Premium
Modeling DEM Errors in Coastal Flood Inundation and Damages: A Spatial Nonstationary Approach
Author(s) -
Karamouz M.,
Fereshtehpour M.
Publication year - 2019
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/2018wr024562
Subject(s) - kriging , digital elevation model , flood myth , computer science , probabilistic logic , terrain , interpolation (computer graphics) , gaussian , elevation (ballistics) , multivariate interpolation , algorithm , data mining , remote sensing , mathematics , geography , artificial intelligence , machine learning , cartography , geometry , archaeology , motion (physics) , physics , quantum mechanics , computer vision , bilinear interpolation
Digital elevation model (DEM) as an essential input to flood risk analyzers is subject to a certain level of uncertainty, which increases as the resolution becomes coarser. To quantify this uncertainty, a probabilistic framework incorporating 2‐D hydrodynamic flood mapping by LISFLOOD‐FP along with a sequential Gaussian simulation (SGS) model is presented in this paper. Based on ordinary kriging (OK) interpolation techniques, spatial uncertainty is modeled through SGS by generating several equiprobable realizations. Furthermore, regression kriging (RK) is used in the SGS algorithm utilizing its ability to explore additional information such as terrain characteristics to estimate the elevation error. A new technique called nonstationary sequential Gaussian simulation is also proposed by introducing the spatial nonstationarity of DEM error into SGS framework based on two approaches of uniform and nonuniform moving window. Different DEM resolutions resampled from a 1‐m light detection and ranging‐derived DEM are used in the hydrodynamic model to derive the accuracy‐efficiency trade‐offs and select the most suitable spatial resolution for probabilistic analysis. A detailed comparison among OK, RK, nonstationary OK, and nonstationary RK models is made considering probabilistic floodplain characteristics. The subsequent risk curves are then derived under floods of different return periods. The methodology is implemented in the coastal areas of Lower Manhattan and Brooklyn in New York City. Results show that the proposed methodology could reduce the uncertainty in risk estimation imposed by the stationarity assumption. The methodology could help decision‐makers to more accurately utilize the existing tools and data for flood risk analysis and preparedness.