Assessment of an ASTER-generated DEM for 2D hydrodynamic flood modeling

https://doi.org/10.1016/j.jag.2010.05.007Get rights and content

Abstract

Flood modeling often provides inputs to flood hazard management. In the present work we studied the flooding characteristics in the data scarce region of the Lake Tana basin at the source of the Blue Nile River. The study required to integrate remote sensing, GIS with a two-dimensional (2D) module of the SOBEK flood model. The resolution of the topographic data in many areas, such as the Lake Tana region, is commonly too poor to support detailed 2D hydrodynamic modeling. To overcome such limitations, we used a Digital Elevation Model (DEM) which was generated from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) image. A GIS procedure is developed to reconstruct the river terrain and channel bathymetry. The results revealed that a representation of the river terrain largely affects the simulated flood characteristics. Simulations indicate that effects of Lake Tana water levels propagate up to 13 km along the Ribb River. We conclude that a 15 m resolution ASTER DEM can serve as an input to detailed 2D hydrodynamic modeling in data scarce regions. However, for this purpose it is necessary to accurately reconstruct the river terrain geometry and flood plain topography based on ground observations by means of a river terrain model.

Introduction

Effective flood hazard management requires reliable information about the characteristics of a flood event such as flood levels, flow velocities and the extent of the inundated areas. Such information can be obtained through hydrodynamic modeling that simulates the dynamics of a flood over a distributed model domain and for selected time periods.

In hydrodynamic flood modeling the availability of data at required spatial and temporal resolution constitute essential input data. Topographic and terrain data, together with roughness data are the most important information to implement a hydrodynamic model. For assessment of the flood simulation results flood observation data is required. Topographic data commonly is represented by means of a Digital Elevation Model (DEM) which, to the hydrodynamic model, serves as a principle source of topographic data for representing river terrain geometry and floodplain topography. The often encountered lack of accurate geometric data of river cross-sections leads to an erroneous representation of the hydraulic conveyance factor, which is calculated on the basis of the channel cross-section geometry. This error introduces uncertainty into the simulated flood dynamics, negatively affecting variables such as water levels, discharges, time to peak, and the size of the inundated areas. Accurate modeling relies on a high-resolution DEM (with a spatial resolution of less than 15 m) which is often missing, thus limiting the utility of advanced flood models in such cases. Nevertheless, in the absence of detailed surveyed cross-sectional data it is common to extract cross-sections from DEMs. For instance, Haile and Rientjes (2005) derive river cross-sections geometries from a Light Detection And Ranging (LiDAR) DEM using the HEC-RAS software application for one-dimensional (1D) modeling, which has been developed by the U.S. Army Corps of Engineers. Tate et al. (2002) proposed a GIS-based technique to import cross-section data created in the HEC-RAS software to reconstruct river terrain topography. To support advanced two-dimensional (2D) modeling, Merwade et al. (2008) proposed GIS techniques that include applications for (i) mapping and analysis of river channel data in a curvilinear coordinate system following the channel axis, (ii) interpolation of river cross-sections to create a three-dimensional mesh for the main channel and (iii) integration of interpolated 3D mesh with surrounding terrain.

In distributed flood modeling a DEM commonly serves to represent riverine terrain in terms of land surface elevations. DEMs are available at various resolutions that may range from 0.1 m for Airborne Light Detection And Ranging (LIDAR) to 90 m for Radar Topography Mission (SRTM). In reconstructing riverine terrain for hydrodynamic flood modeling, critical is to select a DEM with resolution that allows for accurate representation of river terrain characteristics like river banks and lining and flood plain topography. Selected resolutions also should favour the model execution time of the computationally demanding model algorithms not to result in excessive executions times of, for instance, 2–4 weeks. In the present study an ASTER DEM of 15 m spatial resolution is selected that, as we hypothesize, allows reconstruct of the riverine terrain while computation time is expected to be smaller than a maximum of 7 days. A description on the construction of the riverine terrain model as based on the ASTER DEM and limited field data is presented in Section 2.4. According to Cuartero et al. (2004), such DEM as generated by automatic correlation of ASTER stereo image data exhibits a root-mean-square error (RMSE) contained in a range between 10 and 30 m. Further, ASTER DEMs are known to be very accurate in near-flat regions and smoothly sloped areas, but are characterized by larger errors in areas covered by forest, snow and limited solar exposure, while errors as large as few hundred meters can be encountered in areas with steep cliffs and deep valleys (see Eckert et al., 2005). Gonçalves and Oliveira (2004) observe that for open water surfaces and cloud-covered areas stereo matching is not feasible and thus the estimation of surface elevation of wet riverine areas may be incorrect. ASTER DEMs are used for various applications, including geomorphometric analysis (Kamp et al., 2003), volcano topographic mapping (Kervyn et al., 2006), watershed boundary delineation (Pryde et al., 2007) and studies of glaciers and rockglaciers (Bolch, 2004). Their application in flood modeling is not very common.

Hydrodynamic flood model simulations require a series of field measurements on inundated areas and water levels for comparison purposes to simulated counter parts. Traditionally, observed data consist of time series of water levels but with advancements in Remote Sensing technology remotely sensed flood image data has become available which provide a complete coverage of the inundated areas. Commonly spatially distributed observations of water levels and data on the extent of inundated areas are used to calibrate and validate floodplain inundation models (for reference see Werner, 2004, Hunter et al., 2005). Horritt and Bates (2002) used synthetic aperture radar (SAR) images to calibrate and validate flooded area simulations for the River Seven in UK. Unlike optical sensors, SAR sensors are capable of capturing images under any weather condition during day- and nighttime, which makes this technique suitable for round-the-clock observation of inundated areas (see Mason et al., 2009). A setback however consists in the limited number of available images due to long revisit time of the respective satellite, which is of 35 days for the ERS1 and ERS2 satellites and 7–10 days for RADARSAT (see Townsend and Walsh, 1998). Moreover, the presence of vegetation such as forests, single trees, wind or flow turbulence can increase radar back-scatter effects, making the delineation of inundated areas problematic (see Smith, 1997, Imhoff et al., 1987).

In this study we propose to use images from the Moderate Resolution Imaging Spectro-radiometer (MODIS) sensor mounted onboard of the Terra satellite as multiple images of the same target area are available. A constrain for use of MODIS is that acquisitions require cloud free atmospheric conditions by the optical sensors. In Ticehurst et al. (2009), Low et al. (2004) and Zhan et al. (2002) the effective use of MODIS in flood mapping is shown. Further we note that the flood inundation maps produced by the Dartmouth Flood Observatory (see http://www.dartmouth.edu/∼floods) also are MODIS-based products.

Lake Tana, which is located at a latitude of 12°00′N, and longitude of 37°15′E in northern Ethiopia, is the largest open water body of the country and it is considered the source of Blue Nile River. The area of study is the Ribb catchment (1586 km2) which is one of the largest catchments of the Lake Tana basin area that drains into Lake Tana. The flooding of Ribb river plains is a regularly recurrent phenomenon. The flooding in August of 2006 is known for its particular long duration and devastating impact. It has resulted in the displacement of some 8728 people (Legesse and Gashaw, 2008). We simulated the flood dynamics in correspondence of the lower reach of Ribb River (a stretch of 20 km upstream of lake Tana) where flooding occurred by riverbank overflow and backwater effects by high lake water levels. The 2D flood modeling has been performed with the SOBEK hydrodynamic flow model (see Dhondia and Stelling, 2002). The ASTER DEM is used to represent the river terrain and flood plain topography of the riverine areas. In this study, GIS techniques that also include geostatistical interpolation are applied to develop a river terrain model from the DEM to derive the full river terrain geometry and flood plain topography. Subsequently the effect of selected interpolation settings for river terrain model development on simulated flood inundation extent is evaluated. The 2D hydrodynamic flood model is applied to study flood characteristics by river flows in Rib River and the backwater effect as induced by high water levels in Lake Tana. The flood inundation extent simulated by the model is compared with the MODIS inundation extent as well as spatially distributed historic observations of flood plain extensions.

The following objectives are identified for this study. Firstly, effectiveness of use of an ASTER DEM of 15 m resolution for 2D flood modeling has to be assessed. The DEM serves to reconstruct the river terrain geometry and the flood plain topography by advanced GIS processing and limited field data. Secondly, effectiveness of a 2D hydrodynamic flood model for the simulation of extreme flood events in the complex study area has to be assessed.

Section 2 describes the methodology of the flood modeling approach and the procedure to reconstruct the riverine terrain. Results of this research are presented and discussed in Section 3 with emphasis on validation issues for the terrain reconstruction as well as the flood modeling. The conclusions are presented in the fourth and final section.

Section snippets

Model approach

Two-dimensional hydrodynamic models consider the variation of flow in both, the longitudinal and the transverse directions of river channel. The principle underlying this kind of models is the solution the vertically integrated Navier–Stokes equations, also referred to as shallow-water equations. Applicability of advanced 2D hydrodynamic flood models has become within reach due to recent advances in topographic data acquisition technologies and due to the broad emergence of low-cost computing

Effect of interpolation in river terrain accuracy

The ASTER DEM allowed to derive river cross-sections of satisfactorily accuracy only at locations where the river channel carries low discharge. As such, the terrain of the river between these cross-sections should be reconstructed by interpolating elevation values. For this purpose the effects of the selected distance weight factors on the accuracy of the inverse distance interpolation scheme are evaluated. Such evaluation allows selecting the interpolation scheme which gives optimal results.

Conclusions

The floodplain of Ribb River is frequently flooded and the resident population is forced to leave their land each year. An assessment of the flood hazards requires accurate estimation of the flood depth, flow velocity and inundation areas through hydrodynamic flood modeling. For such modeling, river terrain geometry and floodplain topography need to be reconstructed but in this study is restricted by lack of in situ and high-resolution terrain data. To overcome the latter restriction we used an

References (31)

  • J.F. Dhondia et al.

    Application of one dimensional–two dimensional integrated hydraulic model for flood simulation and damage assessment

  • S. Eckert et al.

    Accuracy assessment of automatically derived digital elevation models from aster data in mountainous terrain

    International Journal of Remote Sensing

    (2005)
  • J.A. Gonçalves et al.

    Accuracy analysis of DEMs derived from ASTER imagery

    International Archives of Photogrametry and Remote Sensing

    (2004)
  • A.T. Haile et al.

    Rainfall variability over mountainous and adjacent lake areas: the case of Lake Tana basin at the source of the Blue Nile River

    Journal of Applied Meteorology and Climatology

    (2009)
  • A.T. Haile et al.

    Effects of LIDAR DEM resolution in flood modelling: a model sensitivity study for the city of Tegucigalpa, Honduras

    International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences

    (2005)
  • Cited by (72)

    • A multi-terrain feature-based deep convolutional neural network for constructing super-resolution DEMs

      2023, International Journal of Applied Earth Observation and Geoinformation
    • Urban flood modeling: Perspectives, challenges, and opportunities

      2022, Coastal Flood Risk Reduction: The Netherlands and the U.S. Upper Texas Coast
    View all citing articles on Scopus
    View full text