A deep learning based framework for enhanced reference evapotranspiration estimation: evaluating accuracy and forecasting strategies | JoVE Visualize

Area of Science:

Agricultural Engineering
Data Science
Hydrology

Background:

Optimizing agricultural resource management, including crop yield and water use, is crucial for food security.
Accurate estimation of reference evapotranspiration (ET_o) is vital for precise irrigation, but traditional methods are often costly and data-intensive.
Limited meteorological data poses a significant challenge for reliable ET_o prediction and irrigation planning.

Purpose of the Study:

To evaluate the performance of deep learning sequential models (LSTM, N-BEATS, TCN) for predicting daily ET_o.
To compare the effectiveness of standard vs. recursive prediction strategies using TCN for ET_o estimation.
To identify an efficient method for ET_o time-series prediction, particularly in data-scarce agricultural settings.

Main Methods:

Three deep learning sequential models—Long short-term memory (LSTM), Neural Basis Expansion Analysis for Time Series (N-BEATS), and Temporal Convolutional Network (TCN)—were evaluated for daily ET_o prediction.
The TCN model, identified as the best performer, was further used to assess two prediction strategies: standard prediction using historical data and a recursive approach involving predicted climatological data.
Model performance was quantitatively assessed using metrics such as Nash-Sutcliffe Efficiency (NSE), Theil U2, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE).

Main Results:

The Temporal Convolutional Network (TCN) model demonstrated superior performance compared to LSTM and N-BEATS in predicting daily ET_o.
The TCN model achieved high accuracy with NSE = 0.99, Theil U2 = 0.005, RMSE = 0.092, and MAE = 0.048.
The recursive prediction strategy using TCN yielded more accurate ET_o values than the standard approach, proving beneficial for data-scarce irrigation planning.

Conclusions:

Deep learning models, particularly TCN, offer an efficient and accurate alternative for predicting ET_o time-series.
The TCN model's performance and the effectiveness of its recursive strategy provide a valuable tool for precise water resource management in agriculture, especially where meteorological data is limited.
This study highlights the potential of advanced AI techniques to improve agricultural efficiency and sustainability through better irrigation planning.