Artificial neural network for flood susceptibility mapping in Bangladesh

Future flood susceptibility mapping under climate and land use change

Floods are a significant natural hazard, causing severe damage. Understanding how climate change and land use and land cover (LULC) changes influence flood patterns is crucial for developing sustainable management strategies. This research aims to develop flood susceptibility maps considering the impacts of climate change and land use changes, providing insights into risks from urbanization and climate shifts. Three machine learning models-XGBoost, Random Forest (RF), and Support Vector Machine (SVM)-optimized with Particle Swarm Optimization, were applied to the flood-prone Kashkan watershed in Iran. Results showed that distance from the river, digital elevation model, precipitation, and LULC were the most influential factors. The RF model outperformed others in mapping flood-prone areas, with high-risk zones covering 20% (1908 km2) of the region, primarily in built-up areas. Land use projections for 2050, using the CA-MARKOV model, estimate built-up areas will expand to 859.3 km2. Future precipitation patterns were examined using 8 selected general circulation models under the SSP126 and SSP585 scenarios. Analysis under the SSP585 scenario indicates a 1.9 km2 rise in moderate flood areas, a 36.26 km2 increase in high-risk zones, and a 21.94 km2 decline in very low-risk areas, highlighting expansion of high and moderate flood risk areas.

Leveraging GIS-based AHP, remote sensing, and machine learning for susceptibility assessment of different flood types in peshawar, Pakistan

Due to its diverse topography, Pakistan faces different types of floods each year, which cause substantial physical, environmental, and socioeconomic damage. However, the susceptibility of specific regions to different flood types remains unexplored. To the best of our knowledge for the first time, this study employed an integrated approach by leveraging a GIS-based Analytical Hierarchy Process (AHP), remote sensing, and machine learning (ML) algorithms, to assess susceptibility to three different types of flooding in Peshawar, Pakistan. The study first evaluated the degree of susceptibility to riverine, urban, and flash floods using the GIS-based AHP technique, and then employed ML models, (i.e., specifically Random Forest [RF] and Extreme Gradient Boosting [XG-Boost] to analyze multi-type flood susceptibility in the study region. The performance of the ML models was also evaluated, and the XG-Boost model outperforms RF, demonstrating a higher correlation coefficient (R2 = 0.561-0.922) and lower mean absolute error (MAE = 0.042-0.354), and root-mean-square error (RMSE = 0.119-0.415) for both training and testing datasets. The superior performance of the XG-Boost was further confirmed by the higher value of the area under the curve (AUC) values, which is relatively higher (0.87) than that of the AHP (0.70) and RF (0.86) models. Based on the relative best performance, the XG-Boost model was chosen for further susceptibility assessment of different types of floods, and the generated flood susceptibility maps revealed that 20.9% of the total area is susceptible to riverine flooding, while 30.27% and 48.68% of the total area is susceptible to urban and flash flooding, respectively. The study's findings are significant, offering valuable insights for relevant stakeholders in guiding future flood risk management and sustainable land use plans in the study area.

Delineating the drought vulnerability zones in Bangladesh

The research aims to explore the vulnerability of Bangladesh to drought by considering a comprehensive set of twenty-four factors, classified into four major categories: meteorological, hydrological, agricultural, and socioeconomic vulnerability. To achieve this, the study utilized a knowledge-based multi-criteria method known as the Analytic Hierarchy Process (AHP) to delineate drought vulnerability zones across the country. Weight estimation was accomplished by creating pairwise comparison matrices for factors and different types of droughts, drawing on relevant literature, field experience, and expert opinions. Additionally, online-based interviews and group discussions were conducted with 30 national and foreign professionals, researchers, and academics specializing in drought-related issues in Bangladesh. Results from overall drought vulnerability map shows that the eastern hills region displays a notably high vulnerability rate of 56.85% and an extreme low vulnerability rate of 0.03%. The north central region shows substantial vulnerability at high levels (35.85%), while the north east exhibits a significant proportion (41.68%) classified as low vulnerability. The north west region stands out with a vulnerability rate of 40.39%, emphasizing its importance for drought management strategies. The River and Estuary region displays a modest vulnerability percentage (38.44%), suggesting a balanced susceptibility distribution. The south central and south east regions show significant vulnerabilities (18.99% and 39.60%, respectively), while the south west region exhibits notable vulnerability of 41.06%. The resulting model achieved an acceptable level of performance, as indicated by an area under the curve value of 0.819. Policymakers and administrators equipped with a comprehensive vulnerability map can utilize it to develop and implement effective drought mitigation strategies, thereby minimizing the losses associated with drought.

Geospatial approach to pluvial flood-risk and vulnerability assessment in Sunyani Municipality

Historically, and in recent times, efforts have been to understand, predict, analyze, and quantify floods and their impacts in various countries of the globe. Although recent scientific advances have introduced approaches to assessing the risks presented by flooding, little studies have been carried out in the Sunyani Municipality of Ghana for generating a pluvial flood-risk and vulnerability map for risk identification, resilience, emergency preparedness, and urban spatial planning. In this study, five parameters that influence both pluvial and fluvial flooding were assessed to map flood-prone areas within the Sunyani Municipality. These are precipitation, drainage density, LULC, elevation, and slope, which were integrated in GIS. Using an AHP, weights were assigned to each parameter based on its level of influence on flooding. The findings reveal that 21.32 % of the Sunyani Municipality lies within a highly flood-prone area, 39.65 % in a flood-prone area, while 28.06 % and 10.97 % in slightly flood-prone and not flood-prone areas respectively. Built-up areas close to watersheds with lower elevations and larger drainage density are the places that are highly flood-prone. Some towns within the highly flood-prone and flood-prone areas are Abesim, Newtown, Nkwarbeng, Baakoniaba, Kootokrom, and Penkwase. Highly valued infrastructure such as schools, churches, and hospitals have also been found within these highly flood-prone areas. These findings can aid the government and relevant stakeholders in disaster risk management to be better informed, and to effectively plan and prevent flood challenges in the Sunyani Municipality. Moreover, urban spatial planners in the study setting can consider incorporating the flood hazard maps generated from this study into their spatial plans for proactive physical developments.

Forecasting of flash flood susceptibility mapping using random forest regression model and geographic information systems

Flash floods, rapid and devastating inundations of water, are increasingly linked to the intensifying effects of climate change, posing significant challenges for both vulnerable communities and sustainable environmental management. The primary goal of this research is to investigate and predict a Flood Susceptibility Map (FSM) for the Ibaraki prefecture in Japan. This research utilizes a Random Forest (RF) regression model and GIS, incorporating 11 environmental variables (involving elevation, slope, aspect, distance to stream, distance to river, distance to road, land cover, topographic wetness index, stream power index, and plan and profile curvature), alongside a dataset comprising 224 instances of flooded and non-flooded locations. The data was randomly classified into a 70 % training set for model development, with the remaining 30 % used for model validation through Receiver Operating Characteristics (ROC) curve analysis. The resulting map indicated that approximately two-thirds of the prefecture as exhibiting low to very low flood susceptibility, while approximately one-fifth of the region is categorized as high to very high flood susceptibility. Furthermore, the RF model achieved a noteworthy validation with an area under the ROC curve of 99.56 %. Ultimately, this FSM serves as a crucial tool for policymakers in guiding appropriate spatial planning and flood mitigation strategies.

Future groundwater potential mapping using machine learning algorithms and climate change scenarios in Bangladesh

The aim of the study was to estimate future groundwater potential zones based on machine learning algorithms and climate change scenarios. Fourteen parameters (i.e., curvature, drainage density, slope, roughness, rainfall, temperature, relative humidity, lineament density, land use and land cover, general soil types, geology, geomorphology, topographic position index (TPI), topographic wetness index (TWI)) were used in developing machine learning algorithms. Three machine learning algorithms (i.e., artificial neural network (ANN), logistic model tree (LMT), and logistic regression (LR)) were applied to identify groundwater potential zones. The best-fit model was selected based on the ROC curve. Representative concentration pathways (RCP) of 2.5, 4.5, 6.0, and 8.5 climate scenarios of precipitation were used for modeling future climate change. Finally, future groundwater potential zones were identified for 2025, 2030, 2035, and 2040 based on the best machine learning model and future RCP models. According to findings, ANN shows better accuracy than the other two models (AUC: 0.875). The ANN model predicted that 23.10 percent of the land was in very high groundwater potential zones, whereas 33.50 percent was in extremely high groundwater potential zones. The study forecasts precipitation values under different climate change scenarios (RCP2.6, RCP4.5, RCP6, and RCP8.5) for 2025, 2030, 2035, and 2040 using an ANN model and shows spatial distribution maps for each scenario. Finally, sixteen scenarios were generated for future groundwater potential zones. Government officials may utilize the study's results to inform evidence-based choices on water management and planning at the national level.

Cyclone vulnerability assessment in the coastal districts of Bangladesh

This research aims to assess the vulnerability to cyclones in the coastal regions of Bangladesh, employing a comprehensive framework derived from the Intergovernmental Panel on Climate Change (IPCC, 2007). The study considers a total of eighteen factors, categorized into three critical dimensions: exposure, sensitivity, and adaptive capacity. These factors are crucial in understanding the potential impact of cyclones in the region. In order to develop a cyclone vulnerability map, Principal Component Analysis (PCA) was applied, primarily focusing on the dimensions of sensitivity and adaptive capacity. The findings of this analysis revealed that sensitivity and adaptive capacity components accounted for a significant percentage of variance in the data, explaining 90.00 % and 90.93 % of the variance, respectively. Despite the lack of details about data collection, the study identified specific factors contributing significantly to each dimension. Notably, proximity to the coastline emerged as a highly influential factor in determining cyclone exposure. The results of this research indicate that certain areas, such as Jessore, Khulna, Narail, Gopalgonj, and Bagerhat, exhibit low exposure to cyclones, whereas regions like Chandpur and Lakshmipur face a high level of exposure. Sensitivity was found to be high in most areas, with Noakhali, Lakshmipur, and Chandpur being the most sensitive regions. Adaptive capacity was observed to vary significantly, with low values near the sea, particularly in locations like Cox's Bazar, Shatkhira, Bagerhat, Noakhali, and Bhola, and high values in regions farther from the coast. Overall, vulnerability to cyclones was found to be very high in Noakhali, Lakshmipur, Chandpur, and Bhola, low in Jessore and Khulna, and moderate in Barisal, Narail, Gopalgonj, and Jhalokati. These findings are expected to provide valuable insights to inform decision-makers and authorities tasked with managing the consequences of cyclones in the region.

Quantitative assessment of Land use/land cover changes in a developing region using machine learning algorithms: A case study in the Kurdistan Region, Iraq

The identification of land use/land cover (LULC) changes is important for monitoring, evaluating, and preserving natural resources. In the Kurdistan region, the utilization of remotely sensed data to assess the effectiveness of machine learning algorithms (MLAs) for LULC classification and change detection analysis has been limited. This study monitors and analyzes LULC changes in the study area from 1991 to 2021 using a quantitative approach with multi-temporal Landsat imagery. Five MLAs were applied: Support Vector Machine (SVM), Random Forest (RF), Artificial Neural Network (ANN), K-Nearest Neighbor (KNN), and Extreme Gradient Boosting (XGBoost). The results showed that the RF algorithm produced the most accurate maps of the three-decade study period, accompanied by a high kappa coefficient (0.93-0.97) compared with the SVM (0.91-0.95), ANN (0.91-0.96), KNN (0.92-0.96), and XGBoost (0.92-0.95) algorithms. Consequently, the RF classifier was implemented to categorize all obtainable satellite images. Socioeconomic changes throughout these transition periods were revealed by the change detection results. Rangeland and barren land areas decreased by 11.33 % (-402.03 km2) and 6.68 % (-236.8 km2), respectively. The transmission increases of 13.54 % (480.18 km2), 3.43 % (151.74 km2), and 0.71 % (25.22 km2) occurred in agricultural land, forest, and built-up areas, respectively. The outcomes of this study contribute significantly to LULC monitoring in developing regions, guiding stakeholders to identify vulnerable areas for better land use planning and sustainable environmental protection.

Coupling of machine learning and remote sensing for soil salinity mapping in coastal area of Bangladesh

Soil salinity is a pressing issue for sustainable food security in coastal regions. However, the coupling of machine learning and remote sensing was seldom employed for soil salinity mapping in the coastal areas of Bangladesh. The research aims to estimate the soil salinity level in a southwestern coastal region of Bangladesh. Using the Landsat OLI images, 13 soil salinity indicators were calculated, and 241 samples of soil salinity data were collected from a secondary source. This study applied three distinct machine learning models (namely, random forest, bagging with random forest, and artificial neural network) to estimate soil salinity. The best model was subsequently used to categorize soil salinity zones into five distinct groups. According to the findings, the artificial neural network model has the highest area under the curve (0.921), indicating that it has the most potential to predict and detect soil salinity zones. The high soil salinity zone covers an area of 977.94 km2 or roughly 413.51% of the total study area. According to additional data, a moderate soil salinity zone (686.92 km2) covers 30.56% of Satkhira, while a low soil salinity zone (582.73 km2) covers 25.93% of the area. Since increased soil salinity adversely affects human health, agricultural production, etc., the study's findings will be an effective tool for policymakers in integrated coastal zone management in the southwestern coastal area of Bangladesh.