A Transfer Learning Approach Based on Radar Rainfall for River Water-Level Prediction DOI Open Access

Futo Ueda,

Hiroto Tanouchi,

Nobuyuki EGUSA

et al.

Water, Journal Year: 2024, Volume and Issue: 16(4), P. 607 - 607

Published: Feb. 18, 2024

River water-level prediction is crucial for mitigating flood damage caused by torrential rainfall. In this paper, we attempt to predict river water levels using a deep learning model based on radar rainfall data instead of from upstream hydrological stations. A incorporating two-dimensional convolutional neural network (2D-CNN) and long short-term memory (LSTM) constructed exploit geographical temporal features data, transfer method newly defined flow–distance matrix presented. The results our evaluation the Oyodo basin in Japan show that presented measurements has good accuracy case rain, with Nash–Sutcliffe efficiency (NSE) value 0.86 Kling–Gupta (KGE) 0.83 6-h-ahead forecast top-four peak height cases, which comparable conventional (NSE = 0.84 KGE 0.83). It also confirmed maintains its performance even when amount training site reduced; values NSE 0.82 were achieved reducing torrential-rain-period 12 3 periods (with 105 other rivers learning). demonstrate few rain at location potentially enable us if stations have not been installed location.

Language: Английский

Enhancing Flood Susceptibility Modeling: a Hybrid Deep Neural Network with Statistical Learning Algorithms for Predicting Flood Prone Areas DOI

Motrza Ghobadi,

Masumeh Ahmadipari

Water Resources Management, Journal Year: 2024, Volume and Issue: 38(8), P. 2687 - 2710

Published: March 18, 2024

Language: Английский

Citations

9
Multi-Criteria Assessment of Flood Risk on Railroads Using a Machine Learning Approach: A Case Study of Railroads in Minas Gerais DOI Creative Commons
Fernanda Oliveira de Sousa, Victor Andre Ariza Flores, Christhian Santana Cunha

et al.

Infrastructures, Journal Year: 2025, Volume and Issue: 10(1), P. 12 - 12

Published: Jan. 8, 2025

In a climate change scenario where extreme precipitation events occur more frequently and intensely, risk assessment plays critical role in ensuring the safety operational efficiency of facilities. This case study uses combination multi-criteria analysis approach hydrological studies that use machine learning algorithms to simulate new rainfall order estimate flooding on railroads. Risk variables, including terrain, drainage capability, accumulated flow, land cover, will be weighed using multicriteria approach. A methodical evaluation most vulnerable locations railroad network possible thanks these parameters based geographic information system (GIS) meantime, historical precipitation, balance data used calibrate validate models. The database required for model can created with data. research regions are situated densely rail-networked state Minas Gerais. geographical climatic diversity Gerais makes it perfect place test suggested approaches. models evaluated included linear regression, random forest, decision tree, support vector machines. Among models, Linear Regression emerged as best-performing an R2 value 0.999998, mean squared error (MSE) 0.018672, low tendency overfitting (0.000011).

Language: Английский

Citations

1
Improving flood hazard susceptibility assessment by integrating hydrodynamic modeling with remote sensing and ensemble machine learning DOI Creative Commons

Izhar Ahmad,

Rashid Farooq, Muhammad Ashraf

et al.

Natural Hazards, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 11, 2025

Abstract Floods are natural disasters with significant economic and infrastructural impacts. Assessing flood susceptibility in mountainous urban regions is particularly challenging due to the complicated interaction which structures terrain affect behavior. This study employs two ensemble machine learning algorithms, Extreme Gradient Boosting (XGBoost) Random Forest (RF), develop maps for Hunza-Nagar region, has been experiencing frequent flooding past three decades. An unsteady flow simulation carried out HEC-RAS utilizing a 100-year return period hydrograph as an input boundary condition, output of provided spatial inundation extents necessary developing inventory. Ten explanatory factors, including climatic, geological, geomorphological features namely elevation, slope, curvature, topographic wetness index (TWI), normalized difference vegetation (NDVI), land use cover (LULC), rainfall, lithology, distance roads rivers considered mapping. For inventory, random sampling technique adopted create repository non-flood points, incorporating ten geo-environmental conditioning factors. The models’ accuracy assessed using area under curve (AUC) receiver operating characteristics (ROC). prediction rate AUC values 0.912 RF 0.893 XGBoost, also demonstrating superior performance accuracy, precision, recall, F1-score, kappa evaluation metrics. Consequently, model selected represent map area. resulting will assist national disaster management infrastructure development authorities identifying high susceptible zones carrying early mitigation actions future floods.

Language: Английский

Citations

1
Review and Intercomparison of Machine Learning Applications for Short-term Flood Forecasting DOI Creative Commons

Muhammad Asif,

Monique M. Kuglitsch,

Ivanka Pelivan

et al.

Water Resources Management, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 14, 2025

Abstract Among natural hazards, floods pose the greatest threat to lives and livelihoods. To reduce flood impacts, short-term forecasting can contribute early warnings that provide communities with time react. This manuscript explores how machine learning (ML) support forecasting. Using two methods [strengths, weaknesses, opportunities, threats (SWOT) comparative performance analysis] for different forecast lead times (1–6, 6–12, 12–24, 24–48 h), we evaluate of models in 94 journal papers from 2001 2023. SWOT reveals best was produced by hybrid, random forest (RF), long memory (LSTM), artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS) approaches. The analysis, meanwhile, favors convolutional network, ANFIS, multilayer perceptron, k-nearest neighbors algorithm (KNN), LSTM, ANN, vector (SVM) at 1–6 h; LSTM 6–12 SVM, RF 12–24 hybrid h. In general, approaches consistently perform well across all times. Trends such as hybridization, model selection, input data decomposition seem improve accuracy models. Furthermore, effective stand-alone ML RF, genetic algorithm, KNN, better outcomes through hybridization other By including parameters environmental, socio-economical, climatic parameters, produce more accurate forecasting, making it warning operational purposes.

Language: Английский

Citations

1
Historical precipitation and flood damage in Japan: functional data analysis and evaluation of models DOI Creative Commons
Atsushi Wakai, Yasuaki Hijioka, Masayuki Yokozawa

et al.

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(2), P. e0318335 - e0318335

Published: Feb. 25, 2025

The future increase of large-scale weather disasters resulting from the increased frequency extreme events caused by climate change is a matter concern. Predicting flood damage through statistical analysis requires accurate modeling relationship between historical precipitation and damage. An that considers as time series may be appropriate for this purpose. Functional data was applied to model daily river basins in Kanto Koshin regions Japan. Flood statistics national government 1-km grid past National Agriculture Food Research Organization were used. models obtained functional more than those derived simple linear regression without considering precipitation. new also about four times estimating annual sum damage, compared each event. accuracy prediction higher recent years earlier study period (1993–2020). results showed influence on apparent years. This findings imply progress development project improvement structures along have indirectly affected levels associated with

Language: Английский

Citations

1
Coastal Flood Risk and Smart Resilience Evaluation under a Changing Climate DOI Creative Commons

Ping Shen,

Shilan Wei,

Huabin Shi

et al.

Ocean-Land-Atmosphere Research, Journal Year: 2023, Volume and Issue: 2

Published: Jan. 1, 2023

Coastal areas are highly vulnerable to flood risks, which exacerbated by the changing climate. This paper provides a comprehensive review of literature on coastal risk assessment and resilience evaluation proposes smart-resilient city framework based pre-disaster, mid-disaster, post-disaster evaluations. First, this systematically reviews origin concept development resilience. Next, it introduces social-acceptable criteria level for different phases. Then, management system smart cities is proposed, covering 3 phases disasters (before, during, after). Risk essential in pre-disaster scenarios because understanding potential hazards vulnerabilities an area or system. Big data monitoring during component effective emergency response that can allow more informed decisions thus quicker, responses disasters, ultimately saving lives minimizing damage. Data-informed loss assessments crucial providing rapid, accurate impact. understanding, turn, instrumental expediting recovery reconstruction efforts aiding decision-making processes resource allocation. Finally, impacts climate change summarized. The resilient communities better equipped withstand adapt environmental conditions crucial. To address compound floods, researchers should focus trigging factor interactions, assessing economic social improving systems, promoting interdisciplinary research with openness. These strategies will enable holistic risks context change.

Language: Английский

Citations

22
A breakthrough in fast flood simulation DOI Creative Commons
Bastian van den Bout,

Victor Jetten,

C.J. van Westen

et al.

Environmental Modelling & Software, Journal Year: 2023, Volume and Issue: 168, P. 105787 - 105787

Published: Aug. 11, 2023

The current status of technological advancement does not allow to generate detailed spatial flood forecasts. This hinders warning-systems, interactive planning tools and Our novel method computes hazard maps over three orders magnitude faster than state-of-the-art methods. It applies physically-based principles steady-state flow evade dynamic aspects simulations. estimates the relevant information for hazard, such as peak height, velocity arrival time. Performance indicators show similar or exceeding accuracy compared traditional models depending on type data. In our tests, computation is reduced 1500 times. provides new perspective field hazards, risk reduction through types early-warning systems, user-interactive assessment systems. As climate change expected aggravate presented can bring efficiency simulation. freely available at www.fastflood.org.

Language: Английский

Citations

17
Rapid urban flood inundation forecasting using a physics-informed deep learning approach DOI

Fang Yang,

Ding Wu,

Jianshi Zhao

et al.

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 643, P. 131998 - 131998

Published: Sept. 20, 2024

Language: Английский

Citations

8
Deep learning in water protection of resources, environment, and ecology: achievement and challenges DOI
Xiaohua Fu, Jie Jiang,

Xie Wu

et al.

Environmental Science and Pollution Research, Journal Year: 2024, Volume and Issue: 31(10), P. 14503 - 14536

Published: Feb. 2, 2024

Language: Английский

Citations

6
Real-Time Urban Flood Depth Mapping: Convolutional Neural Networks for Pluvial and Fluvial Flood Emulation DOI
Maelaynayn El Baida, Farid Boushaba, Mimoun Chourak

et al.

Water Resources Management, Journal Year: 2024, Volume and Issue: 38(12), P. 4763 - 4782

Published: May 18, 2024

Language: Английский

Citations

6