Feature Analysis and Classification of Particle Data from Two-Dimensional Video Disdrometer

Abstract

We developed a ground observation system for solid precipitation using two-dimensional video disdrometer (2DVD). Among 16,010 particles observed by the system, around 10% of them were randomly sampled and manually classified into five classes which are snowflake, snowflake-like, intermediate, graupel-like, and graupel. At first, each particle was represented as a vector of 72 features containing fractal dimension and box-count to represent the complexity of particle shape. Feature analysis on the dataset clarified the importance of fractal dimension and box-count features for characterizing particles varying from snowflakes to graupels. On the other hand, performance evaluation of two-class classification by Support Vector Machine (SVM) was conducted. The experimental results revealed that, by selecting only 10 features out of 72, the average accuracy of classifying particles into snowflakes and graupels could reach around 95.4%, which had not been achieved by previous studies.

Solid hydrometeor classification and riming degree estimation from pictures collected with a Multi-Angle Snowflake Camera

“…For this purpose, ground-based snowflake imagers like the twodimensional video disdrometer (2DVD; Kruger and Krajewski 2002), the Hydrometeor Velocity and Shape Detector (HVSD; Barthazy et al, 2004), the Snowflake Video Imager (SVI or PIP in its newest version; Newman et al, 2009) and the Multi-Angle Snowflake Camera (MASC; Garrett et al, 2012) provide relevant information in the form of twodimensional binary or grayscale particle images and in some cases the associated fall speed measurements. Recent investigations have shown the potential of the 2DVD to automatically detect and classify hydrometeors imaged according to their type and riming extent (Grazioli et al, 2014;Gavrilov et al, 2015). Bernauer et al (2016) also proposed a decisiontree approach to distinguish between three degrees of riming by deriving constraints on the particle shape and fall speed parameters measured by the 2DVD.…”

Responses to the reviewers

“…A comparison was conducted with respect to a hydrometeor classification scheme developed for 2DVD data and detailed in Grazioli et al (2014). The method provides an estimate of the dominant type of hydrometeor measured by the instrument on a time interval of 1 min.…”

“…Because it is a priori impossible to know exactly what features are relevant to the target concept (i.e., hydrometeor classification), a large set of 72 descriptors derived from the particle size, shape and textural information was introduced. Several of them have already been used for hydrometeor identification purposes in previous works (e.g., Lindqvist et al, 2012;Nurzyńska et al, 2012Nurzyńska et al, , 2013Grazioli et al, 2014;Schmitt and Heymsfield, 2014). As we experienced some issues with the MASC fall speed measuring unit during the campaign in Davos, this parameter was discarded in the proposed methodology.…”

“…Multiple solid hydrometeor classification schemes have been introduced in the literature (e.g., Korolev and Sussman, 2000;Lindqvist et al, 2012;Grazioli et al, 2014) and it is often difficult to find an equivalence between them, partly because the type of particles probed depends on the observation conditions (ground-based or airborne instrument, measurement location). In the present work, efforts were made in order to develop an exhaustive and versatile classification scheme applying the following procedure.…”

“…For this purpose, ground-based snowflake imagers like the twodimensional video disdrometer (2DVD; Kruger and Krajewski 2002), the Hydrometeor Velocity and Shape Detector (HVSD; Barthazy et al, 2004), the Snowflake Video Imager (SVI or PIP in its newest version; Newman et al, 2009) and the Multi-Angle Snowflake Camera (MASC; Garrett et al, 2012) provide relevant information in the form of twodimensional binary or grayscale particle images and in some cases the associated fall speed measurements. Recent investigations have shown the potential of the 2DVD to automatically detect and classify hydrometeors imaged according to their type and riming extent (Grazioli et al, 2014;Gavrilov et al, 2015). Bernauer et al (2016) also proposed a decisiontree approach to distinguish between three degrees of riming by deriving constraints on the particle shape and fall speed parameters measured by the 2DVD.…”