Predictive modeling of evapotranspiration using LSTM and explainable AI

Embracing smart irrigation management techniques empowers growers to irrigate with greater efficiency, thereby promoting sustainable agricultural production. In this context, growers often rely on crop evapotranspiration (ETc) as a key factor in making informed irrigation decisions, underscoring the significance of accurately determining and spatially mapping crop water status. Technological progress, exemplified by the emergence of unmanned aerial vehicles (UAVs), has brought about a revolutionary shift in agricultural monitoring. UAV platforms can capture high-resolution images with centimeter-level spatial accuracy and offer higher temporal coverage compared to satellite imagery. Considering these advancements, this study introduces a robust method for classifying water stress in cotton using a compact UAV platform and convolutional neural networks (CNN). The experiment was conducted at the USDA-ARS Cropping Systems Research Laboratory (CSRL) in Lubbock, Texas, where the cotton field was divided into 12 drip zones. The study included three replications to evaluate four irrigation treatments: “rainfed”, “full irrigation”, “percent deficit of full irrigation”, and “time delay of full irrigation”. The results demonstrated that the CNN model successfully classified the cotton water stress using the UAV-based RGB image, achieving an overall best prediction accuracy of approximately 91%. By segmenting the original cotton images into separate canopy and soil areas using morphological image processing methods, the authors also isolated and analyzed the individual contributions of these components to cotton water stress. Additionally, a random forest classifier revealed the relative importance of different image features in the classification process through feature importance analysis. These findings highlighted the state-of-the-art performance of the proposed system in cotton water stress classification and provided valuable insights into the key image features contributing to accurate classification. The authors concluded that integrating UAV-based RGB imagery and CNN models had great potential for assessing water stress in cotton.

A new approach to measure spatial variability of soil parameters and field technique to test-value specific fertilizer

Information on the distribution of soil properties is important to know the status of nutrients in the soils based on which fertilizer nutrients are recommended. Given the variability of nutrients in the soils, making a site-specific fertilizer recommendation seems to be a compelling work. To determine the spatial variability of soil nutrients and to make judicious and precise fertilizer recommendations, new measures are designed with this study. These measures are tested against the soil samples (n = 43) for total nitrogen (N), organic matter (OM), phosphorus (P2O5), and potassium (K2O) in the study area. The descriptive statistical analysis indicated an average of low nitrogen and organic matter, while phosphorus was found to be very high and the level of potassium was high. The spread of nutrients across the data sets, however, included low, medium, high, and very high levels of ratings. The Deviation Square Index was developed and applied for the variability measurement and found that the largest variation was with phosphorus distribution, followed by potassium, nitrogen, and organic matter. The coefficient of variation (CV%) analysis also exhibited similar trends in nutrient distributions. Nitrogen was the main determinant explaining the variations in rice yield, while phosphorus and potash were negatively related to the yield. An index of fertilizer nutrient recommendation called Test-Value Specific Dose (TVSD) was developed and used to calculate the nutrient recommendation for each sampled location. This new method gave easy and more accurate doses of fertilizer over the blanket recommendation to fit the variations across the soil samples.

Predictive modeling of evapotranspiration using LSTM and explainable AI

Sasmita Sahoo, Aayush Kumar

Article ID: 3534
Vol 6, Issue 3, 2025

DOI: https://doi.org/10.54517/ama3534

Received: 31 March 2025; Accepted: 6 June 2025; Available online: 26 June 2025; Issue release: 30 September 2025

VIEWS - 0 (Abstract)

Download PDF

Abstract

Evapotranspiration (ET) modeling plays a vital role in water resource management, agriculture, and climate adaptation. Accurate ET prediction is essential for effective irrigation planning and crop management. However, traditional methods often struggle to capture the complex relationships between environmental factors, resulting in less reliable forecasts. To address this, we implemented and optimized the Long Short-Term Memory (LSTM) network model to predict ET with improved accuracy of 98.8%, achieving a Mean Squared Error (MSE) of 0.12. Our approach incorporates SHapley Additive exPlanations (SHAP) to enhance model interpretability, offering insights into how key factors like solar radiation, wind speed, air temperature, and relative humidity impact ET predictions. The results showed that solar radiation had the highest impact on ET, followed by wind speed and air temperature. This improved understanding of key factors can help farmers and water managers make better decisions about irrigation, ensuring efficient water use and supporting sustainable agriculture. This provides a reliable and interpretable solution for ET prediction, aiding smarter irrigation strategies, improving resource efficiency, and supporting sustainable agricultural practices.

Keywords

evapotranspiration; SHAP; LSTM; deep learning; XGBoost; XAI; black-box

References

1. Granata F. Evapotranspiration evaluation models based on machine learning algorithms—A comparative study. Agricultural Water Management. 2019; 217: 303–315. doi: 10.1016/j.agwat.2019.03.015

2. Fereres E, García-Vila M. Irrigation Management for Efficient Crop Production. In: Meyers RA (editor). Encyclopedia of Sustainability Science and Technology. Springer; 2019. pp. 345–360.

3. Ben Abdallah E, Grati R, Boukadi K. Towards an explainable irrigation scheduling approach by predicting soil moisture and evapotranspiration via multi-target regression. Journal of Ambient Intelligence and Smart Environments. 2023; 15: 89–110. doi: 10.3233/AIS-220086

4. Goldstein A, Fink L, Meitin A, et al. Applying machine learning on sensor data for irrigation recommendations: revealing the agronomist’s tacit knowledge. Precision Agriculture. 2018; 19(3): 421–444. doi: 10.1007/s11119-017-9531-8

5. Goap A, Sharma D, Shukla AK, et al. An IoT-based smart irrigation management system using machine learning and open-source technologies. Computers and Electronics in Agriculture. 2018; 155: 41–49. doi: 10.1016/j.compag.2018.09.040

6. Arrieta AB, Díaz-Rodríguez N, Del Ser J, et al. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion. 2020; 58: 82–115. doi: 10.1016/j.inffus.2019.12.012

7. Chakraborty D, Başagaoglu H, Winterle J. Interpretable vs. non-interpretable machine learning models for data-driven hydro-climatological process modeling. Expert Systems with Applications. 2021; 170: 114498. doi: 10.1016/j.eswa.2020.114498

8. Dikshit A, Pradhan B. Explainable AI in drought forecasting. Machine Learning with Applications. 2021; 6: 100192. doi: 10.1016/j.mlwa.2021.100192

9. Di Bucchianico A. Coefficient of Determination (R²). In: Encyclopedia of Statistics in Quality and Reliability. John Wiley & Sons, Ltd; 2008. doi: 10.1002/9780470061572.eqr173

10. Hochreiter S, Schmidhuber J. Long short-term memory. Neural Computation. 1997; 9(8): 1735–1780. doi: 10.1162/neco.1997.9.8.1735

11. Wu T, Zhang W, Jiao X, Guo W, Hamoud YA, et al. Evaluation of stacking and blending ensemble learning methods for estimating daily reference evapotranspiration. Computers and Electronics in Agriculture. 2021; 184: 106039. doi: 10.1016/j.compag.2021.106039

12. Koné BA, Grati R, Bouaziz B, Boukadi K. Explainable Machine Learning for Evapotranspiration Prediction. In: Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2023); 13-15 November 2023; Rome, Italy. pp. 97–104. doi: 10.5220/0012253200003543

13. Zhao X, Zhang L, Zhu G, et al. Exploring interpretable and non-interpretable machine learning models for estimating winter wheat evapotranspiration using particle swarm optimization with limited climatic data. Computers and Electronics in Agriculture. 2023; 212: 108140. doi: 10.1016/j.compag.2023.108140

14. Acharki S, Raza A, Vishwakarma DK, et al. Comparative assessment of empirical and hybrid machine learning models for estimating daily reference evapotranspiration in sub-humid and semi-arid climates. Scientific Reports. 2025; 15(1): 2542. doi: 10.1038/s41598-024-83859-6

15. Shi B, Yuan Y, Zhuang T, et al. Improving water status prediction of winter wheat using multi-source data with machine learning. European Journal of Agronomy. 2022; 139: 126548. doi: 10.1016/j.eja.2022.126548

16. Hadadi F, Moazenzadeh R, Mohammadi B. Estimation of actual evapotranspiration: A novel hybrid method based on remote sensing and artificial intelligence. Journal of Hydrology. 2022; 609: 127774. doi: 10.1016/j.jhydrol.2022.127774

17. Bellido-Jiménez JA, Estévez J, Vanschoren J, et al. AgroML: An open-source repository to forecast reference evapotranspiration in different geo-climatic conditions using machine learning and transformer-based models. Agronomy. 2022; 12(3): 656. doi: 10.3390/agronomy12030656

18. Zhang H, Wang G, Li S, et al. Understanding Evapotranspiration Driving Mechanisms in China with Explainable Machine Learning Algorithms. International Journal of Climatology. 2025; 45(6). doi: 10.1002/joc.8774

19. Vaz PJ, Schütz G, Guerrero C, Cardoso PJS. Hybrid Neural Network Based Models for Evapotranspiration Prediction Over Limited Weather Parameters. IEEE Access. 2023; 11: 963–976. doi: 10.1109/ACCESS.2022.3233301

20. Khairan HE, Zubaidi SL, Muhsen YR, et al. Parameter optimisation-based hybrid reference evapotranspiration prediction models: A systematic review of current implementations and future research directions. Atmosphere. 2022; 14(1): 77. doi: 10.3390/atmos14010077

21. Hu X, Shi L, Lin G, et al. Comparison of physical-based, data-driven and hybrid modeling approaches for evapotranspiration estimation. Journal of Hydrology. 2021; 601: 126592. doi: 10.1016/j.jhydrol.2021.126592

22. Ghosal S, Blystone D, Singh AK, et al. An explainable deep machine vision framework for plant stress phenotyping. Proceedings of the National Academy of Sciences of the United States of America. 2018; 115(18): 4613–4618. doi: 10.1073/pnas.1716999115

23. Goyal P, Kumar S, Sharda R. A review of the Artificial Intelligence (AI) based techniques for estimating reference evapotranspiration: Current trends and future perspectives. Computers and Electronics in Agriculture. 2023; 209: 107836. doi: 10.1016/j.compag.2023.107836

24. Talib A, Desai AR, Huang J, et al. Evaluation of prediction and forecasting models for evapotranspiration of agricultural lands in the Midwest US. Journal of Hydrology. 2021; 600: 126579. doi: 10.1016/j.jhydrol.2021.126579

25. Hameed MM, AlOmar MK, Razali SFM, et al. Application of artificial intelligence models for evapotranspiration prediction along the southern coast of Turkey. Complexity. 2021; 2021: 8850243. doi: 10.1155/2021/8850243

26. Su H, McCabe MF, Wood EF, et al. Modeling evapotranspiration during SMACEX: Comparing two approaches for local-and regional-scale prediction. Journal of Hydrometeorology. 2005; 6(6): 910–922. doi: 10.1175/JHM464.1

27. Sun G, Alstad K, Chen J, et al. A general predictive model for estimating monthly ecosystem evapotranspiration. Ecohydrology. 2011; 4(2): 245–255. doi: 10.1002/eco.194

28. Mostafa RR, Kisi O, Adnan RM, et al. Modeling potential evapotranspiration by improved machine learning methods using limited climatic data. Water. 2023; 15(3): 486. doi: 10.3390/w15030486

29. Zhao L, Xia J, Xu C, et al. Evapotranspiration estimation methods in hydrological models. Journal of Geographical Sciences. 2013; 23: 359–369. doi: 10.1007/s11442

Refbacks

There are currently no refbacks.

This site is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).

Editor-in-Chief

Prof. Zhengjun Qiu

Zhejiang University, China

Honorary Editor-in-Chief

Cheng Sun

Academician of World Academy of Productivity Science; Executive Chairman, World Confederation of Productivity Science China Chapter, China

Indexing & Archiving

- - - - J-Gate

News & Announcements

2025-02-01

Highlight article: Innovations in Modern Agriculture: UAVs and CNNs Transform Cotton Water Stress Monitoring

In the realm of modern agriculture, the integration of cutting-edge technologies is revolutionizing the way we approach sustainable farming practices. A recent study published in Advances in Modern Agriculture titled "Classification of cotton water stress using convolutional neural networks and UAV-based RGB imagery" has garnered significant attention for its innovative approach to precision irrigation management. Conducted by researchers from Institute of Data Science and the AgriLife Research and Extension Center of Texas A&M University (authors's information is below). This study introduces a novel method for classifying cotton water stress using unmanned aerial vehicles (UAVs) and convolutional neural networks (CNNs), offering a powerful solution for optimizing water use in agriculture.

2024-09-11

Smart agriculture is becoming a reality!

Modern agricultural technology is evolving rapidly, with scientists collaborating with leading agricultural enterprises to develop intelligent management practices. These practices utilize advanced systems that provide tailored fertilization and treatment options for large-scale land management.

2024-05-01

Notice on the prohibition of using AI techniques to generate papers!

This journal values human initiative and intelligence, and the employment of AI technologies to write papers that replace the human mind is expressly prohibited. When there is a suspicious submission that uses AI tools to quickly piece together and generate research results, the editorial board of the journal will reject the article, and all journals under the publisher's umbrella will prohibit all authors from submitting their articles.

Readers and authors are asked to exercise caution and strictly adhere to the journal's policy regarding the usage of Artificial Intelligence Generated Content (AIGC) tools.

More Announcements...

Journal Center

Asia Pacific Academy of Science Pte. Ltd. (APACSCI) specializes in international journal publishing. APACSCI adopts the open access publishing model and provides an important communication bridge for academic groups whose interest fields include engineering, technology, medicine, computer, mathematics, agriculture and forestry, and environment.

more

Article Tools

Indexing metadata

How to cite item

Review policy