Accuracy assessment of RS-based DEMs in flood inundation mapping of different morphological types of rivers

Flood is one of the most devastating natural hazards that causes significant losses and damages. Flood inundation maps are initial requirements for river floodplain management and eliminating flood risk in integrated flood management plans.Surveyed topography datasets, together with bathymetry data, were utilized to construct the river terrain for flood inundation mapping, however, such data is rare. Therefore, for tackling this challenging issue, DEMs as an accelerator and free of charge datasets, are widely used in flood modelling. The recent advancement of GIS and RS technologies enabled scientists to benchmark high-accuracy DEMs, such as LIDAR (Light Detection and Ranging), which perform as excellent as driven DEMs from surveyed topographies, for flood modelling. Despite the reliability of these finer datasets, the limited coverage of world and the exorbitant price of provision has forced experts to utilize coarser-resolution DEMs rather than these higher-resolution DEM sources. As a result, the accuracy of open-access and free of charge DEMs used in flood modelling in the data-sparse rivers should be investigated broadly, unless the lower-resolution DEMs hinder flood modelling results by incorrectly reproducing river terrain. In this study, three sets of widely used open-access DEMs’ performance in flood inundation mapping and estimating hydraulic parameters of four various rivers are assessed thoroughly.

In this paper, three sets of free and accessible 30m resolution DEM resources (ALOS, SRTM, ASTER) of four various sorts of rivers in Iran will be utilized as HEC-RAS geometry input file. In short, this study consistsof 5 main steps:1) Creating surveyed topography maps with 1:1000 or 1:2000 scales2) Generating GDEM by interpolating topography maps obtained from the prior step3) Determining channel geometry and extraction of cross-sections by operating GIS-based DEMs and GDEM in HEC-GeoRAS extension on ArcGIS4) Flood modelling of the rivers by geometry input files produced from step 3, plus US and DS boundary conditions and design flow of the rivers5) Assessment of DEMs’ performance in estimating hydraulic parameters based on efficiency measures, such as Root Mean Square Error (RMSE), Mean Absolute Difference (MAD), Mean Absolute Percentage Error (MAPE), Relative Error (RE) and F-statistics.6) Investigating the relation among the accuracy of DEMs and morphology characteristics of various Iran rivers, namely Sojasrud in Zanjan province, Taleghanrud in Alborz province, Gorganrud in Gorgan province and Sarbaz in Sistan-Baluchistan province of Iran.

The river geometry derived from ALOS DEMs were more identical to surveyed cross-sections. According to previous studies, ALOS DEMs captures the more accurate depicting river bathymetry data, the higher performance of this dataset in flood modelling is expected. However, all of these GIS-based DEMs were unable to present the Gorganrud river geometry data.Apart from the better representation of river geometry, the ALOS DEMs dataset is superior to ASTER and SRTM DEMs datasets in terms of fairy prediction of hydraulic parameters (water extents and water surface elevation). For instance, ALOS had the least RMSE (2.3-8.6 m) in predicting flood extent compared to the rest. Moreover, the higher value of F statistics (approximately 80%) proves that this model presented the flood inundation map with the highest agreement. Reversely, the RMSE of ASTER, the least accurate model, in flood extent estimation was from 2.8 to 15m. Identically, the predicted flood pattern map of these rivers using ASTER leads to the more significant discrepancy with the F-statistics of 68% at most. The SRTM performance in generating flood inundation map was better than ASTER and less accurate compared to ALOS (maximum value of 78% in F-statistic was recorded).The accuracy of DEMs in stimulating WSE of the mountainous rivers was similar to each other, and it was higher than predicted WSE of lower-slope rivers. Moreover, the estimated WSE using ALOS led to less disagreement with the benchmark values, whereas the operation of ASTER and SRTM for this purpose resulted in overestimating. Although the Relative Error of these DEMs in predicting Gorganrud river flood WSE was almost 5%, the MAPE of whole DEMs was 1% within all cross-sections in the rest of the study rivers.

The accuracy of flood inundation mapping is highly dependent upon its geometry input file. The high-resolution DEMs, including LIDAR dataset, are often used due to their accuracy, still in most of flood inundation mappings projects, coarser-resolution DEMs are operated owing to their accessibility and free of charge datasets. It is critical to evaluate the influence of these DEMs on hydraulic outputs. In this research, various open-access DEMs configurations were operated in four various rivers located in different regions of Iran. The results reveal that the ALOS DEM dataset is superior to ASTER and SRTM DEM datasets in terms of a better representation of river geometry and fairy prediction of hydraulic parameters (water extents and water surface elevation). For instance, the higher percentage of F statistics (approximately 80%) proves that this model presented the flood inundation map with the highest agreement. However, the maximum values of F-statistics of SRTM and ASTER were nearly 78% and 68%, respectively, showing the flaws of these models in flood extents mapping.The efficiency of all DEMs datasets in estimating WSE of the rivers was excellent, but Gorganrud river is an exemption. The Mean Absolute Percentage Error of these DEMs for estimating WSE, which was under 1% within the majority of cross-sections, was not meaningful.Consequently, ALOS is particularly potent in accurately hydraulic modelling. Additionally, the remote-sensing based DEMs are more applicable in wide and (or) straight river reaches than narrow and meandering rivers. It would be better if DEMs are modified in advance to reduce the level of disparities in flood modelling. In this line, there are several modification methods led to lowering this predictable errors, such as using GCPs (Ground Control Points) or merging cross-sections with DEMs to create an integrated surface.