Assessing demographic and economic vulnerabilities to sea level rise in Bangladesh via a nighttime light-based cellular automata model

The Intergovernmental Panel on Climate Change (IPCC) 6th Assessment Report (AR6) forecasts a sea level rise (SLR) of up to 2 m by 2100, which poses significant risks to regional geomorphology. As a country with a rapidly developing economy and substantial population, Bangladesh confronts unique challenges due to its extensive floodplains and 720 km-long Bay of Bengal coastline. This study uses nighttime light data to investigate the demographic repercussions and potential disruptions to economic clusters arising from land inundation attributable to SLR in the Bay of Bengal. By using geographical information system (GIS)-based bathtub modeling, this research scrutinizes potential risk zones under three selected shared socioeconomic pathway (SSP) scenarios. The analysis anticipates that between 0.8 and 2.8 thousand km2 of land may be inundated according to the present elevation profile, affecting 0.5–2.8 million people in Bangladesh by 2150. Moreover, artificial neural network (ANN)-based cellular automata modeling is used to determine economic clusters at risk from SLR impacts. These findings emphasize the urgency for land planners to incorporate modeling and sea inundation projections to tackle the inherent uncertainty in SLR estimations and devise effective coastal flooding mitigation strategies. This study provides valuable insights for policy development and long-term planning in coastal regions, especially for areas with a limited availability of relevant data.

and economic activities with GIS-based bathtub modeling and ANN-based MOLUSCE modeling, this research investigates potential risk zones under various SSP scenarios.The methodology offers a detailed evaluation of future demographic impacts and disruptions to economic clusters, thus providing valuable guidance for effective policy-making and long-term coastal planning in Bangladesh and demonstrating the potential of nighttime light data in addressing data scarcity issues in similar regions.

Result analysis
Sea level projection under different scenarios.From the IPCC's AR6, two medium-confidence SSP scenarios (SSP1-1.9 and SSP2-4.5)and one low-confidence SSP scenario (SSP5-8.5)are selected for this study.The projected SLR trend was investigated based on the mean and several quantiles (5%, 17%, 50%, 83%, and 95% 7 ) with an interval of 10 years between 2020 and 2150 (Fig. 1).Here, each quantile represents the percentage likelihood of the worst possible scenario and an SLR projection based on the respective scenario in comparison to average emissions from 1995 to 2014.
The mean sea level drops periodically in SSP1-1.9 and SSP2-4.5 (from 2020 to 2040), with a 5% quantile showing a positive impact on the earth (Fig. 1).Other quantiles for different SSPs show substantial sea Figure 1.Potential SLR under selected SSP scenarios 5,35 as per SSP1-1.9,SSP2-4.5, and SSP5-8.5.Each quantile shows the probability of the worst-case scenario and an SLR forecast based on emissions from 1995 to 2014.
level increases with temporal variation.The mean SLR according to SSP1-1.9 shows a yearly increase rate of − 0.84 mm/year to 8.38 mm/year from the 5 to 95% quantiles; slight increases from 1.4 to 12.516 mm/year occur under SSP2-4.5.The likelihood of such an SLR at this rate is comparatively high, as they present medium confidence levels.The mean SLR in SSP1-1.9 is projected to be 0.48 m by 2150.However, SSP5-8.5 shows a SLR of 3.78-42.507mm/year according to the quantile variation, which is much higher than either of the previous two scenarios.Moreover, this scenario shows that the projected mean SLR can be 1.89 (0.72-4.81) m by 2150, compared to only 0.78 (0.39-1.27) m for SSP2-4.5.Overall, SSP5-8.5 shows a very high magnitude of SLR relative to other scenarios for the representative point.
Spatiotemporal pattern of flood inundation in the region.Bangladesh is one of the largest coastal floodplains worldwide; consequently, low-lying parts of the country are more likely to be affected by a very small rise in sea level.Therefore, to demonstrate the possibilities of inundation, a GIS-based bathtub model was used based on the selected SSP scenarios.From the simulation, the projected inundated land area by 2150 can be observed (Fig. 3); the spatial distribution of this potential sea inundation is depicted separately (Fig. 2).Based on present global socioeconomic trends, the SSP1-1.9scenario was considered the highest-possibility scenario, and the SSP5-8.5 scenario was considered the lowest-possibility scenario in this SLR projection.Figure 2 also represents the significant transportation routes throughout the country, which indicate higher population concentrations in urban locations.
Figure 2 shows that a greater portion of the coastal area of the BOB and adjacent plains of the Meghna delta fan is marked as being at a very high risk of sea inundation because its elevation is less than 0.45 ± 0.49 m.This includes the greater Khulna, Bagerhat, Satkhira, Jessore, and Narail from the southwestern region; Borguna, Potuakhali, and Bhola from the southernmost region; and Cox's Bazar and part of Chittagong from the southeastern coast of the BOB.A few fragmented clusters of Lakshmipur, Noakhali, and Sharitpur are also estimated to be affected by sea inundation by 2150 under this projection.Even though the northern part of Bangladesh comprises higher elevations, a cluster of flooded zones may be observed across the lower elevated floodplains near the Surma-Kushiyara-Meghna River system.
The capital city, Dhaka, is projected to be at a partly high and mostly moderate risk of flooding from SLR.This is because Dhaka is next to the Buriganga River, which stages the early stage of Meghna Delta.Faridpur and Rajbari districts are moderate, but the northern parts of Shariatpur, Madaripur, and the southern part of Munshiganj are expected to be flooded by 2150, with a lower level of confidence.Port-city Chittagong, Sitakund, and Sandwip also have moderate-to-low risk factors with an elevation of lower 1.89 ± 0.49 m (Fig. 1).In contrast, Kutubdia, Matarbari, and St. Martin islands are projected to be high-risk areas for sea inundation.
Projected inundation and its impact on the population.As discussed in the previous section, the spatial projections of mean sea inundation under various SSP scenarios are presented (Fig. 3).The potential inundated area was calculated using the number of pixels in the DEM that had lower elevation values than the projected SLR for each SSP scenario.These estimations were further related to the UN's population density projection for Bangladesh.
According to the SSP1-1.9scenario, sea inundation will range to 567.0963 km 2 , affecting approximately 0.83 million people by 2050.This inundation is projected to expand further by 705.4767 km 2 in the 2100s and 835.3926 km 2 by 2150, affecting approximately 0.8-0.85 million Bangladeshi inhabitants.Comparatively, a slightly higher inundated area is featured for SSP2-4.5, which indicates a transition of inundation from 567.16 to 1156.7 km 2 within the study period.The expected affected population due to such inundation ranges between 0.8 and 1.05 million for the following years.However, for both scenarios, a decline in the affected population can be observed, as SSP1-1.9 and SSP2-4.5 estimate a significant decline in population and anthropogenic emissions after 2080-2100.
In contrast, SSP5-8.5 illustrates a very high magnitude of sea inundation, as this scenario models a future with a very high population density and a high frequency of energy consumption and emissions.The scenario depicts that the inundated area will extend approximately 611.32 km 2 , affecting approximately 0.9 million people in the 2050s, which is roughly equal to the other two SSP scenarios.In the 2100s, the inundated area will increase to 1183.2 km 2 , affecting almost 1.2 million habitats, and is projected to extend to 2967.5 km 2 by 2150, affecting more than 2.7 million habitats in Bangladesh due to an expected mean SLR of approximately 1.89 m.

Economic activity simulation under different scenarios.
The MOLUSCE simulation plugin is used to determine the potential economic centers at risk due to sea inundation, as presented (Fig. 4).The method is assessed using selected SSP scenarios at 50-year intervals to emphasize the impact of potential sea inundation on Bangladesh's economic centers and urban settlements.The base years of 2014-2022 are chosen for economic and urban population concentration simulation, and an increase in variable clusters with temporal variation throughout all the districts of Bangladesh is evident.Major cities, such as Dhaka, Tongi, Gazipur, Rajshahi, Comilla, Chittagong, Sylhet, Pabna, Bogura, and Rangpur, have the highest potential for growth since they are important economic centers.Moreover, transition zones such as the Dhaka-Chittagong, Dhaka-Khulna, and Dhaka-Sylhet roads feature a significant increase in nighttime light radiance since they may function as ruralurban fringe borders.A relatively high magnitude of nighttime light radiance may also be seen along the routes of Bogra-Rangpur, Bogra-Rajshahi, and Tangail-Rajshahi since the northern quarter of Bangladesh is expected to be significantly industrialized by 2030.Moreover, coastal cities such as Chittagong, Cox's Bazar, Khulna, and Mongla are anticipated to see growing demographic densities and commercial interest.
However, the potential economic and urban concentration zones have a high risk of sea inundation in different SSP scenarios.The sea inundation is heading from the southwestern coast to the southern coast, alongside the delta plains of the Meghna-Surma-Kushiyara River (Fig. 4).Significant sea inundation can also be observed along the floodplain of the Meghna-Padma stream in the western region and the Sangu-Matamuhuri-Karnaphuli river in the east for SSP5-8.5 starting in 2100.SSP1-1.9,although featuring a very low level of inundation, impacts several districts from the southern coast of Bangladesh.By 2150, rivers such as the Balaswar, Payra, Tetulia, and Arpangasia may become gateways for potential sea inundation and land loss in their adjacent major economic centers.While SSP2-4.5 shows a frequent similar inundation pattern, with time, SLR may potentially move toward the districts of Khulna, Satkhira, Bhola, and Barisal, where economic concentration is expected to increase by 2100.Moreover, in this scenario, a few clusters in Chittagong and Cox's Bazar, as well as a significant percentage of Dhaka Division, are predicted to be vulnerable to sea inundation.Although the SSP5-8.5 scenario is expected to have lower confidence, major land areas where economic concentrations are much higher are projected to be inundated.This projection covers a large portion of the Padma-Meghna confluence and significant scattered clusters in the southern and southernmost parts of Bangladesh.The prediction places vast economic and urban centers, such as Chittagong, Khulna, Barisal, Mongla, and potentially sections of Dhaka and Sylhet, at risk.Remarkably, large plain areas adjacent to the Brahmaputra-Jamuna stream have a lower elevation profile (8 m), which may be critical to the inhabitants of Bangladesh's northern regions, such as Pabna, Kushtia, Rajbari, and  the predicted visualization in Fig. 4 comprises 12.48% of the total land bodies, while SSP2-4.5 impacts 3.82% of the total land bodies in Bangladesh (Fig. 3).

Discussion
Bangladesh is one of the fastest economically growing countries with a vast population.Therefore, changes in economic clusters are obvious 36,37 .Factors such as rural-urban migration, rapid industrialization and urbanization, and geo-economic opportunity often cause changes in the economic orientations of the country 38,39 .As per our simulation (Fig. 4), a significant change in economic concentrations throughout the country, therefore, is evident.However, in the near future, the people of Bangladesh may need to move their settlements and economic centers due to climatic consequences such as SLR.From the above analysis, it is apparent that a significant portion of land area with socioeconomic prospects is at risk for sea inundation.Appropriate planning of the redistribution of the economic infrastructure may also be needed, as potential SLR may inundate and prevent prospectus zones from becoming inhabited.Considering the mean SLR, our study reveals that more than 0.5-2% of inhabited land may be submerged based on the present elevation profile, thus possibly evicting more than 2.8 million people by 2150 38,40 .
As per the IPCC's SSP1-1.9 and SSP2-4.5 scenarios, the population density may be facing a degrading trend after 2080 along with a significant decline in GHG emissions.This may have a positive impact, lowering the ice sheet melting rate or seawater expansion (Fig. 2).However, the results from the worse-case scenarios (considering the 17 to 95% quantile) may cause up to 42.507 mm/year of SLR 2,4,35 .Moreover, contemporary subsidence in Dhaka and Khulna significantly reduces the elevation of land areas under the SLR threshold.A subsidence of 2.08 mm/year in Central Dhaka 17 and 2.83 mm/year in Sundarbans 41 , Khulna, was measured.Furthermore, Rahman et al. 42 reported a potential 170 km 2 area loss in the Sundarbans between 1973 and 2010 due to subsidence, which is extremely concerning for the coastal belt ecosystem in Bangladesh.
The SLR in the BOB may also have a significant effect on the hydrology and morphological characteristics of the Meghna River, as the BOB receives a large amount of discharge from it.As the sea level rises, the rivers mentioned in preceding sections may need a wider catchment during the rainy season for additional discharges to prevent tremendous flooding and long-term rainwater logging.Higher chances of flooding the tributaries and distributaries are also expected and may cause significant inundation to nearby plains, as SSP5-8.5 denotes that the Padma River and upper Meghna River would have higher water levels.CEGIS, Dhaka, previously identified 40 new char areas (acquired lands) totaling approximately 1643 km 2 in the Meghna estuary alone that have been developed during the previous 15 years 43 .Based on our study, these areas can be considered the most vulnerable areas even in the IPCC medium confidence scenarios.Our predictions also indicate that certain clusters of the districts of Kushtia, Meherpur, Pabna, Faridpur, Chandpur, Munshiganj, Narayanganj, Brahmanbaria, and Kishoreganj would face severe floods in their respective economic regions as a result of SLR.These districts comprise the crop hubs of Bangladesh.
The area in southern Bangladesh will be most affected by SLR in 2150, according to SSP5-8.5.By 2150, a significant part of Bangladesh's southern area is expected to be submerged under seawater.As a result, Bangladesh will lose several important economic areas, including Jessore, Faridpur, Madaripur, Chuadanga, Jhenaidah, Bagerhat, Gopalganj, Narail, Barishal, Chittagong City, and Cox's Bazar.The world's largest mangrove, the Sundarbans, may also face a massive catastrophe since a significant number of land bodies are expected to be submerged under seawater while their adjacent distributaries rise.The inundation of Chittagong City would have a substantial adverse impact on Bangladesh's economy.The Port of Chittagong is used to convey more than 92% of imported and exported cargo.This is why the Port of Chittagong is referred to as the "Gateway of Bangladesh."The existence of this port has led the City of Chittagong to become one of several significant commercial centers in Bangladesh.

Conclusion
This work demonstrates the use of an integrated application of sea inundation projections by utilizing hydrodynamic bathtub modeling and spatiotemporal analysis to evaluate the potential for various SLR scenarios in the BOB.The bathtub model was developed in response to the lack of readily available, user-friendly tools for modeling transitory coastal flooding [44][45][46] .Elevation data retrieved from Google Earth and the IPCC's AR6 offer an additional advantage by enabling the evaluation of numerous SLR scenarios 35 .Furthermore, our study utilizes an ANN-based population proxy nighttime simulation to include the trends underlying climate change scenarios, making it easier to identify future susceptible locations 33 .The findings of this paper indicate that considerable coastal (Khulna, Chittagong, and Cox's Bazar) and low-lying floodplain (parts of Dhaka and Sylhet) areas are particularly vulnerable to potential SLR.These cities may be forced to endure significant financial and demographic consequences if severe SLR scenarios occur simultaneously.According to our projections, up to 2% of habitable land might be submerged, resulting in the relocation of approximately 2.8 million individuals by 2150.This submerged land includes the potential economic clusters of Bangladesh.Furthermore, possible changes in salt intrusion, habitat destruction, and ecological imbalance in the world's largest mangrove 2,4,35 can be identified from our spatiotemporal analysis.
However, we are aware that the extent of future climatic consequences is still uncertain in terms of future humanitarian responses.Administrations at all levels must incorporate the implications of climate change in their planning.The decision-support frameworks that rely on this research can be utilized in Bangladesh's coastal zone and lower-elevation floodplain planning.However, our study lacks the potential for practical planning to eradicate the consequences of SLR.Therefore, we recommend that future researchers focus on planning preventive mechanism to mitigate SLR consequences with a concentration on economic clusters.More precise topographic data, such as higher-resolution DEM data and valid statistics regarding mean subsidence in the region, should www.nature.com/scientificreports/also be utilized in future studies to create more credible conclusions.Furthermore, any land-use changes should be accounted for based on local extreme weather occurrences.

Materials and methods
Study area.Bangladesh has a diversified landscape with a population of approximately 169 million people dispersed throughout 148,460 km 2 .The climate is a tropical monsoonal one, with hot, humid summers and moderate winters.The economy depends primarily on agriculture, including significant crops such as rice, jute, tea, and textiles.Its geomorphology is formed by the country's dynamic river systems, coastal activities, and tectonic activity.Hills and mountains are concentrated in the northern and eastern regions, while the Gangetic-Brahmaputra floodplain, one of the world's largest river deltas, dominates the central area.A low-lying deltaic plain caused by the confluence of multiple rivers forms the coastal area 47 .The Ganges-Brahmaputra Delta spans the western sector of the Bay of Bengal, formed by the sedimentation of the Ganges, Brahmaputra, and Meghna rivers (Fig. 5).Furthermore, in addition to the ongoing erosion and shifting caused by the dynamic interplay of river, tidal, and wave processes, the bay receives a large amount of deposition from Bangladeshi rivers.The bay is influenced by monsoonal rainfall, with significant rainfall during the southwest monsoon from June to September and the northeast monsoon from November to February 47 .
Data.The NASA "Sea Level Projection Tool" data from the IPCC AR6 were used to evaluate the future SLR projection in the BOB, considering Cox's Bazar point as the base 7 .This tool enables users to assess sea level projection statistics corresponding to the IPCC AR6 48 and improves access and visualization of the report's consensus projections.It shows both regional and global sea level projections from 2020 to 2150 relative to a 1995-2014 baseline and how they differ considerably by scenario 5,35,49 .Sea level estimation utilizes uncertainty in emissions model temperature shifts as well as unpredictability in temperature-driver interactions, such as temperature expansion, ocean movement, and glacier and ice sheet melting.Here, the average rates of cumulative sea-level change are represented in mm year −1 .IPCC scenarios are statistical techniques used to examine the parameter space of low-and high-emissions 50 .There is also information about how different physical processes will contribute to future SLR, which suggests which processes will be major contributors.In the dataset, quantiles spanning from the 17th to the 83rd are often interpreted as likely intervals, with "probable" meaning a likelihood of at least 66%.Thus, in this study, mean SLR data were used to project inundation in potential economic clusters by using SSP1-1.9 as a low emission scenario with a higher likelihood of uncertainty, SSP2-4.5 as a medium emission scenario with a moderate likelihood of uncertainty, and SSP5-8.5 as a high emission scenario with a low likelihood of uncertainty.Moreover, SLR under different quantiles was also analyzed to understand the magnitude of area loss based on the selected SSP scenarios.The UN's population projection data from IPCC AR6 were used to assess the affected population due to sea inundation 48,51 .The UN analyzes assessments of all countries' populations by age and gender once every 2 years in a report called World Population Prospects (WPP) [51][52][53] .The UN projections are developed based on multiple factors, e.g., future fertility, mortality, and global migration rates.Thus, the UN generates the "medium" prediction, as it is a single value for each prospective population 53 .Furthermore, the UN's population projections use the conventional cohort-component approach 54,55 .
Digital elevation models (DEMs) are often used in water resource studies to associate drainage attributes such as ridges, basin bottoms, channel networks, and surface hydrology with sub-floodplain channel size, length, and slope [56][57][58][59][60] .At present, multiple remotely sensed DEM datasets are available, e.g., NASA Shuttle Radar Topographic Mission (STRM), Thermal Emission and Reflection Radiometer (ASTER), Global Digital Elevation Model (GDEM), and European Space Agency (ESA) DEM over resolutions ranging from 30 to 900 m.However, this dataset may not be ideal for float-scale computations for precise hydrological research.Therefore, a 5 m resolution DEM of Bangladesh was generated for this study by assessing the elevation values in Google Earth Pro.Google Earth Pro is an enhanced version of Google Earth that enables the simultaneous representation of several locations as well as access to the elevation profile.Although Google Earth elevation data use SRTM as its elevation base data, several studies have confirmed its higher accuracy of elevation profiles compared to other available DEM datasets [61][62][63] .Moreover, the accuracy of a DEM improves with the level of denseness and unpredictability of the dots.In this study, the elevation data from Google Earth were extracted using Keyhole Markup Language (KML), with a total of 2.5 million elevation points covering the study region.The assessed point was further used to create a high-resolution (5 m DEM) raster dataset via inverse distance weighted (IDW) interpolation in ArcMap 10.8.Esri's ArcMap software is a complete platform for producing, collecting, analyzing, and demonstrating geographical data, making it popular among GIS professionals and scholars.In this study, ArcMap version 10.8 (downloaded from https:// bit.ly/ 45Vbr PH) was used.IDW specifically assumes that near objects are more similar than remote objects.IDW utilizes metrics around the prediction region to forecast any unrecorded value.Afterward, the modeled elevation profile was projected using the UTM WGS84 grid coordinate system in ArcMap.The resulting elevation map was then visualized in Fig. 5 via QGIS 2.18 to portray the spatial location of Bangladesh and the BOB.Quantum Geographic Information System, or QGIS, is popular open-source software that lets users produce, analyze, and display geographical data.The model used in this study is QGIS 2.18 (downloaded from https:// bit.ly/ 3qCUv 0g), which was selected in accordance with the study's computability.
NPP-VIIRS data were derived as a proxy for determining economic clusters.Numerous researchers employ it to estimate its economic effects on regional and national scales [64][65][66] .The Earth Observations Group (EOG) at the National Centers for Environmental Information (NCEI) produces monthly mean NPP-VIIRS images.This instrument has a resolution of 15 arc seconds.The VIIRS sensor aboard the Suomi National Polar-Orbiting Partnership (NPP) satellite analyzes data in 22 distinct wavelength bands, one of which is the DNB (Day/Night Band).VIIRS DNB statistics have been used to evaluate the population, assess working conditions in remote modernization, monitor natural hazards and disharmony, and comprehend light pollution's biological impacts 32,64,66 .Mean radiance composite images are made with data from the VIIRS Day/Night Band based on nighttime brightness.

Simulation models
Change analysis using a cellular automata model.Satellite-based nighttime light images have been broadly used by economists as a substitution to determine economic interaction in poor countries.Cities and towns with strong social and cultural infrastructures show greater population dynamics [64][65][66] .Furthermore, it is pertinent to forecast the potential consequences of SLR in these economic areas.Therefore, to evaluate spatiotemporal changes and predict the future economic changes between the study years, the MOLUSCE plugin www.nature.com/scientificreports/ on QGIS was used.The MOLUSCE plugin makes use of cellular automata data to simulate hypothetical outcomes based on past ones 67 .It evaluates altering analysis and transformation prospects via four methods: logistic regression, weights of evidence, multicriteria evaluation, and artificial neural network (multilayer perceptron).Such approaches construct a parameter that forms the foundation for the prediction model 34,68 .In this study, several nighttime DNB clusters from 2014 and 2020 were applied with the QGIS MOLUSCE tool to generate a transition matrix.The MP-ANN approach in QGIS 2.18 software is used to train a model of preceding nighttime light transitions 69,70 .The plugin was used to forecast the scenarios for 2050, 2100, and 2150, utilizing nighttime light visuals as a spatial parameter.
The abovementioned Eq. (1) demonstrates the logistic regression that is used to detect the potential changes in the training raster dataset.Here, LU t+1 = land-use class at time t + 1, argmax j = class that maximizes the product of transition likelihood and current land-use pattern, P (LU t+1 = j | LU t = i) = transition probability from land-use class i to j at time t + 1, and LUP t(i,j) = current land-use pattern value for the transition from land-use class i to j at time t.
Furthermore, the MP-ANN model was validated by contrasting the simulated and estimated values for 2018, 2021, and 2022 DNB composites using MOLUSCE QGIS validation.It is a nonlinear statistical analysis approach that prepares urban development drivers and accounts for intricate underlying characteristics during modeling 71 .The statistical alterations of an ANN have generally weighted sums of materials, activation coefficients, and bias variables, and their formulae are dependent on the design of the neural network (e.g., feedforward, recurrent, convolutional) as well as the modeler's stimulation functions and other factors.MOLUSCE's ANN is coupled with the cellular-automata (CA) simulation approach.CA employs the Monte Carlo algorithm technique, which is thought to be compatible with predicting land-use change 72 .
In Eq. ( 2), the Monte Carlo simulation creates independent samples from probabilistic distributions that represent the system's unanticipated variables or characteristics.Here, Y = estimated outcome, N = number of samples, f(x i ) = value of the function at the ith random sample, x i.
Furthermore, to eliminate dataset inconsistencies, a raster calculator was used to scale all the raster images between 1 and 5 with similar intervals before running the model.The neural network learning curve with a maximum iteration of 2000 is presented in Fig. 6.
Figure 6 displays the network's efficacy on the training and test datasets as an accumulation of training epochs from the training data periodically.The red line in this learning curve represents the training set performance, whereas the green line represents the validation set performance.Furthermore, the result of the training dataset will demonstrate the kappa validation value in Table 1.The overall kappa (k i ) (Eq. 3) and % of correctness (C) (Eq.4) were determined as follows: (1) www.nature.com/scientificreports/Here, P 0 = observed proportion of agreement, P e = proportion expected by chance, n ij = diagonal elements in the error matrix, k = total number of classes, and n = total number of samples in the error matrix.
However, the DNB composite simulation accuracy was evaluated using a percent correctness value that exceeded > 85% (Table 1).
Bathtub approach.The bathtub model, or bathtub approach, is a frequently used framework for assessing flooding worldwide 60 .It is a geospatial technique that simulates coastal surface runoff via digital elevation models and is dependent on the accuracy of the geomorphologic input data 73 .A bathtub model considers potentially vulnerable regions that might be below the inundation level and hydrologically associated with the source of flooding (e.g., the ocean or river).Therefore, several researchers have implemented this approach to monitor the flooding prospects of coastal areas or, in conjunction with socioeconomic and infrastructure data, to assess the flood inundation risk 74,75 .The bathtub methodology can be performed in GIS software, which allows for easy incorporation with other geodatabases or even in matrix computational development tools 73 .
However, for the current study in which certain SSP scenarios and years were chosen, the bathtub approach was applied to evaluate areas adjacent to the BOB that could potentially be inundated due to SLR.The predicted submerged land bodies of Bangladesh due to sea inundation (Figs. 2 and 4) in IPCC AR6 35 were analyzed using ArcMap 10.8.The spatial distribution of potential sea inundation is visualized in Fig. 2 by using the same ArcMap version to highlight the potential hazard of substantial manmade buildups in Bangladesh.The publicly-available shapefiles of rail and road networks were collected from https:// geoda sh.gov.bd/.The elevation profile of distinct IPCC-proposed hazard scenarios was used to develop nine raster files for the evolution of SLR scenarios in Fig. 4. For further analysis, reclassified rasters were transformed into polygons. https://doi.org/10.1038/s41598-023-40329-9

Figure 2 .
Figure 2.An inundation simulation to project the potential vulnerability of growing economic clusters in Bangladesh by 2150.A geographic information system (GIS) software package, ArcMap (Version 10.8, downloaded from https:// bit.ly/ 45Vbr PH), was used to generate the figure.

Figure 3 .
Figure 3. Potential inundation and affected population on temporal variation based on SSP scenarios.The calculation of the potential inundated area was based on the pixel count of the digital elevation model (DEM) with elevation values lower than the anticipated rise in sea level for each shared socioeconomic pathway (SSP) scenario.

Figure 4 .
Figure 4. Spatiotemporal variation in sea inundation and its threat to potential economic clusters.The figure depicts the results obtained from the MOLUSCE simulation.The approach is evaluated through the utilization of specific SSP scenarios at 50-year intervals, with the aim of highlighting the potential consequences of sea inundation on Bangladesh's economic centers and urban settlements. https://doi.org/10.1038/s41598-023-40329-9 https://doi.org/10.1038/s41598-023-40329-9

Figure 5 .
Figure 5. Elevation profile of Bangladesh, assessed with the SRTM digital elevation model.The GIS software packages ArcMap (Version 10.8, downloaded from https:// bit.ly/ 45Vbr PH) and QGIS (Version 2.18, downloaded from https:// bit.ly/ 3qCUv 0g) were used to generate the figure.

Figure 6 .
Figure 6.Neural network learning curve of the DNB composite images of the potential nighttime light clusters to project future economic clusters (the red line represents the training set performance, whereas the green line represents the validation set performance).

Table 1 .
Validation parameters (K parameters) and % correctness of the CA-ANN model in QGIS software.