Title:

Multimodal Machine Learning for 3D Characterization of Hidden Groundwater and Geothermal Resources: Case Study, Lāna‘i Hawaii

Authors:

Michael J FRIEDEL., Nicole LAUTZE, Erin WALLIN, Aaron ROTHFOLK

Key Words:

Multimodal machine learning, 3D geothermal stratigraphic units, 3D hidden geothermal resources, 3D hidden groundwater resources, 3D temperature prediction, 3D chloride prediction, 3D specific capacity prediction, 3D geology predictions, dike swarms, pluton, sill, batholith, Moho, Lāna‘i Hawaii

Conference:

Stanford Geothermal Workshop

Year:

2022

Session:

Emerging Technology

Language:

English

Paper Number:

Lautze2

File Size:

1490 KB

View File:

Abstract:

The availability of freshwater and low-cost electricity are limiting factors for sustainable living in Hawaii. This raises the question: Can technology be developed to locate and characterize freshwater and geothermal resources simultaneously? We present a multimodal machine learning (MML) workflow to characterize the 3D distribution of features (physical and geochemical properties and state variables) along a groundwater-geothermal continuum. The success of this MML workflow is based on the availability and assimilation of mutual information to: (1) identify latent features using neural network proposals in a genetic algorithm with feature constraints, (2) organize related information across a hypersurface by competitive learning, (3) predict regionally continuous features by minimizing the quantization and topological errors with competitive learning, and (4) group statistically meaningful features based on the mode of stochastic k-means clusters. The proposed MML workflow is applied to a subset of the Hawaii Play Fairway data for Lāna‘i, Hawaii. These data include direct point measurements, such as borehole water level, geology, temperature, chloride concentration, isotopes, and specific conductivity; and derived 3D volume properties (numerically inverted), such as specific capacity, density, and electrical resistivity. Despite field data characteristics (disparate, scale dependent, spatially limited, sparse, and uncertain), the MML workflow yields a single 3D transdisciplinary data product whose voxels each contain statistical summaries of model features. Five-fold cross-validation (e.g., five randomly shuffled stratified split sets, each split with 80% training and 20% testing) reveals moderate generalizability of the MML model to independent data. Preliminary interpretation of continuous features reveals hidden freshwater (low temperature and low chloride concentration water) and geothermal (high temperature and brackish water) resources. In addition to integrating machine learning expertise and practices into the Play Fairway project outcomes, this study provides new capabilities for characterizing continuous subsurface hydrogeologic and geothermal features in the Hawaiian Islands for sustainable living (including Hawai‘i Island as part of the new Island Heat project sponsored by the Department of Energy) and at other geothermal sites worldwide.


ec2-18-209-66-87.compute-1.amazonaws.com, you have accessed 0 records today.

Press the Back button in your browser, or search again.

Copyright 2022, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.


Attend the nwxt Stanford Geothermal Workshop, click here for details.

Accessed by: ec2-18-209-66-87.compute-1.amazonaws.com (18.209.66.87)
Accessed: Thursday 28th of March 2024 10:32:55 AM