Flood Susceptibility Mapping: Integrating Machine Learning and GIS for Enhanced Risk Assessment

Applied Computing and Geosciences, Journal Year: 2024, Volume and Issue: 23, P. 100183 - 100183

Published: Aug. 3, 2024

Flooding presents a formidable challenge in the United States, endangering lives and causing substantial economic damage, averaging around $5 billion annually. Addressing this issue improving community resilience is imperative. This project employed machine learning techniques publicly available data to explore factors influencing flooding develop flood susceptibility maps at various spatial resolutions. Six algorithms, including Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), K-nearest neighbor (KNN), Adaptive Boosting (Ada Boost), Extreme Gradient (XGB) were used. Geospatial datasets comprising thirteen predictor variables 1528 inventory collected since 1996 analyzed. The are rainfall, elevation, slope, aspect, flow direction, accumulation, Topographic Wetness Index (TWI), distance from nearest stream, evapotranspiration, land cover, impervious surface, surface temperature, hydrologic soil group. Five hundred twenty-eight non-flood points randomly created using stream buffer for two scenarios. A total of 2964 classified into flooded (1) non-flooded (0) categories used as target. Overall, testing results showed that XGB RF models performed relatively well both cases over multiple resolutions compared other models, with an accuracy ranging 0.82 0.97. Variable importance analysis depicted such streams, type, surfaces significantly affected prediction, suggesting strong association underlying driving process. improved performance variation susceptible areas across scenarios considering appropriate non-flooding training critical developing flood-susceptibility models. Furthermore, tree-based ensemble algorithms like XG boost stack generalization approach can help achieve robustness model where being evaluated.

Language: Английский

---

Flood Susceptibility Analysis with Integrated Geographic Information System and Analytical Hierarchy Process: A Multi-Criteria Framework for Risk Assessment and Mitigation

Water, Journal Year: 2025, Volume and Issue: 17(7), P. 937 - 937

Published: March 23, 2025

Flooding is among the most destructive natural disasters globally, and it inflicts severe damage on both environments human-made structures. The frequency of floods has been increasing due to unplanned urbanization, climate change, changes in land use. Flood susceptibility maps help identify at-risk areas, supporting informed decisions disaster preparedness, risk management, mitigation. This study aims generate a flood map for Davidson County Tennessee using an integrated geographic information system (GIS) analytical hierarchical process (AHP). In this study, ten causative factors are employed flood-prone zones. AHP, form weighted multi-criteria decision analysis, applied assess relative impact weights these factors. Subsequently, into ArcGIS Pro (Version 3.3) create area overlay approach. resulting classified county five zones: very low (17.48%), (41.89%), moderate (37.53%), high (2.93%), (0.17%). FEMA hazard used validate created from Ultimately, comparison reinforced accuracy reliability assessment GIS AHP

Language: Английский

---

Water Body Mapping Using Long Time Series Sentinel-1 SAR Data in Poyang Lake

Water, Journal Year: 2022, Volume and Issue: 14(12), P. 1902 - 1902

Published: June 13, 2022

Mapping water bodies with a high accuracy is necessary for resource assessment, and mapping them rapidly flood monitoring. Poyang Lake the largest freshwater lake in China, its wetland one of most important world. affected by floods from Yangtze River basin every year, fluctuation area level directly or indirectly affects ecological environment Lake. Synthetic Aperture Radar (SAR) particularly suitable large-scale body mapping, as SAR allows data acquisition regardless illumination weather conditions. The two-satellite Sentinel-1 constellation, providing C-Band data, passes over about five times month. With temporal-spatial resolution, can be used to accurately monitor body. After acquiring all (1A 1B) ensure consistency processing, we propose use Python SeNtinel Application Platform (SNAP)-based engine (SARProcMod) process construct dataset 10 m resolution. To extract information an automatic classification based on modified U-Net convolutional neural network (WaterUNet), which classifies using artificial sample datasets validation accuracy. results show that maximum minimum areas our study were 2714.08 km2 20 July 2020, 634.44 4 January 2020. Compared gauging station, was highly correlated level, correlation coefficient being up 0.92 R2 quadratic polynomial fitting 0.88; thus, resulting relationship estimate According results, conclude WaterUNet are very monitoring well emergency mapping.

Language: Английский

---

FLOOD RISK MAPPING USING RANDOM FOREST AND SUPPORT VECTOR MACHINE

ISPRS annals of the photogrammetry, remote sensing and spatial information sciences, Journal Year: 2023, Volume and Issue: X-4/W1-2022, P. 201 - 208

Published: Jan. 13, 2023

Abstract. Floods are among the natural disasters that cause financial and human losses all over world every year. By production of a flood risk map determination potential areas, possible damages this phenomenon can be reduced. To extend in Calcasieu Parish, Louisiana, US, conditioning factors affecting occurrence including elevation, slope, plan curvature, land use, distance from rivers, density rainfall, normalized difference vegetation index (NDVI), modified water (MNDWI), built-up (NDBI) were identified their information layers produced using Google Earth Engine (GEE) cloud platform. Then, for mapping, Random Forest (RF) support vector machine (SVM) as two learning models have been implemented results compared. RF SVM validated based on maximum absolute error (MAE) with an accuracy 0.043 0.097, respectively. Visualization predicted values QGIS software confirms model has provided better outputs than model. analysing features importance model, it was verified curvature highest degree influence degrees 0.197, 0.135, 0.123.

Language: Английский

---

Prediction of flood sensitivity based on Logistic Regression, eXtreme Gradient Boosting, and Random Forest modeling methods

Water Science & Technology, Journal Year: 2024, Volume and Issue: 89(10), P. 2605 - 2624

Published: May 7, 2024

Floods are one of the most destructive disasters that cause loss life and property worldwide every year. In this study, aim was to find best-performing model in flood sensitivity assessment analyze key characteristic factors, spatial pattern evaluated using three machine learning (ML) models: Logistic Regression (LR), eXtreme Gradient Boosting (XGBoost), Random Forest (RF). Suqian City Jiangsu Province selected as study area, a random sample dataset historical points constructed. Fifteen different meteorological, hydrological, geographical variables were considered assessment, 12 based on multi-collinearity study. Among results comparing ML models, RF method had highest AUC value, accuracy, comprehensive evaluation effect, is reliable effective risk model. As main output map divided into five categories, ranging from very low high sensitivity. Using (i.e., accuracy model), high-risk area covers about 44% mainly concentrated central, eastern, southern parts old city area.

Language: Английский

---

Introducing a New Index for Flood Mapping Using Sentinel-2 Imagery (SFMI)

Computers & Geosciences, Journal Year: 2024, Volume and Issue: 194, P. 105742 - 105742

Published: Oct. 25, 2024

Language: Английский

---

Spatial Prediction of Current and Future Flood Susceptibility: Examining the Implications of Changing Climates on Flood Susceptibility Using Machine Learning Models

Entropy, Journal Year: 2022, Volume and Issue: 24(11), P. 1630 - 1630

Published: Nov. 10, 2022

The main aim of this study was to predict current and future flood susceptibility under three climate change scenarios RCP2.6 (i.e., optimistic), RCP4.5 business as usual), RCP8.5 pessimistic) employing four machine learning models, including Gradient Boosting Machine (GBM), Random Forest (RF), Multilayer Perceptron Neural Network (MLP-NN), Naïve Bayes (NB). conducted for two watersheds in Canada, namely Lower Nicola River, BC Loup, QC. Three statistical metrics were used validate the models: Receiver Operating Characteristic Curve, Figure Merit, F1-score. Findings indicated that RF model had highest accuracy providing maps (FSMs). Moreover, provided FSMs flooding is more likely occur River watershed than Loup watershed. Following scenario, area percentages classes 2050 2080 have changed by following from year 2020 2050, respectively: Very Low = -1.68%, -5.82%, Moderate +6.19%, High +0.71%, +0.6% -1.61%, +2.98%, -3.49%, +1.29%, +0.83%. Likewise, watershed, changes between years were: -0.38%, -0.81%, -0.95%, +1.72%, +0.42% -1.31%, -1.35%, -1.81%, +2.37%, +2.1%, respectively. impact on flood-prone places revealed regions designated highly very susceptible flooding, grow forecasts both watersheds. contribution lies novel insights it provides concerning British Columbia Quebec over time various scenarios.

Language: Английский

---

Assessment of flood susceptibility prediction based on optimized tree-based machine learning models

Journal of Water and Climate Change, Journal Year: 2022, Volume and Issue: 13(6), P. 2353 - 2385

Published: June 1, 2022

Abstract Due to the physical processes of floods, use data-driven machine learning (ML) models is a cost-efficient approach flood modeling. The innovation current study revolves around development tree-based ML models, including Rotation Forest (ROF), Alternating Decision Tree (ADTree), and Random (RF) via binary particle swarm optimization (BPSO), estimate susceptibility in Maneh Samalqan watershed, Iran. Therefore, implement 370 flood-prone locations case were identified (2016–2019). In addition, 20 hydrogeological, topographical, geological, environmental criteria affecting occurrence area extracted predict susceptibility. under curve (AUC) variety other statistical indicators used evaluate performances models. results showed that RF-BPSO (AUC=0.935) has highest accuracy compared ROF-BPSO (AUC=0.904), ADTree-BPSO (AUC=0.923). findings illustrated chance flooding center question greater than points due lower elevation, slope, proximity rivers. ensemble framework proposed here can also be maps regions with similar geo-environmental characteristics for management prevention.

Language: Английский

---

Supervised and unsupervised machine learning approaches using Sentinel data for flood mapping and damage assessment in Mozambique

Remote Sensing Applications Society and Environment, Journal Year: 2023, Volume and Issue: 32, P. 101015 - 101015

Published: June 22, 2023

Natural hazards, such as flooding, have been negatively impacting developed and emerging economies alike. The effects of floods are more prominent in countries the Global South, where large parts population infrastructure insufficiently protected from natural hazards. From this scope, a lot effort is required to mitigate these impacts by continuously providing new reliable tools aid mitigation preparedness, during or after flood event. Flood mapping followed damage assessment plays an important role all stages. In work we investigate dataset provided DrivenData Labs based on Sentinel-1 (S1) imagery (VH, VV labels) help map city Beira Mozambique. Exploiting Google Earth Engine (GEE), deployed supervised unsupervised machine learning (ML) methods comprising 13 worldwide. We first mapped country-by-country including This part was helpful understand sensitivity each method when applied data different regions with polarizations. then trained model globally (in countries) used it predict Beira. To assess accuracy experiments intersection over union (IoU) metric, results which compared benchmark IoU achieved winner competition for 2021. implementation ML using VH VV+VH produced satisfactory results, showed be better than imagery; Cambodia Bolivia polarization yielded IoUs values ranging 0.819 0.856 above (0.8094). predictions 0.568, reasonable outcome. proposed approach alternative mapping, especially Mozambique due its low cost time effectiveness even approaches, relatively high-quality near real-time. Finally, Sentinel-2 (S2) land cover classification perform integrated enhance quality results. show that 20% agricultural area about 10% built up were flooded. Flooded includes highly populated neighborhoods Chaimite Ponta Gea located center city.

Language: Английский

---

Robustness of machine learning algorithms to generate flood susceptibility maps for watersheds in Jordan

Geomatics Natural Hazards and Risk, Journal Year: 2024, Volume and Issue: 15(1)

Published: July 20, 2024

The study examined three machine learning algorithms (MLAs): random forest (RF), support vector (SVM), and artificial neural networks (ANN) for generating flood susceptibility maps in two watersheds Jordan. Both were selected because they represent climatic regimes: desert mountainous areas. Because of a shortage past floods location, physical model was utilized to generate them based on simulations 100-year rainfall. 10,000 randomly used MLAs training testing. During training, thirteen influential factors identified. Out them, the distance stream, elevation, topographic wetness index have shown an overwhelming effect Zarqa Ma'in watershed (they gained 50% IGR), while stream density, elevation Al-Buaida 44% IGR). For mapping, RF outperformed other both thus mapping. classified into five classes, 11% fell high very 5.2% within these classes. In conclusion, able produce efficiently, can form alternative modeling.

Language: Английский

---