Automatic shape detection of ice crystals

摘要

Clouds have a crucial impact on the energy balance of the Earth-Atmosphere system. They can cool the system by partly reflecting or scattering of the incoming solar radiation (albedo effect); moreover, thermal radiation as emitted from the Earth's surface can be absorbed and partly re-emitted by clouds leading to a warming of the atmosphere (greenhouse effect). The effectiveness of both effects crucially depends on the size and the shape of a cloud's particulate constituents, i.e. liquid water droplets or solid ice crystals. For studying cloud microphysics, in situ measurements on board of aircraft are commonly used. An important class of measurement techniques comprises optical array probes (OAPs) as developed since the 1970s [13]. While water droplets can be assumed as spherical, the shape and size of ice particles are highly variable. The currently used analysis methods to determine the particles' size from OAP detection do rarely consider shape details or fine structures of ice particles, which may lead to artificial biases in the results. In this paper, we present two new computational analysis methods, combined in an hybrid approach, for an automatic classification of ice particles and water droplets. The first method computes the principal components of a cloud particle and uses them to determine an ellipse, which can then be used to filter for spherical particles. The second method uses convolutional neural networks (CNNs) for the classification of columns and rosettes, respectively. Although we currently only classify these two particle types with CNNs, the presented method can be easily adapted for the classification of other particle types. The particularity of our method is that we use a virtual data set to pre-train the networks, which are then further trained with a smaller amount of manually classified real cloud particles in a fine tuning step. We evaluated our models on a small data set of real cloud particles and in a final field test on OAP image data that was not previously classified. The precision of this field test was better than 81% and ranged up to 98%, demonstrating that the new methods are suitable for providing profound shape classifications of cloud particle images obtained by OAP measurements. All methods we describe in this paper have been implemented in Python and C and are fully open source. Code and documentation are available on Github (https://github.com/lcsgrlch/oap).