Artificial intelligence/machine learning estimation of FAO and ASCE standardized reference ET erodes the physical basis and intended purpose of ET standardization
10 affiliations
- 1Department of Soil and Water Systems, University of Idaho, Boise, ID, USA
- 2Department of Soil and Water Systems, University of Idaho, Kimberly, ID, USA
- 3University of Nebraska, Lincoln, USA
- 4Idaho Department of Water Resources, ID, USA
- 5USDA-ARS Water Management and Systems Research Unit, Fort Collins, CO, USA
- 6USDA-ARS Grassland Soil and Water Research Laboratory, Temple, TX, USA
- 7Desert Research Institute, Reno, NV, USA
- 8LEAF—Linking Landscape, Environment, Agriculture and Food Research Center, Associated Laboratory TERRA, School of Agriculture, University of Lisbon, Portugal
- 9Desertification Research Centre (CIDE), CSIC-UV-GVA, Moncada, Valencia, Spain
- 10FAO, Rome, Italy
Abstract
The rapid adoption of artificial intelligence and machine learning (AI/ML) in hydrology has delivered notable gains in prediction, gap filling, and data integration. Recently, these methods have increasingly been applied to estimate FAO and ASCE S tandardized Reference E vapotranspiration (ET ref ), a trend that warrants careful reconsideration. ET ref is not a directly observable natural flux but a deliberately defined climatic index, constructed to represent atmospheric evaporative demand over a fixed reference surface using physically based relationships among solar radiation, air temperature, humidity, and wind speed. Treating ET ref as a learnable target conflates the estimation of a physical phenomenon with the reproduction of a standardized physical calculation. Although AI/ML-based ET ref estimates may achieve high statistical accuracy, these methods risk undermining the intended purpose of ET ref standardization by embedding local covariances, obscuring error attribution, limiting transferability across space and time, and weakening the institutional knowledge that anchors scientific evapotranspiration (ET) practice to physical principles and crop coefficient frameworks. Moreover, reduced-input AI/ML formulations may inadvertently weaken the case for expanding and maintaining high-quality reference weather networks that accurate ET ref depends on. We argue that AI/ML adds the most value in ET science when used to support, rather than replace, the FAO/ASCE ET ref physical equation. Acceptable AI/ML uses include improvement of standardized ET ref frameworks' meteorological datasets, reference-condition screening, estimating missing weather data and estimation of actual evapotranspiration (ET a , not ET ref ) where the modeling complexity truly justifies the flexibility of the AI/ML tools.
Related
- Evaluating hv-block cross-validation for dependence-aware model selection in reference evapotranspiration estimation
- Spatial estimation of standardized precipitation index based drought severity across Türkiye using machine learning models
- An Artificial Intelligence Integrated Geospatial Decision-Support Framework for Scenario-Based Climate-Adaptive Land-Use Optimization and Environmental Risk Reduction
- Strengthening and safeguarding artificial intelligence for a planet under pressure
- WP-MIP: An Artificial Intelligence, Hybrid, and Physically Based Model Intercomparison Project for Weather Prediction
BibTeX
@article{kukal2026artificial,
title = {{Artificial intelligence/machine learning estimation of FAO and ASCE standardized reference ET erodes the physical basis and intended purpose of ET standardization}},
author = {Meetpal Singh Kukal and Richard G. Allen and Ayşe Kiliç and Philip A. Blankenau and Kendall C. DeJonge and Kelly R. Thorp and Justin Huntington and Luis Santos Pereira and R. López-Urrea and Maher Salman and Paula Paredes and Clarence W. Robison},
journal = {Agricultural Water Management},
year = {2026},
month = jul,
doi = {10.1016/j.agwat.2026.110657},
url = {https://doi.org/10.1016/j.agwat.2026.110657},
note = {CC BY},
}
RIS
TY - JOUR
TI - Artificial intelligence/machine learning estimation of FAO and ASCE standardized reference ET erodes the physical basis and intended purpose of ET standardization
AU - Meetpal Singh Kukal
AU - Richard G. Allen
AU - Ayşe Kiliç
AU - Philip A. Blankenau
AU - Kendall C. DeJonge
AU - Kelly R. Thorp
AU - Justin Huntington
AU - Luis Santos Pereira
AU - R. López-Urrea
AU - Maher Salman
AU - Paula Paredes
AU - Clarence W. Robison
PY - 2026
DA - 2026/07/22
JO - Agricultural Water Management
DO - 10.1016/j.agwat.2026.110657
UR - https://doi.org/10.1016/j.agwat.2026.110657
AB - The rapid adoption of artificial intelligence and machine learning (AI/ML) in hydrology has delivered notable gains in prediction, gap filling, and data integration. Recently, these methods have increasingly been applied to estimate FAO and ASCE S tandardized Reference E vapotranspiration (ET ref ), a trend that warrants careful reconsideration. ET ref is not a directly observable natural flux but a deliberately defined climatic index, constructed to represent atmospheric evaporative demand over a fixed reference surface using physically based relationships among solar radiation, air temperature, humidity, and wind speed. Treating ET ref as a learnable target conflates the estimation of a physical phenomenon with the reproduction of a standardized physical calculation. Although AI/ML-based ET ref estimates may achieve high statistical accuracy, these methods risk undermining the intended purpose of ET ref standardization by embedding local covariances, obscuring error attribution, limiting transferability across space and time, and weakening the institutional knowledge that anchors scientific evapotranspiration (ET) practice to physical principles and crop coefficient frameworks. Moreover, reduced-input AI/ML formulations may inadvertently weaken the case for expanding and maintaining high-quality reference weather networks that accurate ET ref depends on. We argue that AI/ML adds the most value in ET science when used to support, rather than replace, the FAO/ASCE ET ref physical equation. Acceptable AI/ML uses include improvement of standardized ET ref frameworks' meteorological datasets, reference-condition screening, estimating missing weather data and estimation of actual evapotranspiration (ET a , not ET ref ) where the modeling complexity truly justifies the flexibility of the AI/ML tools.
SN - 0378-3774
SN - 1873-2283
ER -