Random Forest Predictions of Fine Ash Content Using Properties of Electric Discharges in Shock Tube Experiments

Abstract:

Fundamental knowledge gaps exist pertaining to the physical processes behind the rise, spread, and longevity of ash-rich volcanic jets. However, recent studies show turbulent entrainment and mixing of atmosphere are key underlying processes that are particularly sensitive to fine ash (< 63 µm) concentration. This work shows that reliably characterizing and understanding ongoing volcanic hazards depends critically on high-resolution, real-time measurements of fine ash concentration within erupting jets that are not yet possible due to current limitations of remote sensing techniques, such as doppler radar and thermal imaging. In contrast, real-time measurements of electric discharges often present at the onset of explosive eruptions, vent discharges, are easily obtained. Recent laboratory studies of shock tube generated vent discharges have demonstrated the influence of fine ash concentration on the measured magnitude, timescale, and number of discharges in each experiment, but this relationship has yet to be quantified. We analyze these measured properties with a machine learning algorithm (random forest) to determine the most significant ash-discharge relationships and utilize them to constrain the fine ash concentration of each experiment. Varying fine ash concentration most significantly alters the behavior of positive discharges, which neutralize excess negative charge generated through particle-particle and particle-wall collisions. These relationships allow us to predict the fine ash concentration in each experiment with 93% accuracy. In addition to providing valuable grain size information, these results can provide insight into dominant ash charging mechanisms in plumes that are potentially linked to processes governing turbulent plume dynamics. Since this model is only trained using measured properties of discharges, it shows promise for use in real-time eruption monitoring and hazard assessment.