Numerical tools for interpreting rock surface roughness
MetadataShow full item record
Visual recognition of the irregular shapes of natural objects and surfaces is crucial to the study of geology. Field geologists are aided by a learned ability to, at a glance, differentiate cohesive outcropping rock from talus, pick out fossils from loose clasts on the ground, judge fault displacement and perform any number of other discriminative tasks on the basis of morphology. Interpretation of geology depends upon identifying subtle and ambiguous natural forms that may be widely variable in response to diverse conditions of formation, preservation and weathering. Applying numerical models to these morphological cues empowers geologists to support or refute hypotheses with enhanced objectivity. With the growing prevalence and quality of remotely sensed spatial data in the geosciences, accurate modeling of diagnostic aspects of Earth surface morphology becomes more feasible while its potential benefits increase greatly. Terrestrial laser scanning (TLS) in particular is valued for collecting dense point clouds which enable precise, non-contact measurement of structures in outcrop. Unfortunately, the process of interpreting irregular shapes and textures in point-cloud data is labor intensive and its results subject to user bias. Automated point-cloud classification tools promise improved objectivity in scene interpretation, though their application to natural surface geometry has not yet been studied in depth. The objective of this research is to investigate, apply and refine automated point-cloud classification methodology for geological targets, with discriminating the erosive styles of sedimentary rocks in outcrop as the central theme. To accomplish this goal, three studies investigate range measurement biases in TLS data collection, the scale dependence of rock surface roughness, and 3D multiscale classification methodology. The findings of this research have implications in particular for studies using TLS to measure statistics of target morphology close to the instrument resolution limits. The multiscale classification methodology investigated is immediately applicable to a range of identification and discrimination tasks encountered in the geosciences. Overall, this research will contribute to increased productivity and objectivity of interpretations for the growing population of geoscientists working with remotely sensed spatial data through improved automation of a wider range of interpretive tasks.