Lotfi et al. (2026) Integrating deep learning, drought indices, and CMIP6 climate scenarios for future drought prediction across Iran

Identification

• Journal: Results in Engineering
• Year: 2026
• Date: 2026-09-20
• Authors: Sara Lotfi, Jafar Masoompour Samakosh, Kobra Soltani, Khabat Khosravi
• DOI: 10.1016/j.rineng.2026.113121

Research Groups

• Department of Geography, Faculty of Literature and Humanities, Razi University, Kermanshah, Iran
• Department of Natural Resources, College of Agriculture and Natural Resources, Razi University, Kermanshah, Iran

Short Summary

This study developed a sensitivity-driven framework to assess drought prediction performance across different climatic zones in Iran using observations from 67 synoptic stations, CMIP6 models, and near-future projections under SSP2–4.5 and SSP5–8.5 scenarios.

Objective

• To evaluate the relative performance of SPI and SPEI indices for drought characterization and monitoring in arid and semi-arid regions.
• To investigate the potential of deep-learning architectures to improve drought-index reconstruction from CMIP6 climate projections.
• To assess the impact of climatic heterogeneity on drought prediction performance and future drought characteristics.

Study Configuration

• Spatial Scale: National scale, covering approximately 1648,195 km² in Iran.
• Temporal Scale: Historical period (1992–2014) and near-future projections (2026–2055).

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) networks, feedforward Dense networks, and conventional deterministic calculations.
• Data sources: CMIP6 climate models, historical meteorological observations from 67 synoptic stations across Iran.

Main Results

• The LSTM-12/64 model achieved the highest overall SPEI performance (NSE=0.887; NRMSE=0.370).
• SPI outperformed LSTM configurations for precipitation-based Standardized Precipitation Index (SPI) with an NSE of approximately 0.42.
• Future projections indicate progressive drought intensification under both scenarios, with greater increases in drought frequency, event count, and duration under SSP5–8.5.

Contributions

• This study provides a comprehensive assessment of drought prediction performance across different climatic zones in Iran using a sensitivity-driven framework.
• The results highlight the importance of considering climatic heterogeneity when evaluating drought indices and predictive models.
• The study demonstrates the potential of deep-learning architectures to improve drought-index reconstruction from CMIP6 climate projections.

Funding

• This research was funded by [project name], [program name], and [reference code].

Citation

@article{Lotfi2026Integrating,
  author = {Lotfi, Sara and Samakosh, Jafar Masoompour and Soltani, Kobra and Khosravi, Khabat},
  title = {Integrating deep learning, drought indices, and CMIP6 climate scenarios for future drought prediction across Iran},
  journal = {Results in Engineering},
  year = {2026},
  doi = {10.1016/j.rineng.2026.113121},
  url = {https://doi.org/10.1016/j.rineng.2026.113121}
}