37 D1. Recent Advances in Geoscience and Geoscience Education (Posters)

Friday, 17 May 2024
9:00 AM - 1:30 PM
Grand Ballroom

DEEP LEARNING-BASED CLASSIFICATION OF MIMA MOUND PRESENCE IN PACIFIC NORTHWEST LANDSCAPES USING DIGITAL ELEVATION MODELS

Paper number: 37-1
Program: Technical Sessions
Time: 9:00 AM – 1:30 PM

Authors

Abstract

Mima mounds are widespread across the Pacific Northwest. These roughly circular convex forms have variable morphologies and organization but commonly form striking landscape patterns. Despite more than 150 years of investigation, a consensus on the origin of Mima mounds remains elusive. Compelling hypotheses such as vegetation anchoring and fossorial rodents have gained favor; however, extreme variations in both characteristics and arrangement of mounds pose challenges for understanding mound genesis. A lack of comprehensive mound inventories obscures the identification of patterns and associations with geologic, biotic, and climatic features, further complicating hypothesis testing, particularly in broad regions with large samples. Consequently, these constraints have led to inconsistencies in our understanding of Mima mounds, hindering our ability to correlate variations with potential controlling factors. Here, we utilize widely available high-resolution LiDAR data to develop a machine learning model that classifies the presence of Mima mounds in regularly gridded areas, enhancing inventorying efforts across Oregon and Washington. To generate training data for the model, we first built a tool that streamlines the process of labeling LiDAR imagery with evidence of Mima mounds. This tool crops raster data to regularly sized grids, constructs imagery from derivatives of digital elevation models, prompts users for labels, and compiles a database of the processed imagery. We evaluated the most effective size of gridded visualizations compared to detailed mapping, determining that a grid size of 200x200m captures important detail in the distribution of mound fields while minimizing computational cost. Using a database of thousands of labelled images, we trained a convolutional neural network to classify images constructed from digital elevation models with validation accuracy in the range of 90%. We deploy this model across an expansive region encompassing hundreds of thousands of square kilometers in central Oregon and Washington. The outcome is presented in the form of a heat map of likely mound presence.

Geological Society of America Abstracts with Programs. Vol. 56, No. 4

doi: 10.1130/abs/2024CD-399371

© Copyright 2024 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.