High-resolution trench photomosaics from image-based modeling: Workflow and error analysis

Abstract

Photomosaics are commonly used to construct maps of paleoseismic trench exposures, but the conventional process of manually using image-editing software is time consuming and produces undesirable artifacts and distortions. Herein, we document and evaluate the application of image-based modeling (IBM) for creating photomosaics and 3D models of paleoseismic trench exposures, illustrated with a case-study trench across the Wasatch fault in Alpine, Utah. Our results include a structure-from-motion workflow for the semiautomated creation of seamless, highresolution photomosaics designed for rapid implementation in a field setting. Compared with conventional manual methods, the IBM photomosaic method provides a more accurate, continuous, and detailed record of paleoseismic trench exposures in approximately half the processing time and 15%–20% of the user input time. Our error analysis quantifies the effect of the number and spatial distribution of control points on model accuracy. For this case study, an ∼87 m2 exposure of a benched trench photographed at viewing distances of 1.5–7 m yields a model with <2 cm root mean square error (rmse) with as few as six control points. Rmse decreases as more control points are implemented, but the gains in accuracy are minimal beyond 12 control points. Spreading control points throughout the target area helps to minimize error. We propose that 3D digital models and corresponding photomosaics should be standard practice in paleoseismic exposure archiving. The error analysis serves as a guide for future investigations that seek balance between speed and accuracy during photomosaic and 3D model construction.