Rice Yield Estimation Based on Continuous Wavelet Transform With Multiple Growth Periods.

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Rice yield estimation is crucial for agricultural planning and food security.
• Traditional methods often rely on single growth stages or basic spectral indices, limiting accuracy.
• Accurate, large-scale yield prediction requires integrating data across multiple crop development phases.

Purpose of the Study:

• To develop and validate an improved rice yield estimation model.
• To explore the effectiveness of combining multiple growth stages for yield prediction.
• To introduce and assess the continuous wavelet transform algorithm in rice yield modeling.

Main Methods:

• Collected canopy reflectance spectra at four key rice growth stages: elongation, heading, flowering, and milky.
• Constructed yield estimation models using vegetation index, first derivative, and continuous wavelet transform.
• Employed random forest algorithm and multiple stepwise regression for model development.
• Validated model performance using training and two independent validation sets.

Main Results:

• Combining multiple growth stages significantly enhanced model accuracy compared to single-stage models.
• The optimal model integrated first derivative, wavelet transform, and vegetation index with the random forest algorithm.
• The optimal model achieved high predictive accuracy, with R-squared values of 0.86 (training), 0.85 (validation 1), and 0.80 (validation 2).
• Low RMSE and MAPE values across datasets indicate robust and reliable yield predictions.

Conclusions:

• The developed model accurately predicts rice yield using multi-stage spectral data and advanced analysis.
• This approach provides a robust foundation for large-scale, statistically sound rice yield estimation.
• The integration of continuous wavelet transform offers a promising advancement in crop yield monitoring technology.