Related Experiment Video
Updated: May 13, 2026
08:20
In Situ Soil Moisture Sensors in Undisturbed Soils
Published on: November 18, 2022
Monthly soil temperature modelling across different depths using machine learning algorithms in northeast India.
Abhaya Kumar Pradhan1, Deepak Jhajharia2, Kiran Bala Behura1
1Department of Soil and Water Conservation Engineering, College of Agricultural Engineering & Technology, OUAT, Bhubaneswar, Odisha, 751003, India.
Scientific Reports
|May 11, 2026
Summary
This study models soil temperature in northeast India using machine learning (ML) and meteorological data. Random Forest Regression and Bayesian Neural Network models accurately predicted soil temperature at various depths.
Area of Science:
- Agricultural Science
- Environmental Science
- Data Science
Background:
- Soil temperature is critical for agricultural systems and hydrogeological processes.
- Accurate soil temperature data is scarce in data-limited regions like northeast India.
- Machine learning (ML) offers a promising approach for modeling soil temperature using available meteorological variables.
Purpose of the Study:
- To model soil temperature at 5, 15, and 30 cm depths in Jorhat, Assam (northeast India).
- To evaluate the performance of five ML techniques: Random Forest Regression (RFR), Support Vector Regression (SVR), Boosted Regression Trees (BRT), Classification And Regression Tree (CART), and Bayesian Neural Network (BNN).
- To identify the most influential meteorological variables for soil temperature prediction.
Main Methods:
- Utilized eight long-term meteorological variables: air temperatures (min, mean, max), rainfall, sunshine hours, relative humidity (min, max), and number of rainy days.
- Applied five ML algorithms (RFR, SVR, BRT, CART, BNN) to predict soil temperature at three different depths (5, 15, 30 cm).
- Assessed model performance using R-squared (R²) and Nash-Sutcliffe Efficiency (NSE) metrics.
Main Results:
- RFR and BNN demonstrated superior performance in predicting 5 cm soil temperature (R² ≈ 0.983).
- BNN and RFR were the top models for 15 cm soil temperature prediction (R² ≈ 0.988).
- BNN and CART achieved the highest accuracy for 30 cm soil temperature prediction (R² ≈ 0.975).
- Mean air temperature was identified as the most significant predictor of soil temperature across all depths.
Conclusions:
- Machine learning models, particularly RFR and BNN, can effectively predict soil temperature in data-limited regions.
- Air temperature is the primary driver of soil temperature variations.
- These findings support the development of soil temperature-based crop models in areas lacking direct soil temperature measurements.
