Estimation of paddy rice leaf area index using machine learning methods based on hyperspectral data from multi-year...

Area of Science:

• Agricultural Remote Sensing
• Machine Learning Applications
• Plant Physiology

Background:

• Leaf Area Index (LAI) is a critical parameter for monitoring crop growth and health.
• Accurate LAI estimation is essential for precision agriculture and yield prediction in paddy rice.
• Traditional methods for LAI estimation have limitations in accuracy and robustness.

Purpose of the Study:

• To evaluate the performance of machine learning (ML) methods against traditional methods for paddy rice LAI estimation.
• To compare the effectiveness of raw spectral reflectance versus first derivative spectra in LAI modeling.
• To identify the most accurate and robust ML model for LAI estimation using hyperspectral data.

Main Methods:

• Evaluated Support Vector Regression (SVR), Random Forests (RF), and Artificial Neural Networks (ANN) against Vegetation Index (VI) and Partial Least Squares Regression (PLSR).
• Utilized a four-year field-collected hyperspectral reflectance dataset of paddy rice.
• Compared models built on raw spectral data and first derivative spectra, optimizing model parameters.

Main Results:

• First derivative spectra generally yielded more accurate LAI estimation models than raw spectra.
• Machine learning methods demonstrated superior accuracy and robustness compared to VI and PLSR methods.
• Random Forests (RF) models, utilizing first derivative spectra with specific parameter tuning (mtry=10), achieved the highest accuracy (lowest RMSE).

Conclusions:

• Machine learning approaches, especially RF, offer significant advantages for accurate and robust LAI estimation in paddy rice.
• The use of first derivative spectral data enhances the performance of LAI estimation models.
• Optimized ML models provide a powerful tool for agricultural monitoring and management.