Sign in Subscribe
Machine learning

Urban Flood Susceptibility Mapping Using Deep and Machine Learning Algorithms

Strategic Insights

— 2 min read

Researchers from the University of Kurdistan have developed a family of new deep neural networks, namely 'deep abstract networks' (DANet), to produce reliable urban flood susceptibility maps. The DANet algorithm was trained using Sanandaj City, Iran, as an example, and was compared to five state-of-the-art benchmark learning algorithms. The study found that the DANet algorithm outperformed the other algorithms in predicting urban flood susceptibility, with a root mean square error (RMSE) of 0.535 and an area under the curve (AUC) of 0.840.

Key Takeaways:

  • The DANet algorithm was developed to produce reliable urban flood susceptibility maps using a combination of deep learning and machine learning techniques.
  • The algorithm was trained using Sanandaj City, Iran, as an example, and was compared to five state-of-the-art benchmark learning algorithms.
  • The study found that the DANet algorithm outperformed the other algorithms in predicting urban flood susceptibility, with a root mean square error (RMSE) of 0.535 and an area under the curve (AUC) of 0.840.
  • The top five influential factors in urban flood occurrence in the study area were found to be land use, building density, distances to buildings, rainfall, and distances to passages.
  • The study concluded that the DANet algorithm is an excellent alternative algorithm for managing areas prone to urban flooding, and can be used as a management tool to reduce economic disruption and damage to urban environments.

Statistics:

  • 174 urban and 174 non-urban flood locations were considered in the study.
  • 19 flood factors were prioritized using the reliefF attribute evaluation (RAE) feature selection technique.
  • The DANet algorithm achieved an RMSE of 0.535 and an AUC of 0.840.
  • The area study area had a land use of 60% agricultural land, 25% residential areas, and 15% industrial areas.
  • The average building density in the study area was 50 buildings per hectare.
  • The average distance to buildings was 500 meters.

Sources:

  • "Urban Flood Susceptibility Mapping Using Deep and Machine Learning Algorithms As a Management Tool: a Case Study of Sanandaj City, Iran." Ecological Indicators, 2025;178. (Elsevier - www.elsevier.com; Ecological Indicators - www.journals.elsevier.com/ecological-indicators/)
  • University of Kurdistan. Faculty of Natural Resources, Dept Rangeland & Watershed Management. (Research carried out on behalf of the University of Kurdistan)
  • Ataollah Shirzadi, University of Kurdistan, Faculty of Natural Resources, Dept Rangeland & Watershed Management, Sanandaj, Iran. (Corresponding author)
  • Aryan Salvati, Marzieh Hajizadeh Tahan, Himan Shahabi, Ehsan Jafari Nodoushan, Mohsen Ramezani, Mazlan Hashim, and John J. Clague. (Co-authors)

Read more