Date

2022-2-04

Author

Arghavanian, Azar

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

355
views

0
downloads

Cite This

Efficient management of water resources is of great importance regarding ever increasing population and water droughts related to climate change. Drainage networks are important geomorphologic and hydrologic features which significantly control runoff generation. Recently, it has been feasible to monitor the earth’s surface and natural phenomena with the aid of remote sensing techniques. However, very high spatial resolution is needed for monitoring small waterways and knowledge of accurate drainage patterns is very important for correct modelling of hydrology. LiDAR (Light Detection and Ranging), which is an active remote sensing technique, uses near-infrared laser light to measure the distance from the sensor to a target on the earth’s surface. LiDAR data can be used in many scientific fields including topographic mapping, hydrology and land cover classification. This thesis explores the applicability of high-posting-density LiDAR data for terrain mapping with a focus on automated detection of natural and manmade water channels, especially those on plains and partially covered with trees and bushes. In this thesis, LiDAR data obtained around Bergama, Turkey, with the flying height of 1200 m, is used. LiDAR point measurements are influenced by three components: bare ground, above-ground objects, and noise. Our first purpose is to extract the vi ground surface from the above ground objects. A variety of methods are used for ground filtering of the LiDAR point cloud data, depending on the local environmental conditions. For the first aim of this study, a robust RANSAC based iterative strategy is proposed. The algorithm uses quadratic polynomial function in RANSAC mathematical model for local fitting to the LiDAR points as models of the ground surface. Generated surface patches are accepted based on the slope, residual distance from the inliers and percentage of inliers. Continuity of the surfaces from accepted neighboring patches are also used as a constraint. This way, regions grow from easy locations to complicated ones. Furthermore, for filling the gaps in the cleaned data where no fit could be reconstructed, Bézier surface fitting, which is a spline interpolation method, is iteratively applied to the points resulted from first stage. The second purpose of this study is to identify channel networks in complicated regions. Hence, curvature analysis is implemented using the Digital Terrain Model (DTM) product of first stage and quadratic fit coefficients in each neighborhood. Surface types are extracted from the estimated first and second fundamental forms and principal curvatures. Afterwards, channel hypotheses are formed based on valley surface type and geometric properties of the channel cross sections. When a cross section is accepted as a channel, the waterway is tracked on two perpendicular directions to extract the whole channel. The overall algorithm is tested on difficult areas and the success levels of DTM extraction and channel detection are shown quantitatively.

Subject Keywords

LiDAR, Digital terrain model, Surface fitting, RANSAC, Drainage network, Curvature, Channel detection, Channel tracking

URI

https://hdl.handle.net/11511/96754

Collections

Graduate School of Natural and Applied Sciences, Thesis