Geologic Publications for Mount Rainier

Mapping bedrock outcrops: An assessment of algorithms and their performance across diverse landscapes

[ back to previous page ]

Author(s): Brittany Selander, Matthew W. Rossi, Suzanne P. Anderson

Category: PUBLICATION
Document Type:
Publisher: SSRN
Published Year: 2023
Volume:
Number:
Pages: 32
DOI Identifier:
ISBN Identifier:
Keywords: Bedrock mapping Binary classification Accuracy assessment Imbalanced data

Abstract:
Accurately mapping bedrock outcrops is important to geomorphology, hydrology, ecology, and hazard management. Airborne lidar has proven useful to identify bedrock exposure from topographic data alone. However, there is a need to test these topographic proxies across diverse environments. We developed a new algorithm that uses both slope and curvature to classify bedrock and evaluate its success with respect to two existing classifiers, a slope-threshold method and a roughness-threshold method, for six different landscapes shaped by different erosional processes. Accuracy of binary classifiers are quantified using the F1 score and the normalized Matthews Correlation Coefficient. By analyzing topographic proxies across gradients in bedrock fraction, we show that normalized Matthew Correlation Coefficient is a more robust measure of accuracy than F1 score when the data is imbalanced (i.e., where either soil or bedrock is more prevalent). We found moderate success in mapping bedrock with all three methods for half of the landscapes (flanks of an active stratovolcano, Mount Rainier, Washington; horst and graben landscape, Canyonlands National Park, Utah; bedrock canyon, Boulder Creek, Colorado). Our new algorithm performed slightly better than the other two methods. None of the methods were successful in a sequence of coastal terraces and sea cliffs (Santa Cruz County, California), a glacially scoured landscape (Southern Wind River Range, Wyoming), and a fluvially incised chaparral canyon (Mission Trails, San Diego, California). All classifiers are confounded by erosional processes that produce smooth, low gradient, bedrock surfaces and/or sediment transport processes that increase roughness. While our new algorithm leverages the success of both slope- and roughness-threshold methods, results emphasize the importance of considering the geomorphic context when choosing among algorithms.

View Report:
View Report [External Link]

Suggested Citations:
In Text Citation:
Selander and others (2023) or (Selander et al., 2023)

References Citation:
Selander, B., M.W. Rossi, and S.P. Anderson, 2023, Mapping bedrock outcrops: An assessment of algorithms and their performance across diverse landscapes: SSRN, 32 p..