- Novel algorithm applicable for global river network extraction
- High-precision and complete global river networks products based on remote sensing imagery and hydrodynamic modelling
- Precise river width layer assosiated with river networks that could provide fundamental data source for many geoscience applications
Precise delineation of river networks is important for flood modelling and flood risk management, water resource management, understanding floodplain dynamics, and characterizing biogeochemical cycles. Traditionally, river networks are usually extracted from digital elevation model (DEM), which are implemented based on the raster grid of surface flow directions and rivers delineated by grids with large flow accumulation areas. Flow directions are generally approximated by conceptual or simplified approaches, such as single flow direction (SFD) (e.g. O'Callaghan and Mark 1984) and multiple flow direction (MFD) (e.g. Gallant and Hutchinson 2011) algorithms. The SFD algorithms restrict the flow movement to only one downhill direction. MFD algorithms allow the flow movement to more than one direction and consider flow divergence. However, they are incapable of representing certain common flow phenomena, such as flow over flat slopes, reversing flow, and backwater effect, which would impact the accuracy of extracted rivers. Thus, how to physically depict flow direction and water flow are currently not addressed in DEM-based methods of river extraction. Moreover, DEM-based methods depend on morphologically defined rivers, which cannot ensure the extracted rivers are real and differentiate between the perennial and intermittent flows in river networks. Satellite imagery, on the other hand, can provide continuous global coverage at high temporal and spatial resolution. Satellite imagery can be used to monitor hydrological changes over space and time and develop water surface mask to differentiate water and non-water pixels. And then real river centrelines can be created from water mask using morphological operations (e.g. Allen and Pavelsky, 2018). The morphological methods facilitate the extraction of channelized elements from local morphologies, without considering overall hydrological connectivity. Neglecting hydrological connectivity may lead to the creation of broken and inconsistent river segments, which may not represent the actual connectivity of the real-world river networks. Thus, further research is needed to explore how to take advantage of the river stream dynamics from satellite imagery and ensure the actual connectivity of the river networks. This PhD project therefore aims to develop a new generation of algorithms to integrate DEM-based methods with satellite imagery to delineate real, high-precision, and connected river networks for geoscience applications.
The flow direction and water flow will be physically depicted using the High-Performance Integrated hydrodynamic Modelling System (HiPIMS) developed at Loughborough University (Xia et al. 2019), which is able to physically simulate the dynamics of water flow across watersheds in accordance with continuity and momentum principles. HiPIMS will be used to realistically simulate global surface water hydrodynamics (water depth, inundation area and river discharge) by openly accessible global DEMs.
Cloud-computing technologies and services, such as Google Earth Engine, will be employed to analyse and manage a multi-decadal time series of satellite images over the globe. Machine learning algorithms will be used to process these massive images to delineate global water mask maps under different hydrological conditions. The integration of numerical modelling and remote sensing data analysis enable the development of real, high-precision, and connected river networks all over the globe, by effectively complementing each other's information.
Training and Skills
This project requires a student with a multidisciplinary background, straddling both hydrology and remote sensing. The student should have an aptitude for data processing, data analytics, numerical modelling and programming. For further development of key skills, the student will be able to benefit from in-house courses and NERC Advanced Training Short Courses in topics such as numerical modelling and forecasting in flood risk management and remote sensing application in flood management.
Year 1: The existing approaches and tools for river delineation will be systematically reviewed, and currently available global river maps, such as MERIT Hydro, HydroSHEDS and GRWL, will be quantitatively evaluated. The gaps and challenges to delineate global high-quality river networks will be identified and new method framework will be concepted.
Year 2: Cloud computation platform (Google Earth Engine, GEE) and machine learning (such as Random Forest model) will be used to classify Landsat images to produce global water mask maps under different hydrological conditions. HiPIMS will be set up to physically depict flow direction and water flow and perform scenario simulations across watersheds to simulate global surface water hydrodynamic by accessible global DEMs.
Year 3: Novel methods for river networks will be developed to integrate water mask maps from satellite imagery and surface water hydrodynamic from HiPIMS modelling to delineate high-quality river networks, which can ensure the realness and connectivity of rivers networks. The methods will be used to combine global water mask maps and HiPIMS modelling products to derive global high-precision river networks. Precise river width layer associated with river networks will also be produced according to water mask maps.
For further information, please contact Prof Qiuhua Liang (Q.Liang@lboro.ac.uk), Dr Huili Chen (H.Chen2@lboro.ac.uk), Dr Jiaheng Zhao (J.Zhao@lboro.ac.uk) or Dr Xilin Xia (X.Xia2@lboro.ac.uk). For enquiries about the application process, please contact Berkeley Young email@example.com, School of Civil and Building Engineering, Loughborough University. Please quote CENTA2 when completing the application form: http://www.lboro.ac.uk/study/apply/research/