Acta geographica Slovenica, 62-3, 2022, 47–62
FLOOD MAPPING BASED
ON OPEN-SOuRCE REMOtE SENSING
DAtA uSING AN EFFICIENt BAND
COMBINAtION SYStEM
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto,
Wei-Koon Lee, Mokhammad Farid Maruf
Flooding in the Tempurejo sub-district, Jember Regency, East Java Province, Indonesia.
A
n
G
G
E
R
 P
U
tR
A
n
tO
62-3_acta49-1.qxd  6.3.2023  13:15  Page 47
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
DOI: https://doi.org/10.3986/AGS.10598
UDC: 528:566.166(594)
COBISS: 1.01
Entin Hidayah1, Tedy Pranadiarso1, Gusfan Halik1, Indarto Indarto2, Wei-Koon Lee3, Mokhammad
Farid Maruf1
Flood mapping based on open-source remote sensing data using an efficient band
combination system
ABSTRACT: Flood mapping is an essential component of planning flood mitigation. The availability of
remote sensing data makes rapid flood mapping possible. This article develops an accurate method for
rapid flood mapping using satellite imagery. Sentinel-2 imagery was tested by acquiring data before and
after a flood event in a lowland area. Flooding extraction was performed using the newly developed Flood
Inundation Extraction Index (FIEI) and compared to the Modified Normalized Difference Water Index
(MNDWI), the most commonly used index. Based on the choice of threshold, the results are divided into
flooded and non-flooded areas. Evaluation of the performance accuracy based on the total and kappa coef-
ficients showed that the FIEI approach is more accurate than the MNDWI approach.
KEY WORDS: rapid flood mapping, Sentinel-2, Modified Normalized Difference Water Index, Flood
Inundation Extraction Index, kappa coefficient, accuracy, Indonesia
Kartiranje poplav na podlagi odprtokodnih podatkov daljinskega zaznavanja
z učinkovitim sistemom kombiniranih pasov
POVZETEK: Kartiranje poplav je ključno za načrtovanje blažitev njihovih posledic. Razpoložljivi podat-
ki, zajeti z daljinskim zaznavanjem, nam to omogočajo. V članku razvijamo natančno metodo za hitro
kartiranje poplav na podlagi satelitskih posnetkov. Posnetke Sentinel-2 smo testirali s podatki pred in po
poplavi v nižinskem območju. Poplave smo zaznavali s pomočjo na novo razvitega ekstrakcijskega indek-
sa najvišjih poplav (FIEI) in ga primerjali s prilagojenim normaliziranim vodnim indeksom (MNDWI),
ki je najpogosteje uporabljen indeks v tovrstnih raziskavah. Območja smo glede na izbor določenega praga
razdelili na poplavljena in nepoplavljena. Vrednotenje natančnosti rezultatov na podlagi skupnih in kapa
koeficientov je pokazalo, da je pristop FIEI natančnejši od pristopa MNDWI.
KLJUČNE BESEDE: hitro kartiranje poplav, Sentinel-2, prilagojen normalizirani vodni indeks, ekstrakcijski
indeks najvišjih poplav, kapa koeficient, pravilnost, Indonezija
The article was submitted for publication on 25th February, 2022.
Uredništvo je prejelo prispevek 25. februarja 2022.
48
1 University of Jember, Faculty of Engineering, Department of Civil Engineering, Jember, Indonesia
entin.teknik@unej.ac.id (https://orcid.org/0000-0002-1233-6850), tedyteknik@gmail.com
(https://orcid.org/0000-0003-2740-848X), gusfan.teknik@unej.ac.id (https://orcid.org/0000-0002-
5447-1268), farid.teknik@unej.ac.id (https://orcid.org/0000-0002-5715-935X)
2 University of Jember, Faculty of Agricultural Technology, Department of Agricultural Engineering,
Jember, Indonesia
indarto.ftp@unej.ac.id (https://orcid.org/0000-0001-6319-6731)
3 University Teknologi Mara, School of Civil Engineering, Shah Alam, Malaysia
leewei994@uitm.edu.my (https://orcid.org/0000-0003-3910-4870)
62-3_acta49-1.qxd  6.3.2023  13:15  Page 48
1 Introduction
A flood is a natural disaster in which a place is inundated for a specific duration, leading to damage to
property and loss of life (Huang and Jin 2020; Mahmood et al. 2021). Floods occur due to heavy rainfall,
which increases basin runoff, causing the water levels in municipal drainage, water bodies, and rivers to
overflow beyond their capacity (Gašparovič and Klobučar 2021). Accurate mapping of flooded areas is
a critical step in flood mitigation. However, the conventional approach to predicting flooded areas takes
a long time and cannot provide timely information during disasters (Ferk et al. 2021). Moreover, tradi-
tional land observation methods are expensive and time-consuming (Rahman et al. 2019).
Flood mapping can provide near real-time data and readily accessible information to predict flooded
areas (Sivanpillai et al. 2021). Satellite remote-sensing technology may provide fast information on flood
maps (Chen et al. 2019). Furthermore, the integration of remote sensing data and geographic information
systems (GIS) has been developed to map flooded areas (Wang et al. 2002). The integrated method of water
body mapping is based on indices and mapping flooded areas (Sarp and Ozcelik 2017; Zhou et al. 2017).
The approach has created opportunities for quantitative analysis of disaster events across all geographic and
spatial scales, including flood mapping. Remote-sensing technology has been developed to provide fast infor-
mation needed to produce near real-time predictions of flood maps (Sajjad et al. 2022). The technology
makes possible quick action by emergency response agencies during a flood event (Sivanpillai et al. 2021).
Previously, Feyisa et al. (2014) used Landsat TM5 images to compare the Automated Water Extraction
Index (AWEI), maximum likelihood (ML), Normalized Difference Water Index (NDWI), and Modified NDWI
(MNDWI) methods. These methods were applied to five research locations with varying threshold values
(TVs) for each site. The AWEI method showed greater accuracy than NDWI, MNDWI, and ML. Another
study attempted to combine synthetic aperture radar (SAR) Sentinel-1 and Sentinel-2 imagery data using
the NDWI method to quickly and cheaply predict flooding (Huang and Jin 2020). Fisher et al. (2016) used
several indices from Landsat TM, ETM+, and OLI images for mapping flooded areas using the water index
(WI). The results showed that the WI2015, WI2006, and AWEIshadow methods had the best accuracy. The kappa
and overall coefficients reached 95% to 99%. Sivanpillai et al. (2021) recommended the MNDWI method
for predicting flooded areas and also to validate the maps by exploring spectral bands 3 and 12 of Sentinel-
2. In addition, the MNDWI method was successfully applied to water areas with a background dominated
by built-up land (Sivanpillai et al. 2021).
Flood mapping for complex land cover requires sensitive spectral combinations to detect floods in low-
land areas. Niwas et al. (2015) and Ettehadi Osgouei et al. (2019) explained that near-infrared (NIR) waves
are suitable for detecting water, and green waves are best for detecting vegetation areas. Mid-infrared (MIR)
waves are suitable for detecting open land or built-up areas. Remote sensing technology for rapid flood
mapping is needed to determine evacuation priorities and mitigation in flood events. Flooded areas can
be extracted on the basis of a certain TV of a selected index. However, each image extracted in several loca-
tions has a TV that varies according to the location and time of image acquisition, affecting the resulting
accuracy (Feyisa et al. 2014). This indicates a development gap for further research to ensure that the cor-
rect TV is adopted for the specific geographical region, image type, and date. Therefore, a new approach
is needed to flood mapping with complex land cover. Previous studies using Landsat imagery with the NDWI
and MNDWI approach in the Citarum watershed, the Bengawan Solo watershed, and the lowland area
of Purworejo, Indonesia, have not yielded satisfactory results (Suwarsono et al. 2013).
Based on the spectral bands used in NDWI and MDWI, this study aims to improve the accuracy by
modifying spectral combinations and testing flood thresholds using Sentinel-2 image data. The Flood
Inundation Extraction Index (FIEI) approach has been developed to distinguish between flooded and non-
flooded areas using a suitable TV for application in multiple locations. The rapid flood mapping developed
using remote sensing integrated with GIS aims to determine priorities for handling flood-affected areas.
2 The study area
The study area covers the Tempurejo sub-district (524.46 km²) in the southern part of Jember Regency,
East Java Province, Indonesia. According to the National Disaster Management Agency, flood events in
the sub-district often occur at the beginning of the year and repeat annually (https://gis.bnpb.go.id/). The
Acta geographica Slovenica, 62-3, 2022
49
62-3_acta49-1.qxd  6.3.2023  13:15  Page 49
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
five flood-prone areas included in the study are the villages of Curahnongko, Curahtakir, Sidodadi, Sanenrejo,
and Wonoasri (Figure 1).
On Friday, February 6th, 2021, the five villages in the sub-district experienced one of the most signifi-
cant floods of the year. This was caused by high-intensity rainfall for three consecutive days starting on 4th
February, 2021, with the Tempurejo rain gauge recording 58 mm, 53 mm, and 90 mm. During the event,
the discharge of the Sanenrejo and Curahnongko rivers increased dramatically and overflowed into the
residential settlements. The flood depth reached two meters and lasted about four to six days. Around 12,000
houses were flooded, seven worship facilities were damaged, and five educational facilities were severely
affected.
3 Methods
The image treatment consists of pre-processing, processing, and post-processing. The pre-processing focus-
es on data collected from the national flood inventory, European Space Agency Sentinel-2 imagery, and
digital elevation model (DEM). Sentinel-2 image data were selected before and after the flood event to describe
the flooded area. This study uses a National Digital Elevation Model (DEMNAS) at 8.1 m pixel resolution.
The DEM is used for watershed delineation. The processing includes overlay, cropping, indices calculation,
flooded area extraction, and TV treatment. If an overall accuracy is less than 75% the accuracy has not
reached the best model. To reach the best model, it is necessary to re-treat the TV by changing the TV
treatment. Meanwhile, post-processing involves accuracy assessment and model analysis to derive the flood
map based on the best model, as summarized in the flowchart in Figure 2.
The new method was applied and tested in five villages, and the success rate was assessed using the
overall accuracy, kappa coefficients, user accuracy, and producer accuracy. The performance was bench-
marked and compared against the MNDWI method, which is reported to have high accuracy in rapid flood
mapping (see Section 1).
3.1 Data collection
Table 1 lists the data types used, including the source and function. The flood inventory used for model
validation consists of an in-situ field survey and flood data from the Indonesian National Board for Disaster
Management (Indonesian: Badan Nasional Penanggulangan Bencana, BNPB) website and the social media
(Official Instagram @bpbd_kab.jember) around the Jember Regency Disaster Management Board and the
newspaper Akurat.co (Nurullatifah 2021).
Flood extraction was performed using the Sentinel-2 imagery taken before and after February 5th, 2021.
The corresponding satellite images used are January 21st, 2020, and February 9th, 2021, corresponding to
before and after the flood event. The flood event inventory data were taken in the field, which excludes
permanent water areas. The Sentinel-2 image data used are data before and after the flood incident to com-
pare the flood before and after the incident and determine the flooded area. Pre-flood data are used to
determine the location of permanent water bodies. Post-flood data describe the entire site of the flood. The
administrative boundary and DEMNAS data were obtained from the Indonesian Geospatial Information
Agency (Indonesian: Badan Informasi Geospasial, BIG). In addition, the DEM data use the vertical datum
EGM2008 with a root mean square error (RMSE) value of 2.79 m and a standard error value of −0.13 m.
3.2 Spectral bands
Table 2 shows the four spectral bands of Sentinel-2 datasets, including the wavelength, spatial resolution,
and principal applications (SUHET 2015; Niwas et al. 2015; Bousbih et al. 2019).
Past research using the NDWI (Sentinel-2 bands 3 and 8) and the MNDWI (bands 3 and 12) reported
that, for pixel values ranging from −1 to 1, values from 0 to 1 best represent open water (i.e., flooding;
50
Figure 1: The study area. p p. 51
62-3_acta49-1.qxd  6.3.2023  13:15  Page 50
Acta geographica Slovenica, 62-3, 2022
51
Cu
ra
hn
on
gk
o
Sa
ne
nr
ej
o
Cu
ra
ht
ak
ir
Si
do
da
di
W
on
oa
sr
i
Sa
ne
nr
ejo
Su
co
G
en
to
ng
Ba
jin
g
Gad
un
g
M
and
ilis
Cur
ah 
Be
ng
kon
g
D
am
pa
r
Cu
ra
h 
N
on
gk
o
Pe
rb
al
an
Sumb
ert
en
ga
h
Gu
nu
ng
m
ay
an
g
Curah
 Mend
ilis
JE
M
BE
R
Le
ge
nd
W
at
er
sh
ed
 b
ou
nd
ar
y
St
ud
y 
A
re
a 
(5
 v
ill
ag
es
)
Re
ge
nc
y 
bo
un
da
ry
Ri
ve
r n
et
w
or
k
0
5
10
15
km
Sc
al
e:
 1
:2
00
,0
00
C
on
te
nt
 b
y:
 E
nt
in
 H
id
ay
ah
M
ap
 b
y:
 T
ed
y 
Pr
an
ad
ia
rs
o
So
ur
ce
: I
nd
on
es
ia
n 
G
eo
sp
at
ia
l A
ge
nc
y, 
20
21
©
 2
02
1,
 U
ni
ve
rs
ity
 o
f J
em
be
r, 
Fa
cu
lty
 o
f E
ng
in
ee
rin
g
EA
ST
 JA
VA
CENTR
AL JAV
A
BA
LI
BA
N
YU
W
A
N
G
I
Ja
v
a
 S
e
a
In
d
ia
n
 O
c
e
a
n
St
ud
y 
A
re
a
Ja
v
a
 S
e
a
In
d
ia
n
 O
c
e
a
n
"
ai
la
nd
Si
ng
ap
or
e
M
al
ay
sia
IN
D
O
N
E
S
IA
IN
D
O
N
E
S
IA
IN
D
O
N
E
S
IA
Au
st
ra
lia
S
o
u
th
 C
h
in
a
 S
e
a
62-3_acta49-1.qxd  6.3.2023  13:15  Page 51
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
52
Table 1: Data sources.
No. Data type Source Function
1 Flood inventory • Field survey Validation of model results
• Indonesian National Board for Disaster Management
(www.gis.bnpb.go.id)
• Jember Disaster Management Board
(Instagram @BPBD Jember Regency)
2 Citra Sentinel 2-L1C Sentinels Scientific Data Hub Flooding extraction
imagery (https://scihub.copernicus.eu/)
3 Administrative boundary Indonesian Geospatial Information Agency (BIG) Boundary of villages
(https://tanahair.indonesia.go.id,
https://geoservices.big.go.id)
4 Digital elevation model (DEM) Indonesian Geospatial Information Agency (BIG) Delineation of watershed (boundary
(resolution = 8.1 × 8.1 m) (https://tanahair.indonesia.go.id) of study) and river network system
Flood
inventory
Sentinel 2
imagery
Administrative
boundary
DEM
Watershed elineationd
--------------------------------------------------------------------------------------------------------------------------------------------------
Selecting spectral bands combination
and calculating FIEI and MNDWI
Overlay and roppingc
Treating hreshold aluet v
Accuracy easurementm
--------------------------------------------------------------------------------------------------------------------------------------------------
Best odelm
Export apsm
No
Yes
--------------------------------------------------------------------------------------------------------------------------------------------------
Pre–Proccesing
Post–Proccesing
Proccesing
Figure 2: Three-stage methodology flowchart.
62-3_acta49-1.qxd  6.3.2023  13:15  Page 52
Xu 2006). However, the soil and terrestrial vegetation properties have a negative value because NIR and
MIR reflectance are higher than green. Niwas et al. (2015) and Ettehadi Osgouei et al. (2019) explained
that NIR and green waves are suitable for detecting water and vegetation areas. In contrast, MIR waves are
suitable for detecting open land or built-up areas. Developing a flood mapping method is necessary by com-
bining the green, NIR, and MIR spectral algorithms available in Sentinel-2 satellite imagery. The algorithm
was applied to flood mapping with varied land cover such as vegetation, built-up land, and open water,
which requires additional sensitivity to detect water and obtain more accurate mapping.
3.3 The Flood Inundation Extraction Index
The FIEI approach with a combination of Sentinel-2 bands 3, 8, and 11 were expected to provide better
results in rapid flood mapping. The combinations were assumed to differentiate the best particular land
uses in the study area. The process of flooding extraction and mapping was performed using ArcMap 10.8.2
software.
The new formula can be expressed as:
(1)
It uses the spectral reflectance pattern of three land cover types: vegetation, water, and built-up areas.
The FIEI will minimize the ratio of the spectral components of bands 3 (green), 8 (NIR), and 11 (MIR).
The results of the FIEI approach were evaluated using the Modified Normalized Difference Water Index
(MNDWI; Xu 2006).
3.4 The Modified Normalized Difference Water Index
Development of the MNDWI for flood mapping was motivated by research showing that the reflec-
tion of NIR waves is suitable for detecting water (e.g., Niwas et al. 2015; Ettehadi Osgouei et al. 2019).
Meanwhile, the MNDWI uses green (band 3) and MIR (band 12) to enhance open water features. In MNDWI,
the MIR band is used instead of the NIR band. This approach can distinguish between open land and built-
up area features frequently correlated with open water.
The MNDWI can be calculated as:
(2)
This yields the following results:
1. At MNDWI, water has a positive value because water can absorb more MIR light than NIR;
2. The built-up area has a negative value; and
3. Soil and vegetation have negative values because soil reflects more MIR light than NIR light, and veg-
etation reflects more MIR light than green light (Xu 2006; Sathianarayanan 2018).
Several studies using the MNDWI approach have shown high accuracy for mapping flooded areas,
with an overall accuracy value of 98% to 99% and kappa coefficients of 0.8 to 0.9 (Fisher et al. 2016;
Zhou et al. 2017; Sivanpillai et al. 2021).
Acta geographica Slovenica, 62-3, 2022
53
FIEI= B3–B8+B11
B3+B8+B11
MNDWI= B3–B12
B3+B12
Table 2: Spectral bands of Sentinel-2 L1C imagery (Niwas et al. 2015; SUHET 2015; Bousbih et al. 2019).
No. Spectral band Wavelength (nm) Spatial resolution (m) Principal applications
1 B03–Green 560 10 Green reflectance by healthy vegetation 
2 B08–NIR 842 10 Biomass surveys, water body delineation
3 B11–MIR 1610 20 Vegetation moisture measurement and built-up
4 B12–MIR 2190 20 Hydrothermal mapping, built-up
62-3_acta49-1.qxd  6.3.2023  13:15  Page 53
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
3.6 Accuracy measurement
The results of the flood mapping from the Sentinel-2 Level 1C satellite imagery using the FIEI and MNDWI
were tested for accuracy. For this purpose, ground truth data acquired by field survey were used (328 sites).
The flood model performance was evaluated using the following criteria:
(3)
(4)
(5)
(6)
In equations (3) to (6), Xii is the diagonal value of the confusion (error) matrix, and Xij is used for flood-
ed and non-flooded area classifications. The matrix size r = 2, Xi+ is the number of pixels in row i obtained
from the remotely sensed analysis, X+j is the number of column j obtained from referenced flood data,
and N is the total number of pixels in the sample.
3.5 Threshold value treatment
According to Feyisa et al. (2014), the pixel TV for extraction of water bodies or flooding varies according
to the location and time of image acquisition. In setting the TV of the FIEI, it is necessary to examine the
pixel value in detail to ensure greater accuracy of the final result of flood mapping. The TV treatment used
in this study is described in Table 3.
54
Table 3: Threshold value treatment.
No. Threshold value Pixel value Description
1 TV1 < 0 Not flooded
≥ 0 Flooded
2 TV2 < 0.05 Not flooded
≥ 0.05 Flooded
3 TV3 < 0.1 Not flooded
≥ 0.1 Flooded
Table 4: Categorization based on a range of kappa coefficient (κ; Watson and Petrie 2010; Sivanpillai et al. 2021).
No. Coefficient range Classification
1 κ < 0.00 Very poor
2 0 ≤ κ ≤ 0.20 Poor
3 0.21 ≤ κ ≤ 0.40 Fair
4 0.41 ≤ κ ≤ 0.60 Moderate
5 0.61 ≤ κ ≤ 0.80 Good
6 κ ≥ 0.80 Very good
User accuracy = × 100%
Xij
Xi+
Producer accuracy = × 100%
Xij
X+j
Overall accuracy = × 100%∑ri=1Xii
1
N
κ = × 100%
N ∑ri=1 Xii – ∑
r
i=1 Xi+ + X+i
N2 – ∑ri=1 Xi + X+i
62-3_acta49-1.qxd  6.3.2023  13:15  Page 54
Acta geographica Slovenica, 62-3, 2022
55
The kappa coefficient (κ) measures how the classification results compare to values assigned by chance.
Furthermore, it is computed from the error matrix and incorporates diagonal and off-diagonal elements
(Sivanpillai et al. 2021). Based on the 2015 Indonesian National Institute of Aeronautics and Space guide-
lines on satellite data processing, the minimum level of accuracy required is 75%. Table 4 shows the range
of kappa coefficients and the respective classification.
4 Results
4.1 Flooding extraction
The extraction of flooding for five villages with two approaches (FIEI and MNDWI) and three TVs resulted
in thirty images. As an example, in Figure 3, the left column indicates that the MNDWI approach with TV1
treatment overestimated the flooded area. This can be observed especially in several residential or built-up areas.
4.2 Effect of threshold setting
Applying TV treatment is critical to define the flooded and non-flooded areas accurately. Therefore, the
application of several TVs to reduce pixel error in flooding extraction is necessary. Figure 4 shows the result-
ing flood mapping image based on FIEI, where the spatial extent of floods is reduced from TV1 to TV3
when the TV value is raised.
4.3 Accuracy measurement
The measurement results using the overall accuracy method and the kappa coefficients show that the FIEI
approach is more accurate than the MNDWI in classifying and mapping flooded and non-flooded areas,
as indicated in Table 5. The overall accuracy (OA) of the FIEI approach using three different TVs showed
results above 70%, whereby Sanenrejo achieved the highest based on TV1 at 98.04%. Subsequently, the MNDWI
approach showed overall accuracy results above 50%, with the TV1 treatment yielding the highest value of
79.69% for Curahnongko.
Table 5: Accuracy assessment for flood maps from the MNDWI and FIEI.
No. Study area Threshold value Overall accuracy (%) Kappa coefficients
FIEI MNDWI FIEI MNDWI
1 Wonoasri TV1 90.36 75.90 0.83 0.03
TV2 83.13 64.44 0.54 0.01
TV3 77.11 70.00 0.43 0.01
2 Curahnongko TV1 95.24 79.69 0.8 0.17
TV2 73.44 77.78 0.31 0.10
TV3 75.81 79.37 0.38 0.12
3 Curahtakir TV1 94.92 66.10 0.88 0.12
TV2 77.97 64.41 0.54 0.26
TV3 74.58 76.27 0.50 0.46
4 Sanenrejo TV1 98.04 76.47 0.96 0.43
TV2 90.20 54.90 0.79 0.05
TV3 80.39 52.83 0.61 0.13
5 Sidodadi TV1 95.83 79.17 0.86 0.10
TV2 83.33 62.50 0.57 0.01
TV3 75.00 58.33 0.44 0.06
Average of TV1 94.88 75.47 0.63 0.14
62-3_acta49-1.qxd  6.3.2023  13:15  Page 55
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
56
113 4 30 E° 0' ''
113 4 30 E° 0' ''
8
2
0
S
°
3
'3
' '
8
2
0
S
°
3
'3
' '
8
2
0
S
°
3
'
' '
8
2
0
S
°
3
'
' '
113 4 30 E° 0' ''
113 4 30 E° 0' ''
8
2
0
S
°
3
'3
' '
8
2
0
S
°
3
'3
' '
8
2
0
S
°
3
'
' '
8
2
0
S
°
3
'
' '
a)
113 4 0 E° 4' ''
113 4 0 E° 4' '' 8
° 2
4
'3
' '
0
S
8
° 2
0
S
4
'3
' '
8
2
0
S
°
4
'
' '
8
2
0
S
°
4
'
' '
113 4 0 E° 4' ''
113 4 0 E° 4' '' 8
° 2
4
'3
' '
0
S
8
° 2
0
S
4
'3
' '
8
2
0
S
°
4
'
' '
8
2
0
S
°
4
'
' '
b)
113 4 0 E° 5' ''
113 4 0 E° 5' ''
8
2
0
S
°
1
'3
' '
8
2
0
S
°
1
'3
' '
113 4 0 E° 5' ''
113 4 0 E° 5' ''
8
2
0
S
°
1
'3
' '
8
2
0
S
°
1
'3
' '
c)
62-3_acta49-1.qxd  6.3.2023  13:15  Page 56
Acta geographica Slovenica, 62-3, 2022
57
113 46 30 E° ' ''
113 46 30 E° ' '' 113 47 0° ' ''
113 47 0° ' ''
8
2
2
0
S
°
'
' '
8
2
2
0
S
°
'
' '
8
2
1
3
0
S
°
'
' '
113 46 30 E° ' ''
113 46 30 E° ' '' 113 47 0° ' ''
113 47 0° ' ''
8
2
2
0
S
°
'
' '
8
2
2
0
S
°
'
' '
8
2
1
3
0
S
°
'
' '
d)
113 40 30 E° ' ''
113 40 30 E° ' ''
8
2
2
3
0
S
°
'
' '
8
2
2
3
0
S
°
'
' '
8
2
2
0
S
°
'
' '
8
2
2
0
S
°
'
' '
113 40 30 E° ' ''
113 40 30 E° ' ''
8
2
2
3
0
S
°
'
' '
8
2
2
3
0
S
°
'
' '
8
2
2
0
S
°
'
' '
8
2
2
0
S
°
'
' '
e)
Legend
Possible inundation
Non inundation (field survey)
Inundation (field survey)
0 1 2 3 40.5
Scale: 1:30,000
Cotent by: Entin Hidayah
Map by: Tedy Pranadiarso
Source: Rupa Bumi Indonesia (1:25,000) Indonesian Geospatial Agency
© 2021, University of Jember Faculty of Engineering
Figure 3: Flooding extraction using MNDWI (left) and FIEI (right) with TV1 for (a) Wonoasri, (b) Curahnongko, (c) Curahtakir, (d) Sanenrejo, and (e)
Sidodadi. (p. 56–57)
62-3_acta49-1.qxd  6.3.2023  13:15  Page 57
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
The kappa coefficients show that the FIEI approach results are more accurate than the MNDWI results.
The coefficients of the FIEI approach achieve over 0.8 for the case of TV1 treatment. The highest kappa
coefficient of 0.96 is obtained using the FIEI approach in Sanenrejo. Meanwhile, the highest coefficients
when using the MNDWI approach are seen in Curahtakir with TV3 treatment. The value of 0.46 obtained
58
113 40 0 E° ' '' 113 40 30 E° ' ''
8
2
4
3
0
S
°
'
' '
8
2
2
3
0
S
°
'
' '
8
2
2
0
S
°
'
' '
8
2
2
0
S
°
'
' '
113 40 0 E° ' '' 113 40 30 E° ' ''
TV3
113 40 0 E° ' '' 113 40 30 E° ' ''
8
2
4
3
0
S
°
'
' '
8
2
2
3
0
S
°
'
' '
8
2
2
0
S
°
'
' '
8
2
2
0
S
°
'
' '
113 40 0 E° ' '' 113 40 30 E° ' ''
TV2
113 40 0 E° ' '' 113 40 30 E° ' ''
8
2
4
3
0
S
°
'
' '
8
2
2
3
0
S
°
'
' '
8
2
2
0
S
°
'
' '
8
2
2
0
S
°
'
' '
113 40 0 E° ' '' 113 40 30 E° ' ''
TV1
Possible flooding
River network
Legend 0 1 2 3 40.5
Scale: 1:30,000
Cotent by: Entin Hidayah
Map by: Tedy Pranadiarso
Source: Rupa Bumi Indonesia (1:25,000)
Indonesian Geospatial Agency
© 2021, University of Jember
Faculty of Engineering
Figure 4: Effect of TV treatment for FIEI.
Table 6: User and producer accuracy assessment for the MNDWI and FIEI approaches.
No. Study area Threshold value User accuracy (%) Producer accuracy (%)
FIEI MNDWI FIEI MNDWI
1 Wonoasri TV1 91.43 87.14 98.46 84.72
TV2 81.43 70.13 98.28 85.71
TV3 74.29 77.92 98.11 85.71
2 Curahnongko TV1 94.64 87.50 100.00 90.74
TV2 73.21 83.93 97.62 90.38
TV3 72.73 85.71 100.00 90.57
3 Curahtakir TV1 95.12 82.93 97.50 72.34
TV2 73.17 63.41 93.75 81.25
TV3 65.85 80.49 96.43 84.62
4 Sanenrejo TV1 100.00 93.94 97.06 75.61
TV2 87.88 60.61 96.67 66.67
TV3 72.73 39.39 96.00 72.22
5 Sidodadi TV1 95.00 91.67 100.00 84.62
TV2 80.00 68.33 100.00 83.67
TV3 70.00 60.00 75.00 85.71
62-3_acta49-1.qxd  6.3.2023  13:15  Page 58
is considered a moderate accuracy. The lower accuracy results obtained by the MNDWI approach are due
to the overestimation of flooded areas and misclassifications of flooded and non-flooded regions in resi-
dential areas or built-up land use.
The FIEI approach with TV1 treatment can more accurately map flooding in residential (or built-up
areas) and vegetation areas than the MNDWI approach. It can produce flooding extraction closer to the
actual flood occurrence without overestimation. This is validated by comparing the extracted flooded area
with data from a field survey. In some evacuation locations, non-flooded areas are accurately spotted, facil-
itating evacuation efforts.
For the accuracy of flood maps used by natural disaster emergency response agencies, more detailed
validations were needed to evaluate user accuracy (UA) and producer accuracy (PA).
Table 6 shows that the UA and PA values of the FIEI approach were higher than the MNDWI approach.
Using the FIEI approach, the UA and PA values were above 90%, reaching 100%. This shows that the clas-
sification between flooded and non-flooded areas has been mapped accurately and reliably.
5 Discussion
In East Java Province, from 2017 to 2021, the average flood incidence is 115 events per year. According
to the National Disaster Management Agency’s data for 2022 showed that 111 people died, two were miss-
ing, 10,384 were evacuated, and 1,599 buildings were damaged (https://gis.bnpb.go.id/). Flood mapping
is very expensive and takes a long time, hence the flood mapping technique with the Sentinel-2 image
approach is needed for mitigation efforts (Sivanpillai et al. 2021).
Previously, the NDWI and MNDWI algorithms were applied based on Landsat, Sentinel-1, and Sentinel-
2 images for rapid flood mapping with various features for water bodies and flooded areas. In Shouguang,
both algorithms based on Sentinel-1 and Sentinel-2 with random forest classification for rivers showed OA
values of 85.22% and 95.45%, respectively (Huang and Jin 2020). Furthermore, for rural watershed areas in
Australia, using two algorithms based on Landsat Oli 8 imagery with unsupervised classification proved to
be effective, with OA values of 96.04% and 95.70%, respectively (Ghofrani et al. 2019). Implementation on
rivers and lakes in Canada, the Tennessee River in the US, the Swedish lake Lungsjön, and Mongolia’s Lake
Khar-Us using Sentinel-2 imagery showed outstanding accuracy with consecutive OA values based on the
NDWI (0.97, 0.965, 0.94, and 0.925) and the MNDWI (0.97, 0.962, 0.96, and 0.945; Niu et al. 2022). The
application of the Landsat image–based MNDWI to features of urban land, rural land, vacant land, swamps,
and agricultural areas in the US had an overall accuracy of 93.92% (Sivanpillai et al. 2021). In addition,
the MNDWI algorithm application based on the Landsat image for the same feature also yielded excel-
lent results, as shown in Table 7 (Feyisa et al. 2014). The application of the MNDWI for coastal forests,
agriculture, grasslands, shrubs, and forests in eastern Australia using Landsat imagery also resulted in out-
standing accuracy, with an OA value of 97.95% (Fisher et al. 2016). Implementation of the NDWI and
Acta geographica Slovenica, 62-3, 2022
59
Table 7: Implementation of MNDWI in several countries.
Location Accuracy (%) TV Land use Reference Data sources
Denmark 98.05 (UA) 0.1 Reservoirs, a harbor, and the sea (Øresund and Køge Bay) Feyisa et al. 2014 Landsat 
Switzerland 99.84 (UA) −0.15 Lake Lauerz, Lake Ägeri, Lake Sihl, Lake Wägital, Lake Klöntal Feyisa et al. 2014 Landsat 
Ethiopia 96.60 (UA) 0.1 Gefersa, Dire, Legedadi Feyisa et al. 2014 Landsat 
South Africa 94.46 (UA) 0.6 Berg River, Wemmershoek, Brandvlei Feyisa et al. 2014 Landsat 
New Zealand 98.14 (UA) 0.2 Lake Te Anau Feyisa et al. 2014 Landsat 
Eastern Australia 97.95 (OA) 0.1 Coastal forest, agricultural, grasslands, shrublands, woodlands Fisher et al. 2016 Landsat 
US 93.92 (OA) 0.2 Urban, rural, bare ground, marshes, agricultural area Sivanpillai et al. 2021 Landsat 
China 85.22 (OA) 0.35 Several rivers Huang and Jin 2020 Sentinel-2 
Spain 68 (OA) −0.35 Wetlands Pena-Regueiro et al. Sentinel-2 
2020 
Note: UA = user accuracy, OA = overall accuracy.
62-3_acta49-1.qxd  6.3.2023  13:15  Page 59
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
MNDWI for rivers and lakes yielded excellent results. However, its implementation based on Sentinel-2
for coastal areas in Valencia in eastern Spain resulted a decrease in performance, as indicated by OA val-
ues of 89% and 68% (Pena-Regueiro et al. 2020). Similarly, in our research the application of the MNDWI
for lowland flooding is also unsatisfactory, with an average OA value of 75.47% (Table 6). It is important
to emphasize that the MNDWI has weaknesses in detecting the water layer underlying the built-up area
in lowlands and also in the coastal area.
Therefore, with outstanding results, we are using the FIEI algorithm to develop rapid flood mapping using
Sentinel-2 imagery and changing the band combination in lowland flooding. The performance of the FIEI
approach provides an average increase in the overall accuracy of 19.41% from the MNDWI in consistent flood
mapping for five villages in the Tempurejo sub-district, Jember Regency, Indonesia. The increased average
of overall accuracy (19.41%) is obtained by calculating the mean of the overall accuracy of FIEI’s TV1 minus
the mean of the overall accuracy of MNDWI’s TV1. As a highlight, the FIEI is an improvement from the
MNDWI for lowland flooding by utilizing NIR waves sensitive to water that can capture lowland flooding.
Meanwhile, the MNDWI application with the most optimal TV equaling 0 cannot provide better accuracy.
In addition to the combination of bands, the TV is also influential in achieving the best accuracy. Each
country had a different TV for achieving the best accuracy, as shown in Table 7. Some locations had the
same TV, equal to 0.1, such as Denmark, Ethiopia, and eastern Australia (Feyisa et al. 2014; Fisher et al. 2016).
In addition, New Zealand and the US have the same TV score, at 0.2. In other countries, the best accura-
cy was obtained from a TV less than 0 (Feyisa et al. 2014; Sivanpillai et al. 2021; Xu 2006; Pena-Regueiro
et al. 2020). Whereas this study shows that using an appropriate TV1 greater than or equal to 0 is flood-
ing, the flooding extraction eliminates falsely classified pixels. In our study case, selecting the proper TV
is expected to produce more optimal map accuracy.
The FIEI approach with treatment TV1 can effectively provide crucial and timely information on flood-
ed areas to map out priority evacuation areas affected by disasters. This saves time in allocating resources
to places that require evacuation to facilitate the distribution of logistical assistance. In line with Kaplan
and Avdan’s (2017) opinion, remote sensing techniques are often less costly and time-consuming for large
geographic areas than conventional field mapping. The FIEI algorithm using Sentinel-2 imagery can be
applied for rapid mapping of floods in other countries. This map is classified into flooded and non-flood-
ed areas suitable for lowland applications and beneficial in planning flood mitigation. This flood map is
certainly very useful in mapping flood risk (Ryu et al. 2017), planning flood mitigation (Sipelgas et al. 2021),
or managing small retention ponds (Ferk et al. 2020).
6 Conclusion
The FIEI approach addressed the MNDWI’s limitations and prevented overestimation of the flooded area.
Compared to the MNDWI approach, it analyzes bands 3, 8, and 11 of Sentinel-2 to extract flooded areas
more effectively. Furthermore, it provides higher accuracy with the appropriate TV treatment. The approach
is recommended for quickly mapping flooded areas (if ground truth data are available) in the Tempurejo
sub-district and similar lowland areas susceptible to flooding.
With free spatial data and ground truth data, the FIEI approach is suitable for deployment during dis-
asters to provide quick flood mapping for response, evacuation, and mitigation efforts. The technique can
be easily extended to other locations if the TV is successfully identified. For applying flood mapping out-
side Indonesia, such as in Denmark, Australia, China, New Zealand, and the US, a combination of bands
3 and 12 is recommended.
ACKNOWLEDGMENTS: This article is part of a research program supported by funding from the Indonesia
Endowment Fund for Education (LPDP) in 2021–2022.
7 References
Bousbih, S., Zribi, M., Pelletier, C., Gorrab, A., Lili-Chabaane, Z., Baghdadi, N., Ben Aissa, N., Mougenot,
B. 2019: Soil texture estimation using radar and optical data from Sentinel-1 and Sentinel-2. Remote
Sensing 11-13. DOI: https://doi.org/10.3390/rs11131520
60
62-3_acta49-1.qxd  6.3.2023  13:15  Page 60
Chen, Z., Luo, J., Chen, N., Xu, R., Shen, G. 2019: RFim: A real-time inundation extent model for large
floodplains based on remote sensing big data and water level observations. Remote Sensing 11-13. DOI:
https://doi.org/10.3390/rs11131585
Ettehadi Osgouei, P., Kaya, S., Sertel, E., Alganci, U. 2019: Separating built-up areas from bare land in
Mediterranean cities using Sentinel-2A imagery. Remote Sensing 11-3. DOI: https://doi.org/10.3390/
rs11030345
European space agency 2015: Sentinel-2 user handbook. European Space Agency Standard Document 1-2.
Ferk, M., Ciglič, R., Komac, B., Lóczy, D. 2021: Management of small retention ponds and their impact
on flood hazard prevention in the Slovenske Gorice Hills. Acta geographica Slovenica 61-1. DOI:
https://doi.org/https://doi.org/10.3986/AGS.7675
Feyisa, G. L., Meilby, H., Fensholt, R., Proud, S. R. 2014: Automated Water Extraction Index: A new tech-
nique for surface water mapping using Landsat imagery. Remote Sensing of Environment 140. DOI:
https://doi.org/10.1016/j.rse.2013.08.029
Fisher, A., Flood, N., Danaher, T. 2016: Comparing Landsat water index methods for automated water clas-
sification in eastern Australia. Remote Sensing of Environment 175. DOI: https://doi.org/10.1016/
j.rse.2015.12.055
Gašparovič, M., Klobučar, D. 2021: Mapping floods in low-land forest using Sentinel-1 and Sentinel-2 data
and an object-based approach. Forests 12-5. DOI: https://doi.org/10.3390/f12050553
Ghofrani, Z., Sposito, V., Faggian, R. 2019: Improving flood monitoring in rural areas using remote sensing.
Water Practice and Technology 14-1. DOI: https://doi.org/10.2166/wpt.2018.118
Huang, M., Jin, S. 2020: Rapid flood mapping and evaluation with a supervised classifier and change detec-
tion in Shouguang using Sentinel-1 SAR and Sentinel-2 optical data. Remote Sensing 12-13. DOI:
https://doi.org/10.3390/rs12132073
Kaplan, G., Avdan, U. 2017: Mapping and monitoring wetlands using Sentinel-2 satellite imagery. ISPRS
Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences IV-4/W4. DOI:
https://doi.org/10.5194/isprs-annals-IV-4-W4-271-2017
Mahmood, S., Sajjad, A., Rahman, A.-ur. 2021: Cause and damage analysis of 2010 flood disaster in dis-
trict Muzaffar Garh, Pakistan. Natural Hazards 107. DOI: https://doi.org/10.1007/s11069-021-04652-6
Niu, L., Kaufmann, H., Xu, G., Zhang, G., Ji, C., He, Y. 2022: Triangle Water Index (TWI): An advanced
approach for more accurate detection and delineation of water surfaces in Sentinel-2 data. Remote Sensing
14-21. DOI: https://doi.org/https://doi.org/10.3390/rs14215289
Niwas, R., Grewal, M. S., Khichar, M. L. 2015: Practical manual on Remote Sensing , GIS and Land Use
Planning. Hisar.
Nurullatifah, T. 2021: Ribuan Warga Terdampak Banjir di Jember. Internet: https://akurat.co/ribuan-warga-
terdampak-banjir-di-jember (25. 1. 2023).
Pena-Regueiro, J., Sebastiá-Frasquet, M.-T., Estornell, J., Aguilar-Maldonado, J. A. 2020: Sentinel-2 appli-
cation to the surface characterization of small water bodies in wetlands. Water 12-5. DOI: https://doi.org/
10.3390/w12051487
Rahman, M., Ningsheng, C., Islam, M. M., Dewan, A., Iqbal, J., Washakh, R. M. A., Shufeng, T. 2019: Flood
susceptibility assessment in Bangladesh Using machine learning and multi-criteria decision analysis.
Earth Systems and Environment 3. DOI: https://doi.org/10.1007/s41748-019-00123-y
Ryu, J., Yoon, E. J., Park, C., Lee, D. K., Jeon, S. W. 2017: A flood risk assessment model for companies and
criteria for governmental decision-making to minimize hazards. Sustainability 9-11. DOI: https://doi.org/
10.3390/su9112005
Sajjad, A., Lu, J., Chen, X., Chisenga, C., Mazhar, N., Nadeem, B. 2022: Riverine flood mapping and impact
assessment using remote sensing technique: A case study of Chenab flood-2014 in Multan district, Punjab,
Pakistan. Natural Hazards 110. DOI: https://doi.org/https://doi.org/10.1007/s11069-021-05033-9
Sarp, G., Ozcelik, M. 2017: Water body extraction and change detection using time series: A case study
of Lake Burdur, Turkey. Journal of Taibah University for Science 11-3. DOI: https://doi.org/10.1016/
j.jtusci.2016.04.005
Sathianarayanan, M. 2018: Assessment of surface water dynamics using multiple water indices around Adama
woreda, Ethiopia. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information
Sciences 4-5. DOI: https://doi.org/10.5194/isprs-annals-IV-5-181-2018
Acta geographica Slovenica, 62-3, 2022
61
62-3_acta49-1.qxd  6.3.2023  13:15  Page 61
Entin Hidayah, Tedy Pranadiarso, Gusfan Halik, Indarto Indarto, Wei-Koon Lee, Mokhammad Farid Maruf, Flood mapping …
Sipelgas, L., Aavaste, A., Uiboupin, R. 2021: Mapping flood extent and frequency from Sentinel-1 imagery
during the extremely warm winter of 2020 in boreal floodplains and forests. Remote Sensing 13-23.
DOI: https://doi.org/10.3390/rs13234949
Sivanpillai, R., Jacobs, K. M., Mattilio, C. M., Piskorski, E. V. 2021: Rapid flood inundation mapping by
differencing water indices from pre- and post-flood Landsat images. Frontiers of Earth Science 15. DOI:
https://doi.org/10.1007/s11707-020-0818-0
Suwarsono, Nugroho, J. T., Wiweka. 2013: Identification of inundated area using normalized difference
water index (NDWI) on lowland region of Java island. International Journal of Remote Sensing and
Earth Sciences 10-2. DOI: https://doi.org/10.30536/j.ijreses.2013.v10.a1850
Wang, Y., Colby, J. D., Mulcahy, K. A. 2002: An efficient method for mapping flood extent in a coastal
floodplain using Landsat TM and DEM data. International Journal of Remote Sensing 23-18. DOI:
https://doi.org/10.1080/01431160110114484
Watson, P. F., Petrie, A. 2010: Method agreement analysis: A review of correct methodology. Theriogenology
73-9. DOI: https://doi.org/10.1016/j.theriogenology.2010.01.003
Xu, H. 2006: Modification of normalized difference water index (NDWI) to enhance open water features
in remotely sensed imagery. International Journal of Remote Sensing 27-14. DOI: https://doi.org/10.1080/
01431160600589179
Zhou, Y., Dong, J., Xiao, X., Xiao, T., Yang, Z., Zhao, G., Zou, Z., Qin, Y. 2017: Open surface water mapping
algorithms: A comparison of water-related spectral indices and sensors. Water 9-4. DOI: https://doi.org/
10.3390/w9040256Bousbih, S., Zribi, M., Pelletier, C., Gorrab, A., Lili-Chabaane, Z., Baghdadi, N., Aissa,
N. Ben., Mougenot, B. 2019: Soil texture estimation using radar and optical data from Sentinel-1 and
Sentinel-2. Remote Sensing. DOI: https://doi.org/10.3390/rs11131520
62
62-3_acta49-1.qxd  6.3.2023  13:15  Page 62