Deep learning methods for flood mapping: a review of existing applications and future research directions

Hydrology and earth system sciences, Journal Year: 2022, Volume and Issue: 26(16), P. 4345 - 4378

Published: Aug. 25, 2022

Abstract. Deep learning techniques have been increasingly used in flood management to overcome the limitations of accurate, yet slow, numerical models and improve results traditional methods for mapping. In this paper, we review 58 recent publications outline state art field, identify knowledge gaps, propose future research directions. The focuses on type deep various mapping applications, types considered, spatial scale studied events, data model development. show that based convolutional layers are usually more as they leverage inductive biases better process characteristics flooding events. Models fully connected layers, instead, provide accurate when coupled with other statistical models. showed increased accuracy compared approaches speed methods. While there exist several applications susceptibility, inundation, hazard mapping, work is needed understand how can assist real-time warning during an emergency it be employed estimate risk. A major challenge lies developing generalize unseen case studies. Furthermore, all reviewed their outputs deterministic, limited considerations uncertainties outcomes probabilistic predictions. authors argue these identified gaps addressed by exploiting fundamental advancements or taking inspiration from developments applied areas. graph neural networks operators arbitrarily structured thus should capable generalizing across different studies could account complex interactions natural built environment. Physics-based preserve underlying physical equations resulting reliable speed-up alternatives Similarly, resorting Gaussian processes Bayesian networks.

Language: Английский

---

Urban flood modeling using deep-learning approaches in Seoul, South Korea

Journal of Hydrology, Journal Year: 2021, Volume and Issue: 601, P. 126684 - 126684

Published: July 18, 2021

Identification of flood-prone sites in urban environments is necessary, but there insufficient hydraulic information and time series data on surface runoff. To date, several attempts have been made to apply deep-learning models for flood hazard mapping areas. This study evaluated the capability convolutional neural network (NNETC) recurrent (NNETR) mapping. A flood-inundation inventory (including 295 flooded sites) was used as response variable 10 flood-affecting factors were considered predictor variables. Flooded then spatially randomly split a 70:30 ratio building validation purposes. The prediction quality validated using area under receiver operating characteristic curve (AUC) root mean square error (RMSE). results indicated that performance NNETC model (AUC = 84%, RMSE 0.163) slightly better than NNETR 82%, 0.186). Both terrain ruggedness index most important predictor, followed by slope elevation. Although output had relative up 20% (based AUC), this modeling approach could still be reliable rapid tool generate map areas, provided inundation available.

Language: Английский

---

U-FLOOD – Topographic deep learning for predicting urban pluvial flood water depth

Journal of Hydrology, Journal Year: 2021, Volume and Issue: 603, P. 126898 - 126898

Published: Sept. 4, 2021

This study investigates how deep-learning can be configured to optimise the prediction of 2D maximum water depth maps in urban pluvial flood events. A neural network model is trained exploit patterns hyetographs as well topographical data, with specific aim enabling fast predictions depths for observed rain events and spatial locations that have not been included training dataset. architecture widely used image segmentation (U-NET) adapted this purpose. Key novelties are a systematic investigation which inputs should provided deep learning model, hyper-parametrization optimizes predictive performance, evaluation performance were considered training. We find input dataset only 5 variables describe local terrain shape imperviousness optimal generate depth. Neural architectures between 97,000 260,000,000 parameters tested, 28,000,000 found optimal. U-FLOOD demonstrated yield similar existing screening approaches, even though assessment performed natural unknown network, generated within seconds. Improvements likely obtained by ensuring balanced representation temporal rainfall dataset, further improved datasets, linking dynamic sewer system models.

Language: Английский

---

Multi-hazard susceptibility mapping based on Convolutional Neural Networks

Geoscience Frontiers, Journal Year: 2022, Volume and Issue: 13(5), P. 101425 - 101425

Published: June 17, 2022

Multi-hazard susceptibility prediction is an important component of disasters risk management plan. An effective multi-hazard mitigation strategy includes assessing individual hazards as well their interactions. However, with the rapid development artificial intelligence technology, techniques based on machine learning has encountered a huge bottleneck. In order to effectively solve this problem, study proposes mapping framework using classical deep algorithm Convolutional Neural Networks (CNN). First, we use historical flash flood, debris flow and landslide locations Google Earth images, extensive field surveys, topography, hydrology, environmental data sets train validate proposed CNN method. Next, method assessed in comparison conventional logistic regression k-nearest neighbor methods several objective criteria, i.e., coefficient determination, overall accuracy, mean absolute error root square error. Experimental results show that outperforms algorithms predicting probability floods, flows landslides. Finally, maps three are combined create map. It can be observed from map 62.43% area prone hazards, while 37.57% harmless. hazard-prone areas, 16.14%, 4.94% 30.66% susceptible landslides, respectively. terms concurrent 0.28%, 7.11% 3.13% joint occurrence floods flow, respectively, whereas, 0.18% subject all hazards. The benefit engineers, disaster managers local government officials involved sustainable land mitigation.

Language: Английский

---

A review of recent advances in urban flood research

Water Security, Journal Year: 2023, Volume and Issue: 19, P. 100141 - 100141

Published: July 13, 2023

Due to a changing climate and increased urbanization, an escalation of urban flooding occurrences its aftereffects are ever more dire. Notably, the frequency extreme storms is expected increase, as built environments impede absorption water, threat loss human life property damages exceeding billions dollars heightened. Hence, agencies organizations implementing novel modeling methods combat consequences. This review details concepts, impacts, causes flooding, along with associated endeavors. Moreover, this describes contemporary directions towards flood resolutions, including recent hydraulic-hydrologic models that use modern computing architecture trending applications artificial intelligence/machine learning techniques crowdsourced data. Ultimately, reference utility provided, scientists engineers given outline advances in research.

Language: Английский

---

Flood Susceptibility Assessment with Random Sampling Strategy in Ensemble Learning (RF and XGBoost)

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(2), P. 320 - 320

Published: Jan. 12, 2024

Due to the complex interaction of urban and mountainous floods, assessing flood susceptibility in areas presents a challenging task environmental research risk analysis. Data-driven machine learning methods can evaluate lacking essential hydrological data, utilizing remote sensing data limited historical inundation records. In this study, two ensemble algorithms, Random Forest (RF) XGBoost, were adopted assess Kunming, typical area prone severe disasters. A inventory was created using observations from 2018 2022. The spatial database included 10 explanatory factors, encompassing climatic, geomorphic, anthropogenic factors. Artificial Neural Network (ANN) Support Vector Machine (SVM) selected for model comparison. To minimize influence expert opinions on training, study employed strategy uniformly random sampling historically non-flooded negative sample selection. results demonstrated that (1) algorithms offer higher accuracy than other methods, with RF achieving highest accuracy, evidenced by an under curve (AUC) 0.87, followed XGBoost at 0.84, surpassing both ANN (0.83) SVM (0.82); (2) interpretability highlighted differences potential distribution training data’s positive samples. Feature importance be utilized human bias collection flooded-site samples, more targeted maps area’s road network obtained; (3) exhibited greater stability robustness datasets varied as their performance F1-Score, Kappa, AUC metrics. This paper further substantiates superiority assessment tasks perspectives interpretability, robustness, enhances understanding impact samples such assessments, optimizes specific process data-driven methods.

Language: Английский

---

Toward Explainable Flood Risk Prediction: Integrating A Novel Hybrid Machine Learning Model

Sustainable Cities and Society, Journal Year: 2025, Volume and Issue: unknown, P. 106140 - 106140

Published: Jan. 1, 2025

Language: Английский

---

A review of flood risk assessment frameworks and the development of hierarchical structures for risk components

Discover Water, Journal Year: 2025, Volume and Issue: 5(1)

Published: Feb. 12, 2025

Language: Английский

---

UAVs in Disaster Management: Application of Integrated Aerial Imagery and Convolutional Neural Network for Flood Detection

Sustainability, Journal Year: 2021, Volume and Issue: 13(14), P. 7547 - 7547

Published: July 6, 2021

Floods have been a major cause of destruction, instigating fatalities and massive damage to the infrastructure overall economy affected country. Flood-related devastation results in loss homes, buildings, critical infrastructure, leaving no means communication or travel for people stuck such disasters. Thus, it is essential develop systems that can detect floods region provide timely aid relief stranded people, save their livelihoods, protect key city infrastructure. Flood prediction warning implemented developed countries, but manufacturing cost too high developing countries. Remote sensing, satellite imagery, global positioning system, geographical information are currently used flood detection assess flood-related damages. These techniques use neural networks, machine learning, deep learning methods. However, unmanned aerial vehicles (UAVs) coupled with convolution networks not explored these contexts instigate swift disaster management response minimize Accordingly, this paper uses UAV-based imagery as method based on Convolutional Neural Network (CNN) extract features from images zone. This effective assessing local infrastructures zones. The study area flood-prone Indus River Pakistan, where both pre-and post-disaster collected through UAVs. For training phase, 2150 image patches created by resizing cropping source images. dataset train CNN model regions change has occurred. tested against validate it, which positive an accuracy 91%. Disaster organizations damages other assets worldwide proper responses help smart governance cities all emergent disasters addressed promptly.

Language: Английский

---

Improving urban flood susceptibility mapping using transfer learning

Journal of Hydrology, Journal Year: 2021, Volume and Issue: 602, P. 126777 - 126777

Published: Aug. 3, 2021

Language: Английский

---