Integrating Social Media and Remote Sensing Data for Flood Assessment in Developing Countries: A Case Study in Douala Estuary, Cameroon

Abstract

According to the United Nations Office for Disaster Risk Reduction, 1 the number of global weather-related disasters in the past decades has been estimated at 90% of the total number of disasters. These are likely to increase in severity and frequency with current and forecasted global climate changes. Disaster risk monitoring using space technology and machine learning has become essential for minimizing and managing the consequences of natural disasters. However, flood assessment and data collection remain deficient in many parts of developing countries. Recent studies in Cameroon have identified socioeconomic limitations and low adaptive capacity to manage floods that threaten and expose populations to vulnerability and danger. The increasing availability of smartphones and social media data allows individuals to directly document floods in real-time and otherwise poorly observed areas. However, these data are rarely used for flood assessment purposes in developing countries. This information, in synergy with remote sensing, can help disaster managers and rescuers determine routes and maps to support flood response as well as post-flood activities, such as calibrating flood hydrodynamic models. The paper will focus on integrating community social media with satellite remote sensing data to assess and assist in flood disaster emergency response and preparedness in Douala Estuary in Cameroon.

Full text

72th International Astronautical Congress, Dubai, United Arab Emirates.
IAC–21–B1,5,9
Integrating Social Media and Remote Sensing Data for Flood Assessment
in Developing Countries: A Case Study in Douala Estuary, Cameroon
Desire Muhire1,2,Swarnajyoti Mukherjee1,3,Krittanon Sirorattanakul1,4,Nzeussi Mbouendeu
Charles-aim´e1,5,Victor Hertel1,6 , Ikechukwu Maduako1,7,Chukwuma Okolie1,8,Daniela
Vargas-Sanabria1,9,Lako Stephane1,10,Ikenna Arungwa1,11,Avanija Menon1,12,Abinash
Silwal1,13,Alessandro Novellino1,14,Barthelemy Ndongo15,Marco Romero1,16
1Space Technology for Earth Applications (STEA) Project Group (PG), Space Generation Advisory
Council (SGAC), Vienna, Austria
2Department of physics, Chouaib Doukkali University, El Jadida, Morocco
3Space System Engineer & Business Professional, GP Advanced Projects, Srl, Brescia, Italy
4Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA, USA
5International Space University, 67400 Ilkirch-Graffenstaden; Strasbourg, France
6IRS, University of Stuttgart, Germany
7Department of Geoinformatics and Surveying, Faculty of Environmental Studies, University of Nigeria,
Enugu Campus
8Department of Surveying and Geoinformatics, Faculty of Engineering, University of Lagos, Nigeria
9Laboratorio de Investigaci´on e Innovaci´on Tecnol´ogica, Universidad Estatal a Distancia, Costa Rica
10 Climate and Water Resources Department, Water For Life Cameroon, Yaounde, Cameroon
11 Department of Surveying and Geoinformatics, School of Environmental Science, Federal University of
Technology, Owerri, Nigeria
12 Department of Physics and Astronomy, University College London, United Kingdom
13 Department of Geomatics Engineering, Kathmandu University, Nepal
14 British Geological Survey, United Kingdom
15 Dschang University, Cameroon
16 Angolan Office for Space Affairs, Luanda, Angola
According to the United Nations Office for Disaster Risk Reduction,1the number of global weather-related
disasters in the past decades has been estimated at 90% of the total number of disasters. These are likely
to increase in severity and frequency with current and forecasted global climate changes. Disaster risk
monitoring using space technology and machine learning has become essential for minimizing and managing
the consequences of natural disasters. However, flood assessment and data collection remain deficient in
many parts of developing countries. Recent studies in Cameroon have identified socio-economic limitations
and low adaptive capacity to manage floods that threaten and expose populations to vulnerability and
danger. The increasing availability of smartphones and social media data allows individuals to directly
document floods in real-time and otherwise poorly observed areas. However, these data are rarely used
for flood assessment purposes in developing countries. This information, in synergy with remote sensing,
can help disaster managers and rescuers determine routes and maps to support flood response as well as
post-flood activities, such as calibrating flood hydrodynamic models. The paper will focus on integrating
community social media with satellite remote sensing data to assess and assist in flood disaster emergency
response and preparedness in Douala Estuary in Cameroon.
Keywords: social media, disaster management, remote sensing, Cameroon
1 Introduction
Increasing Climate change will cause frequent nat-
ural disasters, human and material losses worldwide,
and will severely affect developing countries. The
number of global weather disasters during the period
1995-2015 is estimated at 90% of the total number
of disasters.1During this period, floods affected 2.3
billion people worldwide, with more people affected
in Asia and Africa than in other continents.2
Recent studies in northern Cameroon show that
environmental factors, socio-economic constraints
IAC–21–B1,5,9 Page 1 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
and lack of adaptability to flood management endan-
ger and expose the population to vulnerabilities and
risks.3, 4 The international disaster database EM-
DAT reports 16 floods in Cameroon between 1988
and 2017, which killed 131 people and affected nearly
400,000.5The limited risk assessment is mainly due
to the lack of institutions for flood prevention, im-
mediate disaster protection and mitigation, and the
financial capacity of highly trained personnel. How-
ever, the addition of data sources and big data can
enable real-time flood assessment while significantly
reducing costs.6
As demonstrated,7, 8 open access satellite images
widely used for flood monitoring may have some lim-
itations (spatio-temporal resolution, visit time, etc.).
However, they can be complemented by auxiliary
data. Communication via smartphones, social net-
works are emerging as reliable and inexpensive auxil-
iary data used by the scientific community for disaster
assessment, especially floods.9In emergencies, social
media platforms such as Twitter have become a tool
for exchanging information at the community level.10
Twitter can be used to improve the effectiveness of
social reaction, awareness, and relief efforts when
combined with existing crisis management method-
ologies, machine learning algorithms11 and supported
community-level training.12 When satellite imagery
is not available or if images are obscured by clouds,
Twitter data can also provide real-time damage as-
sessments of disaster situations providing advantages
over other existing methods.13, 14
This paper focuses on the use of Sentinel-1
Synthetic Aperture Radar (SAR) data in Douala,
Cameroon to take advantage of its active illumina-
tion to obtain day/night and all-weather data and
also seeks to use new auxiliary data based on social
media which are very abundant on the continent, but
not valued.
The rest of the paper is organized as follows. Sec-
tion 2 describes the study area while highlighting the
anthropological causes of flooding in Douala. Section
3 presents the methods used to process radar images
and social media data. Section 4 presents the results
and Section 5 concludes with perspectives on actions
to mitigate flooding in the region.
2 Study region
Douala is the economic capital of Cameroon lo-
cated at 04°03’N 9°41’E in Central Africa with a high
population over 3 million inhabitants15(Fig.1). The
problem of flooding is widespread in Cameroon.16, 17
For instance, floods events have occurred up to 5-
10 times a year in the capital and 1-5 times a year
in rural areas such as Maga and Lagdo (North-
ern region).17 Since 1975, when the Bamendjin
dam was constructed, the Ndop plain in northeast-
ern Cameroon has experienced periodic flooding, es-
pecially during the rainy season.18 Limbe, a sea-
side town in the southwest region of Cameroon, was
also heavily flooded in 2001, with over 2000 people
homeless, and destroyed properties and infrastruc-
ture.19 In 2008, the geographical area of Nkolbisson,
Cameroon, for example, was hit by two catastrophic
floods.16, 17
Fig. 1: Location of the urban area and the main wa-
tersheds of Douala. (Source: Figure 5 from16)
2.1 Flood triggering mechanisms in
Cameroon
Poor waste management has been identified as a
major cause of flooding in developing countries like
Cameroon.20 In Mefou, in central Cameroon and in
the Dakar district of Douala (Fig.2), drains were ob-
served to be clogged with plastic bottles and other
solid waste.21 In another study by,22 flooding in
Limbe was discovered to be resulted from river chan-
nel blockage caused by indiscriminate dumping of
refuse into the waterway and sediment deposition
from upstream.
IAC–21–B1,5,9 Page 2 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
Fig. 2: Plastic pollution completely blocking a water-
way in the Dakar district of Douala, Cameroon23
In the context of climate change, the increasing
urbanization of the region is known to disregard
drainage systems designed to contain runoff and the
maximum volume of water that must flow through
them during rainy periods. Thus, factors such as
inadequate drains, uncontrolled waste disposal, and
the nature of precipitation were considered common
and important triggers to consider in mitigating and
preparing for flooding in the region.19
2.2 Precipitation in Douala
In Douala, flooding is common during the rainy
season from March to October(Fig.1 and 3). The
Tongo Bassa watershed located in the heart of the
great Cameroonian economic metropolis of Douala,
is one of the most affected urban location of the city.
Tongo Bassa occupies an area of approximately 4200
ha or 42 km2. The Tongo Bassa basin is crossed by
three rivers and is characterized by a gentle slope (0.1
to 0.7%) which exposes it to the daily tide variations.
Bonamoussadi, Bepanda and Makepe Missoke are the
most frequently affected areas, distributed on both
sides of the Tongo Bassa river.
This highly urbanized basin is subject to rapid
runoff towards low-lying areas with limited infiltra-
tion and high sedimentation rate in drains. Floods in
these areas frequently affect residential houses, goods
and services due to their exposure and low coping ca-
pacities of inhabitants, causing damage and loss of
lives. This case is pertinent to the following dates:
August 2nd to 3rd, 2000, August 9th , 2010, and more
recently that of August 21st, 2020, August 11th, 2021,
September 1st, 16th and 18th 2021.
3 Methods
3.1 Synthetic Aperture Radar
Synthetic Aperture Radar (SAR) is an active
microwave remote sensing system in which a sen-
sor sends a beam towards the ground and acquires
the backscattered signals after interacting with the
Earth’s surface. Unlike optical satellite imagery, it is
independent of solar electromagnetic energy and can
capture high-resolution images during the day and
night, in almost all-weather conditions, and through
clouds.26, 27
The scattering of objects on the SAR image
is highly influenced by the terrain (geometry, fea-
tures, etc.) and also acquisition properties (resolu-
tion, incident wave, ascending or descending-pass,
etc.). In addition, the acquisition can be done
by emitting and receiving horizontal (H) or verti-
cal (V) polarization(cross-polarized (VH/HV) or co-
polarized (HH/VV) acquisitions) that interacted dif-
ferently with the ground. It therefore provides an
additional information to characterize the phenom-
ena of the observed region.27 The best accuracy for
flood mapping has been reported to be by using VH
polarization configuration.28
For a given mission of constant incidence angle and
wavelength, the backscattering signal for a targeted
area varies depending on the dielectric properties of
the target, the physical shape of the scatterers in the
target area of the resolution cell.29 Water and met-
als represent objects with higher dielectric content
than other materials and have a very large response.
Therefore, if the geometric shape lies in front of the
signal line of sight (such as the roofs of houses), the
objects will appear bright because a strong signal is
returned (or backscattered) to the sensor. On the
other hand, if the surface is flat as a plane mirror, the
incoming pulses reflect away from the sensor and they
appear as a dark feature (flat water, etc.). Irregular
geometries, such as vegetation cover, are grayed out
because scattering occurs in all directions and only a
small fraction of signals is reflected back to the sen-
sor. Thus before flooding occurs, dry soil or vegeta-
tion would have a lower dielectric response. After an
area has been flooded ,due to the high dielectric con-
stant of water (80), the moisture content increases
the returned signals. This response presents multi-
ple reflections possibilities(specular reflection, double
bounce, etc.) from the surface, which can make it dif-
ficult to extract the flood map, especially in vegetated
(specular reflection or double-bounce within canopy)
and urban areas (double bounce in buildings).
IAC–21–B1,5,9 Page 3 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
32 °C32 °C
33 °C33 °C
32 °C32 °C
31 °C31 °C
29 °C29 °C
27 °C27 °C
26 °C26 °C 26 °C26 °C
27 °C27 °C
28 °C28 °C
30 °C30 °C
31 °C31 °C
23 °C23 °C 23 °C23 °C
24 °C24 °C 24 °C24 °C
23 °C23 °C 23 °C23 °C 23 °C23 °C 23 °C23 °C 23 °C23 °C 23 °C23 °C 23 °C23 °C 23 ° C23 °C
Precipitation Mean daily maxim um Hot days
Mean daily minim um Cold nights Wind speed
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
20 °C
22 °C
24 °C
26 °C
28 °C
30 °C
32 °C
34 °C
36 °C
0 mm
50 mm
100 mm
150 mm
200 mm
meteoblue
Fig. 3: Douala Average temperatures and precipitation (Littoral, Cameroon, 4.05°N 9.7°E)(Source:
www.meteoblue.com).24 The mean daily maximum (solid red line) shows the maximum temperature
of an average day for every month for Douala. Likewise, mean daily minimum (solid blue line) shows the
average minimum temperature. Warm days and cool nights (dashed red and blue lines) show the average
of the hottest day and coldest night of each month of the last 30 years
Fig. 4: Douala floods are frequents in July and Au-
gust with several damage25
Fig. 5: Spatial distribution of floods (31 events
recorded) in Douala districts over the period 1984-
2018. (Source: Figure 5 from16)
The SAR image has two major inherent limita-
tions due to its angular viewing that leads to radio-
metric distortions or foreshortening and the diffrac-
tion induced speckle noises. SAR data exhibited salt
and pepper noise are caused by a phenomenon inher-
ent in the active coherent illumination system called
speckles. These speckles are due to random construc-
tive and destructive interferences in each resolution
cell of the image, resulting in degradation of image
quality and interpretation. Thus, before any applica-
tion, these radar images must be pre-processed to re-
move the noises either by spatial filtering or by multi-
looking operations.30
In general, floods occur under severe weather con-
ditions with heavy rainfall and dense cloud cover.
These clouds hinder the effectiveness of optical satel-
lite imagery,31 hence, the use of SAR data for flood
monitoring has become very common,32 and much
research has demonstrated its effectiveness in flood
events assessment.33
SAR-based flood detection techniques comprise
thresholding-based methods,34 image segmenta-
tion,35 statistical active contouring,36 rule-based clas-
sification,37 interferometric-SAR coherence analysis
and data fusion approaches.38 To improve accuracy,
thresholding-based flood detection techniques have
evolved by merging additional data with the topo-
graphic data.
3.2 Change detection
The United Nations Platform for Space-based In-
formation for Disaster Management and Emergency
Response (UN-SPIDER) has made available an ad-
vanced thresholding-based method that generates
IAC–21–B1,5,9 Page 4 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
flood extent map and assessment of affected areas.39
The extent of a flood event is calculated us-
ing Sentinel-1 SAR data and a change detection
method. This tool also includes an assessment of
the number of people likely to be exposed, crop-
land and metropolitan areas affected, which can
be cross-referenced with the generated flood extent
layer and visualized in minutes. This approach
is suitable for developing countries as it uses the
Google Earth Engine (GEE) cloud computing plat-
form (https://code.earthengine.google.com) to pro-
cess cloud-based remote sensing data. The main ad-
vantage is the speed of the computation, which is
outsourced to Google’s servers, as well as the avail-
ability of a variety of regularly updated datasets that
are accessible directly in the code editor. Thus, it is
possible to access the satellite data archive without
having to download the raw data. The GEE GRD
imagery includes the following steps: thermal noise
removal, radiometric calibration, terrain correction.
Therefore, only a speckle filter needs to be applied
during pre-processing.
A change detection approach was chosen, where
images before and after the flood event are compared.
This is due to the difficulties of detecting the city
of Douala, which is mainly composed of vegetation
and a dense urban area. Using the basic histogram
thresholding method, it is therefore difficult to dis-
tinguish flooded vegetation from urban flooding due
to double-bounce backscatter.40
Several supplemental datasets are used to suppress
false positives in the flood extent layer. The Eu-
ropean Commission’s Joint Research Centre Global
Surface Water dataset (’Source: EC JRC/Google’,
https://global-surface-water.appspot.com/) is used
to mask all areas covered by water for more than 10
months per year with a spatial resolution of 30 m.41
To eliminate pixels with slopes greater than 5%,
the Hydrological data and maps based on SHut-
tle Elevation Derivatives at multiple Scales (Hy-
droSHEDS) digital elevation model of 3 Arc-Seconds
was used.
3.3 Sentinel 1
Sentinel-1 is part of the space missions by the Eu-
ropean Union and carried out by the European Space
Agency (ESA) under the Copernicus program.42, 43
This program aims to establish a global, continuous,
autonomous, high quality and wide-range Earth ob-
servation capability.
The constellation of polar-orbiting Sentinel-1
satellites (Sentinel-1A and Sentinel-1B) provides con-
tinuous SAR data day and night with a revisit time of
6 days. The data provided by the Copernicus Open
Access Hub are mainly Single Look Complex (SLC)
used for interferometry and the Ground Range De-
tected (GRD).44 Sentinel-1 Level 1 GRD products
consist of focused SAR data that are multi-looked
and projected to ground range using an Earth ellip-
soid model. These data are accessible via the GEE
and were used to map a flood event in August 2020,
in Douala. Sentinel-1 in the GEE are provided in dif-
ferent polarizations, modes, passes and resolutions:45
1. Transmitter Receiver Polarization: [”VV”],
[”HH”], [”VV”, ”VH”], or [”HH”, ”HV”]
2. Instrument Mode: ”IW” (Interferometric Wide
Swath), ”EW” (Extra Wide Swath) or ”SM”
(Strip Map).
3. Orbit Properties pass: ”ASCENDING” or ”DE-
SCENDING”
4. Spatial resolution meters: 10, 25 or 40
5. GRD resolution: ”M” (medium) or ”H” (high).
The Sentinel 1 satellite acquired were single polariza-
tion data at a spatial resolution of 5 m ×20 m, a 250
km swath width of view and in VH polarization.
3.4 Twitter
Publicly available tweets are re-
trieved by using python libraries snscrape
(https://github.com/JustAnotherArchivist/snscrape).
In this study, we used two different keywords in
the query – ”Cameroon flood” and ”Cameroun
inundation” (French for ”Cameroon flood” ). For
each tweet, we extract and retain the following
information: tweeted time, content, number of
replies, number of retweets, and number of likes.
The tweets retrieved include both original tweets
and replies, but not retweets. This work reports and
discusses only aggregated statistics of the tweets.
To retrieve useful common terms and conduct sen-
timental analysis of the tweets, we need to pre-process
the content of the tweets using techniques in Natu-
ral Language Processing. Natural Language Toolkit
(NLTK) python library was used to perform the fol-
lowing steps:
1. Remove links, mentions, and hashtag
2. Splitting sentences into words and punctuation
marks, or tokenization
IAC–21–B1,5,9 Page 5 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
3. Remove stopwords such as articles, prepositions,
and punctions that does not contribute to the
meaning of the content
4. Reducing the words into a root form or lemma-
tization, i.e. convert 2nd and 3rd forms of the
verbs to the base verb
5. Remove non-alphabetical characters and keep
only words that contain three or more letters
Using processed content from the tweets, we can
determine the most common terms by using tf-idf vec-
torization. Term frequency of term t in document d
is defined as:
tf(t, d) = n/N [1]
where n is the frequency of the term t in docu-
ment d and N is the frequency of the term t in all
documents in the database of the library used. The
inverse document term frequency is given by:
tdf(t, d) = log(D
d∈D:t∈D) [2]
where D is the total number of documents and
d∈D:t∈Drepresents the number of documents in
which we find the term t. The product of term fre-
quency and inverse document term frequency is called
tf-idf. A more common term would have the tf-idf
value of closer to zero. In our analysis, tf-idf vec-
torization using a machine learning python library,
scikit-learn. Word clouds are then generated based
on tf-idf values.
To conduct sentimental analysis, we use a python
library Textblob. This library contains a trained
models that could determine the polarity and sub-
jectivity of a given text. Polarity ranges between -1
and 1 with positive values reflecting emotionally pos-
itive message and negative values reflecting emotion-
ally negative messages. Those that are neutral would
have polarity of 0. Subjectivity ranges between 0 and
1 with 1 being subjective and 0 being objective. Us-
ing both polarity and subjectivity would allow us to
evaluate the sentiments of twitter users toward flood-
ing issues in Cameroon.
4 Results and Discussions
The combination of several tools can significantly
contribute to contribute to addressing this issue in
developing and underdeveloped countries. developing
countries. One tool would be satellite imagery such
as Synthetic Aperture Radar.
4.1 Flood mapping
The flood map was obtained by processing
Sentinel-1 GRD data with the UN-SPIDER change
detection method in the GEE tool. Prior to process-
ing, four dates are provided for a collection of pre- and
post-flood images (Tab.1. From these dates, the GEE
searches for available images according to the param-
eters namely the region of interest, the polarization
and the satellite ascending pass. In our case of the
flood study of August 8th , 2020 in Douala, two im-
ages S1 of 2020-08-02 were used as the pre-flood and
S1 2020-08-26 as the post-flood (Fig. 6. We have
privileged a common DESCENDING pass to avoid
different geometric distortion due to the angle.
Table 1: Four dates defining a collection of images
before and after the flood
Start End
Before 2020-08-01 2020-08-19
After 2020-08-21 2020-09-01
Figure 6 shows the before and after Sentinel-1
imageres which were subjected to the UN-SPIDER
change detection approach for delineation of po-
tentially flooded areas in Douala. The potentially
flooded areas are displayed on higher resolution im-
agery in Figure 6. Several flooded areas are observed
along the banks of the Wouri River and along the
channels of its tributaries (Fig. 7). From the im-
agery, the streams and tributaries within the flood-
plain exhibit a predominantly dendritic drainage pat-
tern. In this pattern, there are no inner basins
(endorheic drainage basins), and the floodplain is
drained through the main drainage stem of the Wouri
River and its tributaries.
As a coastal city, Douala is also at risk of sea level
rise (SLR) which has been identified by the United
Nation’s Intergovernmental Panel on Climate Change
(IPCC) as a threat to coastal cities. This is one of the
combined factors leading to floods in Douala (Ndongo
et al., 2015). According to the IPCC in its Special
Report on the Ocean and Cryosphere in a Chang-
ing Climate (SROCC), global mean sea levels will
most likely rise between 0.95 feet (0.29m) and 3.61
feet (1.1m) by the end of the 21st century (IPCC,
2019). To portray the likely impact, the simulated
water level from the 1.1m projected SLR was overlaid
with the potentially flood areas from Sentinel-1 SAR
in Figure 5.For validation of the SAR flood extent, a
list of communities where flood incidents occurred in
20218a).
With heavy flooding (e.g., the flooding in Mbanya
IAC–21–B1,5,9 Page 6 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
Fig. 6: Sentinel-1 imagery for flood analysis through UN-SPIDER change detection approach in Douala –
(A) before, and (B) after flood event
Fig. 7: Potential flood extent delineated from Sentinel-1 SAR processing (Imagery backdrop courtesy of
ESRI Map service)
(a) (b)
Fig. 8: A) Flooding in Mbanya, a community in Douala, September 16, 2021 (Source: Field survey, 2021);
B) Channel constriction at a bridge crossing in the channel of the Wouri River
IAC–21–B1,5,9 Page 7 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
community on 16 September 2021; Figure 8a) there
is the risk of overland flow causing collateral stream
channels to emerge from some of the established trib-
utaries. However some structural and non-structural
measures have been put in place to protect the com-
munities against fluctuating water levels and periodic
flooding.
The location of Douala at the south-eastern shore
of the Wouri River estuary makes it particularly sus-
ceptible to flooding which could be exacerbated by
storm surges and sea level rise (SLR). The river chan-
nel is flanked on either side by densely populated
communities. Areas of concern are communities such
as Bonaberi and Deido which are located close to a
constriction around the Bonaberi/Wouri bridge along
the channel of the Wouri River (see Figure 8b). Such
constrictions could arise when confining margins oc-
cur on both sides of a channel at the same reach.46
Channel constrictions could instigate high velocity
river flows which can rapidly diffuse into the sur-
rounding flood plains and communities.
Analysis of Figure 10 shows that several commu-
nities where actual flooding occurred in 2021 co-
incide with the potential flood extent delineated
from Sentinel-1. Communities such as Mabanda,
Bonaloka, Akwa, Bonanjo and Bali are either situated
within or in very close proximity to the potential flood
extents along the riverbanks. Some otheir neighbour-
hoods inland are also concerned by floods from sea
level rise: b´epanda, malangu´e, cite des palmiers. This
is in line with restricted flow from tributaries existing
inland. The simulated impact zones of the projected
SLR are mainly along the banks of the Wouri River
estuary. However, it is unclear how this could change
in an extreme flood event or storm surge. More-
over, a more comprehensive analysis of the vulner-
ability of Douala to coastal flooding would involve
consideration of both offshore and nearshore hydro-
dynamic forces such as tidal currents, wave action,
winds, and ocean currents. Other city architecture,
dwellers waste management practices and meteoro-
logical factors would also enlighten the comprehen-
sion of flood mechanisms in Douala. anjo.
As a coastal city, Douala is also at risk of sea
level rise (SLR) which has been identified by the
United Nation’s Intergovernmental Panel on Climate
Change (IPCC) as a threat to coastal cities. Accord-
ing to the IPCC in its Special Report on the Ocean
and Cryosphere in a Changing Climate (SROCC),
global mean sea levels will most likely rise between
0.95 feet (0.29m) and 3.61 feet (1.1m) by the end of
the 21st century.47 To portray the likely impact, the
simulated water level from the 1.1m projected SLR
was overlaid with the potentially flood areas from
Sentinel-1 SAR in Figure 9.
The simulated impact zones of the projected SLR
are mainly along the banks of the Wouri River es-
tuary. However, it is unclear how this could change
in an extreme flood event or storm surge. Moreover,
a more comprehensive analysis of the vulnerability
of Douala to coastal flooding would involve consider-
ation of both offshore and nearshore hydrodynamic
forces such as tidal currents, wave action, winds, and
ocean currents.
Analysis of Figure 9 shows that several communi-
ties where flooding occurred frequently coincide with
the potential flood extent delineated from Sentinel-
1. Communities such as Mabanda, Bonaloka, Akwa,
Bonanjo and Bali are either situated within or in
very close proximity to the potential flood extents
along the riverbanks. The simulated impact zones
of the projected SLR are mainly along the banks of
the Wouri River estuary. However, it is unclear how
this could change in an extreme flood event or storm
surge. Moreover, a more comprehensive analysis of
the vulnerability of Douala to coastal flooding would
involve consideration of both offshore and nearshore
hydrodynamic forces such as tidal currents, wave ac-
tion, winds, and ocean currents.
4.2 Twitter analysis
Between January 1st, 2010, and September 23rd,
2021, we are able to retrieve 4285 tweets with key-
word “Cameroon flood” and 213 tweets with key-
word “Cameroun inondation” (Figure 11). Dates
with rapid increase in cumulative number of tweets
are likely related to local flooding events. The largest
outbursts for the number of tweets are in Septem-
ber 2012 and August 2015. The event in September
2012 seemed to be related to actual flood events in
Cameroon, while the event in August 2015 was re-
lated to flood events in Nigeria as a resulting of water
releases from a major dam in Cameroon.
Analyzing the interactions between tweets using
number of likes, number of retweets, and number of
replies, we found that the interactions with each tweet
vary over 3 orders of magnitudes. 95% of all tweets
have less than 5 likes, while the tweet with the largest
number of likes has 778 likes. This reflects that some
users have a much larger influence to the public that
others.
Using tf-idf vectorization, we found the follow-
ing words to be among the common words: “dam”,
“Nigeria”, “releases”, “alert”, “water”, “issues”,
IAC–21–B1,5,9 Page 8 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
Fig. 9: Overlay of flooded communities and potential flood extent (between: 2020-08-21 and 2020-09-15) from
Sentinel-1 SAR, with simulated impact of sea level rise in the Wouri River Estuary (Imagery backdrop
courtesy of ESRI Map service)
Fig. 10: Areas affected by floods between 1984 and 2018
IAC–21–B1,5,9 Page 9 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
“disaster”, “warning”, “kill”. We have also gener-
ated word clouds using the most common words for
both keyword “Cameroon flood” and “Cameroun in-
undation” (Figure 12).
Fig. 11: Time evolution of the cumulative number of
tweets retrieved using the keywords “Cameroon
flood” and “Cameroun inondation”
Fig. 12: Word clouds generated from content in the
tweets retrieved using the keywords “Cameroon
flood” (A) and “Cameroun inundation” (B)
With the processed texts, the sentimental analysis
reveals that about half of the tweets are neural in po-
larity. Among those that are not neural, 60% shows
positive polarity and 40% shows negative polarity. In
terms of subjectivity of the texts, we found an average
subjectivity of 0.22 (see Figure 13). Since this value
is closer to 0 than 1, this means that the texts are
generally more objective than subjective. Since the
content is generally more factual rather than opin-
ions, it is not unexpected that the sentimental of the
majority of tweets turns out to be closer to neutral.
We also manually analyzed the content in the
tweets in French retrieved using the keyword “Camer-
oun inundation” between February 17 th, 2018, and
May 31st, 2021. We found tweet content on 19
days that are related to flooding events in Douala,
Cameroon (see Table 2). Among these, 4 days (July
25th, 2018, August 21st, 2020, August 22nd , 2020,
August 24th, 2020) show significantly higher number
of tweets per day.
Most of the tweets on flooding events in Douala are
almost evenly distributed among jokes, alert, sensiti-
zation and information. Some tweets are complaints
(11%) and very few calls to action mentioned (see
Figure 14). Floods from July 25th, 2020 , August
21st, 2020, and August 24th, 2020 recorded the high-
est number of tweets (5 to 14 tweets/day). Analysis
of the descriptive statistics of the tweets shows that
in Cameroon, people are tweeting about floods, but
the number of tweets is still very low compared to
the statistics of tweets about other natural disasters
in developed countries (e.g., hurricanes, floods, fires).
Moreover, most of the tweets are not alerts or direct
mentions on flood management. Therefore, there is a
need to use social media more constructively so that
an increasing number of Twitter users communicate
about floods to improve flood predictability, registra-
tion, and response.
Fig. 13: Sentimental analysis of tweets with keyword
“Cameroon floods”. The tweets are classified
based on polarity as negative (polarity ¡ 0, blue),
neural (polarity = 0, orange), and positive (po-
larity ¿ 0, green)
5 Conclusion
With climate change, floods natural disasters
caused by storms and torrential rains are increas-
ingly common and affect almost all populations and
regions of the planet. In developing countries, man-
aging these large-scale events is amplifying the socio-
economic difficulties that these countries face. Flood
IAC–21–B1,5,9 Page 10 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
N°Date N°Date
1 17/02/2018 11 04/09/2019
2 04/03/2018 12 03/07/2020
3 25/07/2018 13 21/08/2020
4 26/07/2018 14 22/08/2020
5 02/11/2018 15 24/08/2020
6 30/06/2019 16 27/08/2020
7 17/07/2019 17 04/09/2020
8 07/08/2019 18 15/09/2020
9 10/08/2019 19 31/05/2021
10 23/08/2019
Table 2: Flood dates in Douala from Tweets record
from February 2018 to may 2021
Fig. 14: Narrative of Twitter message screening in
2020
maps are powerful disaster response planning tool for
the immediate concern of human life, settlements and
infrastructures. In this work, we used an open access
tool which help in assessing the history of floods and
the affected areas. In addition, we discuss the values,
challenges, and possible advances of social networks
use to leverage flood response strategies.
Twitter usage in Cameroon is still fairly limited
compared to other developed countries. However, the
data still allows us to learn about flood events that
are not formally documented and associated senti-
ments from related individuals. Further analysis to
compare twitter usage during flood events with reg-
ular scenarios would provide us with a better under-
stand of how twitter and other social media can be
adapted to assist with disaster managements, from
early warning and mitigation to response and recov-
ery.
References
[1] UNISDR CRED. The human costs of weather
related disasters. The United Nations Office for
Disaster Risk Reduction (UNISDR) https://doi.
org/10.1017/CBO9781107415324, 4, 2015.
[2] Margareta Wahlstrom and Debarati Guha-Sapir.
The human cost of weather-related disasters
1995–2015. Technical report, The Centre for Re-
search on the Epidemiology of Disasters (CRED)
& The UN Office for Disaster Risk Reduction,
Geneva, 2015.
[3] Henry Bang, Lee Miles, and Richard Gor-
don. The Irony of Flood Risks in African Dry-
land Environments: Human Security in North
Cameroon. World Journal of Engineering and
Technology, 5(3):109–121, August 2017. Num-
ber: 3.
[4] Henry Bang, Lee Miles, and Richard Gordon.
Evaluating local vulnerability and organisational
resilience to frequent flooding in Africa: the case
of Northern Cameroon. foresight, 21(2):266–284,
January 2019. Publisher: Emerald Publishing
Limited.
[5] Guoqiang Shen and Seong Nam Hwang. Spatial–
temporal snapshots of global natural disaster im-
pacts revealed from em-dat for 1900-2015. Geo-
matics, Natural Hazards and Risk, 2019.
[6] Ross Towe, Graham Dean, Liz Edwards, Vatsala
Nundloll, Gordon Blair, Rob Lamb, Barry Han-
kin, and Susan Manson. Rethinking data-driven
decision support in flood risk management for
a big data age. Journal of Flood Risk Manage-
ment, 13(4):e12652, 2020.
[7] Davide Notti, Daniele Giordan, Fabiana Cal´o,
Antonio Pepe, Francesco Zucca, and Jorge Pe-
dro Galve. Potential and Limitations of Open
Satellite Data for Flood Mapping. Remote Sens-
ing, 10(11):1673, November 2018. Number: 11
Publisher: Multidisciplinary Digital Publishing
Institute.
[8] Ujjawal K. Panchal, Hardik Ajmani, and
Saad Y. Sait. Flooding Level Classification
by Gait Analysis of Smartphone Sensor Data.
IEEE Access, 7:181678–181687, 2019. Confer-
ence Name: IEEE Access.
[9] J´erˆome Le Coz, Antoine Patalano, Daniel
Collins, Nicol´as Federico Guill´en, Carlos Marcelo
IAC–21–B1,5,9 Page 11 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
Garc´ıa, Graeme M Smart, Jochen Bind, Antoine
Chiaverini, Rapha¨el Le Boursicaud, Guillaume
Dramais, et al. Crowdsourced data for flood
hydrology: Feedback from recent citizen science
projects in argentina, france and new zealand.
Journal of Hydrology, 541:766–777, 2016.
[10] Robin Lacassin, Maud Dev`es, Stephen P
Hicks, Jean-Paul Ampuero, R´emy Bossu, Lu-
cile Bruhat, Desianto F Wibisono, Laure Fal-
lou, Eric J Fielding, Alice-Agnes Gabriel, et al.
Rapid collaborative knowledge building via twit-
ter after significant geohazard events. Geo-
science Communication, 3(1):129–146, 2020.
[11] Jyoti Prakash Singh, Yogesh K Dwivedi, Nripen-
dra P Rana, Abhinav Kumar, and Kawal-
jeet Kaur Kapoor. Event classification and lo-
cation prediction from tweets during disasters.
Annals of Operations Research, 283(1):737–757,
2019.
[12] Kathleen M Carley, Momin Malik, Peter M
Landwehr, J¨urgen Pfeffer, and Michael
Kowalchuck. Crowd sourcing disaster manage-
ment: The complex nature of twitter usage in
padang indonesia. Safety science, 90:48–61,
2016.
[13] Clarissa C David, Jonathan Corpus Ong, and
Erika Fille T Legara. Tweeting supertyphoon
haiyan: Evolving functions of twitter during and
after a disaster event. PloS one, 11(3):e0150190,
2016.
[14] Dariusz B Baranowski, Maria K Flatau, Pi-
otr J Flatau, Dwikorita Karnawati, Katarzyna
Barabasz, Michal Labuz, Beata Latos, Jerome M
Schmidt, Jaka AI Paski, et al. Social-media and
newspaper reports reveal large-scale meteorolog-
ical drivers of floods on sumatra. Nature com-
munications, 11(1):1–10, 2020.
[15] populationstat.com. douala, cameroon popula-
tion, 2021. Last accessed 16 September 2021.
[16] Laurent Bruckmann, Am´elie Amanejieu, Mau-
rice Olivier Zogning Moffo, and Pierre Ozer.
Analyse g´eohistorique de l’´evolution spatio-
temporelle du risque d’inondation et de sa ges-
tion dans la zone urbaine de douala (camer-
oun). Physio-G´eo. G´eographie physique et en-
vironnement, (Volume 13):91–113, 2019.
[17] Pamela Aka Tangan, Primus Azinwi Tam-
fuh, Alice Magha Mufur, Evine Laure Tanko
Njiosseu, Jules Nfor, Aminatou Fagny Mefire,
Dieudonn´e Bitom, et al. Community-based ap-
proach in the prevention and management of
flood disasters in babessi sub-division (ndop
plain, north west cameroon). Journal of Geo-
science and Environment Protection, 6(04):211,
2018.
[18] Daniel Sighomnou. Cameroon: integrated flood
management in river logone flood-plain. Viewsite
visited on, 15(05):2018, 2005.
[19] Robert Njilla Mengnjo Ngalim and Simbo Ter-
ence Nunyui. Stakeholders’ perception of
the triggering mechanisms and determinants of
flooding in limbe, south west region of cameroon.
Asian Journal of Geographical Research, pages
17–34, 2020.
[20] Ndongo Barthel´emy, Fonteh Mathias
Fru, Ngu Jiofack Ludovic, and Bonguen
Onouck Rolande Carole. Legislative and hy-
draulic weaknesses in the fight against floods
in cities with developing economies: Case
study of yaounde, cameroon. Centre Region of
Cameroon. Current Journal of Applied Science
and Technology, 2016.
[21] Samba Gideon, Z Mofor Gilbert, and Chi-
anebeng Japhet Kuma. The role of urban forest
in flood risks management in yaound´e vii, centre
region of cameroon. Centre Region of Cameroon.
Current Journal of Applied Science and Technol-
ogy, 39(11), 2020.
[22] Gaston Buh Wung and Festus Tongwa Aka. En-
hancing resilience against floods in the lower mo-
towoh community, limbe, southwest cameroon.
Disaster Prevention and Management: An In-
ternational Journal, 2019.
[23] Wamja Nathalie. Plastic pollution completely
blocking a waterway in the dakar district of
douala, cameroon, 2021. Last accessed 16
September 2021.
[24] meteoblue. Douala average temperatures and
precipi-tation, 2021. Last accessed 16 Septem-
ber 2021.
[25] Afrik 21. Cameroon: concern over increased
flooding, 2021. Last accessed 16 September 2021.
[26] Lisa Landuyt, Alexandra Van Wesemael, Guy
J.-P. Schumann, Renaud Hostache, Niko E. C.
Verhoest, and Frieke M. B. Van Coillie. Flood
IAC–21–B1,5,9 Page 12 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
mapping based on synthetic aperture radar: An
assessment of established approaches. IEEE
Transactions on Geoscience and Remote Sens-
ing, 57(2):722–739, Feb 2019.
[27] Yong Wang, Laura L. Hess, Solange Filoso,
and John M. Melack. Understanding the
radar backscattering from flooded and non-
flooded amazonian forests: Results from canopy
backscatter modeling. Remote Sensing of Envi-
ronment, 54(3):324–332, 1995.
[28] Francisco Carre˜no Conde and Mar´ıa
De Mata Mu˜noz. Flood monitoring based
on the study of sentinel-1 sar images: The ebro
river case study. Water, 11(12):2454, 2019.
[29] Tom G Farr. Radar interactions with geologic
surfaces. Guide to Magellan Image Interpreta-
tion, 93:45–56, 1993.
[30] Fabrizio Argenti, Alessandro Lapini, Tiziano
Bianchi, and Luciano Alparone. A tutorial on
speckle reduction in synthetic aperture radar im-
ages. IEEE Geoscience and remote sensing mag-
azine, 1(3):6–35, 2013.
[31] Joy Sanyal and Xi Xi Lu. Application of re-
mote sensing in flood management with special
reference to monsoon asia: a review. Natural
Hazards, 33(2):283–301, 2004.
[32] G Srinivasa Rao, V Brinda, P Manju Sree, and
V Bhanumurthy. Advantage of multipolarized
sar data for flood extent delineation. In Mi-
crowave Remote Sensing of the Atmosphere and
Environment V, volume 6410, page 64100Z. In-
ternational Society for Optics and Photonics,
2006.
[33] Jean-Michel Martinez and Thuy Le Toan. Map-
ping of flood dynamics and spatial distribution
of vegetation in the amazon floodplain using
multitemporal sar data. Remote sensing of En-
vironment, 108(3):209–223, 2007.
[34] Jordi Inglada and Grgoire Mercier. A new sta-
tistical similarity measure for change detection
in multitemporal sar images and its extension to
multiscale change analysis. IEEE Transactions
on Geoscience and Remote Sensing, 45(5):1432–
1445, 2007.
[35] Sandro Martinis, Andr´e Twele, and Stefan
Voigt. Towards operational near real-time flood
detection using a split-based automatic thresh-
olding procedure on high resolution terrasar-x
data. Natural Hazards and Earth System Sci-
ences, 9(2):303–314, 2009.
[36] MS Horritt, DC Mason, and AJ Luckman. Flood
boundary delineation from synthetic aperture
radar imagery using a statistical active contour
model. International Journal of Remote Sens-
ing, 22(13):2489–2507, 2001.
[37] Biswajeet Pradhan, Mahyat Shafapour Tehrany,
and Mustafa Neamah Jebur. A new semiauto-
mated detection mapping of flood extent from
terrasar-x satellite image using rule-based clas-
sification and taguchi optimization techniques.
IEEE Transactions on Geoscience and Remote
Sensing, 54(7):4331–4342, 2016.
[38] Annarita D’Addabbo, Alberto Refice, Guido
Pasquariello, Francesco P Lovergine, Domenico
Capolongo, and Salvatore Manfreda. A bayesian
network for flood detection combining sar im-
agery and ancillary data. IEEE Transactions
on Geoscience and Remote sensing, 54(6):3612–
3625, 2016.
[39] UN-SPIDER. Recommended practice: Flood
mapping and damage assessment using sentinel-
1 sar data in google earth engine, 2021. Last
accessed 16 September 2021.
[40] Ramanuja Manavalan. Review of synthetic aper-
ture radar frequency, polarization, and inci-
dence angle data for mapping the inundated
regions. Journal of Applied Remote Sensing,
12(2):021501, 2018.
[41] Jean-Fran¸cois Pekel, Andrew Cottam, Noel
Gorelick, and Alan S Belward. High-resolution
mapping of global surface water and its long-
term changes. Nature, 540(7633):418–422, 2016.
[42] Aniceto Panetti, Friedhelm Rostan, Michelan-
gelo L’Abbate, Claudio Bruno, Antonio Bauleo,
Toni Catalano, Marco Cotogni, Luigi Galvagni,
Andrea Pietropaolo, Giacomo Taini, Paolo Ven-
ditti, Markus Huchler, Ramon Torres, Svein
Lokas, David Bibby, and Dirk Geudtner. Coper-
nicus sentinel-1 satellite and c-sar instrument.
In 2014 IEEE Geoscience and Remote Sensing
Symposium, pages 1461–1464, 2014.
[43] Dirk Geudtner, Ram´on Torres, Paul Snoeij, Mal-
colm Davidson, and Bj¨orn Rommen. Sentinel-
1 system capabilities and applications. In 2014
IAC–21–B1,5,9 Page 13 of 14

72th International Astronautical Congress, Dubai, United Arab Emirates.
IEEE Geoscience and Remote Sensing Sympo-
sium, pages 1457–1460. IEEE, 2014.
[44] Federico Filipponi. Sentinel-1 grd preprocessing
workflow. In Multidisciplinary Digital Publishing
Institute Proceedings, volume 18, page 11, 2019.
[45] Google Earth Engine. Sentinel-1 algorithms,
2021. Last accessed 16 September 2021.
[46] DA Sear, DH Hornby, J Wheaton, and CT Hill.
Understanding river channel sensitivity to ge-
omorphological changes. Scientific data, 2021.
Last accessed 16 September 2021.
[47] Michael Meredith, Martin Sommerkorn, San-
dra Cassotta, Chris Derksen, A Ekaykin,
A Hollowed, Gary Kofinas, A Mackintosh,
J Melbourne-Thomas, MMC Muelbert, et al.
Polar regions. chapter 3, ipcc special report on
the ocean and cryosphere in a changing climate.
2019.
IAC–21–B1,5,9 Page 14 of 14