SHALOO et al. (2025) Modeling daily reference evapotranspiration and evaluating uncertainty analysis in machine learning under limited meteorological data conditions for Northern India

Identification

• Journal: Journal of Atmospheric and Solar-Terrestrial Physics
• Year: 2025
• Date: 2025-11-27
• Authors: SHALOO SHALOO, Himani Bisht, Bipin Kumar, Jitendra Rajput, P. S. Brahmanand
• DOI: 10.1016/j.jastp.2025.106696

Research Groups

ICAR- Indian Agricultural Research Institute, New Delhi, India

Short Summary

This study evaluated the performance of Random Forest, Artificial Neural Networks, and Long Short-Term Memory models for daily reference evapotranspiration (ET0) estimation under varying meteorological data availability in Northern India, finding LSTM to be the most reliable model, especially in data-scarce conditions.

Objective

• To assess the effectiveness of Random Forest (RF), Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM) machine learning models in predicting daily reference evapotranspiration (ET0) using various input parameter groupings, and to evaluate their uncertainty under limited meteorological data conditions.

Study Configuration

• Spatial Scale: Three districts in Haryana, Northern India.
• Temporal Scale: 38 years of daily meteorological data.

Methodology and Data

• Models used: Random Forest (RF), Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM). The FAO-56 Penman-Monteith (FAO-56-PM) method was used to generate target ET0 values. Hargreaves-Samani and Priestley-Taylor models were also evaluated.
• Data sources: 38 years of daily meteorological data obtained from the India Meteorological Department (IMD) for three districts in Haryana, including maximum temperature (Tmax), minimum temperature (Tmin), relative humidity (RH), wind speed (WS), and solar radiation (SR).

Main Results

• All evaluated machine learning models (RF, ANN, LSTM) achieved accurate daily ET0 predictions.
• The full meteorological dataset was identified as the optimal input combination for ET0 estimation.
• For limited datasets, combinations including temperature, wind speed, and solar radiation were found to be the most effective.
• Under minimal dataset conditions (only Tmax, Tmin), RF showed the best performance during the training phase (R² = 0.93–0.94).
• During testing with minimal temperature data, LSTM outperformed RF and ANN, achieving higher accuracy (R² ≈ 0.77) and lower errors (Mean Absolute Percentage Error ≈ 17–18 %).
• Uncertainty analysis using a Monte Carlo-based approach revealed that LSTM captured extreme ET0 values with broader confidence intervals, indicating higher sensitivity but lower overall prediction uncertainties. ANN produced tighter intervals with lower variability, while RF offered a balanced performance.
• The LSTM model was identified as the most reliable for ET0 estimation under data-scarce conditions, exhibiting lower prediction uncertainties compared to RF and ANN.
• The temperature-based Hargreaves-Samani model outperformed the radiation-based Priestley-Taylor model in ET0 estimation.

Contributions

• Provides a dependable approach for estimating daily ET0 in semi-arid regions, particularly under limited meteorological data availability.
• Offers practical guidance for efficient water management and supports climate-resilient agriculture, contributing to food security in regions with similar agro-climatic conditions.
• Demonstrates the superior reliability of LSTM for ET0 estimation in data-scarce environments through comprehensive performance and uncertainty analyses.

Funding

Not explicitly mentioned in the provided paper text.

Citation

@article{SHALOO2025Modeling,
  author = {SHALOO, SHALOO and Bisht, Himani and Kumar, Bipin and Rajput, Jitendra and Brahmanand, P. S.},
  title = {Modeling daily reference evapotranspiration and evaluating uncertainty analysis in machine learning under limited meteorological data conditions for Northern India},
  journal = {Journal of Atmospheric and Solar-Terrestrial Physics},
  year = {2025},
  doi = {10.1016/j.jastp.2025.106696},
  url = {https://doi.org/10.1016/j.jastp.2025.106696}
}