Flood risk assessment of slums in Dhaka city

Abstract

Slums are characterized as some of the most impoverished and vulnerable communities within urban areas. They often face a multitude of challenges, including extreme poverty, social marginalization, and inadequate infrastructure. The poor quality of buildings and the lack of adequate infrastructure further exacerbate the vulnerability of slum dwellers to the impacts of flooding. This study focuses on accurate assessment of flood risk in slum area of the Dhaka metropolitan area, the capital of Bangladesh. The man-machine interactive method was utilized to extract slum information from Sentinel-2A satellite imagery. Thirteen evaluation indicators were selected from the aspects of hazard, sensitivity, exposure, and prevention and mitigation capacity. The fuzzy analytic hierarchy process (FAHP) was employed to determine the relative weights of these indicators. A comprehensive evaluation model for flood disaster risk was established, enabling the determination of the spatial distribution of flood risk in slums of Dhaka city. The results indicate that the flood risk distribution in the slums of Dhaka city exhibits an overall high flood risk. The slums with very high flood risk constitute about 22.035% of the total area and are concentrated in Badda, Khilkhet, and Dakshinkhan districts. Slums with high-risk constitute 41.600% of the total area and are located mainly in Turag, Pallabi, Badda, and Gulshan districts. Those areas with abundant rainfall, dense river networks, and flat topography have higher flood risk in slums compared to other regions. These research findings can be a reference for regional flood risk zoning, prevention, and sustainable development planning.

Keywords:

• Dhaka city
• slums
• fuzzy analytic hierarchy process
• flood risk assessment

1. Introduction

Rapid urbanization and population growth pose significant challenges to global development in the twenty-first century (Lintelo et al. 2018). In low- and middle-income countries, urbanization often occurs alongside weak health and social infrastructure, overcrowding, and inadequate housing, leading to the rapid emergence and persistence of informal settlements in cities (Guan et al. 2011). One of the major consequences of rapid urbanization is the proliferation of informal settlements, commonly known as squatter settlements or slums (Risi et al. 2013). Slums are characterized by vulnerable housing, high population density, substandard construction quality, and a lack of social infrastructure, all of which contribute to their high vulnerability to flood hazards (Gram-Hansen et al. 2019; Tom et al. 2022).

Dhaka, the capital of Bangladesh, is one of the fastest-growing mega-cities in the world, with an annual growth rate of 4.4% and a population of 14 million (UN-HABITAT 2009). As a result of rapid urbanization, many rural poor migrate to Dhaka city, where the existing housing infrastructure is unable to meet the demands of the influx. Consequently, a significant portion of the impoverished and low-income population turns to slums and informal settlements as a means of seeking shelter, leading to the prolonged existence of slums in Dhaka (Latif et al. 2016; Rahman et al. 2020; Soma et al. 2021). In Dhaka, most of the impoverished residents live in areas with poor land quality, such as slopes prone to landslides or collapses, embankments susceptible to flooding, and densely populated areas highly vulnerable to the impacts of climate change (Khan 2010). This exacerbates the severity of flood disasters in the city. It is estimated that around 7600 households reside in slums located within 50 meters of rivers, making these areas particularly susceptible to frequent flooding (The World Bank 2007).

Flooding is one of the most frequent, wide-ranging, and devastating natural disasters worldwide (Fan et al. 2012), can result in property damage, loss of livelihoods, economic chaos, and loss of critical infrastructure, leading to casualties (Tom et al. 2022), the potential flood losses in urban areas exhibit an exponential increase with the process of urbanization (Moel et al. 2011). The adverse impacts of floods are particularly pronounced on slum dwellers who have limited access to basic material infrastructure and social protection. Flooding is influenced by various factors, including rainfall, topography, and river water levels. Its occurrence and evolution are characterized by uncertainty and complexity, making it nearly impossible to prevent (Pappenberger et al. 2006; Siam et al. 2022). Therefore, conducting rapid and accurate flood risk assessments is crucial for devising appropriate risk management plans and mitigating the adverse impacts of floods (Mojaddadi et al. 2017).

Researchers have conducted extensive studies on urban flood disasters, employing common methods such as historical data statistics, scenario simulation, and index system analysis. Yang et al. (2015) assessed the vulnerability of the Pearl River Delta urban cluster to flood disasters using data on historical flood events and disaster severity. Sheng et al. (2020) analyzed the spatiotemporal characteristics and flood risk in the Xiong’an New Area based on a historical flood dataset. Li (2016) established flood risk zoning by simulating flood disasters in Zhengzhou City and utilized the Storm Water Management Model (SWMM) to study rainfall-runoff processes. Peng et al. (2018) analyzed flood risk in the Mazhou River Basin in Shenzhen using the CLUE-S model, SCS model, and equal-volume submergence algorithm. Huang et al. (2019) studied the risk in the Donghaochong Basin in Guangzhou under different design rainfall scenarios using the ICM model. However, the mathematical models used in scenario simulation methods require detailed data support, which may be limited by data availability in the research area and lacks universal applicability. Hategekimana et al. (2018) utilized geographic information system (GIS) and fuzzy analytic hierarchy process (FAHP) to establish an index system and analyze the identification of flood-prone areas. Cheng et al. (2019) used the index system method, integrating natural and socio-economic factors, to assess the level of flood risk in Wu’an city, Hebei province. Lei et al. (2021) employed deep learning techniques to construct a flood risk assessment system and evaluate the flood risk in Seoul, South Korea. Generally, the index system method, which combines qualitative and quantitative evaluations, has demonstrated effective performance and is widely applied in flood disaster assessments.

In this study, the GEE cloud computing platform and the Random Forest Classification method, coupling the manual interpretation, were employed for extracting the slum with Sentinel-2A satellite imagery in Dhaka metropolitan area. A flood risk assessment model was developed using a fuzzy analytic hierarchy process to conduct risk assessment in Dhaka metropolitan and its slum area, based on hazard, sensitivity, exposure, prevention and mitigation ability analysis. This research can assist in identifying the most vulnerable areas and populations in slum area, providing valuable scientific references for the development of early warning systems, infrastructure improvements, and flood risk mitigation in Dhaka city.

2. Materials and methods

This study was conducted in three steps. First, slums were preliminarily extracted using Sentinel-2A data and the Random Forest Classification method on the GEE platform. Manual interpretation corrections were then applied, incorporating data such as Points of Interest and rooftop information, to obtain a more reliable slum area. Second, a flood risk assessment model was developed to evaluate the flood hazards, sensitivity, vulnerability, and prevention and mitigation capacity of slums. FAHP was employed for this evaluation. Finally, by integrating the four evaluation results obtained in the previous step, a flood risk assessment of slums was conducted, and a flood risk map was generated. Figure 1 illustrates the overall workflow of the entire study.

Figure 1. Flowchart of this study.

2.1. Study area

Dhaka City is the capital of Bangladesh. Serving as the political, economic, and cultural center of the country, Dhaka is the largest city in Bangladesh and ranks as the 11th largest megacity globally. It has experienced significant development in recent years, making it the fastest-growing large city (Hossain 2013). With a population of approximately 14.54 million, Dhaka faces the challenge of high population density. The Dhaka metropolitan area is also home to around 3.42 million slum dwellers (Islam et al. 2006).

Dhaka is located in the north bank of the Buriganga River in the Ganges delta and is surrounded by multiple rivers (Dewan and Corner 2014) (Figure 2). It is bordered by the Buriganga River to the south, the Turag River and Balu River to the north and northeast, and Tongi and Baru mark the northeastern boundary. The altitude of Dhaka ranges from about 0 to 23 m above sea level, with an average elevation of 7 m. The average annual rainfall is about 1854 mm, of which 80% occurs from May to September. And, Dhaka experiences frequent floods during the rainy season period. It has been documented that the city of Dhaka suffered from the worst and most devastating floods in 1988, 1998, 2004, and 2007, and the occurrence of these extreme rainfall events resulted in the inundation of more than two-thirds of the city, with the depth of water ranging from 0.3 to 4.5 m (Ahasan et al. 2011; Sayed and Haruyama 2017).

Figure 2. Location and administrative map of Dhaka city.

2.2. Slum information extraction

2.2.1. Creating training samples

Based on the manual interpretation, 500 slum area polygon samples were collected from the GEE platform. Among them, 400 samples (80%) were allocated as training samples for training the random forest classifier. The remaining 100 samples (20%) were designated as validation samples for assessing classification accuracy.

2.2.2. Slum automatic extraction

The 2020 sentinel-2A remote sensing images were directly called from the GEE public dataset using date filtering, and the spectral feature indices were extracted after de-clouded processing. A few inputs were used for training, which includes six spectral bands (B2, B3, B4, B8, B11, and B12 in Sentinel 2A), and five major spectral indices including the normalized vegetation index (NDVI) (Rudiastuti et al. 2021), the normalized difference water body index (NDWI) (Zurqani et al. 2019), the normalized building index (NDBI) (Rudiastuti et al. 2021), Enhanced Vegetation Index (EVI) (Huete 2012) and Bare Soil Index (BSI) (Abebe et al. 2019). The formulae were calculated as follows: (1) $$\begin{array}{c}\mathit{\text{NDVI}}=\frac{B8-B4}{B8+B4} \end{array}$$(1) (2) $$\begin{array}{c}\mathit{\text{NDWI}}=\frac{B3-B8}{B3+B8}\end{array}$$(2) (3) $$\begin{array}{c}\mathit{\text{NDBI}}=\frac{B11-B8}{B11+B8}\end{array}$$(3) (4) $$\begin{array}{c}\mathit{\text{EVI}}=2.5\times \frac{B8-B4}{B8+6\times B4-7.5\times B2+1}\end{array}$$(4) (5) $$\begin{array}{c}\mathit{\text{BSI}}=\frac{\left(B12+B4\right)-\left(B8+B2\right)}{\left(B12+B4\right)+\left(B8+B2\right)}\end{array}$$(5)

The random forest classifier was chosen for extracting slums in this experiment because it can handle urban classification tasks involving high-dimensional feature spaces and can map informal settlements in complex environments (Wurm et al. 2017). The entire classification process was conducted in the GEE platform, with the construction and tuning of the classification model implemented using the “ee. Classifier” package in code. The overall accuracy (OA) and Kappa coefficient can be calculated from the confusion matrix. OA represents the proportion of correctly classified samples to the total number of samples, while the Kappa coefficient indicates the similarity between the predicted and true classifications.

2.2.3. Manual correction of the spatial distribution of slums

Due to the classification accuracy of slums extracted using random forest being 81%, which did not meet the desired classification performance, this study utilized high-resolution remote sensing imagery to improve the accuracy of slum classification. We obtained slum maps for the years 2006 and 2010 provided by Gruebner et al. (2014), and undergone detailed manual correction with high resolution of remote sensing images from 91 Map Assistant (Google Earth version) for the years 2006, 2010, and 2020 (See Table 1 for details).

Table 1. Information on remote sensing imagery from three periods.

Year	Projection coordinates	Spatial resolution	Date of image capture
2006	CGCS2000_3_degree_Gauss_Kruger_zone_30	2.28 m	2006/12/25
2010		1.14 m	2010/03/23
2020		1.14 m	2020/12/31

First, we selected an appropriate number of control points in each year’s imagery and used the georeferencing tool in ArcGIS to sequentially rectify the images. Subsequently, we transformed the rectified imagery’s projection coordinate system to WGS_1984_UTM_Zone_46N to provide more accurate base map data for the slum calibration. Additionally, we obtained POI and roof data from OpenStreetMap and used these datasets to correct errors in the slum mapping. The specific approach involved using Google imagery as the base map and overlaying the slum data to assess the accuracy of locations or boundaries, making necessary adjustments accordingly. We filtered out POI types such as bookstores, cinemas, bars, schools, and pharmacies that do not belong to slums and extracted roof data that overlapped with these POI points. Finally, we excluded these roof data from the slum data to obtain more precise slum boundary extents and location information.

2.3 Slum area changes analysis

The land use transfer matrix can describe the source and direction of change among land classes, calculate the direction and amount of mutual transfer between different land use types, and reflect the total land use change trend in the region by describing the direction and degree of evolution of various land use types in a certain period (Wang et al. 2021). The matrix adopts an n-order matrix structure, which can not only visualize the area information of each land type at the beginning and end of the study, but also reflect the richer dynamic changes of the area transfer out of each land type at the beginning as well as the area transfer in of each land type at the end of the study (Zhu and Li 2003). The expression is: (6) $$\begin{array}{c}{S}_{\mathit{\text{ij}}}=\left[\begin{array}{ccc}{S}_{11}& \cdots & S_{1n}\\ ⋮& \ddots & ⋮\\ S_{m1}& \cdots & S_{\mathit{\text{mn}}}\end{array}\right]\end{array}$$(6) where: S is the total land area; S_ij denotes the area of land type i before the transfer converted to land type j after the transfer; n is the number of types of land use; i,j (i,j = 1,2,…, n) denote the types of land use at the beginning and the end of the study period, respectively, and if and when i = j, it denotes the area of land where the transfer of the land type did not occur.

2.4. Flood risk assessment models

2.4.1. Establishment of an evaluation indicator system

When selecting indicators for flood risk assessment in slums in Dhaka city, it is crucial to consider the availability, objectivity, and representativeness of data for each indicator (Liu and Wang 2022). In this study, 13 indicator factors were chosen based on the principles of data reliability, objectivity, accessibility, and ease of spatial processing. These indicators were selected from four guideline layers: hazard, sensitivity, exposure, and prevention and mitigation capacity. The establishment of the flood risk assessment indicator system considered the specific conditions of the study area and the focus of the research (Wu et al. 2017). Figure 3 provides a visual representation of the selected indicators in the study.

Figure 3. Indicator system for flood risk assessment of slums in Dhaka city.

2.4.2. Indicator data sources and pre-processing

1. Hazard indicators

   The frequency of inundation data has a significant impact on the accuracy of risk assessment results. Generally, the more frequent the inundation, the higher the flood risk. Therefore, in this study, the frequency of inundation is considered an important indicator for flood hazard assessment.

   For this experiment, Sentinel-1 radar images for the monsoon period (June to September) of each year from 2017 to 2021 were selected. Two images from each year, representing pre and post-flood conditions, were chosen as the data source using the Google Earth Engine platform. The data was first processed using Lee filtering to remove noise interference. Then, the difference between the pre and post-flood images was calculated. The JRC (Global Surface Water) dataset was imported to optimize the initial flood extent using permanent water bodies and remove areas with a slope greater than 5 degrees. This process resulted in obtaining the flood inundation extent for multiple periods during the monsoon season in Dhaka city. Using GIS software, the flood inundation frequency data for these ten periods from 2017 to 2021 was processed, resulting in the assessment of the floodwater inundation extent during the monsoon season in Dhaka city.

   Rainfall is an important hydrological factor in flood mapping (Waldmann et al. 2020). Daily rainfall data for eight meteorological stations in Dhaka city and its vicinity were downloaded from the National Centres for Environmental Information (NCEI) website (https://www.ncei.noaa.gov/maps/daily/). MATLAB software was used to calculate the c, consecutive rainy days, and historical peak rainfall for six years (2015–2020). Kriging interpolation was then applied to obtain the three rainfall indicators for Dhaka city.

2. Sensitivity indicators

   The elevation data used in this study is the global Digital Elevation Model (DEM) provided by the European Space Agency’s (ESA) Copernicus program. This dataset spatial resolution is 30 meters. After cropping and mosaicking the Copernicus 30m DEM data, the slope was calculated using the surface analysis tools in GIS. The fraction of vegetation cover (FVC) was derived from Landsat8 images in 2020 using ENVI software (https://www.gscloud.cn/). Radiometric calibration and atmospheric correction were performed on the images to improve the accuracy of vegetation index calculations. The normalized difference vegetation index (NDVI) was computed based on the red and near-infrared bands, and then the FVC was further calculated. The distribution of river networks is an important factor in the formation of flood disasters, and the proximity to rivers and lakes increases the sensitivity of the disaster-prone environment (Liu and Wang 2022). The river network data was obtained from OpenStreetMap, and the density analysis tool in GIS was used to calculate the river network density.

3. Exposure indicators

   Population density, road distribution, human activities are important indicators for assessing exposure of disaster risk (Pham et al. 2021). The population density data for the year 2020 was obtained from the WorldPop website (https://hub.worldpop.org/). Road network data was downloaded from OpenStreetMap, and the density analysis tool in GIS was used to calculate the road network density. Land Use and Land Cover (LULC) play a crucial role for indicating human activities (Adnan et al. 2020). The LULC data for the year 2020 was obtained from the global land cover classification dataset provided by the European Space Agency (ESA), it was generated based on satellite remote sensing images from Sentinel-1 and Sentinel-2.

4. Prevention and mitigation capacity indicators

   The brightness and distribution of nighttime lights are often correlated with the scale and level of economic activity, making it an indirect economic indicator. Due to the small size of the study area, GDP distribution data with a 1 km resolution is too coarse to provide sufficient detail and accuracy. Therefore, we choose to use nighttime light data as an indirect representation of GDP. Nighttime light data can be obtained from the Luojia-1 Nighttime Light Remote Sensing Satellite Monitoring website (http://59.175.109.173:8888/app/login.html), with a resolution of 130 m on November 3, 2018 and preprocessed by cropping, mosaicking, and masking. Finally, use a raster calculator to convert the radiance to brightness. Hospital data can be extracted based on the name category of building vector data. Building distribution data can be obtained from the GEOFABRIK website. The Euclidean distance algorithm can be used to calculate the distance from each location to the nearest hospital.

2.4.3. Fuzzy analytic hierarchy process

The fuzzy analytic hierarchy process (FAHP) introduces the concept of fuzzy sets into mathematical models. This method converts the qualitative analysis of relevant indicators involved in decision-making into quantitative analysis, overcoming the drawback of the analytic hierarchy process (AHP) that requires multiple adjustments of data to determine matrix consistency. The FAHP significantly reduces the deviation in indicator scores caused by subjective preferences, making the analytical results more scientifically and logically sound.

Based on the membership relationships between the criterion layer and the indicator layer, let’s assume that the weights of the indicators are denoted as $w_i$ (i = 1,2,3,…,n), and the weights of the criteria are denoted as $w_{\mathit{\text{ij}}}$ (j = 1,2,3,…,m). By using Equation (7)(7) $$\begin{array}{c}{W}_n=W_i\times W_{\mathit{\text{ij}}}\end{array}$$(7) , we can calculate the combined weights $w_n$ of the risk factors in the objective layer. (7) $$\begin{array}{c}{W}_n=W_i\times W_{\mathit{\text{ij}}}\end{array}$$(7)

The project leader of this study invited experts from meteorology, sociology and geography neighbourhoods to assess the importance of 13 factors related to flood risk assessment, and finally obtained the weight values of each indicator in the guideline and indicator layers based on the scores of the experts (Table 2).

Table 2. Flood risk indicator weights.

Target layer	Criterion layer	Weight1	Indicator layer	Indicator properties	Weight2	Comprehensive weight
Flood risk assessment	Hazard	0.467	Flood inundation frequency	Positive	0.308	0.144
			Average annual rainfall	Positive	0.250	0.117
			Consecutive rainy days	Positive	0.213	0.099
			Historical peak rainfall	Positive	0.229	0.107
	Sensitivity	0.246	Elevation	Negative	0.263	0.075
			Slope	Negative	0.254	0.064
			FVC	Negative	0.179	0.062
			River network density	Positive	0.304	0.075
	Exposure	0.223	Population density	Positive	0.425	0.095
			Road network density	Positive	0.308	0.069
			LULC	Positive	0.267	0.059
	Prevention and mitigation capacity	0.064	Distance to hospital	Negative	0.375	0.025
			Night light brightness	Negative	0.625	0.040

To facilitate comparison among different indicators, which have different units, it is necessary to standardize the indicators using the range transformation method. This will ensure that the indicator data is normalized within the range of 0 to 1. The projection coordinate system for the data is WGS_1984_UTM_Zone_46N, and the cell size of each grid layer is 30 m × 30m.

The normalized formula for the positive indicator that contributes to the results is: (8) $$\begin{array}{c}X=\frac{x_i-x_{\mathrm{\text{min}}}}{x_{\mathrm{\text{max}}}-x_{\mathrm{\text{min}}}}\end{array}$$(8)

The normalized formula for the negative indicator that has a depressing effect on the results is: (9) $$\begin{array}{c}X=\frac{x_{\mathrm{\text{max}}}-x_i}{x_{\mathrm{\text{max}}}-x_{\mathrm{\text{min}}}}\end{array}$$(9) where $x_i$ represents the original data of the indicator, and X represents the standardized and processed data.

The evaluation of flood risk factors, sensitivity of the environment to disasters, exposure of vulnerable elements, and prevention and mitigation capacity adopts a weighted comprehensive evaluation method. The calculation formula is as follows: (10) $$\begin{array}{c}G=\sum _{i=1}^{i=n}\mathit{\text{Xi}}\times {\mathit{\text{Weight}}}_{\mathit{\text{Xi}}}\end{array}$$(10) where n is the number of evaluation indicators, G is the comprehensive index of each evaluation, $\mathit{\text{Xi}}$ is the standardized value of each indicator, and ${\mathit{\text{Weight}}}_{\mathit{\text{Xi}}}$ represents the comprehensive weight of each indicator.

Based on the constructed indicator system, we can integrate and combine the evaluations of flood hazard, sensitivity, exposure, and prevention and mitigation capacity to establish a risk assessment model for flood disasters. The expression for this model is as follows: (11) $$\begin{array}{c}\mathit{\text{FDRI}}=f\left(\mathit{\text{VH}},\mathit{\text{VE}},\mathit{\text{VS}},\mathit{\text{VR}}\right)\end{array}$$(11) where FDRI is the final result of the flood risk assessment, $f$ is the power exponent model, and VH, VE, VS, and VR represent the evaluation indices of flood hazard, sensitivity, exposure, and prevention and mitigation capacity, respectively.

3. Results and analysis

3.1. Spatial distribution of slum areas

By extracting and correcting the spatial data on slums for the three time periods 2006, 2010, and 2020, the final spatial distribution of slums (Figure 4) was produced.

Figure 4. Spatial distribution of slums in Dhaka city over three periods of time.

Slums in Dhaka were primarily concentrated in the districts of Badda, Lalbagh, Pal-labi, Dakshinkhan, and Khilkhet in 2006, as shown in Figures 4 and 5, but in 2010 and 2020, they are primarily concentrated in the districts of Badda, Lalbagh, Dakshinkhan, Khilkhet, and Kafrul. The similarities in the distribution of slums in the three periods are that the slums are mainly located in the built-up areas of the city, with a small number of fragmented slums along the river in the eastern suburbs adjacent to the city boundary of Dhaka; some of the slums in Gendaria, Shyampur, and Kadamtali in southern Dhaka are located along the railroad line; the largest per-centage of slum area is found in the Badda area of eastern Dhaka, and the areas with the highest density of slums are in Kamrangir Char and Lalbagh in southwestern Dhaka.

Figure 5. Statistical map of the area of each upazila slum in Dhaka city for three periods of time.

According to Table 3, the number of slum patches and the overall area of slums grew and subsequently dropped between 2006 and 2020. Compared to 2006, there were 65 patches and 2.056 km² more slums overall than that in 2010. By contrast, there were 89 patches and 7.140 km² fewer slums overall in 2020 than that in 2010.

Table 3. The number of slum patches and total slum area in Dhaka city.

Year	The number of slum patches (nos.)	Total area of slums (km^2)
2006	1440	22.810
2010	1505	24.866
2020	1416	17.726

Kamrangir Char and Lalbagh are the two areas with the highest density of slums in Dhaka city. While in 2006 Kamrangir Char and Lalbagh had a concentration of contiguous slums with each slum polygon relatively intact, by 2010 the slum polygons in these two areas were gradually fragmenting, and by 2020 the fragmentation of the slums was very pronounced (Figure 6). This is because with the increasing rate of urbanization, some buildings are gradually added to the city, and it is obvious in the high-resolution remote sensing imagery that the floors of these buildings are higher and cast a larger shadow area on the ground, combined with the basic characteristics of the slums, the POI data and the roofing data can be judged to be part of the taller buildings do not belong to the slums, and therefore these taller buildings are deducted from the experiment, so during 2006 and 2020 years, the slums in Kamrangir Char and Lalbagh show an increasing trend of fragmentation.

Figure 6. Spatial distribution of slums in Kamrangir Char and Lalbagh in Dhaka city during three periods of time.

3.2. Analysis of spatial and temporal changes in slums

Based on the slum data of the three periods, their land use transfer matrix table of Dhaka city are shown in Tables 4 and 5, Figure 7.

Figure 7. Map of land use change patterns in Dhaka city. (a) Distribution of land use conversions during 2006–2010 and (b) distribution of land use conversions during 2010–2020.

Table 4. Dhaka City land use transfer matrix 2006–2010 (km²).

Year	2010
	Land use class	Other	Slum	Total
2006	Other	274.610	3.539	278.149
	Slum	1.483	21.327	22.810
	Total	276.093	24.866	300.959

Table 5. Dhaka City land use transfer matrix 2010–2020 (km²).

Year	2020
	Land use class	Other	Slum	Total
2010	Other	273.294	2.799	276.093
	Slum	9.940	14.927	24.866
	Total	283.234	17.726	300.959

It can be seen from Figure 7 that the overall spatial distribution of areas with large concentrations of slums has not changed much over the period 2006–2020, with the total area showing a slight increase followed by a gradual decrease. The relatively obvious increase of slums in 2006–2010 was distributed in the districts of Turag, Dakshinkhan, and Badda; The highest reduction in slum area from 2010 to 2020 is concentrated in Badda, Pallabi, and Kafrul districts.

During 2006–2010, the area of Dhaka city converted from other land to slum land is 3.539 km² while the area of slums converted to other land is 1.483 km² (seen in Table 4) and the net increase in slums is 9.014%. The plots of other land converted to slums in Dhaka city were mainly concentrated in Turag, Dakshinkhan, Badda, Khilkhet, Uttakhan, Khilgaon, Pallabi, Darus Salam; and the plots of slums converted to other land were mainly located in Mohammadpur, Adabor, Tejgaon Industrial Area, Kafrul, Pallabi, Badda, Shyampur, Gulshan (Figure 7). From Table 6, it can be seen that the incidence of poverty/extreme poverty in both Badda and Pallabi showed a decreasing trend in 2010 as compared to 2005, but the population of these two districts increased more in 2011 as compared to 2001 (The World Bank 2005, 2010; BBS, 2011). The population of Khilgaon district was 336895 in 2001, which decreased to 327717 in 2010, and the incidence of poverty by 2010 the incidence of poverty has increased by 2.1% compared to 2005 (The World Bank 2005, 2010; BBS, 2011), so there is a relative increase in the area of slums in three districts of Badda, Khilgaon, and Pallabi. Except for two districts of Adabor, Tejgaon Industrial Area (no relevant statistics), the incidence of poverty/extreme poverty in the remaining six districts in 2010 showed a decreasing trend compared to 2005, and Mohammadpur, Shyampur showed a significant decrease in the population in 2011 as a result compared to 2001. As a result, eight districts of Dhaka city such as Turag, Dakshinkhan, and Badda have shown a relative reduction in the area of slums during 2006–2010.

Table 6. Key regional data on the transfer of slums to other land uses.

Transform	Upazila/Thana	Population by Upazila/Thana by Residence		Incidence of Poverty 2005		Final Upazila level povmap estimates 2010
		2001	2011	% Poor (Upper poverty line)	% Extreme Poor (cc)	% Poor (Upper poverty line)	% Extreme Poor (lower poverty line)
other-slum	Turag	–	157316	–	–	25.1	12.5
	Dakshinkhan	–	255931	–	–	24.6	11.8
	Badda	359256	536621	18	6.4	13.4	1.2
	Khilkhet	–	130053	–	–	14.7	1.8
	Uttar khan	–	78933	–	–	24.9	4.3
	Khilgaon	336895	327717	11.6	4.1	13.7	1.5
	Pallabi	431257	596835	12	4.2	12	1.8
	Darus Salam	–	159139	–	–	14.2	2.1
slum-other	Mohammadpur	456058	355843	11.7	3.9	4	0.3
	Adabor	–	203989	–	–	12.5	1.4
	Tejgaon Industrial Area	–	146732	–	–	6.7	0.8
	Kafrul	289986	396182	13.1	4.7	7	0.4
	Pallabi	431257	596835	12	4.2	12	1.8
	Badda	359256	536621	18	6.4	13.4	1.2
	Shyampur	376545	184062	14.9	5	12.9	1.2
	Gulshan	190590	253050	8.6	2.9	3.3	0.8

Note: The statistical data in the above table is sourced from: population:(BBS, 2011); Incidence of Poverty: (The World Bank 2005); Final Upazila level povmap estimates: (The World Bank 2010).

Table 5 results show that the conversion of other lands to slum land in Dhaka city is 2.799 km² while the conversion of slums to other land is 9.940 km² from 2010 to 2020, which shows a net reduction of slums by 28.714%.

During the period 2010–2020, the plots of other land converted to slums were mainly located in Turag, Lalbagh, Hazaribagh, Pallabi, Mohammadpur, Kafrul, Kamrangir Char, Gulshan; and the plots of slums converted to other land are mainly located in Pallabi, Badda, Turag, Kafrul, Khilgaon, Lalbagh, Mirpur, Darus Salam (Figure 7). According to City Mayors Statistics, the GDP of Dhaka in 2005 was $52 billion and growing annually at 6.1% (Murad et al. 2021), the projected GDP of Dhaka in the year 2020 is $126 billion (District Statistics 2011). In 2010, the Poverty Gap (PG) estimated using the upper poverty line in Dhaka was 6.2%, and the Squared Poverty Gap (SPG) was 1.8%. By 2016, these figures had decreased to 3.2% and 0.9% respectively (HIES 2010). Similarly, in 2010, the Poverty Gap (PG) estimated using the lower poverty line was 2.7%, and the Squared Poverty Gap (SPG) was 0.7%. By 2016, these figures had decreased to 1.2% and 0.3% respectively (HIES 2016). This indicates a reduction in the severity of poverty between 2010 and 2016. With the continuous development of Dhaka’s economy and education levels, the poverty level further declined between 2016 and 2020. Therefore, from 2010 to 2020, there has been a significant reduction in the area of slums in Dhaka City. Most of the slum areas have been converted into other land uses, consistent with the findings shown in Figure 8.

Figure 8. Spatial distribution of flooding disaster hazard in Dhaka city. (a) Flood inundation frequency, (b) average annual rainfall, (c) consecutive rainy days, (d) historical peak rainfall, (e) flood hazard, and (f) flood hazards in slums

3.3. Flood risk assessment in slums

Based on GIS technology, the weighted calculation of flood risk decision values for different grid cells is performed by overlaying standardized data layers of various indicators using indicator weights. Referring to the five-point scale for disaster risk classification (Zhang et al. 2020) used in previous studies, a natural break method is employed to classify the evaluation results of the criterion layer and the target layer into five levels: very low risk, low risk, moderate risk, high risk, and very high risk. Through grid classification rendering, a zoning map of flood hazard, sensitivity, exposure, prevention and mitigation capacity, and flood risk levels in Dhaka City (Figures 8–12) is generated. Finally, the assessment of flood risk in Dhaka City’s slums is conducted by overlaying the evaluation results with the slums data from 2020.

Figure 9. Spatial distribution of flooding disaster sensitivity in Dhaka city. (a) Elevation, (b) slope, (c) FVC, (d) river network density, (e) flood sensitivity, and (f) flood sensitivity in slums

Figure 10. Spatial distribution of flooding disaster exposure in Dhaka city. (a) Population density, (b) road network density, (c) LULC, (d) flood exposure, and (e) flood exposure in slums

Figure 11. Spatial distribution of flooding disaster prevention and mitigation capacity in Dhaka city. (a) Distance to hospital, (b) night light brightness, (c) prevention and mitigation capacity, and (d) flood prevention and mitigation capacity in slums.

Figure 12. Spatial distribution of flood risk in Dhaka city. (a) Flood risk and (b) flood risk in slums

3.3.1. Flood hazard assessment

Hazard indicators include rainfall and several flood inundations. 8 shows the distribution of the four evaluation indicators and Flood hazard, Flood hazards in slums. The eastern part of Dhaka adjacent to the border suffers from a higher number of flood inundation (Figure 8a), the northeast has higher average annual rainfall and historical peak rainfall (Figure 8b, d), and the southwest has more consecutive days of rainfall (Figure 8c). The flood hazard of Dhaka city increases from southwest to northeast, with the number of inundations contributing the most to the flood hazard (Figure 8e). Slums at low risk are concentrated in Kamrangir Char, Lalbagh, and Hazaribagh, and most of the slums in Dakshinkhan, Khilkhet, and Uttar Khan have high flood risk (Figure 8f).

3.3.2. Flood sensitivity assessment

The sensitivity index reflects the topography, vegetation, and river characteristics of the study area. Figure 9 illustrates the spatial distribution of the four evaluation indicators and the flood sensitivity and flood sensitivity in slums. Since elevation, slope, NDVI, and FVC all have a suppressive effect on flood risk, negative normalization is applied to these indicators. Figure 9a–c displays the normalized results. In this study, a river network density map is created using the main river channels in Dhaka city (Figure 9d). The results show that the high flood sensitivity zones are mainly located in the northwest, southwest, east, and southeast areas of Dhaka (Figure 9e). Slums above the high flood sensitivity level are concentrated in Kamrangir Char, Lalbagh, Hazaribagh, and Adabor. Slums in the medium sensitivity zone have a distinct linear distribution extending from north to south (Figure 9f).

3.3.3. Flood exposure assessment

The flood exposure index includes population density, road network density, and land use types (Figure 10a–c). From Figure 10, it can be observed that areas with higher population density and road network density have more frequent human activities and thus higher flood exposure. In the event of flooding, the extent and chance of damage to crops and built-up areas are relatively higher and the economic and social losses will be higher, so these areas have high exposure. The urban areas of Dhaka city are basically in high or very high exposure and the suburban areas are basically in medium or low exposure except for crop growing areas (Figure 10d). The overall exposure of slums in Dhaka city is on the high side, with slums in Badda, Khilgaon, and Shyampur having very high exposure, high exposure slums are concentrated in Kamrangir Char, Lalbagh, and Dakshinkhan, and low exposure slums are found in the eastern suburbs (Figure 10e).

3.3.4. Prevention and mitigation capacity assessment

Prevention and mitigation capacity refers to the comprehensive ability of a region to cope with the damage caused by floods when floods occur and is generally related to the level of regional economic development. Since both indicators Distance to hospital and Night light brightness are negatively correlated with the prevention and mitigation capacity of the disaster-bearing body, it is necessary to normalize the negative indicators, and Figure 11a, b are the results after normalization. The prevention and mitigation capacity of Dhaka city decreases gradually from west to east (Figure 11c) the more hospitals and the brighter the lights, the stronger the prevention and mitigation capacity of the western area, so the urban area is significantly more resilient than the suburban area. Most of the districts have slums with high or above disaster preparedness and mitigation capacity, slums in Badda, Kadamtali, Turag, and Kafrul have high disaster preparedness and mitigation capacity, whereas the slums with high capacity are mainly located in Kamrangir Char, Lalbagh, and Dakshinkhan, and the slums in the eastern districts are in low disaster preparedness and mitigation capacity (Figure 11d).

3.3.5. Flood risk assessment

Flood risk is calculated by combining four influencing factors: hazard intensity of causative factors, sensitivity of the environment to disaster occurrence, exposure of the affected population, and prevention and mitigation capacity of the region. Figure 12 displays the comprehensive flood risk in Dhaka city (Figure 12a) and the distribution of flood risk in slums (Figure 12b). The evaluation results indicate that the overall distribution of flood risk in the study area shows a gradual increase from low-risk areas in the west to high-risk areas in the east. At the district scale, the very high-risk areas of slums in Dhaka city account for 22.035% of the total area and are mainly concentrated in Badda, Khilkhet, and Dakshinkhan districts, which are densely populated, receive abundant rainfall, are subjected to a high number of historical floods, and have a dense network of rivers and a flat topography, which makes them more vulnerable to floods; slums belonging to the high-risk category account for 41.600% of the total area, mainly in Turag, Pallabi, Badda, and Gulshan, where the slums are located in these areas with flat topography and low vegetation cover, which makes them vulnerable to flood hazards; medium-risk slums accounted for 27.342% of the total area, which is concentrated in the southwestern part of Dhaka city in the areas of Kamrangir Char, Lalbagh and Hazaribagh area in south-western Dhaka city, while slums in low-risk and very low-risk zones account for 6.924% and 2.100% of the total area, respectively (Table 7).

Table 7. A Statistical summary of the slum area in different flood risk levels in Dhaka city.

Risk level	Area (km^2)	Proportion (%)
Very low	0.372	2.100
Low	1.227	6.924
Moderate	4.847	27.342
High	7.374	41.600
Very high	3.906	22.035

4. Conclusion and discussion

The slums of Dhaka city exhibit distinct variations in terms of flood hazard, sensitivity, exposure, and prevention and mitigation capacity across spatial areas.The risk of slum floods increases from the southwest to the northeast, with slums east of the diagonal essentially in the moderately high-risk zone and west of the diagonal in the northwest and southeast; the overall flood sensitivity of slums is at a moderately high level, with slums exceeding the high flood sensitivity level situated in Kamrangir Char, Lalbagh, southwestern Dhaka, Hazaribagh, and Adabor; the overall slum flood exposure is at a moderately high level, with a large number of highly exposed slums dispersed in densely populated areas with a high density of roads; and overall slum prevention and mitigation capacity are at a moderately high level. The overall flood exposure of slums is moderately high, with a large number of highly exposed slums distributed in densely populated areas with high road network density; and the overall slum prevention and mitigation capacity is moderately high. The overall distribution of flood risk in the slums of Dhaka city is characterized by a gradual increase from the medium risk zone in the west to the east, with very high-risk slums distributed in the eastern part of Dhaka city. This is due to abundant rainfall, very dense water systems, and flat terrain, which are more prone to flood water catchment, and the fact that vegetation in farming areas is less adsorbent to surface moisture than in forested areas, resulting in a higher level of risk for slums in these areas.

Flood risk assessment in slum areas is a complex and crucial task. The formation and development of floods are the result of the interaction between natural elements such as climate and geography, as well as socio-economic elements like human activities and economic development. All these factors are constantly changing. Slum areas are typically characterized by poor infrastructure, vulnerability of low-lying areas to flooding, and a low quality of life for the residents, which adds to the complexity of flood risk assessment. Conducting a fully quantitative analysis of flood risk in slums is challenging due to limitations in the accuracy and refinement of data collection. Therefore, future research needs to focus on improving data collection and processing methods to enhance the accuracy and completeness of the data. This can be achieved by strengthening collaboration with local governments, community-based organizations, and residents to establish effective data collection mechanisms. This will enable a better assessment of flood risk in slums and the development of appropriate risk management and mitigation measures. Additionally, in the future, the integration of multidisciplinary technologies such as GIS, remote sensing, artificial intelligence, and big data analytics can provide a more accurate and comprehensive approach to flood risk assessment and management in slum areas. This integration can facilitate better adaptation to and mitigation of the impacts of floods, ultimately contributing to the sustainable development of slum communities.

Author contributions

Chunlin Pu: Conceptualization, Data curation, Writing – review & editing. Fei Yang: Conceptualization, Validation, Writing – review & editing. Xiuli Wang: Conceptualization, Review & editing.

Data availability statement

The 2006 and 2010 slum datasets used in this study are openly available at https://doi.org/10.1155/2014/172182/dataset.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This research was funded by the National Natural Science Foundation of China (Grant No. 42171079), and the Construction Project of China Knowledge Center for Engineering Sciences and Technology (Grant NO. CKCEST-2023-1-5-1).