Skip to content

Many Maps are Better Than One: A Random Forest Approach to Estimate Spatial Density in a Distributed Volcanic Field, Eastern Snake River Plain (ID)

Chuck Connor 1, Matthew Connor2, Laura Connor2, Mark Bebbington3, Michael Ort4, Bill Hackett5, Suzette Payne6

  • Affiliations: 1School of Geosciences, University of South Florida, USA 2Desperate Measures International, LLC, Tampa, Florida, USA 3School of Agriculture and Environment, Massey University, Palmerston North, New Zealand 4Geology, School of Natural Sciences, Trinity College Dublin, Dublin, Ireland 5WRH Associates Inc, Ogden, UT, USA 6SMJP Consulting Services, Idaho Falls, ID, USA

  • Presentation type: Talk [Invited]

  • Presentation time: Tuesday 16:15 - 16:30, Room R290

  • Programme No: 6.3.7

  • Theme 6 > Session 3

Abstract

Machine learning offers a powerful approach for estimating the probable location of new volcanic vents. We employ a Random Forest machine learning algorithm that objectively combines multiple statistical models of spatial density with multiple geophysical models across a common area within a distributed volcanic field (DVF). This approach uses a variance weighting scheme to create composite spatial density estimates that directly incorporate relevant geophysical observations with statistical models for the probability of the opening of new vents. The approach is applied and tested in the eastern Snake River Plain (ESRP), Idaho, where 617 Quaternary vents are mapped along the path of the Yellowstone hotspot track. Six spatial density models are applied to this dataset: three kernel density estimators and three adaptive estimators. Four geophysical models and measurements are included as predictor variables: (1) long-wavelength topography, (2) a strain model for a block representing the ESRP lithosphere, (3) an ESRP gravity model, and (4) an ESRP aeromagnetic model. Composite Random Forest spatial density estimates are created using the multiple spatial density and geophysical data grids. A composite spatial density estimate for the opening of new vents is then applied to a tephra fallout hazard assessment for various facilities located within the Idaho National Laboratory on the ESRP. This approach could be used as an alternative to expert judgment when weighting alternative spatial density models for probabilistic volcanic hazard assessments.