Fadaee et al. (2026) Explainable Artificial Intelligence in Hydrology: A Review

Identification

• Journal: Water Resources Management
• Year: 2026
• Date: 2026-02-01
• Authors: Marzieh Fadaee, Marwan Kheimi
• DOI: 10.1007/s11269-025-04435-9

Research Groups

• Department of Civil Engineering, Shahid Bahonar University of Kerman, Kerman, Iran
• Department of Civil and Environmental Engineering, Faculty of Engineering - Rabigh Branch, King Abdulaziz University, Jeddah, 21589, Saudi Arabia

Short Summary

This paper presents the first systematic and critical review of Explainable Artificial Intelligence (XAI) applications in hydrology and hydrogeology, synthesizing over 180 peer-reviewed studies. It concludes that XAI significantly enhances the transparency and trustworthiness of AI models while deepening the understanding of underlying physical hydrological processes.

Objective

• To provide a systematic and critical synthesis of Explainable Artificial Intelligence (XAI) applications in hydrology and hydrogeology, focusing on post-hoc methods.
• To analyze how XAI improves the understanding of influential drivers of hydrological phenomena in AI-driven models.
• To validate hydrological theories through insights gained from XAI.
• To identify current research gaps and propose future directions, including hybrid frameworks for physically consistent interpretations.

Study Configuration

• Spatial Scale: A systematic review of studies covering diverse hydrological domains globally, including evapotranspiration, glacio-hydrology, precipitation, rainfall–runoff, river and streamflow, sediment transport, floods and droughts, water quality, and groundwater.
• Temporal Scale: Literature published from 2015 up to 2024.

Methodology and Data

• Models used: The review focuses on post-hoc XAI methods such as SHAP (Shapley Additive exPlanations), LIME (Local Interpretable Model-Agnostic Explanations), PDP (Partial Dependence Plots), ALE (Accumulated Local Effects), and PFI (Permutation Feature Importance). These XAI methods are applied to various Machine Learning (ML) and Deep Learning (DL) models, including Random Forest (RF), XGBoost, Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Artificial Neural Networks (ANNs), and other ensemble and tree-based models.
• Data sources: A systematic literature search of peer-reviewed articles published between 2015 and 2024 across databases like Scopus, using Boolean keyword combinations (e.g., "hydrology" AND "AI"). Approximately 1,200 abstracts were screened, leading to a full-text review of 450 papers, from which 180 studies were selected for qualitative analysis.

Main Results

• XAI significantly enhances the transparency, trust, and understanding of physical processes in AI-based hydrological models.
• SHAP is the most widely adopted post-hoc XAI method, used in 50% of the reviewed studies, owing to its robust local and global interpretability and ease of use.
• Post-hoc XAI methods are dominant in hydrological applications due to their flexibility and compatibility with complex ML/DL models.
• Water quality (31%) and river/streamflow (17%) are the most frequently studied hydrological domains using XAI, while evapotranspiration (3%) and groundwater (5%) are comparatively understudied.
• Ensemble learning models (Random Forest and boosting methods) are the most common ML models integrated with XAI (56% of applications). LSTM is the most used deep learning model (59% of DL applications), and XGBoost is the most used boosting algorithm (70% of boosting applications).
• XAI insights often validate existing hydrological theories, such as the role of precipitation and soil moisture in runoff generation, temperature and snowpack depth in glacio-hydrology, and discharge and land use in water quality.
• Key challenges include computational scalability for large-scale or real-time applications, uneven XAI application across hydrological domains, and limited integration with physics-informed models.

Contributions

• Provides the first systematic and critical synthesis of Explainable Artificial Intelligence (XAI) applications specifically within hydrology and hydrogeology.
• Offers a comprehensive analysis of various post-hoc XAI methods, detailing their advantages, disadvantages, and suitability for different hydrological domains.
• Demonstrates how XAI can validate and deepen the understanding of physical hydrological processes, moving beyond purely statistical interpretations of AI models.
• Identifies critical research gaps and proposes future directions, including the development of hybrid XAI-physics-informed frameworks and strategies to address computational scalability and domain unevenness.

Funding

No funds, grants, or other support was received for this research.

Citation

@article{Fadaee2026Explainable,
  author = {Fadaee, Marzieh and Kheimi, Marwan},
  title = {Explainable Artificial Intelligence in Hydrology: A Review},
  journal = {Water Resources Management},
  year = {2026},
  doi = {10.1007/s11269-025-04435-9},
  url = {https://doi.org/10.1007/s11269-025-04435-9}
}