Natural Hazards

https://doi.org/10.1007/s11069-021-05033-9

ORIGINAL PAPER

Riverine flood mapping and impact assessment using remote

sensing technique: a case study of Chenab flood‑2014
in Multan district, Punjab, Pakistan

Asif Sajjad1   · Jianzhong Lu1 · Xiaoling Chen1 · Chikondi Chisenga2 ·

Nausheen Mazhar3 · Basit Nadeem4

Received: 22 January 2021 / Accepted: 13 September 2021

© The Author(s), under exclusive licence to Springer Nature B.V. 2021

Abstract
The Multan district is mainly prone to riverine floods but has remained understudied.
Chenab flood-2014 was the worst flood that this district experienced in recorded history.
This study applies remote sensing (RS) techniques to estimate the extent, calculate dura-
tion, assess the major causes and resulting impacts of the flood-2014, using Landsat-8 OLI
images. These images were obtained for pre-flood, during-flood and post-flood instances.
Secondary data of flood causing factors were obtained for comprehensive analysis. Spa-
tially trained and validated datasets were obtained through Google Earth platform and
Global positioning system. The supervised classification with maximum likelihood algo-
rithm was used to classify land use and land cover of the study area. The Modified Normal-
ized Difference Water Index was utilized to detect flood inundation extent and duration,
and Normalized Difference Vegetation Index was utilized to monitor vegetation coverage
and changes. The analysis allowed us to assess flood causes, and calculate the extent of
the flooded areas with duration and recession, as well as damages to standing crops and
built-up areas. The results revealed that the flood-2014 occurred due to heavy rains in early
September in upper Chenab catchment. The flood inundation continued for around two
months, which heavily affected agriculture and built-up areas. The present study introduces
practical use of RS techniques to provide basis for effective flood inundation mapping and
impact assessment, as an application for early flood response and recovery in the world.

Keywords  River Chenab · Remote sensing · Landsat-8 · MNDWI · Riverine flood

mapping · Impact assessment

1 Introduction

Flood disasters are increasing immensely in frequency and magnitude with destructive con-
sequences throughout the world (Syvitski and Brakenridge 2013; Bhatt et al. 2016). These
disasters cannot be completely prevented, but the post-disaster impacts can be reduced by

* Jianzhong Lu
[email protected]
Extended author information available on the last page of the article

13
Vol.:(0123456789)
 Natural Hazards

applying effective disaster risk reduction measures with the help of remote sensing (RS)
techniques. Flood is considered as one of the most destructive hydro-meteorological dis-
aster, (Mahmood et al., 2019a; b; Sajjad et al. 2020a) and it accounts for almost 50% of
all hydro-meteorological disasters that occur across the world (Cann et  al. 2012; Hal-
gamuge and Nirmalathas 2017). Flood disasters are normally caused by high rainfall (Rah-
man and Khan 2012; FFCI 2014; Mahmood et al. 2016a; Khalid et al. 2018), breaching of
embankments and bunds (Gaurav et al. 2011; Mahmood et al. 2019a, 2021), siltation and
deposition in river bed and lack of structural strategies (Cook and Merwade 2009; Sajjad
et al. 2019). In the recent decade, floods have caused over 250,000 human causalities and
brought enormous economic losses of about 4.8 billion US$ across the world (Mahmood
et al. 2016b; Schumann et al. 2018). However, the most flood related life losses and eco-
nomic damages are reported in the developing countries of the Asian subcontinent, includ-
ing Bangladesh, India, and Pakistan (Islam et al. 2010; Rahman and Khan 2011; Hashmi
et al. 2012; Khan et al. 2014; Halgamuge and Nirmalathas 2017).
In Pakistan, Flood is recognized as the most recurrent disaster (Hashmi et  al. 2012;
Tariq and van de Giesen 2012; Mahmood et  al. 2019a). Mostly, flood disasters occur in
the monsoon months of August and September, due to excessive rainfall in upstream of
Himalaya and Hindukush regions. This results in riverine flooding in floodplain areas,
which ultimately causes a massive damage to human lives, properties and country’s econ-
omy (FFCI 2014; PMD 2014; NDMA 2014; Khan et  al. 2014; Mahmood et  al. 2016b,
2019b; Rahman et al. 2017; Sajjad et al. 2020a). For instance, the riverine flood-2014 was
considered the worst flood in term of human and economic losses in the Chenab plains
(FFCI 2014; PPDMA 2014; Rahman et  al. 2017; Sajjad et  al. 2020b). This shows that
frequent flooding events are a serious problem along Chenab plains of Punjab province,
Pakistan, which enhances the need of significant effort to minimize the impacts of future
flood events. Therefore, a detailed study on rapid flood-2014 mapping and damage assess-
ment was carried out using Landsat-8 data with an emphasis on district Multan. District
Multan is exposed to riverine flooding in every monsoon season (PPDMA 2014). In addi-
tion, exponential population growth, land use development and human interventions have
further intensified the potential of the study area toward the flood disaster. There has been
no scientific study related to rapid flood-2014 mapping and impact assessment in district
Multan, and the previous studies related to flood-2014 were mostly carried out in the upper
Chenab regions and not entirely focused on riverine flood mapping and impact assessment
(Chohan et al. 2015; Rahman et al. 2017; Mahmood and Rani 2018). Therefore, this study
is aimed to perform rapid flood-2014 mapping and provide impact assessment for district
Multan.
RS data provide detailed information, which can be used for comprehensive inves-
tigations of flood causes and the delineation of flood inundated areas (Giustarini et al.
2015; Giordan et  al. 2018). For example, Landsat satellite, which provides free data
since 1972 (Wulder et al. 2012), has been utilized in many studies for flood monitoring
and damage assessment (Sanyal and Lu 2004). However, the main limitation of Landsat
data is that they cannot penetrate through clouds (Chignell et al. 2015). In such cases,
Synthetic aperture radar (SAR) satellites can provide data in all weather conditions that
enable early flood extent detection with high accuracy (Pradhan et al. 2010; Notti et al.
2018; Domeneghetti et  al. 2019). But due to cost restraints, Radar data are not used
for this study. However, cloud-free optical Landsat satellite (~ 30  m) data are used for
delineation of flood inundated areas and LULC change detection. The study area lies in
two overlapping Landsat satellite paths of 150 and 151, which enhanced the temporal
resolution of used images (~ 8 days), as compared to a single path temporal resolution

13
Natural Hazards

of 16 days. The use of high temporal Landsat data enabled us to investigate flood inun-
dation extent and duration. Hence, the present study uses RS techniques to delineate
flooded areas extent coupled with impact assessment on built-up and agricultural areas.

1.1 Study area

District Multan was selected for this study. Its geographical area extends from 29.24 to
30.48° N and 71.00 to 71.83° E (Fig. 1). It is bounded by four districts Muzaffargarh,
Lodhran, Khanewal and Bahawalpur. The total district area under study is 3721 k­ m2.
The population of Multan district is 4.74 million (GoP 2017). The mean annual rainfall
is 157 mm, while the highest rainfall recorded is 50 mm in the month of July and the
lowest documented rainfall of 2 mm in the month of October (PMD 2014). Agriculture
is the main economic activity of the local people. Main crops cultivated in Multan are
cotton, sugarcane, wheat, rice and fodder.

Fig. 1  Location of district Multan, Barrages, Chenab River

13
 Natural Hazards

2 Materials and methods

2.1 Rainfall and discharge data

Secondary datasets were provided by Government departments, i.e., rainfall data of 15

meteorological stations of upper Chenab were acquired from Pakistan Meteorological
Department, (PMD), Islamabad. River Chenab discharge data of flood-2014 were collected
from Flood Forecasting Division (FFD), Lahore.

2.2 Remote sensing data

Landsat-8 data were obtained from open access United States Geological Survey (USGS)
database. Furthermore, eight temporal images of following dates: 17th September, 24th
September, 03rd October, 10th October, 19th October, 26th October, 11th November and
27th November, 2014 were used for flood monitoring in the study area. Likewise, three
temporal images of pre-flood (23rd August), during-flood (17th September) and post-
flood images (27th November) were used for flood damage assessment using LULC
classification.

2.3 Spatial training and validation datasets

Google Earth (GE) and Global position system (GPS) platform were used to collect spatial
training and validation datasets for LULC classification and flood mapping. The GE data
were obtained at about the same time from the acquisition date, as the Landsat-8 images
to support LULC classification. Bands of acquired GE images were combined to better
visualize and locate land use classes, i.e., water class, vegetation and agriculture class,
sand class, built-up area class, deposited material class and barren land class. These classes
were further digitized into vector polygon form and saved as shapefile format in ArcGIS.
For each polygon, a point was taken around the center and recorded in Microsoft (MS)
Excel. These points served as the ground truth validation and training dataset for LULC
classes. The validation GPS points were also collected in post-flood occasion, when the
flooded areas had entirely retreated. These points were used to validate MNDWI derived
riverine flood mapping.

3 Methods

The schematic illustration of the used methodology is presented in Fig. 2. Firstly, radiomet-
ric calibration, atmospheric correction and layer mosaicking were applied on the satellite
images to prepare them for further processing. The Landsat radiometric calibration tool
in ENVI 4.8 was used for radiometric calibration and atmospheric correction in order to
convert the digital number of Landsat satellite images into radiance (Khalid et  al. 2018;
Mahmood et al. 2021). Then, the digital number values on every Landsat satellite image
were converted to reflectance. The relevant information for atmosphere corrections, such as
the data acquisition and sun elevation, was obtained from the Landsat header files. In the
end, Layer mosaicking was performed using Arc GISs’ spatial analysis tools (Sajjad et al.,

13
Natural Hazards

Fig. 2  Schematic illustration of the methodological framework used in this study

2020a, b). The processed Landsat-8 images were used to perform supervised classification
for land use and land cover mapping and to calculate water indices for flood inundation
mapping in ArcGIS 10.5, as shown in Fig. 2. Furthermore, the maximum likelihood (ML)
algorithm was utilized for LULC classification of pre-flood (23rd August, 2014), during-
flood (17th September, 2014) and post-flood (27th November, 2014) images. This algo-
rithm is commonly used for the extraction of unseen information from RS images (Alphan
et al. 2008; Notti et al. 2018; Sajjad et al. 2020a). ML method classified six LULC classes
in the study area (i.e., water class, agricultural areas class, built-up areas class, barren land
class, sand class and deposited soil materials class) with the help of 240 spatial training
points in the study area. These spatial training points were compiled in Microsoft Excel
and then imported into the Arc GIS 10.5.

13
 Natural Hazards

The resultant classified images were further converted into shapefile format (Arc GIS)
and utilized as input for spatial change detection analysis. Change detection analysis is
applied on RS data for monitoring land use/cover changes at various temporal scales (Smits
et  al. 1999). In this study, post-classification comparison method was used to calculate
LULC changes during three flood instances (Lu et al. 2007). This approach is commonly
used for LULC change detection (Khalid et al. 2018; Sajjad et al. 2020a). This approach is
applied on pre-flood, during-flood and post-flood classified maps, incorporated with spa-
tial overlay technique, which enables the comparison among classified LULC classes. This
comparison provides a clear picture of flood damage assessment in district Multan.
For damage assessment, a damaged area was defined as an area that has experienced the
changes from ‘vegetation and agriculture’ class and ‘built-up’ areas class in pre-flood stage
to ‘deposited soil material’ in post-flood occasion. A pre-flood classified map was used as a
basic reference for the change detection of LULC classes to assess flood impacts in district
Multan.

3.1 Modified normalized difference water index

MNDWI was utilized to detect flood-2014 inundation extent in the study area. This index
is commonly used for detection of flood inundation extent as per flood monitoring (Revilla-
Romero et al. 2015; Sajjad et al., 2020a, b). This index was developed by Xu (2006), and
it works efficiently to calculate and delineate flood water areas. Basically, it applies Green
Band 3 and Short-wave infrared Band 6, for the delineation of flooded areas, as shown in
Eq. (1)
(Band 3) − (Band 6)
MNDWI = (1)
(Band 3) + (Band 6)
MNDWI index removes built-up area noise while using Green wavelength to highlight
most of the water reflectance, in which the resultant values range from − 1 to + 1. In Short-
wave infrared band, the high built-up areas reflectance and low water areas reflectance gen-
erate the negative values of built-up areas and the positive values of water areas in resulted
images, respectively (Xu 2006).

3.2 Normalized difference vegetation index

This vegetation index is used as an indicator to quantify and monitor the vegetation and
agricultural areas greenness and to evaluate their density. This index is widely applied for
vegetation areas monitoring, since it helps to detect temporal changes in vegetation and
agriculture areas (Lu et  al. 2007; Musa et  al. 2015). This vegetation index applies Red
Band 4 and Near-infrared (NIR) Band 5 for monitoring vegetation areas, as shown in
Eq. (2):
(Band 5) − (Band 4)
NDVI = (2)
(Band 5) + (Band 4)
This index is used to measure the difference between Red Band, which absorbs vegeta-
tion, and NIR Band, which reflects vegetation, and the values range from − 1 to + 1. The
values closer to + 1 show higher possibility of greener vegetation.

13
Natural Hazards

3.3 Classification and flood inundation validation and accuracy assessment

LULC classification and flood-2014 inundation were validated through GE and GPS
verification points. The ArcGIS-based spatial overlay technique was applied to overlay
verification points on LULC classified maps and flood-2014 inundation maps. Further-
more, these point values were assessed by visual image interpretation technique. For
the validation of LULC classification maps, a total of 545 GE verification points were
used. Furthermore, a total of 30 points were used for individual classes of water, agri-
culture and vegetation, built-up, sand, barren and deposited soil materials to each clas-
sified map. The accuracy of the built-up class has remained slightly low, as mentioned
clearly in Table  2. Supervised Maximum Likelihood classification using Landsat data
has this drawback of variations in accuracy of the classified pixels (Smits et  al. 1999;
Alphan et al. 2008; Notti et al. 2018), as misclassified classes due to spectral confusion
that occurs between some wet sand and bare soil areas and built-up areas (Forget et al.
2018; Sajjad et al. 2020a, b). The accuracy of flood extent images was assessed using
600 GPS-based verification points. For each image, a total of 75 GPS points were used
to validate water and non-water classes. The accuracy results were further calculated by
using confusion matrix. This matrix calculate accuracy by comparing the points of the
resulted images with GE- and GPS-based verification points. Overall, user and producer
accuracies were computed for accuracy assessment of resulted images. Overall accuracy
(OA) is determined by the ratio of correctly resulted image points with the total num-
ber of verification data points. User accuracy (UA) is calculated by dividing correctly
resulted image points with the total number of verification data points of particular class
and producer accuracy (PA) is calculated by measuring the corrected resulted image
points of each class (Giordan et al. 2018; Munasinghe et al. 2018).

4 Results

4.1 Hydro‑meteorological phenomenon

Chenab flood-2014 occurred in early September due to a continuous seven-day wet spell
in upper Chenab. It triggered massive flash floods in upper Chenab and riverine floods in
lower Chenab plain. Analysis reveals that highest rainfall of the 15 stations along Chenab
River was above 113% of normal rainfall (Fig.  3). For example, above 500  mm of rain-
fall was recorded in Lahore, Sialkot and Rawalakot, which counts three times above the
average monthly rainfall of these stations. As a result, Chenab and Jhelum rivers carried
extraordinary flow discharge at various barrages (Fig. 1 for location). Furthermore, analy-
sis revealed that the discharge flow at several barrages was not only exceeding the dan-
ger level but also passed very close to designed capacity of barrages (Fig.  4), for, e.g.,
peak discharge of 0.86 million cubic ft./s at Marala barrage, was very close to its designed
capacity. Likewise, the peak of up to 0.95 million cubic ft./s was recorded at Khanki bar-
rage which was exceeding its designed capacity and breached at several locations (Fig. 4).
Similar situation also occurred at Qadirabad barrage. The exceeded flow overbanked the
river bed and inundated the large surrounding areas (Fig. 4). The swollen flow continued
toward Trimmu barrage (located 100 km in the upstream of district Multan) with its dev-
astating consequences. The analysis further showed that about 0.7 million cubic ft./s flow

13
 Natural Hazards

Fig. 3  Seven-day rainfall of selected meteorological stations,topography, study area

Fig. 4  Chenab flood-2014: designed capacity, peak discharge and danger level of barrage

13
Natural Hazards

reached at Trimmu barrage, which exceeded its capacity. Therefore, Jhang protection Left
(JPL) embankment was breached and affected surrounding areas of Jhang District (Fig. 4).
The results conclude that about 0.65 million cubic ft./s flood water reached district
Multan, which not only overtopped river banks, but also breached several protected

(a)

(b)
1200

1000

800
Area in km²

600

400

200

0
17/09/2014 24/09/2014 3/10/2014 10/10/2014 19/10/2014 26/10/2014 11/11/2014 27/11/2014
Dates

Fig. 5  a Multi-temporal flood inundation extent, peak flood extent, flood receded duration. b Recession of
flood inundation from peak flood (17th September) to post-flood (27th November) period in the study area

13
 Natural Hazards

embankments. As a result, extreme flow triggered huge flood inundation and affected
standing crops, houses, livestock and infrastructures at large scale in the study area
(Fig. 5a). Figure 5a, b revealed that about 0.15 million cubic ft./s flood water remained
standing for two months in the study area (Fig. 5a, b).

4.2 Riverine flood mapping

The riverine flood-2014 inundation extent was delineated to determine the most inun-
dated and affected areas in district Multan. The Table  1 shows that the OA acquired
from riverine flood-2014 inundation images is greater than 90%. The maximum OA of
92%, was achieved from 17 and 24th September images, while the minimum accuracy
of 88%, was acquired from 11th November image. Similarly, the obtained average UA
and PA of water and non-water classes were about 90%. The maximum achieved UA of
water class is 96%, from 17th September image, and the minimum of 85%, is got from
11th November image. As well, the highest gained UA of non-water class is 93%, from
27th November image, while the least is 86% from 17th September image. Besides, the
maximum achieved PA of water class is 94%, from 26th October image, and the least
PA is 87%, which is achieved from 17th September image. Also, the maximum PA of
non-water class is acquired from 17th September image, which is 96% and the least PA
is obtained from 11th November image, is 86% (Table 1).
The results, as shown in Fig. 5a, b, reveal that the cumulative peak flood inundation
of 1033.24 ­km2 area was detected on 17th September image. The flood situation almost
remained stable till 24th September, as water flow continued from upstream, and extent
slightly reduced to 1003.12 ­km2 area with the rate of 4.3km2 /day. On 03rd October, the
flood extent reduced to 886.48 k­ m2, with the receding rate of 14.6km2 /day. The flooded
area continued to recede till 10th October as it covered about 780.6 k­ m2 areas. On 19th
October, moderate stage was observed as flood extent was reduced to 665.76 k­ m2 area
and it continued to reduce with same rate of 13.75 ­km2/day till 26th October with an
area of 491.85km2. The normal stage was observed as decreasing trend continued dur-
ing the month of November, so much so that on 11th November, flood extent covered
only 201.56 ­km2 area. Lastly, only 146.56 k­ m2 area was covered by flood water on 27th
November. The variability of flood extent is presented in Fig. 5a, b.

Table 1  Accuracy assessment of riverine flood mapping

Classes 17/9/14 24/9/2014 3/10/2014 10/10/2014
UA PA OA UA PA OA UA PA OA UA PA OA

Water class (%) 96 86 92 92 91 91 91 88 90 90 91 90

Non-Water class (%) 87 97 91 92 88 91 92 90
Classes 19/10/2014 26/10/2014 11/11/2014 27/11/2014

Water class (%) 91 92 90 86 94 90 85 91 88 91 93 91

Non-water class (%) 91 90 90 87 92 86 93 92

13
Natural Hazards

Table 2  Accuracy assessment of LULC classifications

LULC classes Pre-flood (23/08/2014) During-flood Post-flood
(17/9/2014) (27/11/2014)
UA PA OA UA PA OA UA PA OA

Water class (%) 88 99 92 92 92 85 92 95 86

Built-up class (%) 84 98 76 83 80 95
Agriculture class (%) 92 92 80 87 88 70
Barren land class (%) 96 88 88 78 92 85
Sand class (%) 99 83 88 85 80 80
Deposited soil material (%) 92 92 – – – 86 95

4.3 Land use and land cover classifications and changes

The supervised ML classification algorithm was used to classify LULC classes in the
study area. The accuracy of the results has been assessed using confusion matrix. The
Table  2 reveals the average overall accuracy of LULC classification, which is 88%,
from all flood instances images. The maximum OA of 92% is achieved from 23rd
August image, while the least of 85% was acquired from 17th September image. Simi-
larly, the highest UA of 92% was obtained by water, 96% by barren and 99% by sand.
Water, built-up and deposited soil material have the maximum PA of 99%, 98% and
95%, respectively. Further the detail is presented in Table 2.
The classified results identified six LULC classes of water, built-up, vegetation and
agriculture, barren land, sand and deposited soil material from three flood instance
images in the study area (Fig.  6a). The result of the classified images showed that
deposited material was not found during flooding time because of extensive flood inun-
dation. Also, the results further revealed that meaningful variations were observed in
all classified instance images. In pre-flood instance, the water body covered only 7.6%
area, with built-up at 18.4%, vegetation and agriculture land at 55.2%, sand at 8.2%,
barren land at 5.3% and deposited material at 5.1% of the total area of 1681.329 k­ m2.
During flood instance image, water body showed enormous increase to 49.2%. The
built-up area experienced sizeable reduction, which shifted to 12.1%. On the other
hand, the large extent of water increase caused remarkable decrease to agriculture land
area from 55.2 to 32.8%. The sand also revealed a significant decrease from 8.2 to 1%.
But the barren land slightly decreased from 5.3 to 4.7%. The post-flood instance image
showed that water body retained its pre-flood stage, which is reduced to only 6% from
49.2%. Built-up area increased to17.79% as it was in pre-flood instance. In post-flood,
flood water receded back, but it had deposited sediments and bare soil material in the
study area, that resulted in a remarkable increase to 29.2% of total 1681.329 ­km2 area.
Similarly, the vegetation and agriculture land had noticably increased from 32.40% to
37.50%. The barren land had slightly changed, but the sand area increased to 4.4%
from 1.17% in post-flood image, as shown in Fig. 6a, b. The detail of each class is pre-
sented in Fig. 6b.

13
 Natural Hazards

(a)

(b) 1000
900
800
700
600
Area in km²

500
400
300
200
100
0
Water Built-up Vgetation and Barren land Sand Deposited material
Agriculture

Pre-flood (23th August, 2014) During- Flood (17th September, 2014) Post-flood (27th November, 2014)

Fig. 6  a Supervised LULC during flood-2014 instances. b Change detection of LULC in district Multan

5 Flood impact assessment

The results revealed that the agricultural area covering ~ 323.18 ­km2 was the most
severely inundated in the study area. Similarly, from the total inundated area of 828.25
­km2 in the study area, ~ 162.24 k­ m2 of the built-up area was also inundated as shown
in Fig.  7a. The spatial extent revealed that the most inundated agriculture areas were
located in the southern part, while the the most inundated built-up areas were in the cen-
tral and northern part of the study area (Fig. 7a). Furthermore, the damage result shows
that the agricultural areas of about ~ 180.43 k­ m2, and the built‐up areas of about1 ~ 08
­km2, were seriously damaged due to deposited material in the study area. This further
shows that agriculture was the most affected sector in Multan district. Moreover, the
NDVI results evaluated for agriculture change during flood instance images match with
the classified results (Fig. 7a, b).

13
Natural Hazards

(a)

(b)
200
Damaged areas

150
AREA IN KM²

100

50

0
Vegetaon and Agriculture Built-up

Fig. 7  a and b Flood-2014 damaged agricultural and built-up areas in district Multan

13
 Natural Hazards

Fig. 8  Spatio-temporal vegetation and agriculture changes (a) using ML supervised classification (b) using
NDVI index

The agriculture and vegetation areas are compared in Fig. 8a, b. The classified image
results showed that the vegetation and agriculture areas in all flood instances represent only
1% deviation from the NDVI results. This reveals that both ML supervised classification

13
Natural Hazards

Fig. 9  Spatio-temporal vegetation and agriculture monitoring using NDVI index

and NDVI index produce similar vegetation areas in the study area. Furthermore, this fur-
ther validate our impact assessment results on agriculture and vegetation based on classi-
fied LULC images.
The values of the NDVI analysis ranged from − 0.07 to + 0.5, − 0.17 to + 0.5 and − 0.14
to + 0.39 for the pre-flood, during-flood and post-flood, respectively (Fig.  9). The high
NDVI values were observed in pre-flood while least in post-flood. Higher values of NDVI
in pre-flood image revealed the high density of healthy and productive vegetation and agri-
culture. Likewise, the lower values in post-flood showed the unhealthy vegetation, under
deposited material and bare soil. So, the result showed the significant vegetation and agri-
culture changes in NDVI values from pre-flood to post-flood, as depicted in Fig. 9, and this
helped in concluding that the flood-2014 has badly damaged agricultural areas in the study
area.

6 Discussions

This study demonstrates that optical satellite data allow the flood inundation and its associ-
ated impact assessment. In this study, due to the absence of during-flood, free SAR images,
Landsat-8 data (~ 30  m) was used for the flood inundation and flood impact assessment
(Chohan et al. 2015; Khalid et al. 2018; Sajjad et al. 2020a). Firstly, it is evident that Land-
sat data have limitation to the point of not being able to acquire data in a timely man-
ner, which lessens their appropriateness for the delineation of flooded areas inundation and
impact assessment (Munasinghe et al. 2018; Faisal et al. 2018). However, as the study area

13
 Natural Hazards

of the present study lies in two nearby overlapping Landsat paths of 150 and 151, which
enhanced the temporal resolution of used images (~ 8 days), besides a single path tempo-
ral resolution of 16 days for an area (Notti et al. 2018; Bhatt et al. 2016). Thus, the high
temporal Landsat data allowed a detailed study on riverine flood inundation mapping. Sec-
ondly, cloud weather condition can also restrain the ability to acquire during flood Land-
sat images. In comparison, SAR and RADAR, satellites can easily obtain images in all
weather conditions (Pradhan et al. 2010; Musa et al. 2015). In our case, district Multan was
completely cloud-free, as the causative weather events occurred in upper Chenab (Fig. 3)
about 10 days before the flood reached the study area. Thirdly, Flood inundation impacts
on LULC are significant consideration for flood impact assessment in the study area. The
agricultural and built-up area damages can be extracted by SAR, RADAR and multispec-
tral Landsat satellites (Uddin et al. 2019). With Landsat data, flood inundation areas can
be detected after several days of flood (Rokni et al. 2014; Al Faisal et al. 2018). However,
open Landsat-8 data were used because it can also easily detect sediments on built-up areas
and agricultural fields with high accuracy (Chohan et  al. 2015; Khalid et  al. 2018; Saj-
jad et al. 2020a). Despite that, field survey and high-resolution data are needed for flood
impact assessment. However, due to high cost, it is not permitted to use high- resolution
(RADAR, SAR) data, and aerial photographs in the study area. In this study, a field survey
was also conducted to physically observe the flooding situation. Furthermore, The NDVI
index was also used (Fig. 8) for correlation and validation of classified impact assessment
and provided impact assessment information on agriculture and vegetation. Finally, the
accuracy assessment of ML classification and flood inundation was determined to verify
the consistency of the results. The average OA of 90%, is obtained from flood inundation
images (Table 1), while the LULC classification images produced OA of 88% (Table 2). In
this study, the water body produced the highest average accuracy of 90%. Similarly, sand
class also attained better OA, but in fewer occasions, damp sand pixels are mixed with
water body and agricultural areas. Likewise, in some areas, built-up and agriculture areas
are also mixed with each other due to wet vegetation areas.
The highest flood peak was observed on 17th September, 2014, and then the water
remained standing for almost two months, as shown in Fig.  5a, b. The flooded water
receded in three stages; in peak stage, water slightly reduced with a rate of about 1­ 1km2
area per day till 10th October. In moderate stage, the flood water significantly reduced with
a rate of 1­ 8km2 area per day till 26 October. In the normal stage, flood water decreases with
a rate of 11.5km2/day until 27th November and river regained its pre-flood normal flow
(Fig. 5b). However, the interpretation of result indicated that huge inundation and duration
of almost 8  weeks, generated a huge loss to agricultural and built-up areas, as shown in
Fig. 7a, b.
Furthermore, field survey indicated that the rice crop requires plenty of water for its
cultivation, but long flood duration affected the rice crop. The sugarcane and cotton crops
were almost ready for harvest, but extensive inundation left no chance for saving crops. The
long duration also disrupted the upcoming sowing season of Rabi crops, which were not
being timely cultivated and impacted overall production. As the flood duration increases,
the sensitivity of standing and upcoming sowing crops to flood also enhances.
Based on the NDVI index for vegetation and agriculture monitoring in the study area,
higher value in pre-flood shows healthy and cultivated agriculture in the study area. How-
ever, for during flood instance, the central part was inundated, which reduced the vege-
tation and agriculture areas. Finally, in post-flood instance, flood water receded but the
vegetation area remained lowest due to direct flood damages by inundation coupled with
deposited soil and other material on agriculture and vegetation areas, as shown in (Fig. 9).

13
Natural Hazards

The flood impact investigation purposes to offer low-priced damage assessment of built-up
and agricultural areas; however, it cannot give us indirect impacts and monetary loss infor-
mation, which often occur in post-flooding stage.

7 Conclusion

In lower Chenab basin, district Multan is highly susceptible to persistent riverine floods.
Almost every year, local people residing along floodplains of Chenab River are facing this
menace of floods. In 2014, riverine flood due to seven-day wet spell hit the Chenab basins
and a peak flood water discharge of more than 0.6 million cubic ft./s of water reached study
area that caused inundation and affected at large scale. Subsequently, ready to harvest
standing crops, houses, livestock, and all kind of infrastructures were adversely affected.
These repeated riverine floods are a serious problem that require adequate efforts for its
preparedness and reducing its effects, mainly through post-flood monitoring strategies
which are questionable and far too inadequate especially, in district Multan.
This study showcased the flood inundation and the severity of flood impacts in district
Multan with the integration of remote sensing data with jointly exploited appropriate meth-
ods. Specifically, Landsat-8 images of flood instances have been processed with GIS-based
supervised ML classification for inundated and affected LULC as per damage assessment.
MNDWI water index has been used for flood inundation extent monitoring, while NDVI
index was used to temporal monitoring of agriculture and vegetation areas. The LULC
classification and flood inundation images produced OA of 90%, which reveals the excel-
lent results. Thus, these methods can be suggested beneficial for flooded areas monitoring
and impacts assessment. Moreover, the classified images and NDVI results showed that the
agriculture was the most affected LULC in the study area.
The flood inundation mapping further revealed that the flood water remained for almost
two months in district Multan, which further increased and intensified the flood damages
on agriculture and built-up areas. This research concluded that RS techniques are useful
for pursuing proactive flood mapping and monitoring which is needed for early relief and
recovery operations in flood management. Thus, the present study provides another percep-
tion to riverine flood inundation mapping and impact assessment using the open satellite
data and methods.
This study is based on Multan district; however, the methodology can be applied in any
area and it thus contributes to science on riverine floods. In this current study, supervised
ML classification method coupled with MNDWI index and NDVI index has been used
which can resolve the problem of flood monitoring and deliver appropriate resolutions to
flood authorities in order to pursue flood management. The use of ML classification and
the MNDWI provided a reliable flooded area extent mapping using optical Landsat data.
However, in most cases, these methods are used separately to produce flood mapping. Fur-
thermore, despite the unique agriculture characteristic along Chenab basins, the combina-
tion of NDVI and supervised ML classification provide insight into agricultural areas, and
enable temporal investigation of flood impacts on agriculture and vegetation. In this study,
the identification of two nearby Landsat overlapping paths provides a high temporal data,
which might be used for effective flood assessment and management. This exploitation of
two paths increased the temporal resolution, from 15 to 8  days, which lead to effective
flood mapping for emergency response and early relief operations. Thus, this research pro-
vides another avenue to flood mapping and impact assessment using free satellite data.

13
 Natural Hazards

Acknowledgment and Funding  This work was funded by the National Key Research and Development Pro-
gram (2018YFC1506506), the Frontier Project of Applied Foundation of Wuhan (2019020701011502), Key
Research and Development Program of Jiangxi Province (20201BBG71002), and the LIESMARS Special
Research Funding.

Author contributions  SA and JL conceptualized overall research design; image analysis, flood inundation
maps, LULC maps and flood impact assessment were contributed by SA; validation was contributed by SA,
JL and CC; investigation was contributed by CX; writing—original draft preparation was contributed by
SA; writing—review and editing were contributed by SA, JL, NM and BN.

Data availability  The datasets used and/or analyzed during the current study are available from the corre-
sponding author on reasonable request.

Declarations 
Conflict of interest  The authors declare that they have no conflict of interest.

Ethics approval and consent to participate  Not applicable.

Consent for publication  Not applicable.

References
Al Faisal A, Al Kafy A, Roy S (2018) Integration of remote sensing and GIS techniques for flood monitor-
ing and damage assessment: a case study of Naogaon District, Bangladesh. J Remote Sens GIS 7:2
Alphan H, Doygun H, Unlukaplan Y (2008) Post-classification comparison of land cover using multi-
temporal Landsat and ASTER imagery: the case of Kahramanmaras, Turkey. Environ Monit Assess
151:327–336
Atif I, Mahboob AM, Waheed A (2016) Spatio-temporal mapping and multi-sector damage assessment of
2014 flood in Pakistan using remote sensing and GIS. Indian J Sci Technol 9(1):1–11
Bhatt CM, Rao GS, Farooq M, Manjusree P, Shukla A, Sharma SVSP, Kulkarni SS, Begum A, Bhanumur-
thy V, Diwakar PG, Dadhwal VK (2016) Satellite-based assessment of the catastrophic Jhelum floods
of September 2014, Jammu & Kashmir, India. Geomat Nat Haz Risk 8(2):309–327
Cann KF, Thomas D, Salmon RL, Wyn-Jones AP, Kay D (2012) Extreme water-related weather events and
waterborne disease. Epidemiol Infect 141:1–16
Chignell S, Anderson R, Evangelista P, Merritt LM (2015) Multi-temporal independent component analysis
and Landsat 8 for delineating maximum extent of the 2013 Colorado front range flood. Remote Sens
7:9822–9843
Chohan K, Ahmad S, Islam Z, Adrees M (2015) Riverine flood damage assessment of cultivated lands along
Chenab river using GIS and remotely sensed data: a case study of District Hafizabad, Punjab, Pakistan.
J Geogr Inf Syst 07:506–526
Cook A, Merwade V (2009) Effect of topographic data, geometric configuration and modeling approach on
flood inundation mapping. J Hydrol 377:131–142
Domeneghetti A, Schumann GJP, Tarpanelli A (2019) Preface: remote sensing for flood mapping and moni-
toring of flood dynamics. Remote Sens 11(8):943
Federal Flood Commission Islamabad (FFCI). (2014) Annual flood report. Ministry of Water and Power,
Pakistan
Forget Y, Linard C, Gilbert M (2018) Supervised classification of built-up areas in sub-saharan african cities
using landsat imagery and OpenStreetMap. Remote Sens 10(7):1145
Gaurav K, Sinha R, Panda PK (2011) The Indus flood of 2010 in Pakistan: a perspective analysis using
remote sensing data. Nat Hazards 59(3):1815–1826
Giordan D, Notti D, Villa A, Zucca F, Calò F, Pepe A, Dutto F, Pari P, Baldo M, Allasia P (2018) Low cost,
multiscale and multi-sensor application for flooded area mapping. Nat Hazard 18:1493–1516
Giustarini L, Chini M, Hostache R, Pappenberger F, Matgen P (2015) Flood hazard mapping combining
hydrodynamic modeling and multi annual remote sensing data. Remote Sens 7:14200–14226
Government of Pakistan (GoP) (2017) District census report of Multan, Population Census Organization,
Islamabad

13
Natural Hazards

Halgamuge M, Nirmalathas A (2017) Analysis of large flood events: based on flood data during 1985–2016
in Australia and India. Int J Disast Risk Reduc 24:1–11
Hashmi H, Siddiqui Q, Ghumman AR, Ahmed Kamal M, Mughal HUR (2012) A critical analysis of 2010
floods in Pakistan. Afr J Agric Res 7:1054–1067
Islam AS, Bala S, Haque MA (2010) Flood Inundation map of Bangladesh using MODIS time-series
images. J Flood Risk Manag 3:210–222
Khalid B, Cholaw B, Alvim DS, Javeed S, Khan JA, Javed MA, Khan AH (2018) Riverine flood assessment
in Jhang district in connection with ENSO and summer monsoon rainfall over Upper Indus Basin for
2010. Nat Hazards 92(2):971–993
Khan B, Iqbal MJ (2013) Forecasting flood risk in the indus river system using hydrological parameters and
its damage assessment. Arab J Geosci 6:4069–4078
Khan SI, Hong Y, Gourley JJ, Khattak MU, De Groeve T (2014) Multi-sensor imaging and space-ground
cross-validation for 2010 flood along Indus river, Pakistan. Remote Sens 6:2393–2407
Lu D, Mausel P, Brondi Zio E, Moran E (2007) Change detection techniques. Int J Remote Sens
25(12):2365–2407
Mahmood S, Rani R (2018) Extent of flood damages in Chenab Basin, Upper Indus Plain. Nat Hazards Risk
Assess Vulnerabil Reduct. https://​doi.​org/​10.​5772/​intec​hopen.​79687
Mahmood S, Khan AH, Mayo SM (2016a) Exploring underlying causes and assessing damages of 2010
flash flood in the upper zone of Panjkora River. Nat Hazards 83(2):1213–1227
Mahmood S, Khan AH, Ullah S (2016b) Assessment of 2010 flash flood causes and associated damages in
Dir Valley, Khyber Pakhtunkhwa Pakistan. Int J Disaster Risk Reduct 16:215–223
Mahmood S, Rahman A, Sajjad A (2019a) Assessment of 2010 flood disaster causes and damages in district
Muzaffargarh, Central Indus Basin. Pakistan Environ Earth Sci 78(3):63
Mahmood S, Rahman A, Shaw R (2019b) Spatial appraisal of flood risk assessment and evaluation using
integrated hydro-probabilistic approach in Panjkora River Basin. Pakistan Environ Monit Assess
191:573
Mahmood S, Sajjad A, Rahman A-u (2021) Cause and damage analysis of 2010 food disaster in district,
Muzaffar Garh, Pakistan. Nat Hazards 107:1681–1692
Munasinghe D, Cohen S, Huang YF, TsangYP ZJ, Fang Z (2018) Intercomparison of satellite remote sens-
ing based flood inundation mapping techniques. J Am Water Resour Assoc 54:834–846
Musa ZN, Popescu I, Mynett A (2015) A review of applications of satellite SAR, optical, altimetry and
DEM data for surface water modelling, mapping and parameter estimation. Hydrol Earth Syst Sci
19:3755–3769
National Disaster Management Authority (NDMA) (2014) Annual flood report, Islamabad
Notti D, Giordan D, Caló F, Pepe A, Zucca F, Galve J (2018) Potential and limitations of open satellite data
for flood mapping. Remote Sens 10(11):1673
Pakistan Meteorological Department (PMD) (2014) Annual report, Regional Meteorological Observatory,
Lahore
Pradhan B, Shafiee M, Pirasteh S (2010) Maximum flood prone area mapping using RADARSAT images
and GIS: Kelantan River basin. Int J Geoinform 5:11
Punjab Provimcial Disaster Management Authority (PPDMA) (2014) Annual flood report, Lahore
Rahman A, Khan AN (2011) Analysis of flood causes and associated socio-economic damages in the Hindu
Kush region. Nat Hazards 59(3):1239
Rahman A, Khan AN (2012) Analysis of 2010-flood causes, nature and magnitude in the Khyber Pakh-
tunkhwa. Pakistan Nat Hazards 66(2):887–904
Rahman G, Rahman A, Anwar M, Khan M, Ashraf H, Zafar U (2017) Socio-economic damages caused by
the 2014 flood in Punjab Province, Pakistan. Proc Pakistan Acad Sci 54:365–374
Revilla-Romero B, Hirpa F, Thielen-del Pozo J, Salamon P, Brakenridge R, Pappenberger F, De Groeve T
(2015) On the use of global flood forecasts and satellite-derived inundation maps for flood monitoring
in data-sparse regions. Remote Sens 7:15702–15728
Rokni K, Ahmad A, Selamat A, Hazini S (2014) Water feature extraction and change detection using multi-
temporal Landsat imagery. Remote Sens 6:4173–4189
Sajjad A, Lu J, Chen X, Chisenga C, Mahmood S (2019) The riverine flood catastrophe in august 2010
in south Punjab, Pakistan: potential causes, extent and damage assessment. Appl Ecol Environ Res
17(6):14121–14142
Sajjad A, Lu J, Chen X, Chisenga C, Saleem N, Hassan H (2020a) Operational monitoring and damage
assessment of riverine flood-2014 in the lower Chenab Plain, Punjab, Pakistan, using remote sensing
and GIS techniques. Remote Sens 12(4):714
Sajjad A, Lu J, Chen X, Chisenga C, Saleem N (2020b) Rapid riverine flood mapping with different water
indices using flood instances Landsat-8 images. https://​doi.​org/​10.​3390/​ECWS-5-​08049

13
 Natural Hazards

Sanyal J, Lu XX (2004) Application of remote sensing in flood management with special reference to mon-
soon asia: a review. Nat Hazards 33:283–301
Schumann G, Bates P, Apel H, Aronica G (2018) Global flood hazard mapping, modeling, and forecasting,
pp 239–244
Smits PC, Dellepiane SG, Schowengerdt RA (1999) Quality assessment of image classification algo-
rithms for land-cover mapping: a review and a proposal for a cost-based approach. Int J Remote Sens
20:1461–1486
Syvitski J, Brakenridge R (2013) Causation and avoidance of catastrophic flooding along the Indus River,
Pakistan. GSA Today 23:4–10
Tariq M, van de Giesen N (2012) Floods and flood management in Pakistan. Phys Chem Earth 47–48:11–20
Uddink MMA, Meyer FJ (2019) Operational flood mapping using multi-temporal Sentinel-1 SAR images: a
case study from Bangladesh. Remote Sens 11:1581
Wulder M, Masek J, Cohen W, Loveland T, Woodcock C (2012) Opening the archive: how free data has
enabled the science and monitoring promise of Landsat. Remote Sens Environ 122:2–10
Xu H (2006) Modification of Normalized Difference Water Index (NDWI) to enhance open water features in
remotely sensed imagery. Int J Remote Sens 27:3025–3033

Publisher’s Note  Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Authors and Affiliations

Asif Sajjad1   · Jianzhong Lu1 · Xiaoling Chen1 · Chikondi Chisenga2 ·

Nausheen Mazhar3 · Basit Nadeem4
Asif Sajjad
[email protected]
Xiaoling Chen
[email protected]
Chikondi Chisenga
[email protected]
Nausheen Mazhar
[email protected]
Basit Nadeem
[email protected]
1
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing,
Wuhan University, Wuhan 430079, China
2
Department of Earth Sciences, Ndata School of Climate and Earth Sciences, Malawi University
of Science and Technology, Box 5196, Limbe, Malawi
3
Department of Geography, Lahore College for Women University, Lahore, Pakistan
4
Department of Geography, Bahauddin Zakariya University, Multan, Pakistan

13