Geometry of river networks by remote sensing

Abstract

River networks are the most recognizable and widespread branching pattern on the Earth’s surface. Geometry of river networks fundamentally constrains discharge process, and thus has prominent impacts on water resources distribution. Although theories of river networking mechanisms have been proven useful, large-scale measurements of river geometric features are still lacking. With Earth Observation (EO) data increasingly available for the entire globe, the shortage of river geometry measurements essentially means the absence of proper methods to process EO data effectively. Based on Sentinel-2, the first globally covered optical EO mission that freely provides a ground resolution up to 10 meters, this thesis explores an imagery-based methodology for the geometric extraction of river networks at large scales. Automatic surface water detection is the premise of the targeted methodology. To that end, this study firstly devises a new multispetral water index (MuWI) method for the native 10-m surface water detection based on Sentinel-2 imagery. The accuracy of MuWI method on several benchmark data sets for mapping among frequently confused low-albedo features (e.g., shadows) and sunglint outperfroms prior methods. Based upon the proposed automatic surface water detection, methods for extracting three geometric dimensions of river networks are introduced. The horizontal dimension of river networks, including river width and river topology, is extracted based on the graph theory that links edge and node for the cross-sectional measurements and representation of complex braided systems. The comparisons with the same-day ground observations show the robustness of this graph theory-based method and its potential application to the study of small rivers. The vertical dimension or the river bathymetry is investigated from a planform perspective that matches satellites' nadir view. Signifiant correlation between reach-scale river depth model parameter and planform geomorphologic features is revealed from fine-scale topobathymetry, which can be applied in river bathymetric estimation solely based on optical remote sensing. The reach-scale river behavior also contributes to the theoretical development of hydraulic geometry and the discovery of at-a-reach hydraulic geometry (ARHG). The longitudinal dimension focuses on the construction of drainage networks for an entirety of a major river basin (i.e., Lancang-Mekong River basin). A new integration method is developed to leverage optical remote sensing by combining observational river location information with imagery advantages in spatiotemporal resolution and accessibility. The new integration method is able to increase spatial resolution (to the level of used imagery), improve positional accuracy, and temporalize river networks representation. Overall, this thesis proposes an imagery-based methodology for river networks analysis facilitated by growing EO data, and contributes to the advancement of measuring and understanding the geometry of river networks on Earth and possibly on other planets. The new methodology serves as a paradigm in the field of remote sensing hydrology for the Fourth Paradigm, the concept of data-intensive scientific discovery.