Random Forest Predictions of Fine Ash Concentration and Charging Processes From Experimentally Generated Volcanic Discharges

Abstract

AbstractExplosive volcanic plume rise is governed by the rate at which ambient air is ingested and heated by turbulent entrainment and mixing processes. A daunting observational challenge is to constrain the character of the underlying physical processes and their dependence on complex particle‐particle and particle‐gas interactions. Important clues may lie in the particle‐particle momentum exchange that gives rise to related electrical discharges near the vent during supersonic eruptions. Recent laboratory studies of positive and negative shock‐tube‐generated volcanic discharges show a correlation between fine ash concentration and the magnitude and number of positive discharges. Charge generation via collisions (triboelectrification) is hypothesized to be more efficient with high ash concentrations and at high decompression rates because collisions between particles become more frequent under these conditions. To understand the experimental data in greater detail, we train and implement a regression‐based random forest algorithm to quantitatively predict concentrations of fine ash using discharge count, magnitude and polarity as predictors. Using a metric for variable importance, we find in all, high pressure (HP, ≥10 MPa), and high ash mass experiments (HM, >22g) that positive discharge properties are most important when predicting fine ash concentration, consistent with triboelectrification as the predominant charging process under these conditions. This mechanism is not constrained for low‐pressure (LP, <10 MPa) conditions, suggesting a potential threshold decompression rate condition for this class of charging. This result provides insight into why near‐vent lightning is not a ubiquitous feature of explosive eruptions.