Computational approaches to tephra geochemical correlation and deposit volume estimation

Abstract

This thesis presents two studies employing computational approaches to tephra source attribution and tephra volume estimation. Glass major oxide data was used to train 11 classification algorithms. Each classifier probabilistically attributed tephra chemistries to one of ten volcanic sources from Alaska. An ensemble model, combining random forest and artificial neural network predictions, was highly accurate on held-out and new data from Eklutna Lake, Alaska. Results matched visual assessment and identified tephra geochemically consistent with the Pleistocene Emmons Lake Volcanic Center (Dawson tephra) in Holocene-aged sediments. The May 18, 1980 Mount St. Helens eruption’s volume was reassessed. Spline interpolation permitted thickness, area, and cumulative volume to be calculated from a new synthesized thickness map at high resolution. A novel approach and cumulative volume plots show tephra volume for the deposit with clarity and uncertainty estimation. Reassessment indicates comparable volume to past estimates, but distal volume is proportionally larger than previously thought.