Eruption dynamics are largely driven by the degassing of volatiles, especially H2O. Experimental studies are
crucial for understanding these natural processes. To correlate experimental results with the degassing dynamics
of eruptions, the number of vesicles in a certain melt volume (Vesicle Number Density (VND)) is essential,
especially for determining the mechanisms of vesicle formation and the decompression rate. Beyond that, the
Crystal Number Density (CND) can play a significant role, as crystals may act as possible heterogeneous vesicle
nucleation sites.
To determine the VND and CND in experimental products, ImageJ can be used to analyze intersected crystals
and vesicles of 2D images and to create 3D VND and CND information with the program CSDCorrections.
For the quantitative analysis of the crystals and vesicles, ImageJ requires an image in which each class (glass
surface, crystals and vesicles) is assigned a distinct color. One established method is to manually color the
classes using an image editing program. However, this process is highly time-consuming.
Therefore, we developed Trainable Weka Segmentation as a potential alternative to expedite the image process-
ing. This tool is widely used in other fields, such as segmentation in cell biology.
For our application, the method was used to analyze microscopic surface images of vesicles and crystals embedded
in a glass matrix, obtained from vitreous samples of decompression experiments. Results from this method were
compared with those of the manual image processing.
Overall, the vesicle number densities derived from the different methods are consistent within the error range
of 0.1 log units, based on identical sample surface images. The presented method simplifies and accelerates the
vesicle and crystal number density determination and significantly increases the analyzed sample area, enabling
more robust data analysis.