Towards global bare-earth DEM generation for urban flood simulation

Abstract

Flooding is one of the most frequent and costly natural hazards. The concentration of population and investments make urban areas the most vulnerable sites to flooding. The increasing trend of flood hazards and exposure makes the need for accurate flood mapping in urban areas even more pressing. The development of global consistent Digital Elevation Models (DEMs) based on spaceborne platforms is playing a significant role in global-scale flood inundation modelling which was impossible previously. However, in all the DEMs acquired to date by modern remote sensing techniques above-ground objects such as buildings, and vegetations result in artefact errors which have been largely overlooked in urban areas. For most circumstances of large-scale flood inundation modelling, bare-earth DEMs without these artefacts have to be used. This thesis proposed a set of algorithms for bare-earth DEM generation from DEMs with global coverage or having the potential of global coverage. For existing global DEMs (SRTM 1”, MERIT 3”, and TanDEM-X 3”), their artefact errors in urban areas were found to be 2.31-5.98 m, which highlights the clear need to remove these errors when most flood amplitudes are less than these values. To achieve this a regression model was proposed based on widely available datasets, which can reduce the artefact errors of the MERIT DEM by 15-67%. The MERIT-Urban Corrected (MERIT-UC) showed improved inundation results when used in an extreme fluvial flooding event. For the high-resolution photogrammetry DEMs (ArcticDEM, GoogleDEM), the Simple Morphological Filter (SMRF) is proved to be an advantageous filter. A direct application of SMRF in ArcticDEM showed a 70% reduction in artefact errors. To remove artefact errors in densely built-up areas, an iterative-based SMRF was developed and it reduced the artefact errors of GoogleDEM from 6.49 m to 0.98 m. Using the generated bare-earth GoogleDEM for a fluvial flooding scenario of the Yamuna river reduced the simulated water surface level error to 0.16 m which represents a 97% reduction compared to the original GoogleDEM.
The bare-earth DEM generation approaches developed in this thesis pave the way to the bare-earth DEM generation of the present and next generation of global DEMs. The ability to easily scale up these methods provides a useful basis for global-scale bare-earth DEM generation tasks. Moreover, this thesis also sheds light on further optimizing potentially wide-available DEMs for flood inundation modelling. The boosting use of these DEMs will be particularly beneficial to understand flood hazards at a global scale and especially in data-scarce regions.

Date of Award	21 Mar 2023
Original language	English
Awarding Institution	University of Bristol
Supervisor	Paul D Bates (Supervisor) & Jeff Neal (Supervisor)