Scholarlink

<p>Due to shifting climate patterns, it is expected that river flows will undergo alterations, and identifying flood behavior in data-poor regions is crucial. Therefore, accurate flood frequency analysis, which defines the probability of floods of varying magnitudes occurring within specified return periods, is essential for sustainable water resources management. This study assessed future flood frequency for the Haraz River Basin in northern Iran, using downscaled climate projections with machine learning (ML) models. To investigate future climate change, the LARS-WG weather generator and ML models were used to statistically downscale precipitation (pr), minimum temperature (tasmin), and maximum temperature (tasmax) from the CMIP6 dataset under two common socio-economic pathways (SSP1-2.6 and SSP5-8.5) for the period 2031-2050. For this purpose, this study uses advanced ML techniques, including Random Forest (RF), Gradient Boosting Regression Tree (GBRT), and Least Square Support Vector Regression-Particle Swarm Optimization (LSSVR-PSO) models, and their fusion, in addition to the LARS-WG model. The performance of the model fusion in precipitation, maximum temperature, and minimum temperature was obtained as R<sup>2</sup> = 0.76, R<sup>2</sup> = 0.94, and R<sup>2</sup> = 0.94, respectively. The downscaled variables were used as inputs to a multi-layer perceptron (MLP) model fusion to predict river flow, achieving R<sup>2</sup> = 0.84. For uncertainty analysis, the Model Agnostic Prediction Interval Estimator (MAPIE) method is applied, showing 95% prediction intervals. Sensitivity analysis showed that changes in precipitation have the greatest impact on river discharge, followed by changes in minimum and maximum temperatures. Flood frequency analysis was performed using a three-parameter Weibull distribution fitting the maximum daily discharges, which was the best fit compared to other distributions. Finally, flood frequency showed that maximum daily discharges have decreased, especially under the SSP5-8.5 scenarios. These findings provide valuable insights for regional flood management and adaptation planning in the context of climate change.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Flood Frequency Predictions Under Climate Change and Uncertainty Using Machine Learning Model Fusion and Wavelet Transform

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Kiana Yahyazadeh Shourabi,
  • Mohammad Hossein Niksokhan,
  • Mohammad Reza Nikoo

摘要

Due to shifting climate patterns, it is expected that river flows will undergo alterations, and identifying flood behavior in data-poor regions is crucial. Therefore, accurate flood frequency analysis, which defines the probability of floods of varying magnitudes occurring within specified return periods, is essential for sustainable water resources management. This study assessed future flood frequency for the Haraz River Basin in northern Iran, using downscaled climate projections with machine learning (ML) models. To investigate future climate change, the LARS-WG weather generator and ML models were used to statistically downscale precipitation (pr), minimum temperature (tasmin), and maximum temperature (tasmax) from the CMIP6 dataset under two common socio-economic pathways (SSP1-2.6 and SSP5-8.5) for the period 2031-2050. For this purpose, this study uses advanced ML techniques, including Random Forest (RF), Gradient Boosting Regression Tree (GBRT), and Least Square Support Vector Regression-Particle Swarm Optimization (LSSVR-PSO) models, and their fusion, in addition to the LARS-WG model. The performance of the model fusion in precipitation, maximum temperature, and minimum temperature was obtained as R2 = 0.76, R2 = 0.94, and R2 = 0.94, respectively. The downscaled variables were used as inputs to a multi-layer perceptron (MLP) model fusion to predict river flow, achieving R2 = 0.84. For uncertainty analysis, the Model Agnostic Prediction Interval Estimator (MAPIE) method is applied, showing 95% prediction intervals. Sensitivity analysis showed that changes in precipitation have the greatest impact on river discharge, followed by changes in minimum and maximum temperatures. Flood frequency analysis was performed using a three-parameter Weibull distribution fitting the maximum daily discharges, which was the best fit compared to other distributions. Finally, flood frequency showed that maximum daily discharges have decreased, especially under the SSP5-8.5 scenarios. These findings provide valuable insights for regional flood management and adaptation planning in the context of climate change.

Graphical Abstract