Popper-Bayes geophysical hypothesis testing

Knowledge graph centered on Popper-Bayes geophysical hypothesis testing with 48 nodes and 143 connections. Top connected: climate change, groundwater storage dynamics, Crested Butte, porosity, groundwater.

Method synopsis

A framework for testing geological hypotheses using geophysical data that integrates Monte Carlo simulation, dimension reduction, and Bayesian updating to account for uncertainty in both model parameters and measurements. Consists of prior model generation, falsification testing, and posterior probability calculation.

Synthesized from method descriptions across 1 paper using this protocol.

Procedure from a recent peer-reviewed implementation

Steps below were extracted from the most recent peer-reviewed implementation of this protocol in the corpus — Probabilistic Evaluation of Geoscientific Hypotheses With Geophysical Data. Application to Electrical Resistivity Imaging of a Fractured Bedrock Zone (2021), Journal of Geophysical Research. Implementations in other papers (listed below) may differ.

1. Prior model parameter definition
   Defined prior distributions for five subsurface model parameters: dip angle φ with uniform distribution U(25,165), thickness H with normal distribution N(120,40), outcrop X-position xo with normal distribution N(400,40), fracture zone resistivity ρf with lognormal distribution LN(0.45,20,10), and background resistivity ρb with lognormal distribution LN(0.8,40,30).
   Quantities: 5 parameters, uniform and normal/lognormal distributions with specified boundsDuration: Not applicableConditions: Prior model setup phase
   Equipment: statistical modeling software
2. Prior model Monte Carlo simulation
   Generated 2,000 prior realizations of subsurface electrical resistivity models using PyGIMLi software. Created triangular mesh for each realization using tetgen with minimum mesh angle of 22°. Refined mesh around electrodes and conformed exactly to fault zone geometry.
   Quantities: 2,000 prior realizations, 22° minimum mesh angleDuration: Not specifiedConditions: Computational modeling phase
   Equipment: PyGIMLi software, tetgen mesh generator
3. Machine learning model training
   Trained random forest regression models using 1,000 prior realizations as training set to relate ERT measurements to model parameters. Added independent Gaussian noise to each realization using measured standard deviation. Normalized each element with respect to mean vector and standard deviations from prior training set.

Typical Equipment

• PyGIMLi software
• tetgen mesh generator
• random forest regression software
• Bayesian sampling software
• statistical analysis software

Output Measurements

• posterior probabilities of hypotheses
• model parameter distributions

Papers Using This Protocol (1)

Places Where Used