Rani et al. (2026) Soil Heat Flux Dynamics Modeling Using Temporal Deep Learning For Determining Plant Root Zone Temperature

Identification

• Journal: Lecture notes in networks and systems
• Year: 2026
• Date: 2026-01-01
• Authors: N. Gopika Rani, S. Manjusha, R. Mydhili Nayaki, S. Shrinidhi, Swetha Manivasagam, A. Vedavarshini
• DOI: 10.1007/978-3-032-15401-9_21

Research Groups

PSG College of Technology, Coimbatore, India

Short Summary

This study proposes a deep learning ensemble model combining Temporal Convolutional Networks (TCNs) and Artificial Neural Networks (ANNs) to forecast soil heat flux, a key determinant of Root Zone Temperature (RZT), and subsequently optimizes RZT to a near-ideal range for crop health.

Objective

• To develop and evaluate a deep learning ensemble model (TCN-ANN) for forecasting soil heat flux dynamics to accurately determine and optimize plant root zone temperature (RZT) for improved crop health and productivity.

Study Configuration

• Spatial Scale: An agricultural test field in Faisalabad, Pakistan.
• Temporal Scale: One year of historical soil heat flux data.

Methodology and Data

• Models used: Temporal Convolutional Networks (TCNs), Artificial Neural Networks (ANNs), TCN-ANN ensemble model, Bayesian optimization framework. (Compared against individual TCN, ANN, and LSTM models).
• Data sources: Extensive historical soil heat flux dataset collected by UAF Pakistan.

Main Results

• The TCN-ANN ensemble model achieved a Mean Absolute Error (MAE) of 0.1559 and an R-squared error of 0.945 for soil heat flux prediction at 50 epochs.
• The ensemble model significantly outperformed individual TCN, ANN, and LSTM models in forecasting soil heat flux.
• Post-Bayesian optimization, the Root Zone Temperature (RZT) stabilized within a range of 18 °C to 25 °C.
• The optimized RZT achieved an accuracy of 99.44% within a threshold of ±3.45 °C.

Contributions

• Proposes a novel deep learning ensemble model (TCN-ANN) specifically designed for forecasting soil heat flux dynamics.
• Demonstrates superior performance of the TCN-ANN ensemble over individual deep learning models (TCN, ANN, LSTM) for this task.
• Integrates a Bayesian optimization framework to actively target and stabilize Root Zone Temperature (RZT) within an ideal range for crop health.
• Provides a highly accurate method for RZT determination and optimization, crucial for agricultural productivity.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Rani2026Soil,
  author = {Rani, N. Gopika and Manjusha, S. and Nayaki, R. Mydhili and Shrinidhi, S. and Manivasagam, Swetha and Vedavarshini, A.},
  title = {Soil Heat Flux Dynamics Modeling Using Temporal Deep Learning For Determining Plant Root Zone Temperature},
  journal = {Lecture notes in networks and systems},
  year = {2026},
  doi = {10.1007/978-3-032-15401-9_21},
  url = {https://doi.org/10.1007/978-3-032-15401-9_21}
}