CAGUIAT et al. (2025) Machine learning modeling of reference evapotranspiration in Central Luzon, Philippines

Identification

• Journal: Journal of Agrometeorology
• Year: 2025
• Date: 2025-12-01
• Authors: LEA S. CAGUIAT, Ronaldo B. Saludes, Marion Lux Castro, Rubenito M. Lampayan
• DOI: 10.54386/jam.v27i4.2909

Research Groups

• Agrometeorology, Bio-Structures and Environment Engineering Division, Institute of Agricultural and Biosystems Engineering, College of Engineering and Agro-Industrial Technology, University of the Philippines, Los Baños, Laguna, Philippines

Short Summary

This study evaluates various machine learning algorithms for estimating reference evapotranspiration (ETo) in Central Luzon, Philippines, using limited ground-based weather data. It demonstrates that machine learning, especially Gaussian Process Regression, can accurately predict ETo with only two or three input variables, offering a robust alternative to data-intensive empirical models.

Objective

• Evaluate the performance of various machine learning algorithms and input combinations in estimating ETo using ground-based weather data.
• Determine the optimal machine learning algorithm and input combination using established decision thresholds.

Study Configuration

• Spatial Scale: Four synoptic weather stations (Baler, Casiguran, Clark Airport, Cubi Point) in Central Luzon, Philippines.
• Temporal Scale: Daily meteorological data from 1985 to 2018.

Methodology and Data

• Models used:
  • Reference method: FAO Penman-Monteith method.
  • Machine learning algorithms: Linear regression, Regression tree, Support vector machine (SVM), Gaussian progress regression (GPR), Ensemble of trees, Neural network.
• Data sources:
  • Ground-based daily meteorological data (maximum air temperature, minimum air temperature, relative humidity, and wind speed) collected from four synoptic weather stations of the Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) in Central Luzon, Philippines.
  • Solar radiation data was estimated using differences in maximum and minimum temperature.

Main Results

• Gaussian Process Regression (GPR) consistently achieved the highest estimation accuracy across all stations and input combinations, with R² values ranging from 0.71 to 0.95 and RMSE values from 0.27 to 0.64 mm day⁻¹. Optimizable GPR and Matern 5/2 GPR models showed the highest accuracy within GPR.
• The general ranking of algorithms by performance was: GPR > Neural network > Support vector machines > Ensemble of trees > Regression trees > Linear regression.
• Model performance improved with an increased number of input variables; using four variables (maximum air temperature, minimum air temperature, relative humidity, wind speed) resulted in an 18% increase in R² and a decrease in error values (RMSE=0.22 mm day⁻¹, MSE=0.19 mm day⁻², MAE=0.17 mm day⁻¹) compared to single-variable models.
• Optimal models, meeting the decision thresholds (RMSE < 0.39 mm day⁻¹, R² > 0.75, MSE < 0.15 mm day⁻², MAE < 0.30 mm day⁻¹), required only two or three weather variables depending on the station. For Baler and Casiguran, three variables were optimal, while Clark and Cubi Point required only two.
• All models and combinations passed the Wilcoxon test (P < 0.05), and Principal Component Analysis (PCA) confirmed that the relevant variables explained at least 95% of the variance (σ² ≥ 0.95).

Contributions

• Demonstrated the high accuracy and applicability of machine learning algorithms, particularly GPR, for ETo estimation in data-sparse regions like Central Luzon, Philippines.
• Identified optimal machine learning models and minimal input variable combinations (two or three variables) that meet specific accuracy thresholds, providing a practical solution for agricultural water management.
• Validated machine learning as a robust and accurate alternative to data-intensive empirical ETo models when only limited ground-based meteorological data are available.

Funding

• Department of Science and Technology-Engineering Research and Development for Technology scholarship program.

Citation

@article{CAGUIAT2025Machine,
  author = {CAGUIAT, LEA S. and Saludes, Ronaldo B. and Castro, Marion Lux and Lampayan, Rubenito M.},
  title = {Machine learning modeling of reference evapotranspiration in Central Luzon, Philippines},
  journal = {Journal of Agrometeorology},
  year = {2025},
  doi = {10.54386/jam.v27i4.2909},
  url = {https://doi.org/10.54386/jam.v27i4.2909}
}