Mahmoudi et al. (2025) Decoding Iran’s Drought Drivers: An Explainable AI Approach to Unraveling Global Teleconnection Impacts

Identification

• Journal: Earth Systems and Environment
• Year: 2025
• Date: 2025-11-18
• Authors: Peyman Mahmoudi, Pouria Jafari, Alireza Ghaemi, Jun Jian, Fatemeh Firoozi, Jing Yang
• DOI: 10.1007/s41748-025-00925-3

Research Groups

• Department of Physical Geography, Faculty of Geography and Environmental Planning, University of Sistan and Baluchestan, Zahedan, Iran
• Department of Electronic and Electrical Engineering, Faculty of Electrical and Computer Engineering, University of Sistan and Baluchestan, Zahedan, Iran
• Navigation College, Dalian Maritime University, Dalian, China
• Department of Humanities and Social Science, Farhangyan University, Tehran, Iran
• Faculty of Geographical Science, Key Laboratory of Environmental Change and Natural Disaster, Beijing Normal University, Beijing, China

Short Summary

This study utilized an Explainable Artificial Intelligence (XAI) approach, combining Random Forest and SHAP, to decode the complex, nonlinear, and lagged impacts of 36 global teleconnection indices on monthly meteorological droughts across Iran, identifying both known and lesser-known key regional drivers.

Objective

• To elucidate the nonlinear relationships between an extensive set of 36 global teleconnection indices and monthly meteorological drought variability across Iran using an Explainable Artificial Intelligence (XAI) approach.
• To identify and rank the most influential teleconnection indices affecting monthly drought variability in Iran, considering various time lags (0-month, 1-month, 2-month, and 3-month) on a station-specific basis.
• To generate spatial distribution maps of the impacts of these teleconnection indices to understand regional variations.

Study Configuration

• Spatial Scale: Iran (1,648,195 square kilometers, 25–40°N, 44–64°E), using data from 96 meteorological stations distributed across the country, regionalized into five distinct drought regions.
• Temporal Scale: 30-year period (1993–2022) for monthly meteorological drought analysis, investigating teleconnection impacts at 0-month (simultaneous), 1-month, 2-month, and 3-month lags.

Methodology and Data

• Models used: Random Forest (RF) as the primary predictive model, interpreted using Shapley Additive Explanations (SHAP). Other machine learning models evaluated for comparison include Artificial Neural Network (ANN-Tansig, ANN-ReLU), Support Vector Machine (SVM-Poly, SVM-RBF), and eXtreme Gradient Boosting (XGBoost). The Standardized Precipitation Index (SPI) was used for drought quantification.
• Data sources: Monthly precipitation records from 96 meteorological stations across Iran (1993–2022) provided by the Iran Meteorological Organization (IRIMO). 36 global teleconnection indices (1993–2022) from the Physical Sciences Laboratory (PSL) and the Climate Prediction Center (CPC) of NOAA, United States.

Main Results

• The Random Forest (RF) model demonstrated superior predictive performance for monthly Standardized Precipitation Index (SPI) values, achieving a coefficient of determination (R²) of 0.40 and outperforming other machine learning models (ANN, SVM, XGBoost) in 75% of the stations (p < 0.001).
• Iran was regionalized into five distinct homogeneous drought regions (Northwest, West and Southwest, South and Southeast, Center and Northeast, Southern Caspian Sea coast) based on monthly drought patterns.
• Drought characteristics vary regionally: Region C (South and Southeast) experiences longer but less severe droughts, while Regions A (Northwest) and B (West and Southwest) face shorter but more intense events.
• The relationships between teleconnection indices and drought are predominantly nonlinear and exhibit significant spatiotemporal instability, with no single pattern dominating the entire country.
• Key Teleconnection Drivers and Lags:
  • Simultaneous (0-month lag): Darwin (ENSO western pole) is the most influential (94/96 stations), followed by Tahiti (ENSO eastern pole) and Zonal Wind at 200 hectopascals (hPa) (ZWNDz200). High Darwin values (La Niña-like) and low Tahiti values (El Niño-like) correlate with drier conditions. High ZWNDz200 values correlate with drier conditions.
  • 1-month lag: Darwin and Tahiti remain highly significant. The Iceland Sea Level Pressure (SLP) index expands its influence, with high values (negative North Atlantic Oscillation (NAO) phase) associated with drier conditions.
  • 2-month lag: Atlantic patterns become dominant, with Iceland SLP and Atlantic Meridional Mode (AMM) being most influential. High AMM values (warmer northern tropical Atlantic) correlate with drier conditions. The Western Pacific 850 hPa Zonal Wind index (WPAC850) also shows significant impact.
  • 3-month lag: ENSO poles re-emerge, and ZWNDz200 becomes a vital, widespread driver. The Western Hemisphere Warm Pool (WHWP) index also assumes an important role, with low values linked to drier months.

Contributions

• Pioneering the application of Explainable Artificial Intelligence (XAI), specifically the integration of Random Forest with SHAP, to systematically decode complex, nonlinear, and lagged relationships between 36 global teleconnection indices and monthly meteorological droughts across Iran.
• Moving beyond traditional "black-box" machine learning models to provide physically interpretable insights into the magnitude and direction of influence of individual teleconnection indices on drought variability.
• Identifying the significant roles of previously lesser-known teleconnection patterns (e.g., ZWNDz200, WPAC850, WHWP, AMM) as key modulators of Iran’s drought variability, alongside established drivers like ENSO and NAO.
• Demonstrating the high spatiotemporal variability and non-stationarity of teleconnection impacts on drought across different regions of Iran and at various time lags (0-month, 1-month, 2-month, and 3-month).
• Providing a scientifically robust foundation for developing more accurate and regionally tailored operational drought forecasting systems and adaptive water management strategies in arid and semi-arid regions.

Funding

• National Natural Science Foundation of China (NSFC) (Grant No. 42261144671)
• Iran National Science Foundation (INSF) (Grant No. 4013097)

Citation

@article{Mahmoudi2025Decoding,
  author = {Mahmoudi, Peyman and Jafari, Pouria and Ghaemi, Alireza and Jian, Jun and Firoozi, Fatemeh and Yang, Jing},
  title = {Decoding Iran’s Drought Drivers: An Explainable AI Approach to Unraveling Global Teleconnection Impacts},
  journal = {Earth Systems and Environment},
  year = {2025},
  doi = {10.1007/s41748-025-00925-3},
  url = {https://doi.org/10.1007/s41748-025-00925-3}
}