Related Experiment Video
Updated: May 26, 2025
08:49
Manufacturing Simple and Inexpensive Soil Surface Temperature and Gravimetric Water Content Sensors
Published on: December 21, 2019
9.2K
A spatiotemporal CNN-LSTM deep learning model for predicting soil temperature in diverse large-scale regional
Vahid Farhangmehr1, Hanifeh Imanian2, Abdolmajid Mohammadian3
1Department of Mechanical Engineering, University of Bonab, P.O. Box 55517-61167, Bonab, Iran.
The Science of the Total Environment
|February 23, 2025
Summary
Accurate soil temperature prediction is crucial for various environmental applications. A novel CNN-LSTM model forecasts hourly soil temperatures with high accuracy across diverse climates, outperforming traditional methods.
Area of Science:
- Soil Science
- Hydrology
- Agriculture
- Environmental Engineering
- Climatology
Background:
- Soil temperature is a critical factor influencing numerous environmental and agricultural processes.
- Accurate prediction of soil temperatures is essential for effective decision-making in water resource management, agriculture, and climate adaptation.
- Existing models may not fully capture the complex spatiotemporal dynamics of soil temperature across varied climatic conditions.
Purpose of the Study:
- To develop and evaluate an optimized deep learning model for forecasting hourly spatiotemporal soil temperatures at a 0-7 cm depth.
- To assess the model's performance across five distinct climatic zones in Canada and the US.
- To compare the proposed model's accuracy against established machine learning techniques like Random Forest and Support Vector Regression.
Main Methods:
- An optimized two-dimensional convolutional neural network (CNN) integrated with a single-layer long short-term memory (LSTM) model was employed.
- The CNN-LSTM model was trained using annual hourly time-series spatiotemporal soil temperature data.
- Model performance was evaluated using correlation coefficients, Normalized Root Mean Squared Error (NRMSE), and Coefficient of Determination (R²).
Main Results:
- The CNN-LSTM model demonstrated high prediction accuracy, with training correlations between 99.18% and 99.69%, and testing correlations ranging from 93.72% to 99.24%.
- The model achieved superior performance compared to Random Forest (RF) and Support Vector Regression (SVR) models across all evaluated climatic zones.
- CNN-LSTM yielded NRMSE values from 1.42% to 3.63% and R² values from 93.73% to 99.25%, significantly outperforming RF and SVR.
Conclusions:
- The developed CNN-LSTM model provides a reliable and accurate method for forecasting spatiotemporal soil temperatures.
- The model's effectiveness across diverse climates highlights its potential for large-scale regional applications in agriculture, hydrology, and climate adaptation.
- Findings support improved decision-making for crop management, irrigation strategies, environmental monitoring, and climate change mitigation.
