fitEllipseDirect (white) and fitEllipse functions (blue). The particles were\nobserved during MOSAiC on 15 November 2019 05:25 UTC except the particle on the right (Hyytiälä 23 January 2022 04:10 UTC). https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 3.1\nParticle Detection\n140 Hydrometeors need to be detected and sized based on individual frames. First, video frames containing motion are identified\nby a simple threshold-based filter. Except for the MOSAiC dataset, this is done in real-time, which significantly reduces the\ndata volume. Because snow may stick to the camera window, individual particles within a video frame cannot be identified\nby image brightness. Instead, the moving region of interest (ROI) is identified by openCV’s BackgroundSubtractorCNT class Hydrometeors need to be detected and sized based on individual frames. First, video frames containing motion are identified\nby a simple threshold-based filter. Except for the MOSAiC dataset, this is done in real-time, which significantly reduces the\ndata volume. Because snow may stick to the camera window, individual particles within a video frame cannot be identified\nby image brightness. Instead, the moving region of interest (ROI) is identified by openCV’s BackgroundSubtractorCNT class\n(Zeevi, 2016) in the image coordinate system (horizontal dimension X, vertical dimension Y pointing to the ground). This\n145\nroutine is faster than commonly used background detection algorithms, but still works well with the—relatively simple—\ndetection problem of VISSS. The ROI identified by the background subtraction methods cannot be used directly for particle\nsizing because it contains a few blurred pixels around the particle that would introduce a bias. Therefore, we select a 10\npixel padded box around the ROI and use openCV’s Canny filter (after applying a Gaussian blur with a standard deviation\nof 1.5 pixels) to identify the edges of the particles. To fill in small gaps in the contour, we use dilate contour by 1 pixel,\n150\nfill the contour, erode by 1 pixel, and identify the new contour. Since filling the contour also closes potential holes in the\nparticles, the background detection and Canny filter masks are combined. As a result, VISSS can detect even relatively small\nparticle structures, as shown in Fig. 3. The use of only 1 pixel (i.e., 43 to 59 µm) for dilation was found to be sufficient and\nallows to potentially resolve more details of the particles than MASC and PIP, which dilate by 200 µm (Garrett et al., 2012)\nand 300 µm (Helms et al., 2022), respectively. The particle contours are used to estimate the particle’s maximum dimension\n155\n(using openCV’s minEnclosingCircle function), perimeter p (arcLength), area A (contourArea) and aspect ratio AR, as well\nas the canting angle α. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. the brightness of the darkest pixel must be at least 20 steps darker than the median of the entire image. Particle detection is the\n165\nmost computationally intensive processing step and is typically performed on a small cluster. Processing 10 minutes of heavy\nsnowfall for a single VISSS camera can easily take several hours on a single AMD EPYC 7302 core. the brightness of the darkest pixel must be at least 20 steps darker than the median of the entire image. Particle detection is the\n165\nmost computationally intensive processing step and is typically performed on a small cluster. Processing 10 minutes of heavy\nsnowfall for a single VISSS camera can easily take several hours on a single AMD EPYC 7302 core. 3.1\nParticle Detection\n140 To fill in small gaps in the contour, we use dilate contour by 1 pixel,\n150\nfill the contour, erode by 1 pixel, and identify the new contour. Since filling the contour also closes potential holes in the\nparticles, the background detection and Canny filter masks are combined. As a result, VISSS can detect even relatively small\nparticle structures, as shown in Fig. 3. The use of only 1 pixel (i.e., 43 to 59 µm) for dilation was found to be sufficient and\nallows to potentially resolve more details of the particles than MASC and PIP, which dilate by 200 µm (Garrett et al., 2012) and 300 µm (Helms et al., 2022), respectively. The particle contours are used to estimate the particle’s maximum dimension\n155\n(using openCV’s minEnclosingCircle function), perimeter p (arcLength), area A (contourArea) and aspect ratio AR, as well\nas the canting angle α. AR and α are estimated in three different ways, from the smallest rectangle fitted around the contour\n(minAreaRect) or from an ellipse fitted to the contour (fitEllipse and the more stable fitEllipseDirect). Particle complexity c\n(Garrett et al., 2012; Gergely et al., 2017) is derived from the ratio between particle perimeter p to the perimeter of a sphere with same area A\n160 (1) In addition to these size variables, we store variables describing the pixel brightness (min, max, standard deviation, mean,\nskewness), the position of the centroid, and the blur of the particle estimated from the variance of the Laplacian of the ROI. All\nparticles are processed for which Dmax ≥2 px and A ≥2 px holds. To avoid detection of particles completely out of focus, 7 7 3.2\nParticle Matching This process requires matching the time stamps (\"capture time\") of both cameras. The follower camera’s clock can be\noff by more than 1 frame per 10 minutes. The time assigned by the computers (\"recording time\") is sometimes, but not the two cameras follows a normal distribution with mean zero and standard deviation 1.7 px (1.2 px), based on an analysis\n180\nof manually matched particle pairs. The minimum resolution of 1 pixel is accounted for by integrating the probability density\nfunction (PDF) for an interval of +/- 0.5 pixels. This process requires matching the time stamps (\"capture time\") of both cameras. The follower camera’s clock can be\noff by more than 1 frame per 10 minutes. The time assigned by the computers (\"recording time\") is sometimes, but not always, distorted by computer load. Therefore, the continuous frame index (capture id) is used for matching, but this requires\n185\ndetermining the index offset between both cameras. This takes advantage of the fact that only moving frames are recorded. If\nparticles are present in the joint observation volume, both cameras will record a frame. Therefore, for a subset of 500 leader\nframes, pairs of frames less than 1 ms apart in recording time are identified and the most common capture id offset is used. Similar to h and z, the capture id offset ∆i is used as the mean of a normal distribution with a standard deviation value of always, distorted by computer load. Therefore, the continuous frame index (capture id) is used for matching, but this requires\n185\ndetermining the index offset between both cameras. This takes advantage of the fact that only moving frames are recorded. If\nparticles are present in the joint observation volume, both cameras will record a frame. Therefore, for a subset of 500 leader\nframes, pairs of frames less than 1 ms apart in recording time are identified and the most common capture id offset is used. Similar to h and z, the capture id offset ∆i is used as the mean of a normal distribution with a standard deviation value of 0.01, which ensures that only particles observed at the same time are matched. During MOSAiC, the data acquisition computer\n190\nCPUs turned out to be too slow to keep up with processing during heavy snowfall. 3.2\nParticle Matching The particle detection of each camera is completely separate, so the particles observed b The particle detection of each camera is completely separate, so the particles observed by each camera must be combined. The particle detection of each camera is completely separate, so the particles observed by each camera must be combined. This particle combination allows for the particle position to be determined in a three-dimensional reference coordinate system. 170\nAs a side effect, this constrains the observation volume by discarding particles observed by only one camera. We use a right-\nhanded reference coordinate system (x,y,z) with z pointing to the ground to define the position of particles in the observation\nvolume (Fig. 1). In the absence of an absolute reference, we attach the coordinate system to the leader camera (i.e., (xL,yL,zL)\n= (x,y,z)) such that x = XL and z = YL, where XL and YL are the particle positions in the two dimensional leader images. This particle combination allows for the particle position to be determined in a three-dimensional reference coordinate system. 170\nAs a side effect, this constrains the observation volume by discarding particles observed by only one camera. We use a right-\nhanded reference coordinate system (x,y,z) with z pointing to the ground to define the position of particles in the observation\nvolume (Fig. 1). In the absence of an absolute reference, we attach the coordinate system to the leader camera (i.e., (xL,yL,zL)\n= (x,y,z)) such that x = XL and z = YL, where XL and YL are the particle positions in the two dimensional leader images. The missing dimension y is obtained from the follower camera with y = −XF where XF the vertical position in the follower\n175\nimage. The matching of the particles from both cameras is based on the comparison of two variables: The vertical position of the\nparticles and their vertical extent. Due to measurement uncertainties, the agreement of these variables cannot be perfect and\nthey are treated probabilistically. That is, it is assumed that the difference in vertical extent ∆h (vertical position ∆z) between the two cameras follows a normal distribution with mean zero and standard deviation 1.7 px (1.2 px), based on an analysis\n180\nof manually matched particle pairs. The minimum resolution of 1 pixel is accounted for by integrating the probability density\nfunction (PDF) for an interval of +/- 0.5 pixels. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. For each particle, its three-dimensional position is provided and all per-particle variables from the detection are carried\nforward to the matched particle product. The ratio of matched to observed particles from a single camera varies with the\naverage particle size, since larger particles can be identified even when they are out of focus, and varies between approximately\n200\n10% and 90%. 3.2\nParticle Matching With the additional impact of a bug in the\ndata acquisition code and drifting computer clocks when the network connection to the ship’s reference clock were interrupted,\nthe particle matching for the MOSAiC data set often requires manual adjustment. Th j i t\nd\nt f th i t\nt d PDF i t\nl d i\nd f\n∆h ∆\nd ∆i i\nid\nd\nt h\nhi h d\nib 0.01, which ensures that only particles observed at the same time are matched. During MOSAiC, the data acquisition computer\n190\nCPUs turned out to be too slow to keep up with processing during heavy snowfall. With the additional impact of a bug in the\ndata acquisition code and drifting computer clocks when the network connection to the ship’s reference clock were interrupted,\nthe particle matching for the MOSAiC data set often requires manual adjustment. The joint product of the integrated PDF intervals derived from ∆h, ∆z, and ∆i is considered a match score, which describes\nthe quality of the particle match. Manual inspection revealed that the number of false matches increases strongly for match\n195\nscores less than 0.001, which is used as a cut-off criterion. Assuming that the probabilities for ∆h and ∆y are correctly\ndetermined, this implies that 0.1% of particle matches are falsely rejected, resulting in a negligible bias. the quality of the particle match. Manual inspection revealed that the number of false matches increases strongly for match\n195\nscores less than 0.001, which is used as a cut-off criterion. Assuming that the probabilities for ∆h and ∆y are correctly\ndetermined, this implies that 0.1% of particle matches are falsely rejected, resulting in a negligible bias. 8 3.3\nCorrection for camera alignment Although vertical alignment of both observation volumes is a priority during installation, the cameras can be rotated or dis-\nplaced. As a result, the same particle may be observed at different heights and z = YL = YF does not hold. The observed\noffsets are not constant and can change due to wind load or pressure of accumulated snow on the VISSS frame. We could\n205\nsimply ignore the rotation and continue to take z from the leader, but this would make it impossible to use the vertical position\nto match particles from both cameras (see above). Also, offsets in z reduce the common observation volume of both cameras,\nwhich could lead to biases when calibrating the PSDs if not accounted for. Besides a constant offset in the vertical z dimension Ofz, one of the cameras can also be rotated around the optical axis\n(expressed analogously to aircraft coordinate systems with roll φ), around the horizontal axis perpendicular to the optical axis\n210\n(pitch θ), or around the vertical axis (yaw ψ). As a consequence, ∆z = YL −YF depends on the position of the particle in the\nobservation volume. (expressed analogously to aircraft coordinate systems with roll φ), around the horizontal axis perpendicular to the optical axis\n210\n(pitch θ), or around the vertical axis (yaw ψ). As a consequence, ∆z = YL −YF depends on the position of the particle in the\nobservation volume. To account for the rotation, we attach the coordinate system to the reader (i.e., we assume that the leader is perfectly aligned\n(xL,yL,zL) = (x,y,z)) and retrieve the rotation of the follower with respect to the leader in terms of φ, θ and Ofz. We neglect To account for the rotation, we attach the coordinate system to the reader (i.e., we assume that the leader is perfectly aligned\n(xL,yL,zL) = (x,y,z)) and retrieve the rotation of the follower with respect to the leader in terms of φ, θ and Ofz. We neglect ψ because it is not expected to affect the matching significantly. Mathematically, we need to transform the follower coordinate\n215\nsystem (xF ,yF ,zF ) to our leader reference coordinate system (xL,yL,zL) using rotation and shear matrices. 3.3\nCorrection for camera alignment In the appendix A,\nwe show how the transformation matrices can be arranged so that the follower’s vertical measure zF can be converted to zL\ndepending on φ and θ with zL = −\nsinθ\ncosθcosψ xL\n−sinθsinψcosφ −cosψsinφ\ncosθcosψ\nyF\n+sinθsinψsinφ + cosψcosφ\ncosθcosψ\n(zF + Ofz). (2) 220 (3) (4) This equation can be considered as a forward operator that calculates the expected leader observation zL based on a rotation\nstate (Ofz, φ, and θ) and additional parameters (xL, yF , zF ). While we assume that the rotation state is constant for each\n10 minute observation period, the other variables (xL, yF , zF ) are available on a per-particle basis, combining observations\nfrom both cameras. Therefore, we can use a Bayesian inverse Optimal Estimation retrieval (Rodgers, 2000) implemented by\n225\nthe pyOptimalEstimation library (Maahn et al., 2020) to retrieve the rotation state from the actual observed zL. Since the\ndimension of the rotation state is three, this retrieval is overconstrained when solving for more than three observed particles at\na time This equation can be considered as a forward operator that calculates the expected leader observation zL based on a rotation\nstate (Ofz, φ, and θ) and additional parameters (xL, yF , zF ). While we assume that the rotation state is constant for each\n10 minute observation period, the other variables (xL, yF , zF ) are available on a per-particle basis, combining observations 10 minute observation period, the other variables (xL, yF , zF ) are available on a per-particle basis, combining observations\nfrom both cameras. Therefore, we can use a Bayesian inverse Optimal Estimation retrieval (Rodgers, 2000) implemented by\n225\nthe pyOptimalEstimation library (Maahn et al., 2020) to retrieve the rotation state from the actual observed zL. Since the\ndimension of the rotation state is three, this retrieval is overconstrained when solving for more than three observed particles at\na time. from both cameras. Therefore, we can use a Bayesian inverse Optimal Estimation retrieval (Rodgers, 2000) implemented by\n225\nthe pyOptimalEstimation library (Maahn et al., 2020) to retrieve the rotation state from the actual observed zL. Since the\ndimension of the rotation state is three, this retrieval is overconstrained when solving for more than three observed particles at\na time. 9 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 3.3\nCorrection for camera alignment In an iterative process, the retrieved values for φ, θ, and Ofz including uncertainties are used as a priori input for the next iteration of rotation retrieval\n235\nuntil the change in φ, θ, and Ofz is less than the estimated uncertainties. retrieved values for φ, θ, and Ofz including uncertainties are used as a priori input for the next iteration of rotation retrieval\n235\nuntil the change in φ, θ, and Ofz is less than the estimated uncertainties. The rotation parameters must be estimated manually after the instrument frames are set up or adjusted, but fluctuations in\ntime are automatically retrieved by the following procedure: the rotation estimates and uncertainties (inflated by a factor of\n10) estimated during the previous time step (either automatically or manually obtained) are used to match a subset of the data, The rotation parameters must be estimated manually after the instrument frames are set up or adjusted, but fluctuations in\ntime are automatically retrieved by the following procedure: the rotation estimates and uncertainties (inflated by a factor of\n10) estimated during the previous time step (either automatically or manually obtained) are used to match a subset of the data, The rotation parameters must be estimated manually after the instrument frames are set up or adjusted, but fluctuations in\ntime are automatically retrieved by the following procedure: the rotation estimates and uncertainties (inflated by a factor of\n10) estimated during the previous time step (either automatically or manually obtained) are used to match a subset of the data, estimate the current rotation parameters, and re-match the data until stable rotation parameters are obtained, as discussed above. 240\nA drawback of the method is that this processing step requires processing the 10-minute measurement chunks in chronological\norder, creating a serial bottleneck in the otherwise parallel VISSS processing chain. estimate the current rotation parameters, and re-match the data until stable rotation parameters are obtained, as discussed above. 240\nA drawback of the method is that this processing step requires processing the 10-minute measurement chunks in chronological\norder, creating a serial bottleneck in the otherwise parallel VISSS processing chain. 3.3\nCorrection for camera alignment 0\n1\n2\n3\n4\n5\nParticle diameter [mm]\n0.25\n0.50\n0.75\n1.00\n1.25\n1.50\n1.75\nVelocity [m/s]\nHyytiälä, 2021-01-05\nrimed\nunrimed\n1.00\n1.25\n1.50\n1.75\n2.00\n2.25\n2.50\n2.75\n3.00\nComplexity [-]\nFigure 4. Proof of concept of obtaining particle velocity from particle tracking for data obtained in Hyytiälä on 5 January 2022 00:00-14:30\nUTC. The velocity parameterizations of Lumb (1961) (found in Brandes et al., 2008) for unrimed and rimed aggregates are indicated by the\nsolid and dashed lines, respectively. Figure 4. Proof of concept of obtaining particle velocity from particle tracking for data obtained in Hyytiälä on 5 January 2022 00:00-14:30\nUTC. The velocity parameterizations of Lumb (1961) (found in Brandes et al., 2008) for unrimed and rimed aggregates are indicated by the\nsolid and dashed lines, respectively. The retrieved rotation parameters are required for matching, but retrieving the rotation param The retrieved rotation parameters are required for matching, but retrieving the rotation parameters requires matched particles\nto allow comparison of observed and retrieved particles. To solve this dilemma, the matching algorithm is applied to manually\n230\nselected cases for data where only a single, relatively large (> 10 px) particle is detected, so that the matching can be done\nbased on ∆h alone, ignoring ∆z. The found matched particles are used to retrieve the rotation parameters, assuming a priori\nvalues of zero for the rotation coefficients φ, θ, and Ofz, and a large a priori uncertainty of 5°, 5°, and 50 px, respectively. Then,\nall particles are considered for matching using the normal configuration based on both ∆h and ∆z. In an iterative process, the to allow comparison of observed and retrieved particles. To solve this dilemma, the matching algorithm is applied to manually\n230\nselected cases for data where only a single, relatively large (> 10 px) particle is detected, so that the matching can be done\nbased on ∆h alone, ignoring ∆z. The found matched particles are used to retrieve the rotation parameters, assuming a priori\nvalues of zero for the rotation coefficients φ, θ, and Ofz, and a large a priori uncertainty of 5°, 5°, and 50 px, respectively. Then,\nall particles are considered for matching using the normal configuration based on both ∆h and ∆z. 3.4\nParticle Tracking It will take into account that certain properties of a particle, such as Dmax, particle complexity c, or average bright-\nness only change to a certain extent from one frame to the next Also the fact that the particle trajectory is typically a smooth\n255 can be clearly seen that more complex particles (i.e. needles) fall slower than less complex particles (i.e. graupel) at the same\n250\nparticle size. Despite the large uncertainty of the simple velocity estimate, needle particles roughly follow a parameterization\nof (Lumb, 1961) found in (Brandes et al., 2008) for unrimed aggregates, while graupel exceeds the velocity for rimed particles\nin the same study. The final tracking algorithm (under development) will follow a probabilistic approach similar to particle\nmatching. It will take into account that certain properties of a particle, such as Dmax, particle complexity c, or average bright- 250 ness, only change to a certain extent from one frame to the next. Also, the fact that the particle trajectory is typically a smooth\n255\ncurve instead of a zigzag line can be exploited. This can be seen in a composite of a particle (Fig. 5) observed during MOSAiC,\nwhich also shows how the multiple perspectives of the particle help to identify its true shape. The example also shows that\nduring MOSAiC the alignment of the cameras was not perfect, resulting in some of the measurements being slightly out of\nfocus; this has been resolved for later campaigns. ness, only change to a certain extent from one frame to the next. Also, the fact that the particle trajectory is typically a smooth\n255\ncurve instead of a zigzag line can be exploited. This can be seen in a composite of a particle (Fig. 5) observed during MOSAiC,\nwhich also shows how the multiple perspectives of the particle help to identify its true shape. The example also shows that\nduring MOSAiC the alignment of the cameras was not perfect, resulting in some of the measurements being slightly out of\nfocus; this has been resolved for later campaigns. 3.4\nParticle Tracking Tracking a matched particle over time provides its three-dimensional trajectory, from which sedimentation velocity and inter-\naction with turbulence can be determined. Since the natural tumbling of the particles provides new particle perspectives, the\n245\nestimates of Dmax and AR can be further improved. A proof of concept showing the potential of VISSS for velocity measure- 10 Figure 5. Composit of a snow particle recorded by leader (left) and follower (right) during MOSAiC on 15 November 2019 05:31 UTC. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Figure 5. Composit of a snow particle recorded by leader (left) and follower (right) during MOSAiC on 15 November 2019 05:31 UTC. Figure 5. Composit of a snow particle recorded by leader (left) and follower (right) during MOSAiC on 15 November 2019 05:31 UTC. snow particle recorded by leader (left) and follower (right) during MOSAiC on 15 November 2019 05:31 UT ments is shown in Fig. 4 for a case with both needles and small rimed particles (Hyytiälä, 5 January 2022, 00:00-14:30 UTC). Needles and graupel can be distinguished using the particle complexity c (Eq. 1) which is higher for needles than for graupel. In this example, the particle velocity is simply estimated by pairing the particles closest in space of consecutive frames. Still, it ments is shown in Fig. 4 for a case with both needles and small rimed particles (Hyytiälä, 5 January 2022, 00:00-14:30 UTC). Needles and graupel can be distinguished using the particle complexity c (Eq. 1) which is higher for needles than for graupel. In this example, the particle velocity is simply estimated by pairing the particles closest in space of consecutive frames. Still, it In this example, the particle velocity is simply estimated by pairing the particles closest in space of consecutive frames. Still, it\ncan be clearly seen that more complex particles (i.e. needles) fall slower than less complex particles (i.e. graupel) at the same\n250\nparticle size. Despite the large uncertainty of the simple velocity estimate, needle particles roughly follow a parameterization\nof (Lumb, 1961) found in (Brandes et al., 2008) for unrimed aggregates, while graupel exceeds the velocity for rimed particles\nin the same study. The final tracking algorithm (under development) will follow a probabilistic approach similar to particle\nmatching. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. The particle distribution level2 product is currently only available based on matched particles (level2match), but will be\navailable for tracked particles in the future. This means that the multiple observations of the same particle all contribute to the\nPSD. This does not bias the PSD because the number of particles observed is divided by the number of frames, and the PSD\n270\ndescribes how many particles are on average in the observation volume. For cases where only a single camera is available, a\nproduct based on particles detected by a single camera is also possible, using a threshold based on particle blur for defining the\nobservation volume similar to the PIP (Newman et al., 2009). The particle distribution level2 product is currently only available based on matched particles (level2match), but will be\navailable for tracked particles in the future. This means that the multiple observations of the same particle all contribute to the PSD. This does not bias the PSD because the number of particles observed is divided by the number of frames, and the PSD\n270\ndescribes how many particles are on average in the observation volume. For cases where only a single camera is available, a\nproduct based on particles detected by a single camera is also possible, using a threshold based on particle blur for defining the\nobservation volume similar to the PIP (Newman et al., 2009). 3.5\nParticle distributions\n260 To estimate particle distributions, the individual particle data are binned by particle size (1 px spacing, i.e. 43.125 or 58.75 µm)\nand averaged to one minute resolution for particle properties such as size, area, and perimeter. These binned particle properties\nare available either from one of the cameras or using the minimum, mean, or maximum from both cameras for each observed\nparticle property. PSDs, cross section area A, perimeter p, and particle complexity c are binned with both Dmax and the particle\ni\nl\ndi\n(D\n) I\nddi i\nPSD\ni h d\nl\nil bl f\nA AR\nd\ni\nddi i\nh fi To estimate particle distributions, the individual particle data are binned by particle size (1 px spacing, i.e. 43.125 or 58.75 µm)\nand averaged to one minute resolution for particle properties such as size, area, and perimeter. These binned particle properties\nare available either from one of the cameras or using the minimum, mean, or maximum from both cameras for each observed\nparticle property. PSDs, cross section area A, perimeter p, and particle complexity c are binned with both Dmax and the particle area equivalent diameter (Deq). In addition, PSD-weighted mean values are available for A, AR, and c in addition to the first\n265\nto fourth and sixth moments of the distribution that can be used to describe normalized size distributions (Delanoë et al., 2005;\nMaahn et al., 2015). 11 D[px] = (0.016971 ± 0.000015) · D[µm] + (0.349303 ± 0.027170), (5) D[px] = (0.023047 ± 0.000050) · D[µm] + (0.900593 ± 0.078123),\n(6)\n280\nfor the VISSS2 based on 182 samples. The inverse of the slope is 58.92 µm px−1 (43.389 µm px−1) and is close to the\nmanufacturer’s specification of 58.75 µm px−1 (43.125 µm px−1) for the VISSS1 (VISSS2). The non-zero intercept is caused\nby the fact that the Dmax estimator used to process the images often rounds up to the next full pixel. For VISSS2, this effect\nis exacerbated by the slightly blurrier images. Eqs. 5 and 6 are used to calibrate Dmax, but only the slope is used to calibrate D[px] = (0.023047 ± 0.000050) · D[µm] + (0.900593 ± 0.078123),\n(6)\n280\nfor the VISSS2 based on 182 samples. The inverse of the slope is 58.92 µm px−1 (43.389 µm px−1) and is close to the\nmanufacturer’s specification of 58.75 µm px−1 (43.125 µm px−1) for the VISSS1 (VISSS2). The non-zero intercept is caused\nby the fact that the Dmax estimator used to process the images often rounds up to the next full pixel. For VISSS2, this effect\nis exacerbated by the slightly blurrier images. Eqs. 5 and 6 are used to calibrate Dmax, but only the slope is used to calibrate (6) D[px] = (0.023047 ± 0.000050) · D[µm] + (0.900593 ± 0.078123),\n(6)\n280\nfor the VISSS2 based on 182 samples. The inverse of the slope is 58.92 µm px−1 (43.389 µm px−1) and is close to the\nmanufacturer’s specification of 58.75 µm px−1 (43.125 µm px−1) for the VISSS1 (VISSS2). The non-zero intercept is caused\nby the fact that the Dmax estimator used to process the images often rounds up to the next full pixel. For VISSS2, this effect\nis exacerbated by the slightly blurrier images. Eqs. 5 and 6 are used to calibrate Dmax, but only the slope is used to calibrate Deq, perimeter, and area because potential biases from the image processing routines have not been characterized. Analyzing\n285\nreference spheres would not be helpful because the shape complexity of spheres is much smaller than for real snow particles. The calibration is also checked by holding a millimeter pattern in the camera and measuring the pixel distance in the images,\nthe found difference to the reference spheres is less than 2%. Deq, perimeter, and area because potential biases from the image processing routines have not been characterized. D[px] = (0.016971 ± 0.000015) · D[µm] + (0.349303 ± 0.027170), Analyzing\n285\nreference spheres would not be helpful because the shape complexity of spheres is much smaller than for real snow particles. The calibration is also checked by holding a millimeter pattern in the camera and measuring the pixel distance in the images,\nthe found difference to the reference spheres is less than 2%. Part of the calibration is to characterize the observation volume. For perfectly aligned cameras, this would simply be the vol-\nume of a rectangular box with a base of 1280 px x 1280 px and a height of 1024 px. However, due to the imperfect alignment of\n290\nthe cameras, the actual observation volume is slightly smaller than the rectangular cuboid. Therefore, the observation volumes\nare calculated separately for leader and follower, the eight vertices of the follower observation volume are rotated to the leader\ncoordinate system, and the intersection of the two bodies is calculated using the OpenSCAD library. To account for the removal\nof particles detected at the edge of the image, a buffer of Dmax/2 to the edges of the image is used and the observation volume\nis reduced accordingly. Finally, the volume is converted from pixels to m3 using the calibration factor estimated above. 295 ume of a rectangular box with a base of 1280 px x 1280 px and a height of 1024 px. However, due to the imperfect alignment of\n290\nthe cameras, the actual observation volume is slightly smaller than the rectangular cuboid. Therefore, the observation volumes\nare calculated separately for leader and follower, the eight vertices of the follower observation volume are rotated to the leader\ncoordinate system, and the intersection of the two bodies is calculated using the OpenSCAD library. To account for the removal\nof particles detected at the edge of the image, a buffer of Dmax/2 to the edges of the image is used and the observation volume ume of a rectangular box with a base of 1280 px x 1280 px and a height of 1024 px. However, due to the imperfect alignment of\n290\nthe cameras, the actual observation volume is slightly smaller than the rectangular cuboid. Therefore, the observation volumes\nare calculated separately for leader and follower, the eight vertices of the follower observation volume are rotated to the leader\ncoordinate system, and the intersection of the two bodies is calculated using the OpenSCAD library. 3.6\nCalibration The VISSS calibration is tested using reference steel or ceramic spheres with 1 to 3 mm diameter. After processing using the\n275\nstandard VISSS routines, the estimated sizes are compared to the expected ones. A linear least square fit is applied to the 276\nreference sphere observations resulting in The VISSS calibration is tested using reference steel or ceramic spheres with 1 to 3 mm diameter. After processing using the\n275\nstandard VISSS routines, the estimated sizes are compared to the expected ones. A linear least square fit is applied to the 276\nreference sphere observations resulting in D[px] = (0.016971 ± 0.000015) · D[µm] + (0.349303 ± 0.027170),\n(5)\nfor the VISSS1 and D[px] = (0.016971 ± 0.000015) · D[µm] + (0.349303 ± 0.027170), 4\nPilot studies Here, we analyze first generation VISSS (VISSS1) data collected in winter 2021/22 at the Hyytiälä Forestry Field Station\n(61.845◦N, 24.287◦E, 150 m MSL) operated by the University of Helsinki, Finland to show the potential of the instrument. For comparison, we use a co-located PIP (von Lerber et al., 2017; Pettersen et al., 2020) and OTT Parsivel2 laser disdrometer\n(Löffler-Mang and Joss, 2000; Tokay et al., 2014). The distance between the VISSS and PIP was 20 m. The Parsivel was\nlocated inside of the double fence intercomparison reference, which was located 35 m from VISSS. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. D[px] = (0.016971 ± 0.000015) · D[µm] + (0.349303 ± 0.027170), To account for the removal\nof particles detected at the edge of the image, a buffer of Dmax/2 to the edges of the image is used and the observation volume 12 12 4.1\nCase study comparing VISSS, PIP, and Parsivel The variability of N ∗\n0 and D32 as well as the particle complexity c and the number of\nparticles observed throughout the day are depicted in Fig. 6. c is a spectral variable available for each size bin. Because using\na PSD-weighted average over all sizes for c would be heavily weighted to smaller, due to the finite resolution potentially less as shown in (Maahn et al., 2015). The variability of N ∗\n0 and D32 as well as the particle complexity c and the number of\nparticles observed throughout the day are depicted in Fig. 6. c is a spectral variable available for each size bin. Because using\na PSD-weighted average over all sizes for c would be heavily weighted to smaller, due to the finite resolution potentially less as shown in (Maahn et al., 2015). The variability of N ∗\n0 and D32 as well as the particle complexity c and the number of\nparticles observed throughout the day are depicted in Fig. 6. c is a spectral variable available for each size bin. Because using\na PSD-weighted average over all sizes for c would be heavily weighted to smaller, due to the finite resolution potentially less complex particles, we use the 95th percentile for c in the following. The main precipitation event lasted from 10:00 to 17:30\n315\nUTC and shows an anticorrelation between N ∗\n0 and D32: the former increases up to 105 m−3 mm−1 until 13:00 UTC before\ndecreasing to 103 m−3 mm−1 at the end of the event. The number of particles observed ranges between 10,000 and 100,000\nper minute, showing that estimates of N ∗\n0 and D32 are based on sufficient number of observations to limit the impact of random\nerrors. The particle complexity c divides the core period of the event into two parts with c ≈2 before 13:00 UTC and c ≈2.8 after 13:00 UTC. This transition can also be seen in the random selection of matched particles observed by the VISSS (Fig. 7)\n320\nretrieved from the imagesL1detect product. For each particle, a pair of images is available from the two VISSS cameras. Before\n13:00 UTC, a wide variety of different particle types has been observed, including plates, small aggregates and small rimed\nparticles. 4.1\nCase study comparing VISSS, PIP, and Parsivel VISSS level2match data are compared with PIP and Parsivel observations for a snowfall case on 26 January 2022. Because\nParsivel uses something similar to Deq (see discussion in Battaglia et al., 2010, for the predecessor instrument), Deq is also VISSS level2match data are compared with PIP and Parsivel observations for a snowfall case\nParsivel uses something similar to Deq (see discussion in Battaglia et al., 2010, for the prede p\ny\nParsivel uses something similar to Deq (see discussion in Battaglia et al., 2010, for the predecessor instrument), Deq is also\nused as a PIP and VISSS size descriptor in the following. Also, Deq is not affected by the problems of the PIP particle sizing\n305\nalgorithm identified by (Helms et al., 2022). The PSD is characterized by the two variables N ∗\n0 and D32 used to describe\nthe normalized size distributions N(D) = N ∗\n0 F(D/D32) (Testud et al., 2001; Delanoë et al., 2005) where N ∗\n0 is a scaling\nparameter and D32 normalizes the size distribution by size. Assuming a typical value of 2 for the exponent b of the mass-size\nrelation (e.g., Mitchell, 1996), D32 is the proxy for the mean mass-weighted diameter defined as the ratio of the third to the\nsecond measured PSD moments M3/M2 Assuming the same value for b N ∗can be calculated with\n310 used as a PIP and VISSS size descriptor in the following. Also, Deq is not affected by the problems of the PIP particle sizing\n305\nalgorithm identified by (Helms et al., 2022). The PSD is characterized by the two variables N ∗\n0 and D32 used to describe\nthe normalized size distributions N(D) = N ∗\n0 F(D/D32) (Testud et al., 2001; Delanoë et al., 2005) where N ∗\n0 is a scaling\nparameter and D32 normalizes the size distribution by size. Assuming a typical value of 2 for the exponent b of the mass-size\nrelation (e.g., Mitchell, 1996), D32 is the proxy for the mean mass-weighted diameter defined as the ratio of the third to the N ∗\n0 = M 4\n2\nM 3\n3\n27\n2\n(7) N ∗\n0 = M 4\n2\nM 3\n3\n27\n2 (7) as shown in (Maahn et al., 2015). 4.1\nCase study comparing VISSS, PIP, and Parsivel Since particle shape and mean brightness are not used to match particles, the observed image pairs also confirm the\nability of VISSS to correctly match data from the two cameras. After 13:00 UTC, needles and needle aggregates dominate the after 13:00 UTC. This transition can also be seen in the random selection of matched particles observed by the VISSS (Fig. 7)\n320\nretrieved from the imagesL1detect product. For each particle, a pair of images is available from the two VISSS cameras. Before\n13:00 UTC, a wide variety of different particle types has been observed, including plates, small aggregates and small rimed\nparticles. Since particle shape and mean brightness are not used to match particles, the observed image pairs also confirm the\nability of VISSS to correctly match data from the two cameras. After 13:00 UTC, needles and needle aggregates dominate the observations explaining the increase in observed complexity. Towards the end of the event, particles become smaller and more\n325\nirregularly shaped. Around 18:30 UTC, even some ice lolly shaped particles (Keppas et al., 2017) are observed by the VISSS. observations explaining the increase in observed complexity. Towards the end of the event, particles become smaller and more\n325\nirregularly shaped. Around 18:30 UTC, even some ice lolly shaped particles (Keppas et al., 2017) are observed by the VISSS. 13 This means that a needle with a width of 4 px or 0.4 mm is completely removed by the PIP processing\nscheme resulting in an underestimation in the PIP number concentration in the presence of needles. During this event, a large\nnumber of needles was observed as shown in Fig. 7. This effect is not limited to needles, but is expected to affect particles but the observed VISSS - PIP difference seems to be somewhat larger than expected, namely the difference extends to larger\n335\nD32 values. This may be due to the PIP image processing which dilates each picture twice using a 3-pixel by 3-pixel kernel\n(Helms et al., 2022). This means that a needle with a width of 4 px or 0.4 mm is completely removed by the PIP processing\nscheme resulting in an underestimation in the PIP number concentration in the presence of needles. During this event, a large\nnumber of needles was observed as shown in Fig. 7. This effect is not limited to needles, but is expected to affect particles with large aspect ratios, and particles which have sub-parts with large aspect ratios. Around 10:10 UTC, PIP underestimates\n340\nD32 while the N ∗\n0 difference appears to be smaller. During this time, VISSS observed the presence of radiating assemblage of\nplates (Fig. 7). It is possible that PIPs image processing removes some parts of these particles, resulting in underestimation of\ntheir size. To further investigate the differences between the instruments, we compare VISSS, PIP, and Parsivel spectra (Fig. 8) for with large aspect ratios, and particles which have sub-parts with large aspect ratios. Around 10:10 UTC, PIP underestimates\n340\nD32 while the N ∗\n0 difference appears to be smaller. During this time, VISSS observed the presence of radiating assemblage of\nplates (Fig. 7). It is possible that PIPs image processing removes some parts of these particles, resulting in underestimation of\ntheir size. To further investigate the differences between the instruments, we compare VISSS, PIP, and Parsivel spectra (Fig. 8) for To further investigate the differences between the instruments, we compare VISSS, PIP, and Parsivel spectra (Fig. 8) for\nthe three discussed times during the snowfall case. All three instruments have different sensitivities to small particles. This can be seen for the drop in D32 around 17:45 UTC (Fig. 6) where the Parsivel does not report any values, and the PIP N ∗\n0 estimates differ strongly from the VISSS when D32 < 1\n355\nmm. The VISSS reports D32 values as low as 0.16 mm around 19:00 UTC. Although the sample sizes are sufficient (> 10,000\nparticles per minute), the errors are likely large due to the VISSS resolution of ~0.06 mm. In the absence of an instrument\ndesigned to observe small particles, it is not possible to determine how reliably VISSS detects and sizes small particles. Additional insight is provided by comparing the drop size distributions (DSD) observed by the three instruments during (Fig. 6) where the Parsivel does not report any values, and the PIP N ∗\n0 estimates differ strongly from the VISSS when D32 < 1\n355\nmm. The VISSS reports D32 values as low as 0.16 mm around 19:00 UTC. Although the sample sizes are sufficient (> 10,000\nparticles per minute), the errors are likely large due to the VISSS resolution of ~0.06 mm. In the absence of an instrument\ndesigned to observe small particles, it is not possible to determine how reliably VISSS detects and sizes small particles. Additional insight is provided by comparing the drop size distributions (DSD) observed by the three instruments during (Fig. 6) where the Parsivel does not report any values, and the PIP N ∗\n0 estimates differ strongly from the VISSS when D32 < 1\n355\nmm. The VISSS reports D32 values as low as 0.16 mm around 19:00 UTC. Although the sample sizes are sufficient (> 10,000\nparticles per minute), the errors are likely large due to the VISSS resolution of ~0.06 mm. In the absence of an instrument\ndesigned to observe small particles, it is not possible to determine how reliably VISSS detects and sizes small particles. Additional insight is provided by comparing the drop size distributions (DSD) observed by the three instruments during Additional insight is provided by comparing the drop size distributions (DSD) observed by the three instruments during\na drizzle event on 16 October 2021 (Fig. 8.d). While Parsivel and VISSS mostly agree for D > 1 mm for all three cases,\n345\nParsivel observes more particles for 0.6 mm < D < 1 mm (as previously reported by Battaglia et al., 2010) before dropping\nfor D < 0.6 mm, which is likely related to limitations associated with the Parsivel resolution of 125 µm. The comparison of\nVISSS and PIP shows larger discrepancies as explained above. The PSDs tend to agree for Deq > 1 mm for cases where larger\nice particles are more spherical (11:24 UTC). For the needle case (13:00 UTC), PIP reports lower number concentrations than the three discussed times during the snowfall case. While Parsivel and VISSS mostly agree for D > 1 mm for all three cases,\n345\nParsivel observes more particles for 0.6 mm < D < 1 mm (as previously reported by Battaglia et al., 2010) before dropping\nfor D < 0.6 mm, which is likely related to limitations associated with the Parsivel resolution of 125 µm. The comparison of\nVISSS and PIP shows larger discrepancies as explained above. The PSDs tend to agree for Deq > 1 mm for cases where larger\nice particles are more spherical (11:24 UTC). For the needle case (13:00 UTC), PIP reports lower number concentrations than VISSS and Parsivel for almost all sizes. At 10:10 UTC, VISSS and PIP approximately agree for sizes between 0.4 and 0.8 mm,\n350\nbut PIP reports lower values for other sizes. Although no needles are observed at 10:10 UTC, Fig. 7 shows that there were also\nsmall columns that could be affected by the dilation of structures less than 0.4 mm wide by the PIP software, or some parts of\nradiating assemblage of plates were removed by the image processing. All three instruments have different sensitivities to small particles. This can be seen for the drop in D32 around 17:45 UTC VISSS and Parsivel for almost all sizes. At 10:10 UTC, VISSS and PIP approximately agree for sizes between 0.4 and 0.8 mm,\n350\nbut PIP reports lower values for other sizes. Although no needles are observed at 10:10 UTC, Fig. 7 shows that there were also\nsmall columns that could be affected by the dilation of structures less than 0.4 mm wide by the PIP software, or some parts of\nradiating assemblage of plates were removed by the image processing. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. N ∗\n0 and D32 are also calculated from the PSDs observed by PIP and Parsivel. For the core event, N ∗\n0 measured by the PIP is\nabout an order of magnitude smaller than that measured by VISSS and Parsivel. The agreement of VISSS and Parsivel is better,\nbut some peaks in N ∗\n0 are not resolved by the Parsivel when D32 is large. This discrepancy may be related to problems of the\nParsivel with larger particles reported before (Battaglia et al., 2010). The reason for the observed differences between PIP and\n330\nVISSS is more complex. Overall the measured D32 agrees better than N ∗\n0 . Because D32 is a proxy for the mass-weighted mean\ndiameter, larger more massive snowflakes have a larger impact on D32 than more numerous smaller particles. This implies that\nPIP is not capturing as many small ice particles as VISSS, while measurements of larger particles seem to be less affected. Tiira et al. (2016) have studied the effect of the left-side PSD truncation on PIP observations (see Fig. 6 in Tiira et al., 2016), but the observed VISSS - PIP difference seems to be somewhat larger than expected, namely the difference extends to larger\n335\nD32 values. This may be due to the PIP image processing which dilates each picture twice using a 3-pixel by 3-pixel kernel\n(Helms et al., 2022). This means that a needle with a width of 4 px or 0.4 mm is completely removed by the PIP processing\nscheme resulting in an underestimation in the PIP number concentration in the presence of needles. During this event, a large\nnumber of needles was observed as shown in Fig. 7. This effect is not limited to needles, but is expected to affect particles but the observed VISSS - PIP difference seems to be somewhat larger than expected, namely the difference extends to larger\n335\nD32 values. This may be due to the PIP image processing which dilates each picture twice using a 3-pixel by 3-pixel kernel\n(Helms et al., 2022). 4.2\nStatistical comparison of VISSS, PIP, and Parsivel\n370 The results of the case study comparison of VISSS, PIP, and Parsivel also hold when comparing 6661 minutes of joint snowfall\nobservations during the winter of 2021/22 (Fig. 9). The ratio of N∗\n0 observed by VISSS and PIP (Parsivel) is compared to D32,\nN∗\n0, and complexity c. For D32 < 1 mm, the VISSS to PIP (Parsivel) N∗\n0 ratio increases strongly and can reach a value of\n10,000 (10). Therefore, the comparison of the N∗\n0 ratio with N∗\n0 itself and c is limited to data with D32 > 1 mm. For the PIP, the The results of the case study comparison of VISSS, PIP, and Parsivel also hold when comparing 6661 minutes of joint snowfall\nobservations during the winter of 2021/22 (Fig. 9). The ratio of N∗\n0 observed by VISSS and PIP (Parsivel) is compared to D32,\nN∗\n0, and complexity c. For D32 < 1 mm, the VISSS to PIP (Parsivel) N∗\n0 ratio increases strongly and can reach a value of\n10,000 (10). Therefore, the comparison of the N∗\n0 ratio with N∗\n0 itself and c is limited to data with D32 > 1 mm. For the PIP, the difference in N∗\n0 does not depend on N∗\n0 but—as suggested by the needle case above—on complexity c, with higher c values\n375\nindicating larger N∗\n0 differences, probably related to the image dilation problem discussed above. For the VISSS to Parsivel\ncomparison, the N∗\n0 difference depends rather on N∗\n0 instead of c. Because D32 and N∗\n0 are often anti-correlated, this could be\nrelated to size-dependent errors of the Parsivel as identified by (Battaglia et al., 2010). difference in N∗\n0 does not depend on N∗\n0 but—as suggested by the needle case above—on complexity c, with higher c values\n375\nindicating larger N∗\n0 differences, probably related to the image dilation problem discussed above. For the VISSS to Parsivel\ncomparison, the N∗\n0 difference depends rather on N∗\n0 instead of c. Because D32 and N∗\n0 are often anti-correlated, this could be\nrelated to size-dependent errors of the Parsivel as identified by (Battaglia et al., 2010). https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. DSDs differ no more than 10% for 0.55 mm > D > 0.9 mm both showing a dip in the distribution around 0.55 mm. For\nlarger droplets, differences are likely related to the small sample size. For smaller droplets, VISSS (and PIP) report about an\norder of magnitude higher concentrations than the Parsivel. Similarly, (Thurai et al., 2019) found that a 50 µm optical array\nprobe observed more small drizzle droplets than a Parsivel. For these small particle sizes close to the VISSS camera resolution,\n365\ndiscretization errors likely play a role which we investigate by comparing Dmax and Deq for the VISSS. As drizzle droplets can\nbe considered sufficiently spherical (i.e. AR >0.9) for D <1 mm (Beard et al., 2010), we can evaluate whether Dmax = Deq\nholds as expected (Fig. 8.d). As expected, VISSS Dmax and Deq are in almost perfect agreement for D >0.5 mm, but larger\ndifferences occur for D <0.3 mm indicating that discretization errors can become substantial for D <0.3 mm. The use of drizzle allows Parsivel to be used as a reference instrument as it\n360\nhas been shown to provide accurate DSDs for sizes between 0.5 and 5 mm (Tokay et al., 2014). In fact, Parsivel and VISSS a drizzle event on 16 October 2021 (Fig. 8.d). The use of drizzle allows Parsivel to be used as a reference instrument as it\n360\nhas been shown to provide accurate DSDs for sizes between 0.5 and 5 mm (Tokay et al., 2014). In fact, Parsivel and VISSS 14 4.3\nAdvantage of the second VISSS camera The two-camera VISSS setup allows for quantification of the errors in Dmax, aspect ratio AR, cross-sectional area A, and\n380\nperimeter p that would be made if only a single camera were used (Fig. 10). The errors are defined as the normalized difference\nbetween the maximum observation of Dmax, A, and p from the pair of cameras and the observation of the leader camera alone. For AR, the minimum of both observations is used instead. A positive error indicates that the observation of a single camera\nwould be too small. For this assessment, three cases with mostly dendritic-aggregates (6 December 2021, 07:19 - 12:30 UTC), The two-camera VISSS setup allows for quantification of the errors in Dmax, aspect ratio AR, cross-sectional area A, and\n380\nperimeter p that would be made if only a single camera were used (Fig. 10). The errors are defined as the normalized difference\nbetween the maximum observation of Dmax, A, and p from the pair of cameras and the observation of the leader camera alone. For AR, the minimum of both observations is used instead. A positive error indicates that the observation of a single camera\nwould be too small. For this assessment, three cases with mostly dendritic-aggregates (6 December 2021, 07:19 - 12:30 UTC), needles (5 January 2022, 00:00 - 14:30 UTC), and graupel (6 December 2021, 00:00 - 04:50; 13:30 - 14:20; 21:15-24:00 and\n385\n5 January 2022, 15:00 - 16:40; 19:40 -20:50 UTC) are analyzed using the level1match product, which contains properties for\neach observed matched particle. As expected from the highly irregular shape, the errors are largest for needles. The errors peak\naround 0.7 mm for Dmax, AR, A, and p with mean values of 15, -88, 18, and 14 %, respectively. Due to aggregation forming\nmore spherical needle aggregates, the error decreases for larger sizes. For graupel particles, the error is typically less than 10% (for AR about -30%) and slightly larger for dendritic aggregates. 390\nTo analyze how the error in Dmax affects the simulated radar reflectivity, we use the the PAMTRA radar simulator (Passive\nand Active Microwave radiative TRAnsfer tool, Mech et al., 2020) with the riming-dependent parameterization of the particle\nscattering properties (Maherndl et al., 2023). The error in Dmax translates into mean errors between 0.8 dB (aggregates) and\n2.11 dB (needles). This is less than the 3.2 dB found by Wood et al. 4.3\nAdvantage of the second VISSS camera (2013) using idealized particles, but this is likely related to (for AR about -30%) and slightly larger for dendritic aggregates. 390\nTo analyze how the error in Dmax affects the simulated radar reflectivity, we use the the PAMTRA radar simulator (Passive\nand Active Microwave radiative TRAnsfer tool, Mech et al., 2020) with the riming-dependent parameterization of the particle\nscattering properties (Maherndl et al., 2023). The error in Dmax translates into mean errors between 0.8 dB (aggregates) and\n2.11 dB (needles). This is less than the 3.2 dB found by Wood et al. (2013) using idealized particles, but this is likely related to 15 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 101\n103\n105\nN *\n0 [1/m3/mm]\na)\nHyytiälä 2022-01-26\n0\n1\n2\n3\n4\nD32 [mm]\nb)\nVISSS\nPIP\nParsivel\n1\n2\n3\n4\ncomplexity [-]\nc)\n10:00\n12:00\n14:00\n16:00\n18:00\n20:00\n22:00\ntime\n101\n103\n105\nobserved particles [1/min]\nd)\nure 6 Comparison of VISSS (blue) PIP (orange) and Parsivel (green) for a snowfall case on 26 January 2022 at Hyytiälä 101\n103\n105\nN *\n0 [1/m3/mm]\na)\nHyytiälä 2022-01-26\n0\n1\n2\n3\n4\nD32 [mm]\nb)\nVISSS\nPIP\nParsivel\n1\n2\n3\n4\ncomplexity [-]\nc)\n10:00\n12:00\n14:00\n16:00\n18:00\n20:00\n22:00\ntime\n101\n103\n105\nobserved particles [1/min]\nd)\nFigure 6. Comparison of VISSS (blue), PIP (orange), and Parsivel (green) for a snowfall case on 26 January 2022 at Hy\nD23 (b), complexity c (c), and the number of observed particles (d). The three vertical black lines indicate the sample PS Hyytiälä 2022-01-26 101\n103\n105\nN *\n0 [1/m3/mm]\na)\nHyytiälä 2022-01-26\n0\n1\n2\n3\n4\nD32 [mm]\nb)\nVISSS\nPIP\nParsivel\n1\n2\n3\n4\ncomplexity [-]\nc)\n10:00\n12:00\n14:00\n16:00\n18:00\n20:00\n22:00\ntime\n101\n103\n105\nobserved particles [1/min]\nd)\nFigure 6. Comparison of VISSS (blue), PIP (orange), and Parsivel (green) for a snowfall case on 26 January 2022 at Hyytiälä using N∗\n0 (a),\nD23 (b), complexity c (c), and the number of observed particles (d). The three vertical black lines indicate the sample PSDs shown in Fig. 8. the fact that two perspectives as provided by the VISSS are not sufficient to provide the true Dmax. Also, the use of idealized\n95\nparticles might lead to an overestimation of the bias. Figure 6. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Image pairs of particles observed by the two VISSS cameras on 26 January 2022 between 10:00 and 19:00 UTC. The (R) indicates\nparticles than shown were observed by the VISSS and only a random selection is presented in the panel. Only particles with\n0.59 mm (10 px) are shown. pairs of particles observed by the two VISSS cameras on 26 January 2022 between 10:00 and 19:00 UTC. The (R) in\nes than shown were observed by the VISSS and only a random selection is presented in the panel. Only particl\n(10\n)\nh Figure 7. Image pairs of particles observed by the two VISSS cameras on 26 January 2022 between 10:00 and 19:00 UTC. The (R) indicates\nthat more particles than shown were observed by the VISSS and only a random selection is presented in the panel. Only particles with\nDmax ≥0.59 mm (10 px) are shown. 4.3\nAdvantage of the second VISSS camera Comparison of VISSS (blue), PIP (orange), and Parsivel (green) for a snowfall case on 26 January 2022 at Hyytiälä using N∗\n0 (a),\nD23 (b), complexity c (c), and the number of observed particles (d). The three vertical black lines indicate the sample PSDs shown in Fig. 8. the fact that two perspectives as provided by the VISSS are not sufficient to provide the true Dmax. Also, the use of idealized\n395\nparticles might lead to an overestimation of the bias. the fact that two perspectives as provided by the VISSS are not sufficient to provide the true Dmax. Also, the use of idealized\n395\nparticles might lead to an overestimation of the bias. 16 The hardware and data processing of the open source Video In Situ Snowfall Sensor (VISSS) has been introduced. The VISSS\nconsists of two cameras with telecentric lenses oriented at a 90° angle to a common observation volume. Both cameras are 5\nConclusions The hardware and data processing of the open source Video In Situ Snowfall Sensor (VISSS) has been introduced. The VISSS\nconsists of two cameras with telecentric lenses oriented at a 90° angle to a common observation volume. Both cameras are 17 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 100\n101\n102\n103\n104\n105\nPSD [1/m3/mm]\na) snow: 2022-01-26 10:10:00\nb) snow: 2022-01-26 11:24:00\n10\n1\n100\n101\nD [mm]\n100\n101\n102\n103\n104\n105\nPSD [1/m3/mm]\nc) snow: 2022-01-26 13:00:00\n10\n1\n100\n101\nD [mm]\nd) drizzle: 2021-10-16 14:30:00\nVISSS Deq\nVISSS Dmax\nPIP Deq\nParsivel Deq\nFigure 8. (a-c) Particle size distributions of VISSS, PIP, and Parsivel for the three cases indicated in Fig. 6 on 26 January 2022. Deq is used\nas a size descriptor. (d) Same as (a-c), but showing the drop size distribution of a drizzle case on 16 October 2021. In addition, the VISSS\ndrop size distribution is also shown with Dmax as the size descriptor. snow: 2022-01-26 11:24:00 10\n1\n100\n101\nD [mm]\nd) drizzle: 2021-10-16 14:30:00\nVISSS Deq\nVISSS Dmax\nPIP Deq\nParsivel Deq drizzle: 2021-10-16 14:30:00 d) Figure 8. (a-c) Particle size distributions of VISSS, PIP, and Parsivel for the three cases indicated in Fig. 6 on 26 January 2022. Deq is used\nas a size descriptor. (d) Same as (a-c), but showing the drop size distribution of a drizzle case on 16 October 2021. In addition, the VISSS\ndrop size distribution is also shown with Dmax as the size descriptor. illuminated by LED backlights (see Table 1 for specifications). The goal of the VISSS design was to combine a large, well\n400\ndefined observation volume and relatively high resolution with a design that limits wind disturbance and allows accurate sizing. The VISSS was initially developed for MOSAiC, but additional deployments at Hyytiälä, Finland and Gothic, Colorado USA\nfollowed. An advanced version of the instrument has been installed at Ny-Ålesund, Svalbard. The VISSS processing scheme\nconsists of a series of products with per-particle (level 1) and size distribution (level 2) properties. Required processing steps include particle detection and sizing, particle matching between the two cameras considering the exact alignment of the cameras\n405\nto each other, and integration of particle properties over a size distribution. For estimating sedimentation velocity, particle\ntracking over time is required as well (under development). 5\nConclusions 0\n2\n4\n100\n102\n104\n106\nVISSS N *\n0 /PIP N *\n0 [-]\na)\nall D32\n101\n103\n105\n10\n1\n100\n101\n102\nb)\nD32 > 1 mm\n1\n2\n3\n4\n10\n1\n100\n101\n102\nc)\nD32 > 1 mm\nmean\n101\n102\n103\n104\n# observations\n0\n2\n4\nVISSS D32 [mm]\n10\n1\n100\n101\nVISSS N *\n0 /Parsivel N *\n0 [-]\nd)\n101\n103\n105\nVISSS N *\n0 [1/m3/mm]\n10\n1\n100\n101\ne)\n1\n2\n3\n4\nVISSS complexity [-]\n10\n1\n100\n101\nf)\n102\n103\n104\n# observations 0\n2\n4\n100\n102\n104\n106\nVISSS N *\n0 /PIP N *\n0 [-]\na)\nall D32\n101\n103\n105\n10\n1\n100\n101\n102\nb)\nD32 > 1 mm\n1\n2\n3\n4\n10\n1\n100\n101\n102\nc)\nD32 > 1 mm\nmean\n101\n102\n103\n104\n# observations\n0\n2\n4\nVISSS D32 [mm]\n10\n1\n100\n101\nVISSS N *\n0 /Parsivel N *\n0 [-]\nd)\n101\n103\n105\nVISSS N *\n0 [1/m3/mm]\n10\n1\n100\n101\ne)\n1\n2\n3\n4\nVISSS complexity [-]\n10\n1\n100\n101\nf)\n102\n103\n104\n# observations\nFigure 9. Statistical analysis of the ratio of VISSS to PIP N∗\n0 as a function of (a) VISSS D23, (b) VISSS N∗\n0, and (c) VISSS complexity c. (d-f) Same as (a-c), but comparing the VISSS to the Parsivel. The color indicates the number of particles observed by VISSS, the orange line\nindicates the mean ratio. The analysis for N∗\n0 (b, e) and c (c, f) is restricted to cases with D23 >1 mm. Figure 9. Statistical analysis of the ratio of VISSS to PIP N∗\n0 as a function of (a) VISSS D23, (b) VISSS N∗\n0, and (c) VISSS complexity c. (d-f) Same as (a-c), but comparing the VISSS to the Parsivel. The color indicates the number of particles observed by VISSS, the orange line\nindicates the mean ratio. The analysis for N∗\n0 (b, e) and c (c, f) is restricted to cases with D23 >1 mm. tively spherical particles, agreement was only found for sizes larger than 1 mm. Because the Parsivel is well characterized for\n415\nliquid precipitation (Tokay et al., 2014), a drizzle case is also used for comparison. 5\nConclusions The initial analysis shows the potential of the instrument. The relatively large observation volume of the VISSS leads to\nrobust statistics based on up to 100,000 particles observed per minute. The data set from Hyytiälä obtained in the winter of include particle detection and sizing, particle matching between the two cameras considering the exact alignment of the cameras\n405\nto each other, and integration of particle properties over a size distribution. For estimating sedimentation velocity, particle\ntracking over time is required as well (under development). tracking over time is required as well (under development). The initial analysis shows the potential of the instrument. The relatively large observation volume of the VISSS leads to\nrobust statistics based on up to 100,000 particles observed per minute. The data set from Hyytiälä obtained in the winter of The initial analysis shows the potential of the instrument. The relatively large observation volume of the VISSS leads to\nrobust statistics based on up to 100,000 particles observed per minute. The data set from Hyytiälä obtained in the winter of 2021/22 is used to compare the VISSS with collocated PIP and Parsivel instruments. While the comparison with the Parsivel\n410\nshows—given the known limitations of the instrument for snowfall (Battaglia et al., 2010)—excellent agreement, the compar-\nison with the PIP is more complicated. The differences in the observed PSDs increase with increasing particle complexity c\n(Eq. 1) which may be related to the image dilation used during particle sizing for the PIP, which inadvertently removes needle\nparticles. But differences remain even for non-needle cases and for a case with a relatively high concentration of large, rela- 2021/22 is used to compare the VISSS with collocated PIP and Parsivel instruments. While the comparison with the Parsivel\n410\nshows—given the known limitations of the instrument for snowfall (Battaglia et al., 2010)—excellent agreement, the compar-\nison with the PIP is more complicated. The differences in the observed PSDs increase with increasing particle complexity c\n(Eq. 1) which may be related to the image dilation used during particle sizing for the PIP, which inadvertently removes needle\nparticles. But differences remain even for non-needle cases and for a case with a relatively high concentration of large, rela- 18 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 5\nConclusions The case shows an excellent agreement\nbetween Parsivel and VISSS for droplets larger than 0.5 mm, confirming the general accuracy of VISSS. Compared to both PIP\nand Parsivel, VISSS observes a larger number of small particles that can drastically change the retrieved PSD coefficients in\nsome cases. However, the first generation VISSS resolution of 0.06 mm is likely to introduce discretization errors for particles tively spherical particles, agreement was only found for sizes larger than 1 mm. Because the Parsivel is well characterized for\n415\nliquid precipitation (Tokay et al., 2014), a drizzle case is also used for comparison. The case shows an excellent agreement\nbetween Parsivel and VISSS for droplets larger than 0.5 mm, confirming the general accuracy of VISSS. Compared to both PIP\nand Parsivel, VISSS observes a larger number of small particles that can drastically change the retrieved PSD coefficients in\nsome cases. However, the first generation VISSS resolution of 0.06 mm is likely to introduce discretization errors for particles tively spherical particles, agreement was only found for sizes larger than 1 mm. Because the Parsivel is well characterized for\n415\nliquid precipitation (Tokay et al., 2014), a drizzle case is also used for comparison. The case shows an excellent agreement\nbetween Parsivel and VISSS for droplets larger than 0.5 mm, confirming the general accuracy of VISSS. Compared to both PIP\nand Parsivel, VISSS observes a larger number of small particles that can drastically change the retrieved PSD coefficients in\nsome cases. However, the first generation VISSS resolution of 0.06 mm is likely to introduce discretization errors for particles smaller than 0.3 mm (i.e. 5 px), potentially leading to errors in the sizing of very small particles. Furthermore, we analyzed the\n420\nadvantage of the VISSS due to the availability of a second camera. Depending on the particle type, the availability of a second\ncamera avoids underestimation errors in Dmax of up to 15% and overestimation errors in aspect ratio AR of up to -88%. VISSS product development will continue. After finalizing the particle tracking algorithm to estimate particle sedimentation\nvelocity, machine learning based particle classifications (Praz et al., 2017; Leinonen and Berne, 2020; Leinonen et al., 2021) will be implemented. Also, we will work on making VISSS data acquisition and processing more efficient by handling some\n425\nprocessing steps on the data acquisition system in real-time. 5\nConclusions We invite also the community to contribute to the development\nof the open source instrument. This applies not only to the software products, but allows also for other groups to build the\ninstrument for approximately 22,000 EUR. It could even mean to advance the VISSS hardware concept further, by e.g. adding\na third camera to observe snow particles from below or—given the extended 1300 mm working distance of VISSS3—from will be implemented. Also, we will work on making VISSS data acquisition and processing more efficient by handling some\n425\nprocessing steps on the data acquisition system in real-time. We invite also the community to contribute to the development\nof the open source instrument. This applies not only to the software products, but allows also for other groups to build the\ninstrument for approximately 22,000 EUR. It could even mean to advance the VISSS hardware concept further, by e.g. adding\na third camera to observe snow particles from below or—given the extended 1300 mm working distance of VISSS3—from 19 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 0\n10\n20\n30\n40\n50\nDmax error [%]\na)\n200\n150\n100\n50\n0\naspect ratio error [%]\nb)\nGraupel, N=489,750\nNeedles, N=5,871,437\nAggregates, N=258,297\n0\n2\n4\n6\nDmax [mm]\n0\n20\n40\n60\ncross-sectional area error [%]\nc)\n0\n2\n4\n6\nDmax [mm]\n0\n10\n20\n30\n40\nperimeter error [%]\nd)\nFigure 10. Mean errors of (a) Dmax, (b) aspect ratio AR, (c) cross-sectional area A, and (d) perimeter p as a function of Dmax when using\nonly a single VISSS camera instead of combining the two cameras. Three cases with mostly observations of graupel (blue), needles (orange),\nand aggregates (green) are used. The shaded area indicates the 10th to 90th percentile range. aspect ratio error [%] Figure 10. Mean errors of (a) Dmax, (b) aspect ratio AR, (c) cross-sectional area A, and (d) perimeter p as a function of Dmax when using\nonly a single VISSS camera instead of combining the two cameras. Three cases with mostly observations of graupel (blue), needles (orange),\nand aggregates (green) are used. The shaded area indicates the 10th to 90th percentile range. above. Code and data availability. VISSS hardware plans (Maahn et al., 2023), data acquisition software (Maahn, 2023a), and data processing\n440\nlibraries (Maahn, 2023b) have been released under an open source license. VISSS, PIP, and Parsivel observations used for the pilot study are\navailable at https://zenodo.org/record/7797286 (Maahn and Moisseev, 2023). Appendix A: Coordinate system transformation We use a right handed coordinate system (x,y,z) to define the position of particles in the observation volume, where z points\nto the ground (see Fig. 1). The follower coordinate system (xF ,yF ,zF ) can be transformed into the leader coordinate system\n445\n(xL,yL,zL) by the standard transformation matrix \n\n\n\nxL\nyL\nzL\n\n\n\n=\n\n\n\n\ncosθcosψ\nsinφsinθcosψ −cosφsinψ\ncosφsinθcosψ + sinφsinψ\ncosθsinψ\nsinφsinθsinψ + cosφcosψ\ncosφsinθsinψ −sinφcosψ\n−sinθ\nsinφcosθ\ncosφcosθ\n\n\n\n\n\n\n\n\nx′\nF\ny′\nF\nz′\nF\n\n\n\n\n(A1) \n\n\n\nxL\nyL\nzL\n\n\n\n=\n\n\n\n\ncosθcosψ\nsinφsinθcosψ −cosφsinψ\ncosφsinθcosψ + sinφsinψ\ncosθsinψ\nsinφsinθsinψ + cosφcosψ\ncosφsinθsinψ −sinφcosψ\n−sinθ\nsinφcosθ\ncosφcosθ\n\n\n\n\n\n\n\n\nx′\nF\ny′\nF\nz′\nF\n\n\n\n\n(A1) \n\n\n\nxL\nyL\nzL\n\n\n\n=\n\n\n\n\ncosθcosψ\nsinφsinθcosψ −cosφsinψ\ncosφsinθcosψ + sinφsinψ\ncosθsinψ\nsinφsinθsinψ + cosφcosψ\ncosφsinθsinψ −sinφcosψ\n−sinθ\nsinφcosθ\ncosφcosθ\n\n\n\n\n\n\n\n\nx′\nF\ny′\nF\nz′\nF\n\n\n\n\n(A1) (A1) using the follower’s roll φ, yaw ψ, and pitch θ, analogous to airborne measurements, and w using the follower’s roll φ, yaw ψ, and pitch θ, analogous to airborne measurements, and with x′\nF = xF + Ofx, y′\nF =\nyF + Ofy, and z′\nF = zF + Ofz, where Ofx, Ofy, and Ofz are the offsets of the follower coordinate system in the x, y, and\nz directions, respectively (see Fig. 1) Offsets in Ofx and Ofy are neglected, because they would only materialize in reduced\n450\nparticle sharpness, but not in the retrieved three-dimensional position. The opposite transformation can be described by: yF + Ofy, and z′\nF = zF + Ofz, where Ofx, Ofy, and Ofz are the offsets of the follower coordi yF + Ofy, and zF = zF + Ofz, where Ofx, Ofy, and Ofz are the offsets of the follower coordinate system in the x, y, and\nz directions, respectively (see Fig. 1) Offsets in Ofx and Ofy are neglected, because they would only materialize in reduced\n450\nparticle sharpness, but not in the retrieved three-dimensional position. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 5\nConclusions The VISSS hardware plans (2nd generation VISSS, Maahn et al., 2023), data acquisition software (Maahn, 2023a),\n430\nand data processing libraries (Maahn, 2023b) have been released under an open source license so that reverse engineering as\ndone by Helms et al. (2022) is not required to analyze the VISSS data processing. The only limitation of the used licenses is\nthat modification of the VISSS need to be made publicly available under the same license. above. The VISSS hardware plans (2nd generation VISSS, Maahn et al., 2023), data acquisition software (Maahn, 2023a),\n430\nand data processing libraries (Maahn, 2023b) have been released under an open source license so that reverse engineering as\ndone by Helms et al. (2022) is not required to analyze the VISSS data processing. The only limitation of the used licenses is\nthat modification of the VISSS need to be made publicly available under the same license. There are many potential applications for VISSS observations. It can be used for model evaluation with advanced micro- There are many potential applications for VISSS observations. It can be used for model evaluation with advanced micro-\nphysics (e.g., Hashino and Tripoli, 2011; Milbrandt and Morrison, 2015), characterization of PSDs as a function of snowfall\n435\nformation processes, or retrievals combining in situ and remote sensing observations. Tracking of a particle in three dimensions\ncan be used to understand the impact of turbulence on particle trajectories. Beyond atmospheric science, the VISSS shows po-\ntential for quantifying the occurrence of flying insects, as standard insect counting techniques (e.g., suction traps) are typically\ndestructive and labor-intensive. physics (e.g., Hashino and Tripoli, 2011; Milbrandt and Morrison, 2015), characterization of PSDs as a function of snowfall\n435\nformation processes, or retrievals combining in situ and remote sensing observations. Tracking of a particle in three dimensions\ncan be used to understand the impact of turbulence on particle trajectories. Beyond atmospheric science, the VISSS shows po-\ntential for quantifying the occurrence of flying insects, as standard insect counting techniques (e.g., suction traps) are typically\ndestructive and labor-intensive. 20 Appendix A: Coordinate system transformation The opposite transformation can be described by: 450 \n\n\n\nx′\nF\ny′\nF\nz′\nF\n\n\n\n=\n\n\n\n\ncosθcosψ\ncosθsinψ\n−sinθ\nsinφsinθcosψ −cosφsinψ\nsinφsinθsinψ + cosφcosψ\nsinφcosθ\ncosφsinθcosψ + sinφsinψ\ncosφsinθsinψ −sinφcosψ\ncosφcosθ\n\n\n\n\n\n\n\n\nxL\nyL\nzL\n\n\n\n\n(A2) (A2) Since we have only one measurement in the x and y dimensions, but two in z, we use the difference between the measured\nzL and the estimated zL from matched particles to retrieve the rotation angles and offsets Since we have only one measurement in the x and y dimensions, but two in z, we use the difference between the measured\nzL and the estimated zL from matched particles to retrieve the rotation angles and offsets (A3) In this equation, x′\nF is unknown so it is derived from In this equation, x′\nF is unknown so it is derived from known so it is derived from ψyL −sinθzL\n(A4) x′\nF = cosθcosψxL + cosθsinψyL −sinθzL\n(A4) x′\nF = cosθcosψxL + cosθsinψyL −sinθzL x′\nF = cosθcosψxL + cosθsinψyL −sinθzL\n(A4) (A4) x′\nF = cosθcosψxL + cosθsinψyL −sinθzL\n(A4)\nwhere, in turn yL is not observed. Therefore, yL is obtained from d. Therefore, yL is obtained from where, in turn yL is not observed. Therefore, yL is obtained from where, in turn yL is not observed. Therefore, yL is obtained from yL = cosθsinψx′\nF + (sinφsinθsinψ + cosφcosψ)y′\nF + (cosφsinθsinψ −sinφcosψ)z′\nF . (A5) yL = cosθsinψx′\nF + (sinφsinθsinψ + cosφcosψ)y′\nF + (cosφsinθsinψ −sinφcosψ)z′\nF . (A5) (A5) Inserting equations A5 into A4 yields after a couple of simplifications\n460 Inserting equations A5 into A4 yields after a couple of simplifications\n460 460 460 x′\nF =\ncosθcosψ\n1 −cos2 θsin2 ψ xL\n+ (cosθsinφsinθsin2 ψ + cosφcosψcosθsinψ)\n1 −cos2 θsin2 ψ\ny′\nF\n+ (cosθcosφsinθsin2 ψ −sinφcosψcosθsinψ)\n1 −cos2 θsin2 ψ\nz′\nF\n−\nsinθ\n1 −cos2 θsin2 ψ zL. (A6) x′\nF =\ncosθcosψ\n1 −cos2 θsin2 ψ xL\n+ (cosθsinφsinθsin2 ψ + cosφcosψcosθsinψ)\n1 −cos2 θsin2 ψ\ny′\nF\n+ (cosθcosφsinθsin2 ψ −sinφcosψcosθsinψ)\n1 −cos2 θsin2 ψ\nz′\nF\n−\nsinθ\n1 −cos2 θsin2 ψ zL. Appendix A: Coordinate system transformation (A6) x′\nF =\ncosθcosψ\n1 −cos2 θsin2 ψ xL\n+ (cosθsinφsinθsin2 ψ + cosφcosψcosθsinψ)\n1 −cos2 θsin2 ψ\ny′\nF\n+ (cosθcosφsinθsin2 ψ −sinφcosψcosθsinψ)\n1 −cos2 θsin2 ψ\nz′\nF\n−\nsinθ\n1 −cos2 θsin2 ψ zL. (A6) 21 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Inserting equations A6 into A3 yields: Inserting equations A6 into A3 yields:\nzL = −\nsinθ\ncosθcosψ xL\n−sinθsinψcosφ −cosψsinφ\ncosθcosψ\ny′\nF\n+sinθsinψsinφ + cosψcosφ\ncosθcosψ\nz′\nF . 465 (A7) 465 Author contributions. MM acquired funding, developed the instrument, processed the VISSS data, analyzed the data of the case study,\nand wrote the manuscript. DM processed PIP and Parsivel data and contributed to data analysis. NM and IS contributed to instrument\ncalibration and particle tracking development, respectively. MS supported funding acquisition and was responsible for the VISSS deployment\nat MOSAiC. All authors reviewed and edited the draft. Competing interests. MM is a member of the editorial board of Atmospheric Measurement Techniques. 470 Competing interests. MM is a member of the editorial board of Atmospheric Measurement Techniques. 470 Acknowledgements. Funded by the German Research Foundation (DFG, Deutsche Forschungsgemeinschaft) Transregional Collaborative\nResearch Center SFB/TRR 172 (Project-ID 268020496), DFG Priority Program SPP2115 “Fusion of Radar Polarimetry and Numerical At-\nmospheric Modelling Towards an Improved Understanding of Cloud and Precipitation Processes” (PROM) under grant PROM-CORSIPP\n(Project-ID 408008112), and the University of Colorado Boulder CIRES (Cooperative Institute for Research in Environmental Sciences) Acknowledgements. Funded by the German Research Foundation (DFG, Deutsche Forschungsgemeinschaft) Transregional Collaborative\nResearch Center SFB/TRR 172 (Project-ID 268020496), DFG Priority Program SPP2115 “Fusion of Radar Polarimetry and Numerical At-\nmospheric Modelling Towards an Improved Understanding of Cloud and Precipitation Processes” (PROM) under grant PROM-CORSIPP\n(Project-ID 408008112), and the University of Colorado Boulder CIRES (Cooperative Institute for Research in Environmental Sciences) Acknowledgements. Funded by the German Research Foundation (DFG, Deutsche Forschungsgemeinschaft) Transregional Collaborative\nResearch Center SFB/TRR 172 (Project-ID 268020496), DFG Priority Program SPP2115 “Fusion of Radar Polarimetry and Numerical At-\nmospheric Modelling Towards an Improved Understanding of Cloud and Precipitation Processes” (PROM) under grant PROM-CORSIPP\n(Project-ID 408008112), and the University of Colorado Boulder CIRES (Cooperative Institute for Research in Environmental Sciences) (Project-ID 408008112), and the University of Colorado Boulder CIRES (Cooperative Institute for Research in Environmental Sciences)\nInnovative Research Program. Appendix A: Coordinate system transformation MDS was supported by the National Science Foundation (OPP-1724551) and National Oceanic and Atmo-\n475\nspheric Administration (NA22OAR4320151). The deployment at Hyytiälä was supported by an ACTRIS-2 TNA funded by the European\nCommission under the Horizon 2020 – Research and Innovation Framework Programme, H2020-INFRADEV-2019-2, Grant Agreement\nnumber: 871115. The PIP deployment at the University of Helsinki station is supported by the NASA Global Precipitation Measurement\nMission ground validation program. We thank all persons involved in the MOSAiC expedition (MOSAiC20192020) of the Research Vessel\nPolarstern during MOSAiC in 2019–2020 (Project ID: AWI_PS122_00) as listed in (Nixdorf et al., 2021), in particular Christopher Cox,\n480\nMichael Gallagher, Jenny Hutchings, and Taneil Uttal. In Hyytiälä, the VISSS was taken care of by Lauri Ahonen, Matti Leskinen, and Anna Innovative Research Program. MDS was supported by the National Science Foundation (OPP-1724551) and National Oceanic and Atmo-\n475\nspheric Administration (NA22OAR4320151). The deployment at Hyytiälä was supported by an ACTRIS-2 TNA funded by the European\nCommission under the Horizon 2020 – Research and Innovation Framework Programme, H2020-INFRADEV-2019-2, Grant Agreement\nnumber: 871115. The PIP deployment at the University of Helsinki station is supported by the NASA Global Precipitation Measurement\nMission ground validation program. We thank all persons involved in the MOSAiC expedition (MOSAiC20192020) of the Research Vessel Innovative Research Program. MDS was supported by the National Science Foundation (OPP-1724551) and National Oceanic and Atmo-\n475\nspheric Administration (NA22OAR4320151). The deployment at Hyytiälä was supported by an ACTRIS-2 TNA funded by the European\nCommission under the Horizon 2020 – Research and Innovation Framework Programme, H2020-INFRADEV-2019-2, Grant Agreement\nnumber: 871115. The PIP deployment at the University of Helsinki station is supported by the NASA Global Precipitation Measurement\nMission ground validation program. We thank all persons involved in the MOSAiC expedition (MOSAiC20192020) of the Research Vessel Polarstern during MOSAiC in 2019–2020 (Project ID: AWI_PS122_00) as listed in (Nixdorf et al., 2021), in particular Christopher Cox,\n480\nMichael Gallagher, Jenny Hutchings, and Taneil Uttal. In Hyytiälä, the VISSS was taken care of by Lauri Ahonen, Matti Leskinen, and Anna\nTrosits. In Ny-Ålesund, the VISSS installation was made possible by the AWIPEV team including Guillaume Hérment, Fieke Rader, and\nWilfried Ruhe. During SAIL, we were supported by the Operations team from Rocky Mountain Biological Laboratory team and the DOE\nAtmospheric Radiation Measurement technicians who took great care of the VISSS. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. References Delanoë, J., Protat, A., Testud, J., Bouniol, D., Heymsfield, A. J., Bansemer, A., Brown, P. R. A., and Forbes, R. M.: Statistical Properties of\n500\nthe Normalized Ice Particle Size Distribution, J. Geophys. Res., 110, D10 201, https://doi.org/10.1029/2004JD005405, 2005. Delanoë, J., Protat, A., Testud, J., Bouniol, D., Heymsfield, A. J., Bansemer, A., Brown, P. R. A., and Forbes, R. M.: Statistical Properties of\n500\nthe Normalized Ice Particle Size Distribution, J. Geophys. Res., 110, D10 201, https://doi.org/10.1029/2004JD005405, 2005. Dunnavan, E. L., Jiang, Z., Harrington, J. Y., Verlinde, J., Fitch, K., and Garrett, T. J.: The Shape and Density Evolution of Snow Aggregates, Delanoë, J., Protat, A., Testud, J., Bouniol, D., Heymsfield, A. J., Bansemer, A., Brown, P. R. A., and Forbes, R. M.: Statistical Properties of\n500\nthe Normalized Ice Particle Size Distribution, J. Geophys. Res., 110, D10 201, https://doi.org/10.1029/2004JD005405, 2005. Dunnavan, E. L., Jiang, Z., Harrington, J. Y., Verlinde, J., Fitch, K., and Garrett, T. J.: The Shape and Density Evolution of Snow Aggregates,\nJ. Atmos. Sci., https://doi.org/10.1175/JAS-D-19-0066.1, 2019. Feldman, D., Aiken, A., Boos, W., Carroll, R., Chandrasekar, V., Collins, W., Collis, S., Deems, J., DeMott, P., and Fan, J.: Surface At- Dunnavan, E. L., Jiang, Z., Harrington, J. Y., Verlinde, J., Fitch, K., and Garrett, T. J.: The Shape and Density Evolution of Snow Aggregates,\nJ. Atmos. Sci., https://doi.org/10.1175/JAS-D-19-0066.1, 2019. Feldman, D., Aiken, A., Boos, W., Carroll, R., Chandrasekar, V., Collins, W., Collis, S., Deems, J., DeMott, P., and Fan, J.: Surface At-\nmosphere Integrated Field Laboratory (Sail) Science Plan, Tech. rep., Oak Ridge National Lab.(ORNL), Oak Ridge, TN (United States). 505\nAtmospheric . . ., 2021. Feldman, D., Aiken, A., Boos, W., Carroll, R., Chandrasekar, V., Collins, W., Collis, S., Deems, J., DeMott, P., and Fan, J.: Surface At- mosphere Integrated Field Laboratory (Sail) Science Plan, Tech. rep., Oak Ridge National Lab.(ORNL), Oak Ridge, TN (United States). 505\nAtmospheric . . ., 2021. mosphere Integrated Field Laboratory (Sail) Science Plan, Tech. rep., Oak Ridge National Lab.(ORNL), Oak Ridge, TN (United States). 505\nAtmospheric . . ., 2021. Field, P. R. and Heymsfield, A. J.: Importance of Snow to Global Precipitation, Geophys. Res. Lett., 42, 2015GL065 497,\nhttps://doi.org/10.1002/2015GL065497, 2015. Fitch K E and Garrett T J : Gra pel Precipitating From Thin Arctic Clo ds With Liq id Water Paths Less Than 50 g M 2 Geoph s Res Atmospheric . . ., 2021. Field, P. R. References Battaglia, A., Rustemeier, E., Tokay, A., Blahak, U., and Simmer, C.: PARSIVEL Snow Observations: A Critical Assessment, J. Atmospheric\nOcean. Technol., 27, 333–344, https://doi.org/10.1175/2009JTECHA1332.1, 2010. Ocean. Technol., 27, 333 344, https://doi.org/10.1175/2009JTECHA1332.1, 2010. Beard, K. V., Bringi, V. N., and Thurai, M.: A New Understanding of Raindrop Shape, Atmospheric Research, 97, 396–415,\nhttps://doi.org/10.1016/j.atmosres.2010.02.001, 2010. 490\nBrandes, E. A., Ikeda, K., Thompson, G., and Schönhuber, M.: Aggregate Terminal Velocity/Temperature Relations, J. Appl. Meteorol. Climatol., 47, 2729–2736, https://doi.org/10.1175/2008JAMC1869.1, 2008. Cooper, S. J., Wood, N. B., and L’Ecuyer, T. S.: A Variational Technique to Estimate Snowfall Rate from Coincident Radar, Snowflake, and\nFall-Speed Observations, Atmos. Meas. Tech., 10, 2557–2571, https://doi.org/10.5194/amt-10-2557-2017, 2017. Beard, K. V., Bringi, V. N., and Thurai, M.: A New Understanding of Raindrop Shape, Atmospheric Research, 97, 396–415,\nhttps://doi.org/10.1016/j.atmosres.2010.02.001, 2010. 0 Brandes, E. A., Ikeda, K., Thompson, G., and Schönhuber, M.: Aggregate Terminal Velocity/Temperature Relations, J. Appl. Meteorol. Climatol., 47, 2729–2736, https://doi.org/10.1175/2008JAMC1869.1, 2008. Cooper, S. J., Wood, N. B., and L’Ecuyer, T. S.: A Variational Technique to Estimate Snowfall Rate from Coincident Radar, Snowflake, and Brandes, E. A., Ikeda, K., Thompson, G., and Schönhuber, M.: Aggregate Terminal Velocity/Temperature Relations, J. Appl. Meteorol. Climatol., 47, 2729–2736, https://doi.org/10.1175/2008JAMC1869.1, 2008. Cooper, S. J., Wood, N. B., and L’Ecuyer, T. S.: A Variational Technique to Estimate Snowfall Rate from Coincident Radar, Snowflake, and\nFall-Speed Observations, Atmos. Meas. Tech., 10, 2557–2571, https://doi.org/10.5194/amt-10-2557-2017, 2017. Cooper, S. J., L’Ecuyer, T. S., Wolff, M. A., Kuhn, T., Pettersen, C., Wood, N. B., Eliasson, S., Schirle, C. E., Shates, J., Hellmuth, F., Engdahl,\n495\nB. J. K., Vásquez-Martín, S., Ilmo, T., and Nygård, K.: Exploring Snowfall Variability through the High-Latitude Measurement of Snowfall\n(HiLaMS) Field Campaign, Bull. Am. Meteorol. Soc., 103, E1762–E1780, https://doi.org/10.1175/BAMS-D-21-0007.1, 2022. Del Guasta, M.: ICE-CAMERA: A Flatbed Scanner to Study Inland Antarctic Polar Precipitation, Atmospheric Meas. Tech., 15, 6521–6544,\nhttps://doi.org/10.5194/amt-15-6521-2022, 2022. B. J. K., Vásquez-Martín, S., Ilmo, T., and Nygård, K.: Exploring Snowfall Variability through the High-Latitude Measurement of Snowfall Ilmo, T., and Nygård, K.: Exploring Snowfall Variability through the High-Latitude Measurement of Snowfall gn, Bull. Am. Meteorol. Soc., 103, E1762–E1780, https://doi.org/10.1175/BAMS-D-21-0007.1, 2022. (HiLaMS) Field Campaign, Bull. Am. Meteorol. Soc., 103, E1762–E1780, https://doi.org/10.1175/BAMS-D-21-0007.1, 2022. Del Guasta, M.: ICE-CAMERA: A Flatbed Scanner to Study Inland Antarctic Polar Precipitation, Atmospheric Meas. Tech., 15, 6521–6544,\nhttps://doi.org/10.5194/amt-15-6521-2022, 2022. Del Guasta, M.: ICE-CAMERA: A Flatbed Scanner to Study Inland Antarctic Polar Precipitation, Atmospheric Meas. Tech., 15, 6521–6544,\nhttps://doi.org/10.5194/amt-15-6521-2022, 2022. Appendix A: Coordinate system transformation Thanks to Donald David, Rainer Haseneder-Lind, Jim Kastengren, Pavel Krobot, and Steffen Wolters for assembling VISSS instruments. 485 22 References and Heymsfield, A. J.: Importance of Snow to Global Precipitation, Geophys. Res. Lett., 42, 2015GL065 497,\nhttps://doi.org/10.1002/2015GL065497, 2015. Field, P. R. and Heymsfield, A. J.: Importance of Snow to Global Precipitation, Geophys. Res. Lett., 42, 2015GL065 497,\nhttps://doi.org/10.1002/2015GL065497, 2015. Fitch, K. E. and Garrett, T. J.: Graupel Precipitating From Thin Arctic Clouds With Liquid Water Paths Less Than 50 g M-2, Geophys. Res. Lett., 49, e2021GL094 075, https://doi.org/10.1029/2021GL094075, 2022. 510 Lett., 49, e2021GL094 075, https://doi.org/10.1029/2021GL094075, 2022. 510\nGarrett, T. J. and Yuter, S. E.: Observed Influence of Riming, Temperature, and Turbulence on the Fallspeed of Solid Precipitation, Geophys. Res. Lett., 41, 6515–6522, https://doi.org/10.1002/2014GL061016, 2014. Garrett, T. J., Fallgatter, C., Shkurko, K., and Howlett, D.: Fall Speed Measurement and High-Resolution Multi-Angle Photography of\nHydrometeors in Free Fall, Atmos. Meas. Tech., 5, 2625–2633, https://doi.org/10.5194/amt-5-2625-2012, 2012. Garrett, T. J. and Yuter, S. E.: Observed Influence of Riming, Temperature, and Turbulence on the Fallspeed of Solid Precipitation, Geophys. Res. Lett., 41, 6515–6522, https://doi.org/10.1002/2014GL061016, 2014. Garrett, T. J., Fallgatter, C., Shkurko, K., and Howlett, D.: Fall Speed Measurement and High-Resolution Multi-Angle Photography of Garrett, T. J. and Yuter, S. E.: Observed Influence of Riming, Temperature, and Turbulence on the Fallspeed of Solid Precipitation, Geophys. Res. Lett., 41, 6515–6522, https://doi.org/10.1002/2014GL061016, 2014. Garrett, T. J., Fallgatter, C., Shkurko, K., and Howlett, D.: Fall Speed Measurement and High-Resolution Multi-Angle Photography of\nHydrometeors in Free Fall, Atmos. Meas. Tech., 5, 2625–2633, https://doi.org/10.5194/amt-5-2625-2012, 2012. T. J., Fallgatter, C., Shkurko, K., and Howlett, D.: Fall Speed Measurement and High-Resolution Multi-An meteors in Free Fall, Atmos. Meas. Tech., 5, 2625–2633, https://doi.org/10.5194/amt-5-2625-2012, 2012. Gergely, M. and Garrett, T. J.: Impact of the Natural Variability in Snowflake Diameter, Aspect Ratio, and Orientation on Modeled Snowfall\n515\nRadar Reflectivity, J. Geophys. Res. Atmos., 121, 2016JD025 192, https://doi.org/10.1002/2016JD025192, 2016. Gergely, M. and Garrett, T. J.: Impact of the Natural Variability in Snowflake Diameter, Aspect Ratio, and Orientation on Modeled Snowfall\n515\nRadar Reflectivity, J. Geophys. Res. Atmos., 121, 2016JD025 192, https://doi.org/10.1002/2016JD025192, 2016. Gergely, M., Cooper, S. J., and Garrett, T. J.: Using Snowflake Surface-Area-to-Volume Ratio to Model and Interpret Snowfall Triple-\nFrequency Radar Signatures, Atmospheric Chem. Phys., 17, 12 011–12 030, https://doi.org/10.5194/acp-17-12011-2017, 2017. Grazioli, J., Tuia, D., Monhart, S., Schneebeli, M., Raupach, T., and Berne, A.: Hydrometeor Classification from Two-Dimensional Video Gergely, M., Cooper, S. J., and Garrett, T. J.: Using Snowflake Surface-Area-to-Volume Ratio to Model and Interpret Snowfall Triple-\nFrequency Radar Signatures, Atmospheric Chem. Kennedy, A., Scott, A., Loeb, N., Sczepanski, A., Lucke, K., Marquis, J., and Waugh, S.: Bringing Microphysics to the Masses: The Blowing\nSnow Observations at the University of North Dakota: Education through Research (BLOWN-UNDER) Campaign, Bull. Am. Meteorol. Soc., 103, E83–E100, https://doi.org/10.1175/BAMS-D-20-0199.1, 2022. 535 Soc., 103, E83–E100, https://doi.org/10.1175/BAMS-D-20-0199.1, 2022. 535\nKeppas, S. Ch., Crosier, J., Choularton, T. W., and Bower, K. N.: Ice Lollies: An Ice Particle Generated in Supercooled Conveyor Belts,\nGeophys. Res. Lett., 44, 5222–5230, https://doi.org/10.1002/2017GL073441, 2017. Kuhn, T. and Vázquez-Martín, S.: Microphysical Properties and Fall Speed Measurements of Snow Ice Crystals Using the Dual Ice Crystal\nImager (D-ICI), Atmospheric Meas. Tech., 13, 1273–1285, https://doi.org/10.5194/amt-13-1273-2020, 2020. Keppas, S. Ch., Crosier, J., Choularton, T. W., and Bower, K. N.: Ice Lollies: An Ice Particle Generated in Supercooled Conveyor Belts,\nGeophys. Res. Lett., 44, 5222–5230, https://doi.org/10.1002/2017GL073441, 2017. Kuhn, T. and Vázquez-Martín, S.: Microphysical Properties and Fall Speed Measurements of Snow Ice Crystals Using the Dual Ice Crystal\nImager (D-ICI), Atmospheric Meas. Tech., 13, 1273–1285, https://doi.org/10.5194/amt-13-1273-2020, 2020. uhn, T. and Vázquez-Martín, S.: Microphysical Properties and Fall Speed Measurements of Snow Ice Crystals Imager (D-ICI), Atmospheric Meas. Tech., 13, 1273–1285, https://doi.org/10.5194/amt-13-1273-2020, 2020. Kulie, M. S., Pettersen, C., Merrelli, A. J., Wagner, T. J., Wood, N. B., Dutter, M., Beachler, D., Kluber, T., Turner, R., Mateling, M.,\n540 Kulie, M. S., Pettersen, C., Merrelli, A. J., Wagner, T. J., Wood, N. B., Dutter, M., Beachler, D., Kluber, T., Turner, R., Mateling, M.,\n540\nLenters, J., Blanken, P., Maahn, M., Spence, C., Kneifel, S., Kucera, P. A., Tokay, A., Bliven, L. F., Wolff, D. B., and Petersen,\nW. A.: Snowfall in the Northern Great Lakes: Lessons Learned from a Multi-Sensor Observatory, Bull. Amer. Meteor. Soc., -1, 1–61,\nhttps://doi.org/10.1175/BAMS-D-19-0128.1, 2021. Leinonen, J. and Berne, A.: Unsupervised Classification of Snowflake Images Using a Generative Adversarial Network and K-medoids W. A.: Snowfall in the Northern Great Lakes: Lessons Learned from a Multi-Sensor Observatory, Bull. Amer. Meteor. Soc., -1, 1–61,\nhttps://doi.org/10.1175/BAMS-D-19-0128.1, 2021. https://doi.org/10.1175/BAMS-D-19-0128.1, 2021. Leinonen, J. and Berne, A.: Unsupervised Classification of Snowflake Images Using a Generative Adversarial Network and K-medoids Leinonen, J. and Berne, A.: Unsupervised Classification of Snowflake Images Using a Generative Adversarial Network and K-medoidsi Leinonen, J. and Berne, A.: Unsupervised Classification of Snowflake Images Using a Generative Ad Leinonen, J. and Berne, A.: Unsupervised Classification of Snowflake Images Using a Generative Adversarial Network and K-medoids\nClassification, Atmos. Meas. Tech., 13, 2949–2964, https://doi.org/10.5194/amt-13-2949-2020, 2020. References Phys., 17, 12 011–12 030, https://doi.org/10.5194/acp-17-12011-2017, 2017. Grazioli, J., Tuia, D., Monhart, S., Schneebeli, M., Raupach, T., and Berne, A.: Hydrometeor Classification from Two-Dimensional Video\nDisdrometer Data, Atmos. Meas. Tech., 7, 2869–2882, https://doi.org/10.5194/amt-7-2869-2014, 2014. 520 razioli, J., Tuia, D., Monhart, S., Schneebeli, M., Raupach, T., and Berne, A.: Hydrometeor Classification fro Disdrometer Data, Atmos. Meas. Tech., 7, 2869–2882, https://doi.org/10.5194/amt-7-2869-2014, 2014. 520\nGrazioli, J., Genthon, C., Boudevillain, B., Duran-Alarcon, C., Guasta, M. D., Madeleine, J.-B., and Berne, A.: Measurements of Precipitation\nin Dumont d’Urville, Adélie Land, East Antarctica, Cryosphere, 11, 1797–1811, https://doi.org/10.5194/tc-11-1797-2017, 2017. Grazioli, J., Genthon, C., Boudevillain, B., Duran-Alarcon, C., Guasta, M. D., Madeleine, J.-B., and Berne, A.: Measurements of Precipitation\nin Dumont d’Urville, Adélie Land, East Antarctica, Cryosphere, 11, 1797–1811, https://doi.org/10.5194/tc-11-1797-2017, 2017. 23 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Hashino, T. and Tripoli, G. J.: The Spectral Ice Habit Prediction System (SHIPS). Part III: Description of the Ice Particle Model and the\nHabit-Dependent Aggregation Model, J. Atmospheric Sci., 68, 1125–1141, https://doi.org/10.1175/2011JAS3666.1, 2011. Helms, C. N., Munchak, S. J., Tokay, A., and Pettersen, C.: A Comparative Evaluation of Snowflake Particle Shape Estimation Techniques\n525\nUsed by the Precipitation Imaging Package (PIP), Multi-Angle Snowflake Camera (MASC), and Two-Dimensional Video Disdrometer\n(2DVD), Atmospheric Meas. Tech., 15, 6545–6561, https://doi.org/10.5194/amt-15-6545-2022, 2022. Hicks A and Notaroš B M : Method for Classification of Snowflakes Based on Images by a Multi Angle Snowflake Camera Using Con Helms, C. N., Munchak, S. J., Tokay, A., and Pettersen, C.: A Comparative Evaluation of Snowflake Particle Shape Estimation Techniques\n525\nUsed by the Precipitation Imaging Package (PIP), Multi-Angle Snowflake Camera (MASC), and Two-Dimensional Video Disdrometer\n(2DVD), Atmospheric Meas. Tech., 15, 6545–6561, https://doi.org/10.5194/amt-15-6545-2022, 2022. Hicks, A. and Notaroš, B. M.: Method for Classification of Snowflakes Based on Images by a Multi-Angle Snowflake Camera Using Con-\nvolutional Neural Networks, J. Atmos. Oceanic Technol., 36, 2267–2282, https://doi.org/10.1175/JTECH-D-19-0055.1, 2019. Hicks, A. and Notaroš, B. M.: Method for Classification of Snowflakes Based on Images by a Multi-Angle Snowflake Camera Using Con-\nvolutional Neural Networks, J. Atmos. Oceanic Technol., 36, 2267–2282, https://doi.org/10.1175/JTECH-D-19-0055.1, 2019. Huang, G.-J., Bringi, V. N., Moisseev, D., Petersen, W. A., Bliven, L., and Hudak, D.: Use of 2D-video Disdrometer to Derive Mean\n530\nDensity–Size and Ze–SR Relations: Four Snow Cases from the Light Precipitation Validation Experiment, Atmos. Res., 153, 34–48,\nhttps://doi.org/10.1016/j.atmosres.2014.07.013, 2015. Kennedy A Scott A Loeb N Sczepanski A Lucke K Marquis J and Waugh S : Bringing Microphysics to the Masses: The Blowing Huang, G.-J., Bringi, V. N., Moisseev, D., Petersen, W. A., Bliven, L., and Hudak, D.: Use of 2D-video Disdrometer to Derive Mean\n530\nDensity–Size and Ze–SR Relations: Four Snow Cases from the Light Precipitation Validation Experiment, Atmos. Res., 153, 34–48,\nhttps://doi.org/10.1016/j.atmosres.2014.07.013, 2015. Kennedy, A., Scott, A., Loeb, N., Sczepanski, A., Lucke, K., Marquis, J., and Waugh, S.: Bringing Microphysics to the Masses: The Blowing\nSnow Observations at the University of North Dakota: Education through Research (BLOWN-UNDER) Campaign, Bull. Am. Meteorol. 545 Classification, Atmos. Meas. Tech., 13, 2949–2964, https://doi.org/10.5194/amt-13-2949-2020, 2020. 545\nLeinonen, J., Grazioli, J., and Berne, A.: Reconstruction of the Mass and Geometry of Snowfall Particles from Multi-Angle Snowflake\nCamera (MASC) Images, Atmos. Meas. Tech., 14, 6851–6866, https://doi.org/10.5194/amt-14-6851-2021, 2021. Li, H., Moisseev, D., and von Lerber, A.: How Does Riming Affect Dual-Polarization Radar Observations and Snowflake Shape?, J. Geophys. Res. Atmospheres, 123, 6070–6081, https://doi.org/10.1029/2017JD028186, 2018. Classification, Atmos. Meas. Tech., 13, 2949–2964, https://doi.org/10.5194/amt-13-2949-2020, 2020. 545\nLeinonen, J., Grazioli, J., and Berne, A.: Reconstruction of the Mass and Geometry of Snowfall Particles from Multi-Angle Snowflake\nCamera (MASC) Images, Atmos. Meas. Tech., 14, 6851–6866, https://doi.org/10.5194/amt-14-6851-2021, 2021.l Leinonen, J., Grazioli, J., and Berne, A.: Reconstruction of the Mass and Geometry of Snowfall Particles from Multi-Angle Snowflake\nCamera (MASC) Images, Atmos. Meas. Tech., 14, 6851–6866, https://doi.org/10.5194/amt-14-6851-2021, 2021. Li, H., Moisseev, D., and von Lerber, A.: How Does Riming Affect Dual-Polarization Radar Observations and Snowflake Shape?, J. Geophys. Res. Atmospheres, 123, 6070–6081, https://doi.org/10.1029/2017JD028186, 2018. Li, J., Abraham, A., Guala, M., and Hong, J.: Evidence of Preferential Sweeping during Snow Settling in Atmospheric Turbulence,\n550\nArXiv210904662 Phys., 2021. Löffler-Mang, M. and Joss, J.: An Optical Disdrometer for Measuring Size and Velocity of Hydrometeors, J. Atmos. Oceanic Technol., 17,\n130–139, https://doi.org/10.1175/1520-0426(2000)017<0130:AODFMS>2.0.CO;2, 2000. Lopez-Cantu, T., Prein, A. F., and Samaras, C.: Uncertainties in Future U.S. Extreme Precipitation From Downscaled Climate Projections, Li, J., Abraham, A., Guala, M., and Hong, J.: Evidence of Preferential Sweeping during Snow Settling in Atmospheric Turbulence,\n550\nArXiv210904662 Phys., 2021. ffl\nd\nA O i\nl\ni d\nf\ni\nSi\nd\nl\ni\nf\nd\nA\nO\ni\nh\nl 1 ArXiv210904662 Phys., 2021. Löffler-Mang, M. and Joss, J.: An Optical Disdrometer for Measuring Size and Velocity of Hydrometeors, J. Atmos. Oceanic Technol., 17,\n130–139, https://doi.org/10.1175/1520-0426(2000)017<0130:AODFMS>2.0.CO;2, 2000. Löffler-Mang, M. and Joss, J.: An Optical Disdrometer for Measuring Size and Velocity of Hydrometeors, J. Atmos. Oceanic Technol., 17,\n130–139, https://doi.org/10.1175/1520-0426(2000)017<0130:AODFMS>2.0.CO;2, 2000. Lopez-Cantu, T., Prein, A. F., and Samaras, C.: Uncertainties in Future U.S. Extreme Precipitation From Downscaled Climate Projections,\nGeophys Res Lett 47 e2019GL086 797 https://doi org/10 1029/2019GL086797 2020\n555 Lopez-Cantu, T., Prein, A. F., and Samaras, C.: Uncertainties in Future U.S. Extreme Precipitation From Downscaled Climate Projections,\nGeophys. Res. Lett., 47, e2019GL086 797, https://doi.org/10.1029/2019GL086797, 2020. 555 Geophys. Res. Lett., 47, e2019GL086 797, https://doi.org/10.1029/2019GL086797, 2020. 555\nLuke, E. P., Yang, F., Kollias, P., Vogelmann, A. 570\nMaahn, M., Haseneder-Lind, R., and Krobot, P.: Hardware Design of the Video In Situ Snowfall Sensor v2 (VISSS2), Maahn, M., Turner, D. D., Löhnert, U., Posselt, D. J., Ebell, K., Mace, G. G., and Comstock, J. M.: Optimal Estimation Re-\ntrievals and Their Uncertainties: What Every Atmospheric Scientist Should Know, Bull. Amer. Meteor. Soc., 101, E1512–E1523,\nhttps://doi.org/10.1175/BAMS-D-19-0027.1, 2020. 570 https://doi.org/10.1175/BAMS-D-19-0027.1, 2020. 570\nMaahn, M., Haseneder-Lind, R., and Krobot, P.: Hardware Design of the Video In Situ Snowfall Sensor v2 (VISSS2),\nhttps://doi.org/10.5281/zenodo.7640821, 2023. Maherndl, N., Maahn, M., Tridon, F., Leinonen, J., Ori, D., and Kneifel, S.: A Riming-Dependent Parameterization of Scattering by\nSnowflakes Using the Self-Similar Rayleigh–Gans Approximation, Q.J.R. Meteorol. Soc., in revision, 28, 2023. Maahn, M., Haseneder-Lind, R., and Krobot, P.: Hardware Design of the Video In Situ Snowfall Sensor v2 (VISSS2),\nhttps://doi.org/10.5281/zenodo.7640821, 2023. https://doi.org/10.5281/zenodo.7640821, 2023. Maherndl, N., Maahn, M., Tridon, F., Leinonen, J., Ori, D., and Kneifel, S.: A Riming-Dependent Parameterization of Scattering by\nSnowflakes Using the Self-Similar Rayleigh–Gans Approximation, Q.J.R. Meteorol. Soc., in revision, 28, 2023. Maherndl, N., Maahn, M., Tridon, F., Leinonen, J., Ori, D., and Kneifel, S.: A Riming-Dependent Parameterization of Scattering by\nSnowflakes Using the Self-Similar Rayleigh–Gans Approximation, Q.J.R. Meteorol. Soc., in revision, 28, 2023. Matrosov, S. Y., Ryzhkov, A. V., Maahn, M., and de Boer, G.: Hydrometeor Shape Variability in Snowfall as Retrieved from Polarimetric\n575\nRadar Measurements, J. Appl. Meteor. Climatol., 59, 1503–1517, https://doi.org/10.1175/JAMC-D-20-0052.1, 2020. Mech, M., Maahn, M., Kneifel, S., Ori, D., Orlandi, E., Kollias, P., Schemann, V., and Crewell, S.: PAMTRA 1.0: The Passive and Active\nMicrowave Radiative TRAnsfer Tool for Simulating Radiometer and Radar Measurements of the Cloudy Atmosphere, Geosci. Model\nDev., 13, 4229–4251, https://doi.org/10.5194/gmd-13-4229-2020, 2020. Matrosov, S. Y., Ryzhkov, A. V., Maahn, M., and de Boer, G.: Hydrometeor Shape Variability in Snowfall as Retrieved from Polarimetric\n575\nRadar Measurements, J. Appl. Meteor. Climatol., 59, 1503–1517, https://doi.org/10.1175/JAMC-D-20-0052.1, 2020. Mech, M., Maahn, M., Kneifel, S., Ori, D., Orlandi, E., Kollias, P., Schemann, V., and Crewell, S.: PAMTRA 1.0: The Passive and Active\nMicrowave Radiative TRAnsfer Tool for Simulating Radiometer and Radar Measurements of the Cloudy Atmosphere, Geosci. Model\nDev., 13, 4229–4251, https://doi.org/10.5194/gmd-13-4229-2020, 2020. Milbrandt, J. A. and Morrison, H.: Parameterization of Cloud Microphysics Based on the Prediction of Bulk Ice Particle Properties. Part III:\n580\nIntroduction of Multiple Free Categories, J. Atmos. Sci., 73, 975–995, https://doi.org/10.1175/JAS-D-15-0204.1, 2015. Milbrandt, J. A. M., and Maahn, M.: New Insights into Ice Multiplication Using Remote-Sensing Observa-\ntions of Slightly Supercooled Mixed-Phase Clouds in the Arctic, PNAS, 118, https://doi.org/10.1073/pnas.2021387118, 2021. Lumb, F. E.: Relation between the Terminal Velocity and Dimensions of Snowflakes, Meteor. Mag, 90, 344–347, 1961. Maahn, M.: Video In Situ Snowfall Sensor (VISSS) Data Acquisition Software V0.3.1, Zenodo, https://doi.org/10.5281/zenodo.7640801,\n2023a. 560 Geophys. Res. Lett., 47, e2019GL086 797, https://doi.org/10.1029/2019GL086797, 2020. 555\nLuke, E. P., Yang, F., Kollias, P., Vogelmann, A. M., and Maahn, M.: New Insights into Ice Multiplication Using Remote-Sensing Observa-\ntions of Slightly Supercooled Mixed-Phase Clouds in the Arctic, PNAS, 118, https://doi.org/10.1073/pnas.2021387118, 2021. Lumb, F. E.: Relation between the Terminal Velocity and Dimensions of Snowflakes, Meteor. Mag, 90, 344–347, 1961. Luke, E. P., Yang, F., Kollias, P., Vogelmann, A. M., and Maahn, M.: New Insights into Ice Multiplication Using Remote-Sensing Observa-\ntions of Slightly Supercooled Mixed Phase Clouds in the Arctic PNAS 118 https://doi org/10 1073/pnas 2021387118 2021 upercooled Mixed-Phase Clouds in the Arctic, PNAS, 118, https://doi.org/10.1073/pnas.2021387118, 2021. Relation between the Terminal Velocity and Dimensions of Snowflakes, Meteor. Mag, 90, 344–347, 1961. Maahn, M.: Video In Situ Snowfall Sensor (VISSS) Data Acquisition Software V0.3.1, Zenodo, https://doi.org/10.5281/zenodo.7640801,\n2023a. 560 24 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Maahn, M.: Video In Situ Snowfall Sensor (VISSS) Data Processing Library V2023.1.6, Zenodo, https://doi.org/10.5281/zenodo.7650394,\n2023b. Maahn, M.: Video In Situ Snowfall Sensor (VISSS) Data Processing Library V2023.1.6, Zenodo, https://doi.org/10.5281/zenodo.7650394,\n2023b. Maahn, M. and Moisseev, D.: VISSS, PIP, and Parsivel Snowfall Observations from Winter 2021/22 in Hyytiälä, Finland, Maahn, M.: Video In Situ Snowfall Sensor (VISSS) Data Processing Library V2023.1.6, Zenodo, https://doi.org/10.5281/zenodo.7650394,\n2023b. Maahn, M. and Moisseev, D.: VISSS, PIP, and Parsivel Snowfall Observations from Winter 2021/22 in Hyytiälä, Finland,\nhttps://doi.org/10.5281/zenodo.7797286, 2023. Maahn, M. and Moisseev, D.: VISSS, PIP, and Parsivel Snowfall Observations from Winter 2021/22 in Hyytiälä, Finland,\nhttps://doi.org/10.5281/zenodo.7797286, 2023. Maahn, M., Löhnert, U., Kollias, P., Jackson, R. C., and McFarquhar, G. M.: Developing and Evaluating Ice Cloud Parameteriza-\n565\ntions for Forward Modeling of Radar Moments Using in Situ Aircraft Observations, J. Atmos. Oceanic Technol., 32, 880–903,\nhttps://doi.org/10.1175/JTECH-D-14-00112.1, 2015. Maahn, M., Turner, D. D., Löhnert, U., Posselt, D. J., Ebell, K., Mace, G. G., and Comstock, J. M.: Optimal Estimation Re-\ntrievals and Their Uncertainties: What Every Atmospheric Scientist Should Know, Bull. Amer. Meteor. Soc., 101, E1512–E1523,\nhttps://doi.org/10.1175/BAMS-D-19-0027.1, 2020. 570\nMaahn, M., Haseneder-Lind, R., and Krobot, P.: Hardware Design of the Video In Situ Snowfall Sensor v2 (VISSS2),\nhttps://doi.org/10.5281/zenodo.7640821, 2023. Maherndl, N., Maahn, M., Tridon, F., Leinonen, J., Ori, D., and Kneifel, S.: A Riming-Dependent Parameterization of Scattering by\nS\nfl k\nU i\nth S lf Si il\nR\nl i h G\nA\ni\nti\nQ J R M t\nl S\ni\ni i\n28 2023 Maahn, M., Löhnert, U., Kollias, P., Jackson, R. C., and McFarquhar, G. M.: Developing and Evaluating Ice Cloud Parameteriza-\n565\ntions for Forward Modeling of Radar Moments Using in Situ Aircraft Observations, J. Atmos. Oceanic Technol., 32, 880–903,\nhttps://doi.org/10.1175/JTECH-D-14-00112.1, 2015. q\np g\ng\ntions for Forward Modeling of Radar Moments Using in Situ Aircraft Observations, J. Atmos. Oceanic Technol., 32, 880–903,\nhttps://doi.org/10.1175/JTECH-D-14-00112.1, 2015. Maahn, M., Turner, D. D., Löhnert, U., Posselt, D. J., Ebell, K., Mace, G. G., and Comstock, J. M.: Optimal Estimation Re-\ntrievals and Their Uncertainties: What Every Atmospheric Scientist Should Know, Bull. Amer. Meteor. Soc., 101, E1512–E1523,\nhttps://doi.org/10.1175/BAMS-D-19-0027.1, 2020. and Morrison, H.: Parameterization of Cloud Microphysics Based on the Prediction of Bulk Ice Particle Properties. Part III:\n580\nIntroduction of Multiple Free Categories, J. Atmos. Sci., 73, 975–995, https://doi.org/10.1175/JAS-D-15-0204.1, 2015. Mitchell, D. L.: Use of Mass- and Area-Dimensional Power Laws for Determining Precipitation Particle Terminal Velocities, J. Atmos. Sci.,\n53, 1710–1723, https://doi.org/10.1175/1520-0469(1996)053<1710:UOMAAD>2.0.CO;2, 1996. Moisseev, D., von Lerber, A., and Tiira, J.: Quantifying the Effect of Riming on Snowfall Using Ground-Based Observations, J. Geophys. Mitchell, D. L.: Use of Mass- and Area-Dimensional Power Laws for Determining Precipitation Particle Terminal Velocities, J. Atmos. Sci.,\n53, 1710–1723, https://doi.org/10.1175/1520-0469(1996)053<1710:UOMAAD>2.0.CO;2, 1996. Moisseev, D., von Lerber, A., and Tiira, J.: Quantifying the Effect of Riming on Snowfall Using Ground-Based Observations, J. Geophys. Res. Atmos., 122, 2016JD026 272, https://doi.org/10.1002/2016JD026272, 2017. 585 Moisseev, D., von Lerber, A., and Tiira, J.: Quantifying the Effect of Riming on Snowfall Using Ground-B Res. Atmos., 122, 2016JD026 272, https://doi.org/10.1002/2016JD026272, 2017. 585\nMorrison, H., van Lier-Walqui, M., Fridlind, A. M., Grabowski, W. W., Harrington, J. Y., Hoose, C., Korolev, A., Kumjian, M. R.,\nMilbrandt, J. A., Pawlowska, H., Posselt, D. J., Prat, O. P., Reimel, K. J., Shima, S.-I., van Diedenhoven, B., and Xue, L.:\nConfronting the Challenge of Modeling Cloud and Precipitation Microphysics, J. Adv. Model. Earth Syst., 12, e2019MS001 689,\nhttps://doi.org/10.1029/2019MS001689, 2020. Mülmenstädt, J., Sourdeval, O., Delanoë, J., and Quaas, J.: Frequency of Occurrence of Rain from Liquid-, Mixed-, and Ice-Phase Clouds\n590\nDerived from A-Train Satellite Retrievals, Geophys. Res. Lett., p. 2015GL064604, https://doi.org/10.1002/2015GL064604, 2015. Newman, A. J., Kucera, P. A., and Bliven, L. F.: Presenting the Snowflake Video Imager (SVI), J. Atmos. Oceanic Technol., 26, 167–179,\nhttps://doi org/10 1175/2008JTECHA1148 1 2009 Mülmenstädt, J., Sourdeval, O., Delanoë, J., and Quaas, J.: Frequency of Occurrence of Rain from Liquid-, Mixed-, and Ice-Phase Clouds\n590\nDerived from A-Train Satellite Retrievals, Geophys. Res. Lett., p. 2015GL064604, https://doi.org/10.1002/2015GL064604, 2015. Mülmenstädt, J., Sourdeval, O., Delanoë, J., and Quaas, J.: Frequency of Occurrence of Rain from Liquid-, Mixed-, and Ice-Phase Clouds\n590\nDerived from A-Train Satellite Retrievals, Geophys. Res. Lett., p. 2015GL064604, https://doi.org/10.1002/2015GL064604, 2015. Newman, A. J., Kucera, P. A., and Bliven, L. F.: Presenting the Snowflake Video Imager (SVI), J. Atmos. Oceanic Technol., 26, 167–179,\nhttps://doi.org/10.1175/2008JTECHA1148.1, 2009. Nixdorf, U., Dethloff, K., Rex, M., Shupe, M., Sommerfeld, A., Perovich, D. K., Nicolaus, M., Heuzé, C., Rabe, B., Loose, B., Damm, E., Derived from A-Train Satellite Retrievals, Geophys. Res. Lett., p. 2015GL064604, https://doi.org/10.1002/2015GL064604, 2015. Newman, A. Petäjä, T., O’Connor, E. J., Moisseev, D., Sinclair, V. A., Manninen, A. J., Väänänen, R., von Lerber, A., Thornton, J. A., Nicoll, K., Petersen,\nW., Chandrasekar, V., Smith, J. N., Winkler, P. M., Krüger, O., Hakola, H., Timonen, H., Brus, D., Laurila, T., Asmi, E., Riekkola, M.-L.,\nMona, L., Massoli, P., Engelmann, R., Komppula, M., Wang, J., Kuang, C., Bäck, J., Virtanen, A., Levula, J., Ritsche, M., and Hickmon, , T., O’Connor, E. J., Moisseev, D., Sinclair, V. A., Manninen, A. J., Väänänen, R., von Lerber, A., Thornton, J. W., Chandrasekar, V., Smith, J. N., Winkler, P. M., Krüger, O., Hakola, H., Timonen, H., Brus, D., Laurila, T., Asmi, E., Riekkola, M.-L., na, L., Massoli, P., Engelmann, R., Komppula, M., Wang, J., Kuang, C., Bäck, J., Virtanen, A., Levula, J., Rit Mona, L., Massoli, P., Engelmann, R., Komppula, M., Wang, J., Kuang, C., Bäck, J., Virtanen, A., Levula, J., Ritsche, M., and Hickmon,\nN.: BAECC: A Field Campaign to Elucidate the Impact of Biogenic Aerosols on Clouds and Climate, Bull. Am. Meteorol. Soc., 97,\n610\n1909–1928 https://doi org/10 1175/BAMS-D-14-00199 1 2016 610 N.: BAECC: A Field Campaign to Elucidate the Impact of Biogenic Aerosols on Clouds and Climate, Bull. Am. Meteorol. Soc., 97,\n610\n1909–1928, https://doi.org/10.1175/BAMS-D-14-00199.1, 2016. N.: BAECC: A Field Campaign to Elucidate the Impact of Biogenic Aerosols on Clouds and Climate, Bull. Am. Meteorol. Soc., 97,\n610\n1909–1928, https://doi.org/10.1175/BAMS-D-14-00199.1, 2016. Pettersen, C., Bliven, L. F., von Lerber, A., Wood, N. B., Kulie, M. S., Mateling, M. E., Moisseev, D. N., Munchak, S. J., Petersen, W. A.,\nand Wolff, D. B.: The Precipitation Imaging Package: Assessment of Microphysical and Bulk Characteristics of Snow, Atmosphere, 11,\n785, https://doi.org/10.3390/atmos11080785, 2020. Pettersen, C., Bliven, L. F., von Lerber, A., Wood, N. B., Kulie, M. S., Mateling, M. E., Moisseev, D. N., Munchak, S. J., Petersen, W. A.,\nand Wolff, D. B.: The Precipitation Imaging Package: Assessment of Microphysical and Bulk Characteristics of Snow, Atmosphere, 11,\n785, https://doi.org/10.3390/atmos11080785, 2020. Praz, C., Roulet, Y.-A., and Berne, A.: Solid Hydrometeor Classification and Riming Degree Estimation from Pictures Collected with a\n615\nMulti-Angle Snowflake Camera, Atmos. Meas. Tech., 10, 1335–1357, https://doi.org/10.5194/amt-10-1335-2017, 2017. Praz, C., Roulet, Y.-A., and Berne, A.: Solid Hydrometeor Classification and Riming Degree Estimation from Pictures Collected with a\n615\nMulti-Angle Snowflake Camera, Atmos. Meas. Tech., 10, 1335–1357, https://doi.org/10.5194/amt-10-1335-2017, 2017. Quante, L., Willner, S. J., Kucera, P. A., and Bliven, L. F.: Presenting the Snowflake Video Imager (SVI), J. Atmos. Oceanic Technol., 26, 167–179,\nhttps://doi.org/10.1175/2008JTECHA1148.1, 2009. Nixdorf, U., Dethloff, K., Rex, M., Shupe, M., Sommerfeld, A., Perovich, D. K., Nicolaus, M., Heuzé, C., Rabe, B., Loose, B., Damm, E., Newman, A. J., Kucera, P. A., and Bliven, L. F.: Presenting the Snowflake Video Imager (SVI), J. Atmos. Oceanic Technol., 26, 167–179,\nhttps://doi.org/10.1175/2008JTECHA1148.1, 2009. Nixdorf, U., Dethloff, K., Rex, M., Shupe, M., Sommerfeld, A., Perovich, D. K., Nicolaus, M., Heuzé, C., R Gradinger, R., Fong, A., Maslowski, W., Rinke, A., Kwok, R., Spreen, G., Wendisch, M., Herber, A., Hirsekorn, M., Mohaupt, V., Frick-\n595\nenhaus, S., Immerz, A., Weiss-Tuider, K., König, B., Mengedoht, D., Regnery, J., Gerchow, P., Ransby, D., Krumpen, T., Morgenstern,\nA., Haas, C., Kanzow, T., Rack, F. R., Saitzev, V., Sokolov, V., Makarov, A., Schwarze, S., Wunderlich, T., Wurr, K., and Boetius, A.:\nMOSAiC Extended Acknowledgement, https://doi.org/10.5281/zenodo.5541624, 2021. 25 https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Nomokonova, T., Ebell, K., Löhnert, U., Maturilli, M., Ritter, C., and O’Connor, E.: Statistics on Clouds and Their Relation\nto Thermodynamic Conditions at Ny-Ålesund Using Ground-Based Sensor Synergy, Atmospheric Chem. Phys., 19, 4105–4126, Nomokonova, T., Ebell, K., Löhnert, U., Maturilli, M., Ritter, C., and O’Connor, E.: Statistics on Clouds and Their Relation\nto Thermodynamic Conditions at Ny-Ålesund Using Ground-Based Sensor Synergy, Atmospheric Chem. Phys., 19, 4105–4126,\nhttps://doi.org/10.5194/acp-19-4105-2019, 2019. Nomokonova, T., Ebell, K., Löhnert, U., Maturilli, M., Ritter, C., and O’Connor, E.: Statistics on Clouds and Their Relation omokonova, T., Ebell, K., Löhnert, U., Maturilli, M., Ritter, C., and O’Connor, E.: Statistics on C 600 to Thermodynamic Conditions at Ny-Ålesund Using Ground-Based Sensor Synergy, Atmospheric Chem. Phys., 19, 4105–4126,\n0\nhttps://doi.org/10.5194/acp-19-4105-2019, 2019. https://doi.org/10.5194/acp-19-4105-2019, 2019. Nurzy´nska, K., Kubo, M., and Muramoto, K.-i.: Shape Parameters for Automatic Classification of Snow Particles into Snowflake and Grau\npel, Meteor. Appl., 20, 257–265, https://doi.org/10.1002/met.299, 2013. Nurzy´nska, K., Kubo, M., and Muramoto, K.-i.: Shape Parameters for Automatic Classification of Snow Particles into Snowflake and Grau-\npel, Meteor. Appl., 20, 257–265, https://doi.org/10.1002/met.299, 2013. Pasquier, J. T., Henneberger, J., Korolev, A., Ramelli, F., Wieder, J., Lauber, A., Li, G., David, R. O., Carlsen, T., Gierens, R., Maturilli, M.,\nand Lohmann, U.: Understanding the History of Two Complex Ice Crystal Habits Deduced From a Holographic Imager, Geophys. Res. 605\nLett., 50, e2022GL100 247, https://doi.org/10.1029/2022GL100247, 2023. Pasquier, J. T., Henneberger, J., Korolev, A., Ramelli, F., Wieder, J., Lauber, A., Li, G., David, R. O., Carlsen, T., Gierens, R., Maturilli, M.,\nand Lohmann, U.: Understanding the History of Two Complex Ice Crystal Habits Deduced From a Holographic Imager, Geophys. Res. 605\nLett., 50, e2022GL100 247, https://doi.org/10.1029/2022GL100247, 2023. Petäjä, T., O’Connor, E. J., Moisseev, D., Sinclair, V. A., Manninen, A. J., Väänänen, R., von Lerber, A., Thornton, J. A., Nicoll, K., Petersen,\nW., Chandrasekar, V., Smith, J. N., Winkler, P. M., Krüger, O., Hakola, H., Timonen, H., Brus, D., Laurila, T., Asmi, E., Riekkola, M.-L., Pasquier, J. T., Henneberger, J., Korolev, A., Ramelli, F., Wieder, J., Lauber, A., Li, G., David, R. O., Carlsen, T., Gierens, R., Maturilli, M., 605 and Lohmann, U.: Understanding the History of Two Complex Ice Crystal Habits Deduced From a Holographic Imager, Geophys. Res. 605\nLett., 50, e2022GL100 247, https://doi.org/10.1029/2022GL100247, 2023. N., Middelanis, R., and Levermann, A.: Regions of Intensification of Extreme Snowfall under Future Warming, Sci Praz, C., Roulet, Y.-A., and Berne, A.: Solid Hydrometeor Classification and Riming Degree Estimation from Pictures Collected with a\n615\nMulti-Angle Snowflake Camera, Atmos. Meas. Tech., 10, 1335–1357, https://doi.org/10.5194/amt-10-1335-2017, 2017. Quante, L., Willner, S. N., Middelanis, R., and Levermann, A.: Regions of Intensification of Extreme Snowfall under Future Warming, Sci\nRep, 11, 16 621, https://doi.org/10.1038/s41598-021-95979-4, 2021. Rodgers, C. D.: Inverse Methods for Atmospheric Sounding : Theory and Practice, World Scientific Publishing Company, 2000. yi\ng\ng\nMulti-Angle Snowflake Camera, Atmos. Meas. Tech., 10, 1335–1357, https://doi.org/10.5194/amt-10-1335-2017, 2017. Quante, L., Willner, S. N., Middelanis, R., and Levermann, A.: Regions of Intensification of Extreme Snowfall under Future Warming, Sci\nRep, 11, 16 621, https://doi.org/10.1038/s41598-021-95979-4, 2021. Quante, L., Willner, S. N., Middelanis, R., and Levermann, A.: Regions of Intensification of Extreme Snowfall under Future Warming, Sci\nRep, 11, 16 621, https://doi.org/10.1038/s41598-021-95979-4, 2021. ers, C. D.: Inverse Methods for Atmospheric Sounding : Theory and Practice, World Scientific Publishing Com Schönhuber, M., Lammer, G., and Randeu, W. L.: One Decade of Imaging Precipitation Measurement by 2D-video-distrometer, Adv. Geosci.,\n620\n10, 85–90, https://doi.org/10.5194/adgeo-10-85-2007, 2007. Shupe, M. D., Rex, M., Blomquist, B., Persson, P. O. G., Schmale, J., Uttal, T., Althausen, D., Angot, H., Archer, S., Bariteau, L., Beck, I.,\nBilberry, J., Bucci, S., Buck, C., Boyer, M., Brasseur, Z., Brooks, I. M., Calmer, R., Cassano, J., Castro, V., Chu, D., Costa, D., Cox, C. J.,\nCreamean, J., Crewell, S., Dahlke, S., Damm, E., de Boer, G., Deckelmann, H., Dethloff, K., Dütsch, M., Ebell, K., Ehrlich, A., Ellis, Schönhuber, M., Lammer, G., and Randeu, W. L.: One Decade of Imaging Precipitation Measurement by 2D-video-distrometer, Adv. Geosci.,\n620\n10, 85–90, https://doi.org/10.5194/adgeo-10-85-2007, 2007. Shupe, M. D., Rex, M., Blomquist, B., Persson, P. O. G., Schmale, J., Uttal, T., Althausen, D., Angot, H., Archer, S., Bariteau, L., Beck, I.,\nBilb\nJ B\ni S B\nk C B\nM B\nZ B\nk\nI M C l\nR C\nJ C\nV Ch\nD C\nD C\nC J Schönhuber, M., Lammer, G., and Randeu, W. L.: One Decade of Imaging Precipitation Measurement by 2D-video-distrometer, Adv. Geosci.,\n620\n10, 85–90, https://doi.org/10.5194/adgeo-10-85-2007, 2007. Shupe, M. D., Rex, M., Blomquist, B., Persson, P. O. Takami, K., Kamamoto, R., Suzuki, K., Yamaguchi, K., and Nakakita, E.: Relationship between Newly Fallen Snow Density and\nDegree of Riming Estimated by Particles’ Fall Speed in Niigata Prefecture, Japan, Hydrological Research Letters, 16, 87–92,\n635\nhttps://doi.org/10.3178/hrl.16.87, 2022. Posman, K., Powers, H., Pratt, K. A., Preußer, A., Quéléver, L., Radenz, M., Rabe, B., Rinke, A., Sachs, T., Schulz, A., Siebert, H., Silva,\n630\nT., Solomon, A., Sommerfeld, A., Spreen, G., Stephens, M., Stohl, A., Svensson, G., Uin, J., Viegas, J., Voigt, C., von der Gathen, P.,\nWehner, B., Welker, J. M., Wendisch, M., Werner, M., Xie, Z., and Yue, F.: Overview of the MOSAiC Expedition—Atmosphere, Elem\nSci Anth, 10, 00 060, https://doi.org/10.1525/elementa.2021.00060, 2022.\nTakami, K., Kamamoto, R., Suzuki, K., Yamaguchi, K., and Nakakita, E.: Relationship between Newly Fallen Snow Density and Degree of Riming Estimated by Particles’ Fall Speed in Niigata Prefecture, Japan, Hydrological Research Letters, 16, 87–92,\n635\nhttps://doi.org/10.3178/hrl.16.87, 2022. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-655\nPreprint. Discussion started: 20 April 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Testik, F. Y. and Rahman, M. K.: High-Speed Optical Disdrometer for Rainfall Microphysical Observations, J. Atmospheric Ocean. Technol.,\n33, 231–243, https://doi.org/10.1175/JTECH-D-15-0098.1, 2016. Testud, J., Oury, S., Black, R. A., Amayenc, P., and Dou, X.: The Concept of Normalized Distribution to Describe Raindrop\nSpectra: A Tool for Cloud Physics and Cloud Remote Sensing, J. Appl. Meteorol., 40, 1118–1140, https://doi.org/10.1175/1520-\n0450(2001)040<1118:TCONDT>2.0.CO;2, 2001. 640 0450(2001)040<1118:TCONDT>2.0.CO;2, 2001. Thurai, M., Bringi, V., Gatlin, P. N., Petersen, W. A., and Wingo, M. T.: Measurements and Modeling of the Full Rain Drop Size Distribution,\nAtmosphere, 10, 39, https://doi.org/10.3390/atmos10010039, 2019. Tii\nJ\nM i\nD N\nL b\nA\nO i D\nT k\nA\nBli\nL F\nd P t\nW\nE\nbl\nM\nD\nit\nd It\nC Thurai, M., Bringi, V., Gatlin, P. N., Petersen, W. A., and Wingo, M. T.: Measurements and Modeling of the Full Rain Drop Size Distribution,\nAtmosphere, 10, 39, https://doi.org/10.3390/atmos10010039, 2019. Tiira, J., Moisseev, D. N., von Lerber, A., Ori, D., Tokay, A., Bliven, L. F., and Petersen, W.: Ensemble Mean Density and Its Con-\nnection to Other Microphysical Properties of Falling Snow as Observed in Southern Finland, Atmos. Meas. Tech., 9, 4825–4841,\n645\nhttps://doi.org/10.5194/amt-9-4825-2016, 2016. Tiira, J., Moisseev, D. N., von Lerber, A., Ori, D., Tokay, A., Bliven, L. F., and Petersen, W.: Ensemble Mean Density and Its Con\nnection to Other Microphysical Properties of Falling Snow as Observed in Southern Finland, Atmos. Meas. Tech., 9, 4825–4841,\n645\nhttps://doi.org/10.5194/amt-9-4825-2016, 2016. Tokay, A., Wolff, D. B., and Petersen, W. A.: Evaluation of the New Version of the Laser-Optical Disdrometer, OTT Parsivel2, J. Atmospheric\nOcean. Technol., 31, 1276–1288, https://doi.org/10.1175/JTECH-D-13-00174.1, 2014. Tokay, A., von Lerber, A., Pettersen, C., Kulie, M. S., Moisseev, D. N., and Wolff, D. B.: Retrieval of Snow Water Equivalent by the 645 https://doi.org/10.5194/amt-9-4825-2016, 2016. Tokay, A., Wolff, D. B., and Petersen, W. A.: Evaluation of the New Version of the Laser-Optical Disdrometer, OTT Parsivel2, J. Atmospheric\nOcean. Technol., 31, 1276–1288, https://doi.org/10.1175/JTECH-D-13-00174.1, 2014. Tokay, A., von Lerber, A., Pettersen, C., Kulie, M. S., Moisseev, D. N., and Wolff, D. B.: Retrieval of Snow Water Equivalent by the Tokay, A., Wolff, D. B., and Petersen, W. A.: Evaluation of the New Version of the Laser-Optical Disdrometer, OTT Parsivel2, J. Atmospheric\nOcean. G., Schmale, J., Uttal, T., Althausen, D., Angot, H., Archer, S., Bariteau, L., Beck, I.,\nBilberry, J., Bucci, S., Buck, C., Boyer, M., Brasseur, Z., Brooks, I. M., Calmer, R., Cassano, J., Castro, V., Chu, D., Costa, D., Cox, C. J., Bilberry, J., Bucci, S., Buck, C., Boyer, M., Brasseur, Z., Brooks, I. M., Calmer, R., Cassano, J., Castro, V., Chu, D., Costa, D., Cox, C. J., Creamean, J., Crewell, S., Dahlke, S., Damm, E., de Boer, G., Deckelmann, H., Dethloff, K., Dütsch, M., E J., Engelmann, R., Fong, A. A., Frey, M. M., Gallagher, M. R., Ganzeveld, L., Gradinger, R., Graeser, J., Greenamyer, V., Griesche, H.,\n625\nGriffiths, S., Hamilton, J., Heinemann, G., Helmig, D., Herber, A., Heuzé, C., Hofer, J., Houchens, T., Howard, D., Inoue, J., Jacobi, H.-\nW., Jaiser, R., Jokinen, T., Jourdan, O., Jozef, G., King, W., Kirchgaessner, A., Klingebiel, M., Krassovski, M., Krumpen, T., Lampert, A.,\nLanding, W., Laurila, T., Lawrence, D., Lonardi, M., Loose, B., Lüpkes, C., Maahn, M., Macke, A., Maslowski, W., Marsay, C., Maturilli,\nM., Mech, M., Morris, S., Moser, M., Nicolaus, M., Ortega, P., Osborn, J., Pätzold, F., Perovich, D. K., Petäjä, T., Pilz, C., Pirazzini, R., J., Engelmann, R., Fong, A. A., Frey, M. M., Gallagher, M. R., Ganzeveld, L., Gradinger, R., Graeser, J., Greenamyer, V., Griesche, H.,\n625 , Heinemann, G., Helmig, D., Herber, A., Heuzé, C., Hofer, J., Houchens, T., Howard, D., Inoue, J., Jacobi, H Landing, W., Laurila, T., Lawrence, D., Lonardi, M., Loose, B., Lüpkes, C., Maahn, M., Macke, A., Maslowski, W., Marsay, C., Maturilli, Posman, K., Powers, H., Pratt, K. A., Preußer, A., Quéléver, L., Radenz, M., Rabe, B., Rinke, A., Sachs, T., Schulz, A., Siebert, H., Silva,\n630\nT., Solomon, A., Sommerfeld, A., Spreen, G., Stephens, M., Stohl, A., Svensson, G., Uin, J., Viegas, J., Voigt, C., von der Gathen, P.,\nWehner, B., Welker, J. M., Wendisch, M., Werner, M., Xie, Z., and Yue, F.: Overview of the MOSAiC Expedition—Atmosphere, Elem\nSci Anth, 10, 00 060, https://doi.org/10.1525/elementa.2021.00060, 2022. Takami, K., Kamamoto, R., Suzuki, K., Yamaguchi, K., and Nakakita, E.: Relationship between Newly Fallen Snow Density and Degree of Riming Estimated by Particles’ Fall Speed in Niigata Prefecture, Japan, Hydrological Research Letters, 16, 87–92,\n635\nhttps://doi.org/10.3178/hrl.16.87, 2022. 26 Technol., 31, 1276–1288, https://doi.org/10.1175/JTECH-D-13-00174.1, 2014. Tokay, A., von Lerber, A., Pettersen, C., Kulie, M. S., Moisseev, D. N., and Wolff, D. B.: Retrieval of S Tokay, A., von Lerber, A., Pettersen, C., Kulie, M. S., Moisseev, D. N., and Wolff, D. B.: Retrieval of Snow Water Equivalent by the Precipitation Imaging Package (PIP) in the Northern Great Lakes, J. Atmospheric Ocean. Technol., -1, https://doi.org/10.1175/JTECH-D-\n650\n20-0216.1, 2021. Vázquez-Martín, S., Kuhn, T., and Eliasson, S.: Mass of Different Snow Crystal Shapes Derived from Fall Speed Measurements, Atmospheric\nChem. Phys., 21, 18 669–18 688, https://doi.org/10.5194/acp-21-18669-2021, 2021a. Vázquez-Martín, S., Kuhn, T., and Eliasson, S.: Shape Dependence of Snow Crystal Fall Speed, Atmos. Chem. Phys., 21, 7545–7565, Precipitation Imaging Package (PIP) in the Northern Great Lakes, J. Atmospheric Ocean. Technol., -1, https://doi.org/10.1175/JTECH-D-\n650\n20-0216.1, 2021. Precipitation Imaging Package (PIP) in the Northern Great Lakes, J. Atmospheric Ocean. Technol., -1, https://doi.org/10.1175/JTECH-D-\n650\n20-0216.1, 2021. Vázquez-Martín, S., Kuhn, T., and Eliasson, S.: Mass of Different Snow Crystal Shapes Derived from Fall Speed Measurements, Atmospheric\nCh\nPh\n21 18 669 18 688 htt\n//d i\n/10 5194/\n21 18669 2021 2021 Vázquez-Martín, S., Kuhn, T., and Eliasson, S.: Mass of Different Snow Crystal Shapes Derived from Fall Speed Measurements, Atmospheric\nChem. Phys., 21, 18 669–18 688, https://doi.org/10.5194/acp-21-18669-2021, 2021a. Vázquez-Martín, S., Kuhn, T., and Eliasson, S.: Shape Dependence of Snow Crystal Fall Speed, Atmos. Chem. Phys., 21, 7545–7565,\nhttps://doi.org/10.5194/acp-21-7545-2021, 2021b. 655 Vázquez-Martín, S., Kuhn, T., and Eliasson, S.: Shape Dependence of Snow Crystal Fall Speed, Atmos. Vázquez-Martín, S., Kuhn, T., and Eliasson, S.: Shape Dependence of Snow Crystal Fall Speed, Atmos. Chem. Phys., 21, 7545–7565,\nhttps://doi org/10 5194/acp 21 7545 2021 2021b\n5 https://doi.org/10.5194/acp-21-7545-2021, 2021b. 655\nVignon, E., Besic, N., Jullien, N., Gehring, J., and Berne, A.: Microphysics of Snowfall Over Coastal East Antarctica Simulated by Polar\nWRF and Observed by Radar, J. Geophys. Res. Atmos., 124, 11 452–11 476, https://doi.org/10.1029/2019JD031028, 2019. Vogl, T., Maahn, M., Kneifel, S., Schimmel, W., Moisseev, D., and Kalesse-Los, H.: Using Artificial Neural Networks to Predict Riming\nfrom Doppler Cloud Radar Observations, Atmospheric Meas. Tech., 15, 365–381, https://doi.org/10.5194/amt-15-365-2022, 2022. Vignon, E., Besic, N., Jullien, N., Gehring, J., and Berne, A.: Microphysics of Snowfall Over Coastal East Antarctica Simulated by Polar\nWRF and Observed by Radar, J. Geophys. Res. Atmos., 124, 11 452–11 476, https://doi.org/10.1029/2019JD031028, 2019. Zeevi, S.: BackgroundSubtractorCNT: A Fast Background Subtraction Algorithm, 2016. 2013.\n665\nZeevi, S.: BackgroundSubtractorCNT: A Fast Background Subtraction Algorithm, 2016. Vogl, T., Maahn, M., Kneifel, S., Schimmel, W., Moisseev, D., and Kalesse-Los, H.: Using Artificial Neural Networks to Predict Riming\nfrom Doppler Cloud Radar Observations Atmospheric Meas Tech 15 365 381 https://doi org/10 5194/amt 15 365 2022 2022 Vignon, E., Besic, N., Jullien, N., Gehring, J., and Berne, A.: Microphysics of Snowfall Over Coastal East Antarctica Simulated by Polar\nWRF and Observed by Radar, J. Geophys. Res. Atmos., 124, 11 452–11 476, https://doi.org/10.1029/2019JD031028, 2019. Vogl, T., Maahn, M., Kneifel, S., Schimmel, W., Moisseev, D., and Kalesse-Los, H.: Using Artificial Neural Networks to Predict Riming\nfrom Doppler Cloud Radar Observations, Atmospheric Meas. Tech., 15, 365–381, https://doi.org/10.5194/amt-15-365-2022, 2022. Vogl, T., Maahn, M., Kneifel, S., Schimmel, W., Moisseev, D., and Kalesse-Los, H.: Using Artificial Neural Networks to Predict Riming\nfrom Doppler Cloud Radar Observations Atmospheric Meas Tech 15 365 381 https://doi org/10 5194/amt 15 365 2022 2022 S., Schimmel, W., Moisseev, D., and Kalesse-Los, H.: Using Artificial Neural Networks to Predict Riming Radar Observations, Atmospheric Meas. Tech., 15, 365–381, https://doi.org/10.5194/amt-15-365-2022, 2022. von Lerber, A., Moisseev, D., Bliven, L. F., Petersen, W., Harri, A.-M., and Chandrasekar, V.: Microphysical Properties of Snow and Their\n660\nLink to Ze–S Relations during BAECC 2014, J. Appl. Meteor. Climatol., 56, 1561–1582, https://doi.org/10.1175/JAMC-D-16-0379.1,\n2017. Wood, N. B., L’Ecuyer, T. S., Bliven, F. L., and Stephens, G. L.: Characterization of Video Disdrometer Uncertainties and Impacts on\nEstimates of Snowfall Rate and Radar Reflectivity, Atmospheric Meas. Tech., 6, 3635–3648, https://doi.org/10.5194/amt-6-3635-2013, von Lerber, A., Moisseev, D., Bliven, L. F., Petersen, W., Harri, A.-M., and Chandrasekar, V.: Microphysical Properties of Snow and Their\n660\nLink to Ze–S Relations during BAECC 2014, J. Appl. Meteor. Climatol., 56, 1561–1582, https://doi.org/10.1175/JAMC-D-16-0379.1,\n2017. 2013. 665\nZeevi, S.: BackgroundSubtractorCNT: A Fast Background Subtraction Algorithm, 2016. 27 27"