Fong et al. (2025) Advancing evapotranspiration estimation with remote sensing and artificial intelligence – A review

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-12-01
• Authors: Terry Fong, Yuk Feng Huang, Ren Jie Chin, Chai Hoon Koo
• DOI: 10.1016/j.agwat.2025.110023

Research Groups

• Lee Kong Chian Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, Malaysia

Short Summary

This review paper comprehensively synthesizes the state-of-the-art in evapotranspiration (ET) estimation by integrating remote sensing (RS) data with artificial intelligence (AI) techniques, including machine learning, deep learning, explainable AI, and emerging geospatial foundation models. It highlights how RS addresses data limitations of conventional methods and how AI enhances accuracy and efficiency for sustainable water management.

Objective

• To provide a comprehensive overview of existing research on ET estimation utilizing remote sensing data integrated with AI techniques.
• To enhance the understanding of the current state-of-the-art in AI-driven remote sensing approaches for efficient and accurate ET estimation.
• To discuss various categories of AI models (ML, DL, XAI, GFMs), highlighting their structures, strengths, and limitations, as well as commonly used remote sensing satellites and key parameters relevant to ET estimation.

Study Configuration

• Spatial Scale: Global, regional, and site-specific (as covered by the reviewed literature).
• Temporal Scale: Daily, monthly, hourly, and long-term (as covered by the reviewed literature).

Methodology and Data

• Models used: Machine Learning (ML) algorithms (Artificial Neural Networks (ANN), Multilayer Perceptron (MLP), Support Vector Machine (SVM), Decision Tree, M5 model tree, Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost)), Deep Learning (DL) algorithms (Deep Neural Networks (DNN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Gated Recurrent Unit (GRU)), Hybrid AI models (e.g., CNN-RF, CNN-SVM, ConvLSTM, SA-ConvLSTM, CNN-GRU, CNN-LSTM), Explainable AI (XAI) and Interpretable Machine Learning (IML) techniques (SHAP, LIME, Surrogate Models), Geospatial Foundation Models (GFMs) (e.g., NASA–IBM Prithvi).
• Data sources: Remote sensing satellite imagery (Landsat series (2, 3, 4, 5, 6, 7, 8, 9), Terra/Aqua (MODIS, AIRS), Sentinel series (1, 2, 3A, 3B), Meteosat series (MFG, MSG, SEVIRI), ASTER, VIIRS (Suomi NPP, NOAA-20)), reanalysis data (ERA5, ERA5-Land), and ground-based meteorological observations (for validation/comparison in reviewed studies). Key parameters derived from remote sensing include Land Surface Temperature (LST), Solar Radiation, and Vegetation Indices (Normalized Difference Vegetation Index (NDVI), Leaf Area Index (LAI), Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), Fraction of Photosynthetically Active Radiation (FPAR)).

Main Results

• Remote sensing (RS) effectively overcomes data limitations of conventional ET estimation methods by providing spatially continuous data over large areas, with various satellites offering diverse spatial, temporal, and spectral resolutions.
• Essential RS parameters for ET estimation include Land Surface Temperature (LST), solar radiation, and vegetation indices (NDVI, LAI, EVI, SAVI), which can serve as alternatives or complements to ground-based meteorological data.
• AI models, encompassing machine learning (ML), deep learning (DL), and hybrid approaches, significantly enhance ET estimation accuracy and efficiency, consistently outperforming traditional empirical methods and demonstrating robustness even with limited input data.
• Deep learning models like CNN, LSTM, and GRU are increasingly utilized for their ability to capture complex non-linear relationships and spatiotemporal dependencies in ET data.
• Hybrid AI models (e.g., CNN-RF, ConvLSTM) show superior performance by leveraging the complementary strengths of different architectures, improving accuracy and robustness in regional ET estimation.
• Explainable AI (XAI) and Interpretable Machine Learning (IML) techniques are crucial for increasing the transparency and trustworthiness of AI models, identifying the relative importance of input variables (e.g., temperature, wind speed, solar radiation) in ET predictions.
• Geospatial Foundation Models (GFMs) represent an emerging paradigm for scalable, generalizable, and transferable Earth observation, offering a unified framework for diverse geospatial tasks, including advanced ET mapping.

Contributions

• Provides a comprehensive and up-to-date review of the integration of remote sensing data with various artificial intelligence techniques (ML, DL, XAI, GFMs) for evapotranspiration estimation.
• Systematically discusses the characteristics and applications of common remote sensing satellites relevant to ET, along with essential image pre-processing steps.
• Identifies key remote sensing-derived parameters (LST, solar radiation, vegetation indices) and reanalysis-based methods as crucial inputs for AI-driven ET models.
• Consolidates existing literature, highlights the strengths and limitations of different AI models, and outlines current challenges and future research directions for improving ET estimation accuracy, scalability, and interpretability.
• Emphasizes the emerging role of Geospatial Foundation Models and Explainable AI in advancing the field towards more robust, transparent, and globally consistent ET predictions.

Funding

• Ministry of Higher Education (MOHE) Malaysia: Fundamental Research Grant Scheme (FRGS/1/2022/TK06/UTAR/02/4)
• Universiti Tunku Abdul Rahman: UTAR Research Scholarship Scheme grant (Vote No. 8160/0020)
• Universiti Tunku Abdul Rahman: UTARRF Top-up Scheme (Vote No. 6235/K04)

Citation

@article{Fong2025Advancing,
  author = {Fong, Terry and Huang, Yuk Feng and Chin, Ren Jie and Koo, Chai Hoon},
  title = {Advancing evapotranspiration estimation with remote sensing and artificial intelligence – A review},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.110023},
  url = {https://doi.org/10.1016/j.agwat.2025.110023}
}