Scholarlink

<p>Practical flood-risk assessment requires methods that represent both baseline flooding and amplification from levees, berms, road embankments, and other discontinuities controlling floodplain connectivity. This study presents FARHAD, a cloud-native machine-learning workflow for weakline-aware flood-risk mapping, demonstrated in two Iranian basins: Aqqala and Poldokhtar. Open geospatial and Earth-observation data were processed in Google Earth Engine to derive terrain, hydrological, hydrographic, land-surface, exposure, and vulnerability predictors. An Extreme Gradient Boosting (XGBoost) depth surrogate was trained against GLOFAS-derived design-event targets to produce 30&#xa0;m depth rasters for 30-, 100-, 300-, and 1000-year average recurrence intervals. Pixel-wise cross-validation showed high internal agreement, while spatial-block cross-validation indicated more conservative transferability, supporting interpretation as screening-level products rather than independently validated hydraulic simulations. Hazard intensity was characterised using depth and a derived Depth × Velocity proxy, then classified using the Swiss FOEN danger concept and the Australian Disaster Resilience Handbook 7. Hazard layers were combined with gridded population, building footprints, and a composite vulnerability index using Risk = Hazard × Exposure × Vulnerability to generate continuous and classified risk maps. Weaklines were delineated from machine-learning-derived flood extents and terrain context, and closure configurations were ranked using a depth-weighted exposure index. Results show that weakline effects differ by basin morphology. Dense weakline clusters in mountainous Poldokhtar produced corridor-bound increases in upper hazard and risk tiers, whereas selected closures in low-gradient Aqqala generated localised escalation near the urban fringe with limited basin-wide footprint change. FARHAD therefore provides a reproducible screening framework for comparing Baseline and conservative weakline-closure risk patterns in data-scarce catchments.</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

Fully Automated Risk and Hazard Assessment for Decision-support (FARHAD): A Cloud-Native Machine-Learning Framework for Worst-Case Scenario Flood-Risk Mapping via Weakline Analysis

  • 期刊论文

  • 发表日期：

    2026年6月19日

查看全文
  • Afshin Fouladi Semnan,
  • Mohammad Javad Ostad Mirza Tehrani

摘要

Practical flood-risk assessment requires methods that represent both baseline flooding and amplification from levees, berms, road embankments, and other discontinuities controlling floodplain connectivity. This study presents FARHAD, a cloud-native machine-learning workflow for weakline-aware flood-risk mapping, demonstrated in two Iranian basins: Aqqala and Poldokhtar. Open geospatial and Earth-observation data were processed in Google Earth Engine to derive terrain, hydrological, hydrographic, land-surface, exposure, and vulnerability predictors. An Extreme Gradient Boosting (XGBoost) depth surrogate was trained against GLOFAS-derived design-event targets to produce 30 m depth rasters for 30-, 100-, 300-, and 1000-year average recurrence intervals. Pixel-wise cross-validation showed high internal agreement, while spatial-block cross-validation indicated more conservative transferability, supporting interpretation as screening-level products rather than independently validated hydraulic simulations. Hazard intensity was characterised using depth and a derived Depth × Velocity proxy, then classified using the Swiss FOEN danger concept and the Australian Disaster Resilience Handbook 7. Hazard layers were combined with gridded population, building footprints, and a composite vulnerability index using Risk = Hazard × Exposure × Vulnerability to generate continuous and classified risk maps. Weaklines were delineated from machine-learning-derived flood extents and terrain context, and closure configurations were ranked using a depth-weighted exposure index. Results show that weakline effects differ by basin morphology. Dense weakline clusters in mountainous Poldokhtar produced corridor-bound increases in upper hazard and risk tiers, whereas selected closures in low-gradient Aqqala generated localised escalation near the urban fringe with limited basin-wide footprint change. FARHAD therefore provides a reproducible screening framework for comparing Baseline and conservative weakline-closure risk patterns in data-scarce catchments.