Afrin et al. (2026) Uncertainty Analysis of Artificial Neural Network based Regional Flood Modelling in New South Wales, Australia

Identification

• Journal: Lecture notes in civil engineering
• Year: 2026
• Date: 2026-01-01
• Authors: Nilufa Afrin, Sadia T. Mim, Ataur Rahman, Khaled Haddad
• DOI: 10.1007/978-3-032-18708-6_3

Research Groups

• Western Sydney University, Sydney, Australia
• EnviroWater Sydney, Sydney, Australia

Short Summary

This study conducts an in-depth uncertainty analysis of Artificial Neural Network (ANN)-based Regional Flood Frequency Analysis (RFFA) using data from 88 gauged stations in New South Wales, Australia, finding that ANN performance and uncertainty vary with return period, with mid-range quantiles showing lower uncertainty.

Objective

• To perform an in-depth uncertainty analysis of Artificial Neural Network (ANN)-based Regional Flood Frequency Analysis (RFFA) in New South Wales, Australia, and quantify the uncertainty of modelled design flood quantiles.

Study Configuration

• Spatial Scale: 88 gauged stations across New South Wales, Australia.
• Temporal Scale: Modelling of six design flood quantiles (Q2, Q5, Q10, Q20, Q50, Q100), implying analysis over historical flood records to derive these frequencies.

Methodology and Data

• Models used: Artificial Neural Network (ANN) for Regional Flood Frequency Analysis (RFFA). Monte Carlo simulation technique with random 70/30 train–test splits for robustness testing and uncertainty quantification.
• Data sources: Data from 88 gauged stations in New South Wales, Australia, utilizing eight hydrological and physiographical predictors.

Main Results

• The median relative error ratio (REr) for the modelled flood quantiles ranged from 54.13% to 59.81%.
• Specific REr values were: 59.62 ± 10.73% (Q2), 55.86 ± 10.6% (Q5), 54.13 ± 10.76% (Q10), 55.86 ± 11.44% (Q20), 57.72 ± 12.37% (Q50), and 59.81 ± 10.58% (Q100).
• ANN performance and associated uncertainty varied with the return period, with mid-range quantiles (e.g., Q10, Q20) generally achieving lower uncertainty.

Contributions

• Provides a comprehensive uncertainty analysis of ANN-based RFFA specifically within the Australian context.
• Demonstrates that the performance and uncertainty of ANN models in RFFA are dependent on the flood return period.
• Reinforces the critical importance of conducting uncertainty analysis before applying AI-driven RFFA models to ungauged catchments.

Funding

• Not specified in the provided text.

Citation

@article{Afrin2026Uncertainty,
  author = {Afrin, Nilufa and Mim, Sadia T. and Rahman, Ataur and Haddad, Khaled},
  title = {Uncertainty Analysis of Artificial Neural Network based Regional Flood Modelling in New South Wales, Australia},
  journal = {Lecture notes in civil engineering},
  year = {2026},
  doi = {10.1007/978-3-032-18708-6_3},
  url = {https://doi.org/10.1007/978-3-032-18708-6_3}
}