Rice yield prediction through integration of biophysical parameters with SAR and optical remote sensing data using...

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Global population growth and climate variability necessitate enhanced food security measures.
• Rice is a staple food for billions, making accurate yield prediction vital for global food security.

Purpose of the Study:

• To predict rice yield at 45, 60, and 90 days after transplanting (DAT) using machine learning models.
• To evaluate the effectiveness of combining optical and Synthetic Aperture Radar (SAR) data with crop biophysical parameters for yield prediction.
• To assess the performance of various machine learning models for early rice yield estimation.

Main Methods:

• The study utilized optical and SAR data combined with crop biophysical parameters.
• Machine learning models, including eXtreme gradient boosting (XGB), Neural Network (NNET), Cubist, Support Vector Regression, and Random Forest, were employed.
• Yield predictions were made at 45, 60, and 90 DAT across two rice seasons in Uttarakhand, India.

Main Results:

• Machine learning models demonstrated relatively accurate early rice yield estimates.
• For summer rice, XGB consistently performed best at all prediction stages (45, 60, 90 DAT).
• For kharif rice, XGB, NNET, and Cubist were the top models at 45, 60, and 90 DAT, respectively. Prediction accuracy improved closer to harvest, with 90 DAT yielding the best results for both seasons.

Conclusions:

• The integration of remote sensing data and biophysical parameters with machine learning models significantly enhances early rice yield prediction.
• This approach supports informed decision-making for farmers, policy planners, and researchers, thereby strengthening food security planning and resource management.
• The study highlights the potential of advanced analytical techniques for optimizing agricultural practices and ensuring stable food supplies.