Patro et al. (2025) Collaborative Station Learning for Rainfall Forecasting

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Atmosphere
• Year: 2025
• Date: 2025-10-16
• Authors: Bagati Sudarsan Patro, Prashant Bartakke
• DOI: 10.3390/atmos16101197

Research Groups

[Information not provided in the paper text.]

Short Summary

This study proposes a novel framework combining geometry-based weather station selection with deep learning to enhance extreme rainfall predictions. The Bi-GRU model, utilizing a linear station topology, achieved the highest predictive accuracy (R2 = 0.9548, RMSE = 2.2120 mm) for real-time, location-specific early warning systems.

Objective

• To address the research gap in spatial configuration-aware modeling by proposing a novel framework that combines geometry-based weather station selection with advanced deep learning architectures to enhance the precision and reliability of extreme rainfall predictions using real-time data from Automatic Weather Stations.

Study Configuration

• Spatial Scale: Local, station-centric, focused on a target station (Chinchwad, Pune, India) and its surrounding Automatic Weather Stations configured in linear, triangular, quadrilateral, and circular topologies.
• Temporal Scale: Real-time prediction, enabling alerts up to 2 hours before extreme rainfall events.

Methodology and Data

• Models used: Six deep learning models were trained and assessed, including Bi-GRU and Transformer models.
• Data sources: Real-time data from Automatic Weather Stations, organized into twelve unique datasets based on four geometric topologies centered around the target station Chinchwad.

Main Results

• The proposed Bi-GRU model, when applied under a linear geometric topology, achieved the highest predictive accuracy with an R2 of 0.9548 and a Root Mean Square Error (RMSE) of 2.2120 mm.
• This performance significantly outperformed other deep learning models and geometric configurations tested.
• The Transformer model demonstrated poor generalization capabilities, indicated by high Mean Absolute Percentage Error (MAPE) values.
• The findings highlight the critical importance of both geometric topology in weather station networks and the chosen model architecture for improving extreme rainfall prediction accuracy.
• The developed framework can enable real-time, location-specific early warning systems, capable of issuing alerts 2 hours prior to extreme rainfall events.

Contributions

• Proposes a novel framework that integrates geometry-based weather station selection with advanced deep learning architectures for extreme rainfall prediction.
• Demonstrates the significant impact of geometric topology in Automatic Weather Station network design on prediction accuracy.
• Provides practical guidance for optimizing AWS network design, particularly in data-sparse regions.
• Enables the development of real-time, location-specific early warning systems for extreme rainfall, capable of issuing alerts 2 hours in advance.

Funding

[Information not provided in the paper text.]

Citation

@article{Patro2025Collaborative,
  author = {Patro, Bagati Sudarsan and Bartakke, Prashant},
  title = {Collaborative Station Learning for Rainfall Forecasting},
  journal = {Atmosphere},
  year = {2025},
  doi = {10.3390/atmos16101197},
  url = {https://doi.org/10.3390/atmos16101197}
}