Uncertainty in terrestrial laser scanning for measuring surface movements at a local scale

Abstract

Terrestrial laser scanning (TLS) is a remote sensing tool that can record a large amount of accurate topographical information with a fine spatial resolution over a short period of time. It has been used increasingly for measuring ground surfaces (i.e. topographical survey) and monitoring surface movements, such as those caused by landslides. However, the capability of this technique in these applications has not been fully explored in the literature, and thus forms the focus of this thesis. A quantitative study has been carried out to investigate the major error sources that affect the accuracy of digital elevation models (DEMs) derived from TLS survey data, and the magnitude of deformation that can be detected by repeated TLS surveys, at a local scale.

In this research, vegetation-induced elevation errors in TLS measurements and the ways in which they can be minimised have been investigated experimentally. The presence of short vegetation was found to be a significant limiting factor for TLS surveys of terrain surfaces, with the average grass-induced elevation error being roughly 65% of the grass height. A finer resolution scan with a lower incidence angle (greater visibility) can effectively reduce vegetation error, as will scanning the same area from multiple scanner locations.

The influence of measurement errors in source data points (or a point cloud) on a triangulated irregular network (TIN) with linear interpolation has been analysed. Based on the law of error
propagation, an analytical solution was derived to calculate the error variance at any location within a TIN model, due to vertical and horizontal errors in source data points. For the special case of equal and independent error variances in source data points, the maximum, average and minimum values of propagated error variance within a TIN were found to be equal to unity, a half and a third respectively of the error variance in source data points.

Errors in DEMs created from the TLS data points representing four terrain surfaces of different characteristics have been quantified using a statistical resampling method. The results show that terrain surface complexity can considerably affect the accuracy of DEMs. The effects of data point density (equivalent point spacing) on the DEM errors have also been analysed. For the data point spacings (35-100 mm) considered in the analyses, the DEM errors increased almost linearly with increasing data point spacing. The results also show that the DEM errors can be decomposed into two parts: a noise-related part and a data-density dependent part.

Repeat TLS surveys of some fixed objects have been carried out, to seek to empirically quantify the georeferencing-induced positional errors involved in repeated TLS surveys. The results indicate that repeated TLS surveys can measure millimetric deformations of smooth surfaces if a high georeferencing accuracy is achieved. The DEM errors, along with the georeferencing-induced positional errors, were used to infer the minimum magnitude of movements that can be measured by multi-temporal TLS surveys of rough terrain surfaces. In the case of the Newbury cutting considered in this study, the minimum level of detection was approximately 20 mm (at a 95% confidence level) for the data point spacing of 35 mm.
The findings in this research can aid in assessing the fitness of TLS surveys of terrain surfaces for a particular project, and thus are of use in the survey planning. The methods presented in this thesis can be applied to analyse errors in DEMs for making more meaningful interpretations of DEMs or surface variations derived from repeated TLS surveys.

More information

Identifiers

ORCID for J. Smethurst: orcid.org/0000-0001-8175-985X

ORCID for W. Powrie: orcid.org/0000-0002-2271-0826

Catalogue record

Export record