Deep Learning-Based Water Crystal Classification

Much of the earth’s surface is covered by water. As was pointed out in the 2020 edition of the World Water Development Report, climate change challenges the sustainability of global water resources, so it is important to monitor the quality of water to preserve sustainable water resources. Quality of water can be related to the structure of water crystal, the solid-state of water, so methods to understand water crystals can help to improve water quality. As a first step, a water crystal exploratory analysis has been initiated with the cooperation with the Emoto Peace Project (EPP). The 5K EPP dataset has been created as the first world-wide small dataset of water crystals. Our research focused on reducing the inherent limitations when fitting machine learning models to the 5K EPP dataset. One major result is the classification of water crystals and how to split our small dataset into several related groups. Using the 5K EPP dataset of human observations and past research on snow crystal classification, we created a simple set of visual labels to identify water crystal shapes, in 13 categories. A deep learning-based method has been used to automatically do the classification task with a subset of the label dataset. The classification achieved high accuracy when using a fine-tuning technique.

Accuracy in starphotometry

Starphotometry, the night-time counterpart of sunphotometry, has not yet achieved the commonly sought observational error level of 1 %: a spectral optical depth (OD) error level of 0.01. In order to address this issue, we investigate a large variety of systematic (absolute) uncertainty sources. The bright-star catalogue of extraterrestrial references is noted as a major source of errors with an attendant recommendation that its accuracy, particularly its spectral photometric variability, be significantly improved. The small field of view (FOV) employed in starphotometry ensures that it, unlike sun- or moonphotometry, is only weakly dependent on the intrinsic and artificial OD reduction induced by scattering into the FOV by optically thin clouds. A FOV of 45 arcsec (arcseconds) was found to be the best trade-off for minimizing such forward-scattering errors concurrently with flux loss through vignetting. The importance of monitoring the sky background and using interpolation techniques to avoid spikes and to compensate for measurement delay was underscored. A set of 20 channels was identified to mitigate contamination errors associated with stellar and terrestrial atmospheric gas absorptions, as well as aurora and airglow emissions. We also note that observations made with starphotometers similar to our High Arctic instrument should be made at high angular elevations (i.e. at air masses less than 5). We noted the significant effects of snow crystal deposition on the starphotometer optics, how pseudo OD increases associated with this type of contamination could be detected, and how proactive techniques could be employed to avoid their occurrence in the first place. If all of these recommendations are followed, one may aspire to achieve component errors that are well below 0.01: in the process, one may attain a total 0.01 OD target error.

Assessing fall velocity-maximum dimension relationships and particle size distributions for snowfall

Snowfall is an atmospheric phenomenon that can cause significant impacts to many aspects of daily life in Missouri. Further, no two snowfall events are exactly the same, as even small differences in environmental characteristics can result in differing snow crystal types dominating the event, which in turn can result in differing impacts from event to event. Therefore, it is necessary to understand snowfall behavior so that better forecasts and in situ analyses may be made. In this study, snowflake maximum dimension and fall velocity measurements were recorded using the OTT Parsivel Laser Disdrometer. In conjunction with distribution of measured maximum dimensions, RAP Analysis soundings were used to determine snow crystal type. From there, the relationships between fall velocity and maximum dimension and the particle size distributions of snowflakes from many snowfall events were analyzed. Observed relationships between fall velocity and maximum dimension were compared with previously derived relationships, and it was found that, in most cases, no single curve represented the relationship in the observed data well, with discrepancies caused by instrumentation error and lack of a single dominant crystal type. To analyze particle size distributions, several distribution functions were fit to the observed distribution using a least-squares regression method in MATLAB. It was found that, overall, the triple Gaussian distribution function performed the best in modeling particle size distributions in snow, but there were some instances where the gamma function modeled the distribution best. Further study, especially with the inclusion of field observations in addition to instrument observations, is necessary to develop a better understanding of these snowfall events.