Sharbaf et al. (2025) Rain gauge network design via implementation of global sensitivity analysis coupled with geostatistics and principal component analysis

Identification

• Journal: Stochastic Environmental Research and Risk Assessment
• Year: 2025
• Date: 2025-10-17
• Authors: Mohammad Ali Mohammad Jafar Sharbaf, Mohammad Javad Abedini
• DOI: 10.1007/s00477-025-03115-9

Research Groups

• Department of Civil and Environmental Engineering, School of Engineering, Shiraz University, Shiraz, Iran

Short Summary

This study introduces a novel rain gauge network design methodology that integrates global sensitivity analysis (variance decomposition) with geostatistics (Block Ordinary Kriging) and Principal Component Analysis. The proposed approach prioritizes stations based on their contribution to the uncertainty of mean annual areal rainfall estimates, demonstrating comparable performance to existing variance-minimization methods while offering computational efficiency.

Objective

• To develop and evaluate a novel rain gauge network design methodology that prioritizes stations based on their contribution to the uncertainty of mean annual areal rainfall estimates, by coupling variance-based Global Sensitivity Analysis (GSA) with Block Ordinary Kriging (BOK) and Principal Component Analysis (PCA).

Study Configuration

• Spatial Scale: Southwestern Iran (Khouzestan and Kohkiloyeh-Bouyerahmad provinces), spanning approximately 25,000 square kilometers, geographically demarcated by longitudes 49°17′ to 51°22′ east and latitudes 30°2′ to 31°56′ north. The study area was discretized into a 5 km x 5 km grid.
• Temporal Scale: Monthly rainfall data spanning at least a decade, focused on annual time scale for areal estimation of mean annual rainfall.

Methodology and Data

• Models used:
  • Global Sensitivity Analysis (GSA) using variance decomposition (FAST method).
  • Geostatistical modeling: Block Ordinary Kriging (BOK), Variogram modeling (Exponential model with variance of 37,511 mm² and effective range of 206,991 m).
  • Principal Component Analysis (PCA).
  • Cluster Analysis (k-means algorithm).
  • Comparative analysis with: Time-consuming approach, Bastin’s simplified approach, Kassim-Kottegoda’s simplified approach, and Artificial Bee Colony (ABC) optimization coupled with geostatistics.
• Data sources:
  • Monthly rainfall data from 34 rain gauge stations in Southwestern Iran, spanning at least a decade.
  • Universal Transverse Mercator (UTM) coordinates, elevation (m), and mean annual rainfall (mm) for each station.

Main Results

• Principal Component Analysis (PCA) revealed significant redundancy in the existing network, indicating that 8 to 9 stations could capture over 80% of the total variance in observations.
• The proposed method, which couples GSA (variance decomposition) with BOK, PCA, and cluster analysis (prioritizing stations with the highest variance decomposition index), demonstrated performance comparable to established variance-minimization techniques for achieving a desired information level (e.g., 80% variance capture).
• GSA identified stations 7, 21, 24, 28, 30, and 31 as significant contributors to response surface uncertainty, while stations 1, 2, 6, 8, 9, 10, 12, 23, and 27 were deemed non-important.
• An approximate inverse relationship was observed between the GSA index of selected stations and the associated variance of residuals, suggesting that stations with a more substantial influence on response surface variability tend to exhibit lower residual variance for a given information level.
• Eliminating non-influential stations (e.g., 9 stations contributing only 1.5% to overall uncertainty) significantly reduced computational costs, particularly for exhaustive search methods.
• The proposed method achieved greater accuracy when clustering was based solely on spatial coordinates rather than combining location and rainfall data.

Contributions

• First implementation of variance-based Global Sensitivity Analysis (GSA) (specifically variance decomposition) in the domain of rain gauge network design.
• Introduction of a novel criterion for selecting representative stations within cluster analysis (maximum variance decomposition index) when coupled with PCA, offering an alternative to traditional criteria like maximum average annual rainfall.
• Development of a computationally efficient approach for optimal rain gauge network design, particularly valuable for intermediate data set sizes where the "curse of dimensionality" makes exhaustive searches impractical.
• Provides a new framework for prioritizing rain gauge stations based on their individual contribution to the uncertainty of model output, complementing existing methods focused on minimizing residual variance or maximizing information content.

Funding

• This manuscript was developed autonomously, without reliance on any external funding or grant.

Citation

@article{Sharbaf2025Rain,
  author = {Sharbaf, Mohammad Ali Mohammad Jafar and Abedini, Mohammad Javad},
  title = {Rain gauge network design via implementation of global sensitivity analysis coupled with geostatistics and principal component analysis},
  journal = {Stochastic Environmental Research and Risk Assessment},
  year = {2025},
  doi = {10.1007/s00477-025-03115-9},
  url = {https://doi.org/10.1007/s00477-025-03115-9}
}