Chaturvedi et al. (2026) Harnessing Deep Learning Methods for Rainfall Downscaling and Projections Over India: A CMIP6 Based Assessment of Future Shifts

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

Identification

• Journal: International Journal of Climatology
• Year: 2026
• Date: 2026-09-20
• Authors: Manisha Chaturvedi, Anjali, Mrinalini Srivastava, Prashant K. Srivastava, Rajesh Kumar Mall
• DOI: 10.1002/joc.70598

Research Groups

Not specified in the provided text.

Short Summary

The study employs BiLSTM and GRU deep learning models to downscale CMIP6 precipitation data over India, demonstrating that BiLSTM significantly improves the spatial fidelity of rainfall projections and predicts a future shift of monsoon rainfall toward southern India.

Objective

• To address the scale mismatch between global climate models (GCMs) and localized hydroclimatic variability by applying deep learning methods to downscale daily precipitation projections over India.

Study Configuration

• Spatial Scale: Regional (India, specifically highlighting the Western Ghats, Northeast India, the Western Himalayas, and central India).
• Temporal Scale: Historical period (1951–2010) and future projection periods (2040–2069 and 2070–2100).

Methodology and Data

• Models used: Bidirectional Long Short-Term Memory (BiLSTM), Gated Recurrent Unit (GRU), and 12 CMIP6 global climate models.
• Data sources: Daily precipitation data from CMIP6 under SSP2-4.5 and SSP5-8.5 scenarios.

Main Results

• Model Performance: BiLSTM outperformed GRU in reproducing observed rainfall, achieving a correlation coefficient ($r$) of 0.85–0.94, RMSE $\le 4.5$, and MAE between 1.16 and 1.63.
• Variability Capture: The models effectively captured annual variability (~14 mm) and monsoon (JJAS) intensity (~24 mm) while reducing systematic overestimation inherent in raw CMIP6 data.
• Future Projections (SSP5-8.5, 2070–2100):
  • Precipitation increases of ~20% in western India.
  • Precipitation increases of ~40% in southern India, the west coast, and the western Himalayas.
  • Precipitation reductions approaching ~50% in central India.
• General Trend: A coherent shift of monsoonal rainfall toward southern India is projected under both SSP scenarios, indicating increased spatial polarization.

Contributions

• Provides a high-fidelity downscaling approach using deep learning that substantially improves the usability of CMIP6 projections for regional adaptation planning in monsoon-dependent regions.
• Identifies specific regional hydroclimatic contrasts and a southward shift in monsoon patterns that were previously obscured by the coarse resolution of GCMs.

Funding

Not specified in the provided text.

Citation

@article{Chaturvedi2026Harnessing,
  author = {Chaturvedi, Manisha and Anjali and Srivastava, Mrinalini and Srivastava, Prashant K. and Mall, Rajesh Kumar},
  title = {Harnessing Deep Learning Methods for Rainfall Downscaling and Projections Over India: A CMIP6 Based Assessment of Future Shifts},
  journal = {International Journal of Climatology},
  year = {2026},
  doi = {10.1002/joc.70598},
  url = {https://doi.org/10.1002/joc.70598}
}