Int J Disaster Risk Sci (2021) 12:890–902 www.ijdrs.

com
https://doi.org/10.1007/s13753-021-00377-z www.springer.com/13753

ARTICLE

The Importance of Digital Elevation Model Selection in Flood

Simulation and a Proposed Method to Reduce DEM Errors:
A Case Study in Shanghai
Kepeng Xu1,2 • Jiayi Fang1,2,3 • Yongqiang Fang1,2 • Qinke Sun1,2 • Chengbo Wu1,2 •

Min Liu1,2,3

Accepted: 15 October 2021 / Published online: 9 November 2021

Ó The Author(s) 2021

Abstract Digital Elevation Models (DEMs) play a critical management; and (2) the accuracy of a DEM corrected by
role in hydrologic and hydraulic modeling. Flood inunda- the Dixon criterion for predicting inundation extent is
tion mapping is highly dependent on the accuracy of improved, in addition to reducing errors in extreme water
DEMs. Various vertical differences exist among open depths—this indicates that the corrected datasets have
access DEMs as they use various observation satellites and some performance improvement in the accuracy of flood
algorithms. The problem is particularly acute in small, flat simulation. A freely available, accurate, high-resolution
coastal cities. Thus, it is necessary to assess the differences DEM is needed to support robust flood mapping. Flood-
of the input of DEMs in flood simulation and to reduce related researchers, practitioners, and other stakeholders
anomalous errors of DEMs. In this study, we first con- should pay attention to the uncertainty caused by DEM
ducted urban flood simulation in the Huangpu River Basin quality.
in Shanghai by using the LISFLOOD-FP hydrodynamic
model and six open-access DEMs (SRTM, MERIT, Keywords Digital elevation models  Dixon
CoastalDEM, GDEM, NASADEM, and AW3D30), and criterion  Hydraulic modeling  Shanghai  Urban
analyzed the differences in the results of the flood inun- flooding
dation simulations. Then, we processed the DEMs by using
two statistically based methods and compared the results
with those using the original DEMs. The results show that: 1 Introduction
(1) the flood inundation mappings using the six original
DEMs are significantly different under the same simulation Flooding has caused tremendous economic losses and
conditions—this indicates that only using a single DEM fatalities around the world (Najibi and Devineni 2018), and
dataset may lead to bias of flood mapping and is not ade- coastal cities are areas that are at significant risk from
quate for high confidence analysis of exposure and flood flooding (Aerts et al. 2014). The world population in
coastal cities has increased 4.5 times in the last 70 years
and coastal urbanization will likely continue in the coming
& Jiayi Fang decades (Barragán and de Andrés 2015). Urban areas are
jyfang@geo.ecnu.edu.cn expanding to receive large influxes of people into cities
& Min Liu (Dang et al. 2020). In recent decades there has been a
mliu@geo.ecnu.edu.cn massive migration of people towards the coastal regions of
1
School of Geographic Sciences, East China Normal
China (Niu and Zhao 2018), which has led to a large
University, Shanghai 200241, China concentration of wealth and population in these areas
2 (Dovern et al. 2014). This change increases coastal cities’
Key Laboratory of Geographic Information Science (Ministry
of Education), East China Normal University, exposure to risks (Surjan et al. 2016), especially flood risk.
Shanghai 200241, China The flood threat to coastal cities is getting worse due to
3
Institute of Eco-Chongming, East China Normal University, climate change, which leads to a high degree of uncertainty
Shanghai 200241, China in future flood risk (Deng et al. 2016; Lin et al. 2016; Gu

123
Int J Disaster Risk Sci 891

et al. 2019; Fang et al. 2020). The threat of flooding in its locations. Some studies have attempted to improve the
diverse forms ultimately results in economic losses and DEM, including the possibility of reducing DEM noises
human casualties. Therefore, an accurate assessment of through systematic editing, such as eliminating spurious
flood inundation in urban areas in coastal cities is vital to pits or sinks (Hutchinson 1989; Soille et al. 2003). Noise
understanding flood risks and providing precise disaster can also be reduced by filtering, for example by applying
forecasting and emergency response (Yang et al. 2020; Yin an adaptive-scale smoothing filter to remove random
et al. 2020). speckle noise (Gallant 2011). Nonetheless, such a revised
To better understand flood risk, flood inundation map- DEM does not necessarily perform well in any study area,
ping is critically important for identifying potential impact as different DEMs have distinct performances in different
areas and assessing inundation depths to evaluate the regions (Wong et al. 2014; Elkhrachy 2018). Some meth-
severity of flood hazards. Due to the improvement of ods revised errors based on the original product by deriving
computer performance, as well as the simplification and a completely new DEM, but high precision DEMs on a
development of the algorithms, hydrodynamic models have global scale still have the potential of having significant
been widely used for flood simulations (Lesser et al. 2004; errors at local scales (Holmes et al. 2000). Some current
Yu and Lane 2006). As the primary topographic input data, studies attempt to obtain more accurate elevation data
Digital Elevation Models (DEMs) have proven to be a vital through field investigation, such as the LiDAR with higher
element in controlling hydrodynamic model accuracy terrain accuracy or ground control point interpolation with
(Kenward et al. 2000; Cobby et al. 2001). Open access the help of unmanned aerial vehicles (Karamuz et al. 2020;
DEM products have been widely used in flood simulation Kim et al. 2020). However, the high economic cost and
and mapping (Pedrozo-Acuña et al. 2015). However, the time required for acquiring data make such methods chal-
relatively poor resolution and accuracy of open access lenging to apply to a large-scale area in flood simulations
DEMs at present significantly limit the ability to estimate (Aguilar et al. 2010).
the inundation areas and relevant risks (Sampson et al. Biases or artifacts in the DEMs are systematic errors due
2016). It has been demonstrated that low DEM data quality to the procedures used in the DEM generation (Wechsler
can lead to severe flood prediction biases (Hawker, Bates, 2007). Some studies focused on the vertical differences
et al. 2018). This is mainly affected by the spatial resolu- between different DEMs (Sanders 2007; Bhuyian and
tion and vertical error of DEMs. Low spatial resolution Kalyanapu 2018). However, most studies do not spatially
affects the delineation of surface features and the accuracy quantify how the uncertainty reflected in the vertical error
of the flood simulation (Vaze et al. 2010; Saksena and affects the results (Gesch 2018). Some studies evaluated
Merwade 2015). Elevation errors in the vertical direction better performing DEMs by comparing the quality of the
can also affect the accuracy of the terrain simulation and data within one or more study areas (Du et al. 2016; Zhang
hence the flooding simulation (Mukherjee et al. 2013; et al. 2019), but this does not mean that DEMs can have the
Talchabhadel et al. 2021). It has been recognized that same performance in other areas of interest. There is also a
accurate DEMs are critical for high precision flood mod- category of studies based on global error metrics, such as
eling and management (Cook and Merwade 2009; Coveney determining upper and lower bounds on elevation error
and Fotheringham 2011). (Kyriakidis et al. 1999). Although the error in the DEM is
Plenty of research has been done on improving DEMs not reduced, the boundaries of the error are determined.
for flood modeling. Digital Elevation Model noise correc- Some of the studies conducted error analysis through
tion and systematic DEM bias correction are two common probabilistic studies, and the relevant tools included using
control methods. In high-precision flood modeling, the the Monte Carlo technique (Wechsler and Kroll 2006) and
error of terrain attributes reflected by DEMs will affect the sequential Gaussian simulation (Fereshtehpour and Kar-
simulation of channel morphology and bathymetry in flu- amouz 2018). Such operations do not produce a precise
vial floods. Coarse resolution of DEMs is problematic for result but a map that contains the spatial distribution of
determining channel morphology, signal attenuation of possible errors, thus indicating the likelihood of any posi-
water bodies, or radar reflections, and this issue cannot be tion falling above or below a specified elevation. A mod-
sufficiently addressed. ified deterministic and probabilistic approach to vertical
Errors in measurements appear as noise in the elevation uncertainty is considered a better option than a simple
data, including stripe noise or random speckle noise. Stripe deterministic approach that ignores the effects of elevation
noise affects extensive range but can be easily identified errors (Gesch 2018).
and eliminated, such as by using the two-dimensional In this study, we generated a new DEM by taking
Fourier filtering technique to detect unrealistic terrain advantage of several sets of DEMs through considering the
undulations and remove stripe noise (Yamazaki et al. interconnections and differences between current open
2017). Speckle noise occurs randomly and in uncertain access DEMs. We considered whether existing datasets can

123
892 Xu et al. DEM Selection in Flood Simulation and a Method to Reduce DEM Errors

be corrected to eliminate errors and, based on this idea, The Huangpu River is a typical lowland river in
provide a statistically based method and apply it in the Shanghai, which originates from the Dianshan Lake—one
DEM. This is an entirely new DEM but takes full advan- of the lakes in the Taihu Lake Basin—and flows into the
tage of every raster information of the original DEMs data. Yangtze River Estuary. It is the largest river in the Taihu
The significant benefits of this method are the convenience Lake Basin, carrying 70% of Taihu Lake’s water flow (Yin
of the technique and the fact that missing or incorrect et al. 2013). The mainstream of the Huangpu River extends
partial DEM data do not limit the results generated for the from Mishidu gauge station in the upper reaches to the
particular study area. The results generated by the method Wusongkou gauge station in the lower reaches, with a total
can also be continuously updated as the number of DEM length of about 75 km, running through the city center of
products increases, and theoretically will be closer to the Shanghai (Fig. 1). The mainstream of the Huangpu River
accurate elevation values. This study investigated the fea- was selected as the study area to illustrate the impact of
sibility of the new statistically based method for elimi- different DEMs on inundation simulation. The study area
nating DEM error. This was achieved by: (1) simulating constitutes the central part of Shanghai Municipality,
and comparing the flooding results from the currently covering over 3,000 km2, where more than 13 million
available open access DEMs; (2) analyzing and evaluating people live.
the performance of the newly generated DEM versus the
original DEMs in flood simulation; and (3) discussing the 2.2 Digital Elevation Models
error factors present in typical DEMs in the study area.
We considered seven global DEM datasets that are freely
accessible: SRTM, MERIT, CoastalDEM, GDEM,
2 Materials and Methods AW3D30, NASADEM, and TanDEM-X. As there is a lot
of missing information in TanDEM-X for the study area,
The following subsections present a brief introduction of the first six datasets were chosen for the study (Fig. 2).
the study area and the employed datasets and their prop- SRTM: The first version of the Shuttle Radar Topog-
erties, followed by the methods. raphy Mission (SRTM) products was released in 2003 with
30 m and 90 m horizontal resolutions. The published
2.1 The Shanghai Study Area datasets have been processed using the interpolation algo-
rithm to fill the data hole of SRTM (Reuter et al. 2007).
Shanghai is located on the alluvial plain of the Yangtze The absolute elevation errors of SRTM at the 90% quantile
River Delta in China, with a dense population of more than (LE90) ranged from 5.6 cm to 9.0 m (Rodrigue et al.
24 million and produced over USD 550 billion (RMB 3.8 2006). Both 30 m and 90 m resolution DEMs were adopted
trillion) of gross domestic product (GDP) in 2019, which is in this study, with the 90 m data used for error correction.
3.8% of China’s national GDP (SMSB 2020). It is a low- MERIT: The MERIT was developed by removing
lying area, with an average elevation of 4 m. However, the various error components, including absolute deviation,
estuary system’s average tidal amplitude can reach up to fringe noise, speckle noise, and tree height deviation from
4.6 m (Wang et al. 2012). Due to the low-lying terrain and existing DEMs (SRTM, AW3D, Panoramas DEM). Data
the high tidal amplitude, Shanghai suffers complicated for all regions except Antarctica are provided. The absolute
types and frequent high flood events and is one of the most elevation error at the 90% quantile is 5 m (Yamazaki et al.
vulnerable cities affected by floods in the world (Balica 2017). We chose MERIT version 1.0.3 with a 90 m reso-
et al. 2012). More than 1,800 flood-related casualties were lution for this study.
reported from 1949 to 2005 (Wen and Xu 2006). These CoastalDEM: The CoastalDEM dataset is exclusively
devastating flood disasters occurred in association with developed for coastal areas, which is based on SRTM data.
various meteorological hazards, such as typhoons, heavy Using machine learning techniques, the accuracy of coastal
rains, prolonged precipitation, and riverine flooding. terrain in this dataset has been effectively improved by
Typhoon Winnie in 1997 caused the Huangpu River to cutting *50% root mean square error (RMSE) compared
reach an extreme water level of 5.99 m (Du et al. 2015), with SRTM (Kulp and Strauss 2018). Moreover, more
with an inundation area of 495 km2 in the city center, coastal areas are exposed to floods based on this dataset
causing seven deaths and direct economic losses of USD (Kulp and Strauss 2019). This dataset has 30 m and 90 m
80 million (RMB 670 million). Although the Typhoon versions, and only 90 m horizontal resolution has free
Winnie event is extremely rare, Shanghai is still likely to access for non-commercial use, which was adopted in this
suffer from severe flooding events in the future due to study.
global sea level rise and land subsidence. GDEM: ASTER GDEM is a global one-arc-second
elevation dataset based on the products of the new Earth

123
Int J Disaster Risk Sci 893

Fig. 1 Location of the study

area—the Huangpu River
Basin—in Shanghai, China

observation satellite Terra released by METI and NASA. It Four sets of 30 m resolution (AW3D30, NASADEM,
has the advantage of broad data coverage, covering most of GDEM, STRM 30 m) and three kinds of 90 m resolution
the Earth’s regions between 83°N and 83°S latitude. The DEM data (SRTM 90 m version, MERIT, CoastalDEM)
RMSE of ASTER elevations was estimated to be 8.68 m were selected for this study. Figure 2 shows the six sets of
(Tachikawa, Hato, et al. 2011, Tachikawa, Kaku, et al. open access DEMs and their basic information. Among
2011). The GDEM version 2 released in October 2011 them, MERIT, CoastalDEM, and NASADEM are based on
improves the spatial resolution by using 260,000 additional SRTM data modified by the algorithm. The GDEM data
stereo pairs, and improves horizontal and vertical accuracy source is from the ASTER satellite, and AW3D30 is from
(Tachikawa, Hato, et al. 2011), and this version was the ALOS satellite.
adopted in this study. The table in Fig. 2 shows the basic properties of the six
NASADEM: NASADEM reprocesses SRTM data and datasets in the study area. CoastalDEM has the lowest
merges with ASTER GDEM elevations to improve height average elevation and GDEM has the highest average
accuracy. The new and improved SRTM heights in height in the study area, and CoastalDEM’s average ele-
NASADEM come from better vertical control by referring vation is more than three times of GDEM. In terms of
to the Ice, Cloud, and Land Elevation Satellite (ICESat), elevation distribution, the five sets of DEMs, except
and gaps in SRTM are reduced by using interferometric CoastalDEM, have the largest proportion of elevation in
unwrapping algorithm (Crippen et al. 2016). NASADEM the more than 4 m range. In contrast, in the CoastalDEM
covers 80% of Earth’s land regions and includes land only 2.0% of the study area is higher than 4 m, and nearly
between 60°N and 56°S latitude. The version used in this half of the area (43.8%) has less than 2 m elevation.
study was released in February 2020 with 30 m horizontal
resolution. 2.3 Methodology
AW3D30: ALOS World 3D 30 m data are obtained by
the Panchromatic Remote-sensing Instrument for Stereo The following subsections present the three main processes
Mapping (PRISM) on the advanced land observation of the study including Digital Elevation Model processing,
satellite ALOS (Tadono et al. 2014). The elevation values simulating flooding events, and assessing inundation sim-
are obtained by software calculations of the position of the ulation accuracy.
same feature imaged on three cameras with different
viewing angles, which effectively improves accuracy. 2.3.1 Digital Elevation Model Processing
AW3D30 is a resampled version for non-commercial use
from AW3D5 with a grid size of 30 m with global coverage We first extracted DEMs of the study area including
from 83°N to 82°S. The RMSE of AW3D versus 5,121 coordinate transformation and clips. Then we used two
points distributed across 127 image tiles were 4.40 m methods to process the DEMs. Before processing, resam-
(Takaku et al. 2016). The AW3D30 data in this study is pling was done to ensure that the spatial resolution of the
from version 2.3 released in April 2019 with 30 m hori- six datasets was consistent. The resampling method was
zontal resolution. nearest neighbor, which has been found to lead to the
highest accuracy in DEM resampling (Takagi 1998).

123
894 Xu et al. DEM Selection in Flood Simulation and a Method to Reduce DEM Errors

Fig. 2 Six Digital Elevation Model (DEM) datasets used for the study in Shanghai, China (Inserted table: Basic information of the six DEMs
used in this study)

The first method to process the DEMs is to directly take the application to the processing of DEM data and analyzed
the mean of the same row and column elevation values for the elevation data at the same location (rows, columns) of
the six datasets (Mean). The second one is to remove error different DEM products, eliminating data with large errors
elevations by the Dixon method before direct averaging. and retaining data with elevations in the normal range
The revision principle is based on the statistical method compared to all DEM data at that location.
of eliminating bad values from the data, reducing errors, Mean: The original high-resolution data was first
and thus improving the accuracy of the data. We extended resampled to a lower resolution to ensure consistency

123
Int J Disaster Risk Sci 895

across the datasets. Then a new elevation value z was of the one-dimensional channel and two-dimensional flood
generated by averaging the elevation data from several area with two corresponding solutions. For one-dimen-
datasets of the original DEMs at the same row (x) and sional channel water simulation, the simplified St. Venant
column (y), assigning that value to the same location, and equations are used in the model.
so on to generate a new DEM. The hydrological simulation of a two-dimensional flood
Dixon criterion: Compared with the direct average, this area needs to use the continuity equation and momentum
method adds the process of eliminating the error value by equation based on the terrain data of the DEM. Considering
the Dixon method, which is suitable for checking the the water balance of adjacent grids:
consistency of a set of measured values when the amount dV
of data is limited (Dixon 1950). This method screens out ¼ Qup þ Qdown þ Qleft þ Qright ð1Þ
dt
error values by using the extreme difference as a metric to 1=2
estimate the difference between adjacent data and to Aij Rij Sij
Qij ¼ ð2Þ
identify excessive differences as anomalous data. Dixon’s n
test can directly detect outlier values by using the range where V is the total flow per grid, Qup, Qdown, Qleft, and
ratio without calculating the arithmetic mean and standard Qright respectively relate to the flow rates of upstream,
deviation (SD) of the sample, and therefore is suitable for downstream, left, and right units adjacent to the grid,
small-sample data. It performs well with small sample sizes t represents time, and Qij represents the flow between grid
and does not require assumptions about the normality of i and grid j; Aij and Rij represent the cross-sectional area
the data. This method is not ideal when the maximum and and hydraulic radius at the junction of adjacent grids i and
minimum values are both suspicious, or two suspicious j respectively; Sij represents the water slope between i and
values exist on the same side of the maximum (minimum) j, and n is the manning coefficient.
value. To assess the uncertainty in inundation simulation result
In this study, the Dixon criterion thresholds refer to the caused by DEM errors, we carried out the sensitivity
national standards of the People’s Republic of China analysis by changing the input DEM while keeping other
(SAMR 2008), as shown in Table 1. Since the number of input conditions the same. We did not take the uncertain-
samples (DEMs) is less than 10, the table lists only n values ties such as sea level rise, ground subsidence, and
up to 10. embankment into consideration. In the water level height
design, we referred to the water level height designed in
2.3.2 Flooding Simulation relevant research (Yin et al. 2013). In the inundation sim-
ulation, we evaluated the differences between two sets of
We chose the hydrodynamic model LISFLOOD-FP to water level data of 50-year (50a) and 100-year (100a)
simulate flood inundation, which is widely used in flood return periods and compared the two scenarios’ average
simulation and mapping (Fewtrell et al. 2008; Zhao et al. values.
2020). LISFLOOD-FP is a simplified two-dimensional
hydrodynamic raster-based inundation model (Bates and 2.3.3 Assessment of Inundation Simulation Accuracy
De Roo 2000). It can simulate floodplain inundation in a
computationally efficient manner over complex topogra- We used historical disaster conditions and relevant litera-
phy, which can also be used for hydrodynamic simulation ture (Yuan 1999) to determine areas prone to flooding
along the Huangpu River and assessed the accuracy of the
inundation simulation by using the binary classification.
Table 1 Dixon criterion threshold, national standards of the People’s
Republic of China The binary classification method is applicable in areas with
small slope variations, with good performance in flood
n Statistics 1-a
simulation (Stephens et al. 2014; Samela et al. 2017), and
Low-Level Noise High-Level Noise 0.99 0.95 is therefore appropriate in this study. The two-dimensional
5 0.821 0.710
matrix includes four scenarios, with accuracy based on a
x x
c0 ¼ xð2Þ ð1Þ
c¼
xðnÞ xðn1Þ comparison with the observed and actual results, including
6 ðnÞ xð1Þ xðnÞ xð1Þ 0.740 0.628
dry and wet indicators, as shown in Table 2.
7 0.680 0.569
Once the observed area and simulation area are deter-
8 0.717 0.608 mined, the accuracy of the submergence simulation should
x x xðnÞ xðn1Þ
9 c0 ¼ xðn1Þ
ð2Þ ð1Þ
xð1Þ c¼ xðnÞ xð2Þ 0.672 0.564 be evaluated by calculating the F1 Score. The F1 Score is a
10 0.635 0.530 measure of both precision and recall, representing the mean

123
896 Xu et al. DEM Selection in Flood Simulation and a Method to Reduce DEM Errors

Table 2 Contingency table of the confusion matrix with true positive 3.1 Effect of Horizontal Resolution
(TP), false negative (FN), false positive (FP), and true negative (TN)
values
We used the same data but with two resolutions, that is, the
Simulated 30 m and 90 m SRTM datasets, to compare flood inunda-
Wet Dry tion under the same scenario setting. The results are shown
in Fig. 4. Comparing the results of the two datasets, we
Observed Wet True Positive (TP) False Negative (FN) found that the SD, MAX, and inundation area of the flood
Dry False Positive (FP) True Negative (TN) simulation results increase as the return period increases.
The most significant change is in inundation extent. The
change in inundation area is 20.7% between the two return
periods for the 30 m resolution DEM and 18.6% for the 90
of the reconciliation of accuracy and recall rates. A higher m resolution DEM. The smallest change is in SD, where
F1 value indicates better accuracy: the change is 2.9% for the 30 m resolution DEM and 2.6%
  for the 90 m resolution DEM. This result illustrates that the
1 1 1 1
¼ þ ð3Þ flooding simulation error increases as the inundation level
F1 2 Precision Recall
  increases with the increase of the return period. But the
TP
Precision = ð4Þ
TP + FP
 
TP
Recall = ð5Þ
FN + TP
Figure 3 shows a simple diagram that illustrates how to
determine the quality of flooding simulation results by
binary classification (Jafarzadegan and Merwade 2017).

3 Results

This section compares the inundation simulation results

using six sets of original DEMs and two sets of processed
DEMs. Considering that the proposed method resamples
the DEMs to reduce the spatial resolution, we also compare
two sets of DEMs with the same data source but different
spatial resolutions and analyze these results.

Fig. 4 Comparison of the results of flood inundation simulation using

two SRTM datasets under two inundation scenarios (50a and 100a
return periods) on the Huangpu River in Shanghai, where 4a and 4c
represent results using the original 30 m SRTM dataset, and 4b and 4d
represent results using the original 90 m SRTM dataset. SD represents
Fig. 3 An example for understanding the binary classification terms the standard deviation (in meters), Max represents the simulated
including the rate of true positive (TP) and false positive (FP), and the maximum water depth (in meters), and Area refers to the inundation
F1 score used to validate flood inundation simulation accuracy area (km2)

123
Int J Disaster Risk Sci 897

choice of different spatial resolutions may lead to differ- water level scenario is 1.7 times that of the 50a water level
ences in the results, where coarser resolutions lead to a scenario, and CoastalDEM is less sensitive, with a pre-
degradation of SD and maximum water depth while dicted range of 1.1 times under the 100a water level sce-
increasing the inundation area, but the overall change is not nario compared with the 50a.
significant. Comparing the two sets of processed DEMs, the inun-
dation range of the simulated results, after directly aver-
3.2 Cross-comparison of Eight Sets of Digital aging elevation, is smaller than that of the processed results
Elevation Models of the Dixon method. This is due to the excessive height of
GDEM affecting the final results, while the difference
The results of the simulation using the eight sets of DEMs between the results of the Dixon method and the Mean
under two flooding scenarios are shown in Fig. 5. The shows the superior performance of the Dixon method in
results using the six sets of original data show an increase outlier identification and error elimination.
in the maximum inundation depth, SD, and inundation area
with the increase of the return period, which indicates that 3.3 Assessment of Inundation Simulation Accuracy
the error of the simulation increases with the increase of the
return period. The results of the simulation using DEMs Based on the binary approach presented in Sect. 2.3.3, F1
processed by the Mean method and the Dixon method also values for using different DEMs under the two return
follow this rule. Comparing the results of the same DEM periods are shown in Table 3, where a higher F1 value
under different flooding scenarios, and the effect of dif- indicates better overall accuracy.
ferent DEMs used under the same flooding scenario, shows Comparing the F1 values of simulation using the six sets
that the effect of different DEMs on the results of the of raw DEM data, AW3D30, MERIT, NASADEM, and
flooding simulation is significantly greater than that of the SRTM perform better with water level deepening. GDEM
scenario selection, which also shows the importance of remains unchanged and CoastalDEM performance
selecting a suitable DEM for flooding simulation. decreases due to further expansion of the inundation extent,
The maximum inundation depth using CoastalDEM is resulting in an overprediction. MERIT performs best
the largest of the eight datasets for both the 50a and 100a among the original DEM datasets in determining the
flooding scenarios, whereas GDEM shows the smallest inundation area. Although using CoastalDEM the simula-
inundation depth. The maximum inundation depths for the tion identified most waterlogging-prone regions, the accu-
remaining six datasets ranked in the middle and are less racy of the result ranks fairly low due to the high
different from each other. The results obtained using the inundation range and large error. GDEM identifies most
two sets of processed DEM significantly reduce the depth areas as dry, however, and its accuracy is the lowest due to
of extreme water levels compared to the original DEMs, considerable ignorance in inundation area identification.
with the results using the DEM generated from the direct Comparing the two modification methods, there is a
averaging process being more pronounced. The Dixon considerable gap between the simulation accuracy of the
criterion also significantly reduces the maximum depth of DEM processed by the two methods. Except for GDEM,
inundation compared to the original data. the F1 values of the direct averaging elevation method are
The inundated area results in the study area are shown in even lower than all the original data, indicating that the
Fig. 6. The simulated inundation area varies significantly process ignores the error of elevation blindly and averaging
between the different DEMs. The smallest inundation area not only cannot improve the value of the anomalous areas,
results from using GDEM—the average inundation area is but also result in error of some accurate elevation values. In
less than 2 km2. In comparison, the most massive inun- contrast, the data corrected by the Dixon criterion obtained
dation area results from using CoastalDEM, with an aver- the highest F1 values among all groups of data, which
age inundation area of more than 1600 km2. The reflected the positive effect of DEM correction.
CoastalDEM simulation results show that the inundation
area is significantly higher than from the other datasets,
with inundation areas exceeding 50% of the study area for 4 Discussion
both the 50a and 100a scenarios, while other datasets are
less than 15%. The difference in the predicted area between Comparing the results between 30 m and 90 m SRTM
the maximum inundation area and the minimum inundation datasets, we found that coarser resolution leads to an
area is more than 1200 times. In addition, inundation ran- increase in inundation extent, which is consistent with
ges from using different DEMs have various sensitivities to previous research (Saksena and Merwade 2015; Lim and
water level height settings, with the Mean being the most Brandt 2019). Coarser resolution reduces the maximum
sensitive. The predicted inundation range under the 100a depth of inundation, which may be due to more

123
898 Xu et al. DEM Selection in Flood Simulation and a Method to Reduce DEM Errors

Fig. 5 Results of flood inundation simulation using six sets of raw Shanghai. SD represents the standard deviation (in meters), Max
DEM data and two sets of processed DEM data under two inundation represents the simulated maximum water depth (in meters), and Area
scenarios (50a and 100a return periods) on the Huangpu River in refers to the inundation area (km2)

123
Int J Disaster Risk Sci 899

high-rise buildings corresponds to the area with the lowest

elevation in CoastalDEM, which illustrates the anomaly of
CoastalDEM. It may overestimate the inundation areas and
exposure.
The SRTM continues to use the old boundary, neglect-
ing on-going coastal land reclamation in the past two
decades. It makes coastal flood simulation more difficult as
the current DEM does not show the real land boundary.
Thus, particular caution is needed in flood modeling,
especially in areas with significant topographic changes,
coastal areas, and low-lying flat areas.
This article presents a statistically based approach
designed to provide flood modelers with a fresh approach
Fig. 6 Simulated inundation area of the Shanghai study area under
two return periods (50a and 100a) to combining current open access datasets to complement
each other to achieve possible performance improvements.
Table 3 F1 scores for using different Digital Elevation Models Although multiple DEM correction methods are available,
(DEMs) under two return periods (50a and 100a) in Shanghai they have been limited to correcting one DEM. Our
Dataset 50a 100a
approach makes correction to the same region possible
based on various sets of currently available DEMs. It can
AW3D30 0.35 0.44 combine the advantages of different datasets, screen out the
MERIT 0.52 0.60 outliers with errors, and can be used in plug and play
NASADEM 0.47 0.49 mode—it only needs to modify the region of interest before
CoastalDEM 0.45 0.40 simulation, and is less time and cost consuming.
GDEM 0.02 0.02 Although highly accurate terrain data improve the pre-
SRTM 0.29 0.42 cision in terrain simulation, the data are not suitable for
Mean 0.23 0.34 direct use in flood simulation due to the limited computa-
Dixon Criterion 0.55 0.64 tional performance. Halving the resolution of the simula-
tion input data resulted in a 10-fold increase in
computational costs (Savage et al. 2016). Thus, even if
considerable anomalous heights being smoothed into higher resolution DEMs can be used, they may only be
coarse height values by the surrounding lower altitudes. It modelled in a coarser resolution to run the simulation. This
may also neglect the extreme inundation areas if relatively means that flood modeling still requires the use of coarse-
local low-lying regions exist. The use of different DEMs scale DEMs at present, which sacrifices the advantages of
for the same scenario simulation has a significant effect on highly accurate topographic data and exposes a disadvan-
the inundation simulation, where the GDEM with the tage to the time and economic cost due to highly accurate
smallest inundation area differs by a factor of over 500 topographic data in flood modeling. Thus, relatively pre-
times from the CoastalDEM with the largest inundation cise terrain data that maintain current computing perfor-
area. This indicates that the choice of DEM input has a mance is a more cost-effective research direction. The idea
significant impact on urban flood simulation and demon- proposed in this article deals with removing some error
strates the critical importance of selecting a suitable DEM points present in the DEMs at the current resolution and
for flood modeling. getting some performance improvement. We hope that the
The quality of DEMs in flood modeling has also been idea can help flood modelers promote the quality
widely considered (Schumann and Bates 2018). Previous improvement of terrain data in flood simulations. It
studies pointed out that those openly accessible and widely requires more effort to understand the DEM effects in flood
used DEMs, SRTM for instance, were acquired in the simulation as well as in other Earth system simulations.
2000s and have numerous errors (Hawker, Bates, et al.
2018, Hawker, Rougier, et al. 2018). Although constantly
updated DEMs, such as MERIT, NASADEM, and Coast- 5 Conclusion
alDEM, have performance improvements, it does not mean
that they are entirely accurate. For example, in our study, In this study, we focused on comparing the inundation
the CoastalDEM misidentifies large areas of high-rise simulation results of six open access DEMs and proposed a
buildings in downtown Shanghai as being below sea new method to eliminate DEM errors in the study area of
level—as shown in Fig. 7, the area with a high density of Shanghai. The study resulted in three main findings:

123
900 Xu et al. DEM Selection in Flood Simulation and a Method to Reduce DEM Errors

Fig. 7 Comparison of CoastalDEM elevation (left), OpenStreetMap (OSM) buildings data (middle), and kernel density estimation result based
on OSM data (right) in Shanghai

1. From the sensitivity analysis, by setting the same checking all DEM data that can be obtained against
simulation conditions but only changing DEMs for each other in a horizontal comparison. At the same
flooding simulation, the results show significant time, the idea has operability and can be used by
differences, such as the difference between the inun- researchers to autonomously determine the study area
dation results of CoastalDEM and GDEM. This as well as the error eliminating method without waiting
implies that more care needs to be taken in DEM for upgraded products to achieve the error reduction.
selection in flood simulation. It is necessary to select a In future research, we will continue to maintain the
more appropriate DEM in the preparation phase of advantages of this idea of convenience and flexibility,
future flooding simulations. and further investigate the methods of error identifi-
2. Our study also found that even when selecting the cation such as machine learning to achieve more
same DEM, different spatial resolutions for flooding accurate results.
simulations also lead to differences in results. As the
spatial resolution of the DEM decreases, the predicted Acknowledgments This work is funded by the National Natural
flood inundation area and the maximum inundation Science Foundation of China (42001096, 41730646); the Shanghai
depth increases for all DEMs. This implies that the Sailing Program (19YF1413700); the China Postdoctoral Science
selection of a coarser DEM may lead to more errors in Foundation (2019M651429); East China Normal University Institute
of Belt and Road & Global Development (ECNU-BRGD-202106),
the inundation results. Nevertheless, the effect of and the National Key R&D Program of China (2017YFC1503001,
spatial resolution on the difference of inundation 2017YFE0100700).
results is much smaller compared to the choice of
DEM in flooding simulation. Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
3. Although the inundation depth results are difficult to adaptation, distribution and reproduction in any medium or format, as
compare due to the scarcity of historical disaster long as you give appropriate credit to the original author(s) and the
records, the predictive performance of the elevation source, provide a link to the Creative Commons licence, and indicate
data for inundation areas after Dixon criterion and if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless
error elimination processing is improved compared to indicated otherwise in a credit line to the material. If material is not
all six sets of original data. The potential of the method included in the article’s Creative Commons licence and your intended
is demonstrated, and a new way of thinking is use is not permitted by statutory regulation or exceeds the permitted
proposed for flooding simulations that suffer from use, you will need to obtain permission directly from the copyright
holder. To view a copy of this licence, visit http://creativecommons.
current DEM limitations. This method can have the org/licenses/by/4.0/.
effect of reducing topographic errors by cross-

123
Int J Disaster Risk Sci 901

References Fereshtehpour, M., and M. Karamouz. 2018. DEM resolution effects

on coastal flood vulnerability assessment: Deterministic and
Aerts, J.C.J.H., W.J.W. Botzen, K. Emanuel, N. Lin, H. de Moel, and probabilistic approach. Water Resources Research 54(7):
E.O. Michel-Kerjan. 2014. Evaluating flood resilience strategies 4965–4982.
Fewtrell, T., P. Bates, M. Horritt, and N. Hunter. 2008. Evaluating the
for coastal megacities. Science 344(6183): 473–475.
Aguilar, F.J., J.P. Mills, J. Delgado, M.A. Aguilar, J.G. Negreiros, and effect of scale in flood inundation modelling in urban environ-
J.L. Pérez. 2010. Modelling vertical error in LiDAR-derived ments. Hydrological Processes 22(26): 5107–5118.
digital elevation models. ISPRS Journal of Photogrammetry and Gallant, J. 2011. Adaptive smoothing for noisy DEMs. Geomor-
phometry 2011, 07–11 September 2011, Redlands, California.
Remote Sensing 65(1): 103–110.
Balica, S.F., N.G. Wright, and F. van der Meulen. 2012. A flood https://gis-lab.info/docs/gallant2011_adaptive_smoothing_for_
vulnerability index for coastal cities and its use in assessing noisy_dems.pdf. Accessed 18 Oct 2021.
climate change impacts. Natural Hazards 64(1): 73–105. Gesch, D. 2018. Best practices for elevation-based assessments of
Barragán, J.M., and M. de Andrés. 2015. Analysis and trends of the sea-level rise and coastal flooding exposure. Frontiers in Earth
world’s coastal cities and agglomerations. Ocean & Coastal Science 6: Article 230.
Management 114: 11–20. Gu, X., Q. Zhang, V.P. Singh, C. Song, P. Sun, and J. Li. 2019.
Bates, P.D., and A.P.J. De Roo. 2000. A simple raster-based model Potential contributions of climate change and urbanization to
for flood inundation simulation. Journal of Hydrology 236(1): precipitation trends across China at national, regional and local
54–77. scales. International Journal of Climatology 39(6): 2998–3012.
Bhuyian, M., and A. Kalyanapu. 2018. Accounting digital elevation Hawker, L., P. Bates, J. Neal, and J. Rougier. 2018. Perspectives on
Digital Elevation Model (DEM) simulation for flood modeling in
uncertainty for flood consequence assessment. Journal of Flood
Risk Management 11(S2): S1051–S1062. the absence of a high-accuracy open access global DEM.
Cobby, D.M., D.C. Mason, and I.J. Davenport. 2001. Image Frontiers in Earth Science 6: Article 233.
processing of airborne scanning laser altimetry data for Hawker, L., J. Rougier, J. Neal, P. Bates, L. Archer, and D.
improved river flood modelling. ISPRS Journal of Photogram- Yamazaki. 2018. Implications of simulating global digital
metry and Remote Sensing 56(2): 121–138. elevation models for flood inundation studies. Water Resources
Cook, A., and V. Merwade. 2009. Effect of topographic data, Research 54: 7910–7928.
geometric configuration and modeling approach on flood inun- Holmes, K.W., O.A. Chadwick, and P.C. Kyriakidis. 2000. Error in a
dation mapping. Journal of Hydrology 377(1): 131–142. USGS 30-meter digital elevation model and its impact on terrain
Coveney, S., and A.S. Fotheringham. 2011. The impact of DEM data modeling. Journal of Hydrology 233(1): 154–173.
source on prediction of flooding and erosion risk due to sea-level Hutchinson, M.F. 1989. A new procedure for gridding elevation and
stream line data with automatic removal of spurious pits. Journal
rise. International Journal of Geographical Information Science
25(7): 1191–1211. of Hydrology 106(3–4): 211–232.
Crippen, R., S. Buckley, P. Agram, E. Belz, E. Gurrola, S. Hensley, Jafarzadegan, K., and V. Merwade. 2017. A DEM-based approach for
M. Kobrick, M. Lavalle, et al. 2016. NASADEM global large-scale floodplain mapping in ungauged watersheds. Journal
of Hydrology 550: 650–662.
elevation model: Methods and progress. ISPRS – International
Archives of the Photogrammetry, Remote Sensing and Spatial Karamuz, E., R. Romanowicz, and J. Doroszkiewicz. 2020. The use
Information Sciences XLI–B4: 125–128. of unmanned aerial vehicles in flood hazard assessment. Journal
Dang, Y., L. Chen, W. Zhang, D. Zheng, and D. Zhan. 2020. How of Flood Risk Management 13(4): Article e12622.
does growing city size affect residents’ happiness in urban Kenward, T., D.P. Lettenmaier, E.F. Wood, and E. Fielding. 2000.
China? A case study of the Bohai rim area. Habitat International Effects of Digital Elevation Model accuracy on hydrologic
97: Article 102120. predictions. Remote Sensing of Environment 74(3): 432–444.
Deng, J.-L., S.-L. Shen, and Y.-S. Xu. 2016. Investigation into pluvial Kim, Y., Y. Tak, M. Park, and B. Kang. 2020. Improvement of urban
flooding hazards caused by heavy rain and protection measures flood damage estimation using a high-resolution digital terrain.
in Shanghai, China. Natural Hazards 83(2): 1301–1320. Journal of Flood Risk Management 13(S1): Article e12575.
Dixon, W.J. 1950. Analysis of extreme values. The Annals of Kulp, S., and B. Strauss. 2018. Coastal DEM: A global coastal digital
elevation model improved from SRTM using a neural network.
Mathematical Statistics 21(4): 488–506.
Dovern, J., M.F. Quaas, and W. Rickels. 2014. A comprehensive Remote Sensing of Environment 206(2): 231–239.
wealth index for cities in Germany. Ecological Indicators 41: Kulp, S., and B. Strauss. 2019. New elevation data triple estimates of
79–86. global vulnerability to sea-level rise and coastal flooding. Nature
Du, S., H. Gu, J. Wen, K. Chen, and A. Van Rompaey. 2015. Communications 10(1): Article 4844.
Detecting flood variations in Shanghai over 1949–2009 with Kyriakidis, P., M. Ashton, and F. Michael. 1999. Geostatistics for
Mann-Kendall tests and a newspaper-based database. Water conflation and accuracy assessment of digital elevation models.
7(5): 1808–1824. International Journal of Geographical Information Science
Du, X., H. Guo, X. Fan, J. Zhu, Z. Yan, and Q. Zhan. 2016. Vertical 13(7): 677–707.
accuracy assessment of freely available digital elevation models Lesser, G., J. Roelvink, J. Kester, and G. Stelling. 2004. Development
over low-lying coastal plains. International Journal of Digital and validation of a three-dimensional morphological model.
Coastal Engineering 51(8): 883–915.
Earth 9(3): 252–271.
Elkhrachy, I. 2018. Vertical accuracy assessment for SRTM and Lim, N., and S. Brandt. 2019. Flood map boundary sensitivity due to
ASTER Digital Elevation Models: A case study of Najran city, combined effects of DEM resolution and roughness in relation to
Saudi Arabia. Ain Shams Engineering Journal 9(4): 1807–1817. model performance. Geomatics, Natural Hazards and Risk
10(1): 1613–1647.
Fang, J., D. Lincke, S. Brown, R. Nicholls, C. Wolff, J. Merkens, J.
Hinkel, A. Vafeidis, et al. 2020. Coastal flood risks in China Lin, N., R.E. Kopp, B.P. Horton, and J.P. Donnelly. 2016. Hurricane
through the 21st century – An application of DIVA. Science of Sandy’s flood frequency increasing from year 1800 to 2100.
the Total Environment 704: Article 135311. Proceedings of the National Academy of Sciences 113(43):
Article 12071.

123
902 Xu et al. DEM Selection in Flood Simulation and a Method to Reduce DEM Errors

Mukherjee, S., P.K. Joshi, S. Mukherjee, A. Ghosh, R.D. Garg, and A. PRISM. ISPRS Annals of the Photogrammetry, Remote Sensing
Mukhopadhyay. 2013. Evaluation of vertical accuracy of open and Spatial Information Sciences II–4: 71–76.
source Digital Elevation Model (DEM). International Journal of Takagi, M. 1998. Accuracy of digital elevation model according to
Applied Earth Observation and Geoinformation 21(1): 205–217. spatial resolution. International Archives of Photogrammetry
Najibi, N., and N. Devineni. 2018. Recent trends in the frequency and and Remote Sensing 32(4): 613–617.
duration of global floods. Earth System Dynamics 9(2): 757–783. Takaku, J., T. Tadono, K. Tsutsui, and M. Ichikawa. 2016. Validation
Niu, G., and G. Zhao. 2018. Living condition among China’s rural– of ‘‘AW3D’’ global DSM generated from ALOS PRISM. ISPRS
urban migrants: Recent dynamics and the inland–coastal differ- Annals of the Photogrammetry, Remote Sensing and Spatial
ential. Housing Studies 33(3): 476–493. Information Sciences III–4: 25–31.
Pedrozo-Acuña, A., J. Rodrı́guez-Rincón, M. Arganis-Juárez, R. Talchabhadel, R., H. Nakagawa, K. Kawaike, K. Yamanoi, and B.R.
Domı́nguez-Mora, and F. González Villareal. 2015. Estimation Thapa. 2021. Assessment of vertical accuracy of open source
of probabilistic flood inundation maps for an extreme event: 30m resolution space-borne digital elevation models. Geomatics,
Pánuco River, México. Journal of Flood Risk Management 8(2): Natural Hazards and Risk 12(1): 939–960.
177–192. Vaze, J., J. Teng, and G. Spencer. 2010. Impact of DEM accuracy and
Reuter, H.I., A. Nelson, and A. Jarvis. 2007. An evaluation of void- resolution on topographic indices. Environmental Modelling &
filling interpolation methods for SRTM data. International Software 25(10): 1086–1098.
Journal of Geographical Information Science 21(9): 983–1008. Wang, J., W. Gao, S. Xu, and L. Yu. 2012. Evaluation of the
Rodriguez, E., C.S. Morris, and J.E. Belz. 2006. A global assessment combined risk of sea level rise, land subsidence, and storm
of SRTM performance. Photogrammetric Engineering & Remote surges on the coastal areas of Shanghai, China. Climatic Change
Sensing 72(3): 249–260. 115(3): 537–558.
Saksena, S., and V. Merwade. 2015. Incorporating the effect of DEM Wechsler, S.P. 2007. Uncertainties associated with digital elevation
resolution and accuracy for improved flood inundation mapping. models for hydrologic applications: A review. Hydrology and
Journal of Hydrology 530: 180–194. Earth System Sciences 11(4): 1481–1500.
Samela, C., T.J. Troy, and S. Manfreda. 2017. Geomorphic classifiers Wechsler, S.P., and C.N. Kroll. 2006. Quantifying DEM uncertainty
for flood-prone areas delineation for data-scarce environments. and its effect on topographic parameters. Photogrammetric
Advances in Water Resources 102: 13–28. Engineering & Remote Sensing 72(9): 1081–1090.
Sampson, C., A. Smith, P. Bates, J. Neal, and M. Trigg. 2016. Wen, K., and Y. Xu. 2006. China meteorological disaster dictionary
Perspectives on open access high resolution Digital Elevation Shanghai volume. Beijing: Meteorological Press. (in Chinese).
Models to produce global flood hazard layers. Frontiers in Earth Wong, W., S. Tsuyuki, K. Ioki, and M. Phua. 2014. Accuracy
Science 3: Article 85. assessment of global topographic data (SRTM & ASTER
SAMR (State Administration for Market Regulation). 2008. Statis- GDEM) in comparison with lidar for tropical montane forest.
tical interpretation of data – Detection and treatment of outliers In Proceedings of the 35th Asian Conference on Remote Sensing
in the normal sample. Beijing: SAMR. (in Chinese). 2014, 27–31 October 2014, Nay Pyi Taw, Myanmar, 722–727.
Sanders, B.F. 2007. Evaluation of on-line DEMs for flood inundation Yamazaki, D., D. Ikeshima, R. Tawatari, T. Yamaguchi, F.
modeling. Advances in Water Resources 30(8): 1831–1843. O’Loughlin, J.C. Neal, C. Sampson, S. Kanae, and P.D. Bates.
Savage, J., F. Pianosi, P. Bates, J. Freer, and T. Wagener. 2016. 2017. A high-accuracy map of global terrain elevations.
Quantifying the importance of spatial resolution and other Geophysical Research Letters 44(11): 5844–5853.
factors through global sensitivity analysis of a flood inundation Yang, Y., J. Yin, M. Ye, D. She, and J. Yu. 2020. Multi-coverage
model. Water Resources Research 52(11): 9146–9163. optimal location model for emergency medical service (EMS)
Schumann, G., and P. Bates. 2018. The need for a high-accuracy, facilities under various disaster scenarios: A case study of urban
open-access global DEM. Frontiers in Earth Science 6: Article fluvial floods in the Minhang district of Shanghai, China. Natural
225. Hazards and Earth System Sciences 20(1): 181–195.
SMSB (Shanghai Municipal Statistics Bureau). 2020. Shanghai Yin, J., D. Yu, and B. Liao. 2020. A city-scale assessment of
statistical yearbook 2020. Beijing: China Statistics Press. (in emergency response accessibility to vulnerable populations and
Chinese). facilities under normal and pluvial flood conditions for Shanghai,
Soille, P., J. Vogt, and R. Colombo. 2003. Carving and adaptive China. Environment and Planning B: Urban Analytics and City
drainage enforcement of grid digital elevation models. Water Science. https://doi.org/10.1177/2399808320971304.
Resources Research 39(12): Article 1366. Yin, J., D. Yu, Z. Yin, J. Wang, and S. Xu. 2013. Multiple scenario
Stephens, E., G. Schumann, and P. Bates. 2014. Problems with binary analyses of Huangpu River flooding using a 1D/2D coupled
pattern measures for flood model evaluation. Hydrological flood inundation model. Natural Hazards 66(2): 577–589.
Processes 28(18): 4928–4937. Yu, D., and S.N. Lane. 2006. Urban fluvial flood modelling using a
Surjan, A., G.A. Parvin, Atta-ur-Rahman, and R. Shaw. 2016. two-dimensional diffusion-wave treatment, part 1: Mesh resolu-
Expanding coastal cities: An increasing risk. In Urban disasters tion effects. Hydrological Processes 20(7): 1541–1565.
and resilience in Asia, ed. R. Shaw, Atta-ur-Rahman, A. Surjan, Yuan, Z. 1999. Floods and drought in Shanghai. Nanjing: Hohai
and G.A. Parvin, 79–90. Amsterdam: Elsevier. University Press. (in Chinese).
Tachikawa, T., M. Hato, M. Kaku, and A. Iwasaki. 2011. Charac- Zhang, K., D. Gann, M. Ross, Q. Robertson, J. Sarmiento, S. Santana,
teristics of ASTER GDEM version 2. In Proceedings of the 2011 J. Rhome, and C. Fritz. 2019. Accuracy assessment of ASTER,
IEEE International Geoscience and Remote Sensing Symposium, SRTM, ALOS, and TDX DEMs for Hispaniola and implications
24–29 July, Vancouver, BC, Canada, 3657–3660. for mapping vulnerability to coastal flooding. Remote Sensing of
Tachikawa, T., M. Kaku, A. Iwasaki, D. Gesch, M. Oimoen, Z. Environment 225: 290–306.
Zhang, J. Danielson, and T. Krieger et al. 2011. ASTER global Zhao, G., P. Bates, and J. Neal. 2020. The impact of dams on design
Digital Elevation Model version 2 – Summary of validation floods in the conterminous US. Water Resources Research 56(3):
results. Washington, DC: NASA. e2019WR025380.
Tadono, T., H. Ishida, F. Oda, S. Naito, K. Minakawa, and H.
Iwamoto. 2014. Precise global DEM generation by ALOS

123