Toh et al. (2026) Enhanced IMERG SPE Using LSTM with a Novel Adaptive Regularization Method

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

Identification

• Journal: Water
• Year: 2026
• Date: 2026-04-10
• Authors: Seng Choon Toh, Wan Zurina Wan Jaafar, Cia Yik Ng, Eugene Zhen Xiang Soo, Majid Mirzaei, Fang Yenn Teo, S. M. F. LAI
• DOI: 10.3390/w18080905

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study develops an Adaptive Regularization framework for an LSTM model to improve satellite-gauge rainfall fusion. It dynamically adjusts learning rate and weight decay, demonstrating superior performance in refining daily IMERG-Final rainfall estimates over flood-prone Peninsular Malaysia.

Objective

• To enhance satellite-gauge rainfall fusion by developing an Adaptive Regularization framework integrated within a Long Short-Term Memory (LSTM) model, aiming to improve the reliability of satellite-based precipitation estimates (SPE) in flood-prone regions.

Study Configuration

• Spatial Scale: Regional scale, specifically the flood-prone east coast of Peninsular Malaysia.
• Temporal Scale: Daily.

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) model integrated with an Adaptive Regularization framework.
• Data sources: Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG-Final) daily rainfall estimates (satellite-based). Gauge observations are implicitly used for fusion and validation.

Main Results

• The Adaptive Regularization framework consistently outperformed ten benchmark optimization algorithms.
• Achieved a Mean Absolute Error (MAE) of 6.87 mm/day, a Correlation Coefficient (CC) of 0.68, a Normalized Root Mean Squared Error (NRMSE) of 1.84, and a Kling–Gupta Efficiency (KGE) of 0.56.
• Enhanced spatial consistency and robustness of rainfall estimates across monsoon seasons.

Contributions

• Introduced an Adaptive Regularization framework for LSTM models to improve satellite-gauge rainfall fusion, moving beyond conventional optimization strategies.
• Developed a dynamic adjustment mechanism for learning rate and weight decay based on validation performance and overfitting indicators, leading to improved training stability, data efficiency, and model generalization.
• Provided a scalable solution for enhancing satellite-based precipitation estimates (SPE) specifically in flood-prone regions.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Toh2026Enhanced,
  author = {Toh, Seng Choon and Jaafar, Wan Zurina Wan and Ng, Cia Yik and Soo, Eugene Zhen Xiang and Mirzaei, Majid and Teo, Fang Yenn and LAI, S. M. F.},
  title = {Enhanced IMERG SPE Using LSTM with a Novel Adaptive Regularization Method},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18080905},
  url = {https://doi.org/10.3390/w18080905}
}