Leaf Anthocyanin Content Retrieval with Partial Least Squares and Gaussian Process Regression from Spectral...

Area of Science:

• Remote Sensing
• Plant Physiology
• Spectroscopy

Background:

• Leaf pigment content retrieval is vital in remote sensing.
• Anthocyanin retrieval studies are less common than for chlorophylls and carotenoids.
• Anthocyanins play a critical physiological role in plants.

Purpose of the Study:

• To explore the use of partial least squares regression (PLSR) and Gaussian process regression (GPR) for leaf anthocyanin content retrieval.
• To establish new methods for accurate anthocyanin estimation in plants.
• To investigate the effectiveness of different regression models in characterizing anthocyanin-related spectral signatures.

Main Methods:

• Utilized leaf reflectance data.
• Applied partial least squares regression (PLSR) and Gaussian process regression (GPR) models.
• Investigated logarithmic transformation of reflectance (log(1/R)) at specific wavelengths (564 nm and 705 nm).

Main Results:

• Gaussian process regression (GPR) achieved the highest accuracy (R²=0.93, RMSE=2.18 in calibration; R²=0.93, RMSE=2.20 in validation).
• Partial least squares regression (PLSR) showed lower accuracy (R²=0.87, RMSE=2.88 in calibration; R²=0.88, RMSE=2.89 in validation).
• GPR's non-linear approach was more effective than PLSR's linear approach for anthocyanin content prediction.

Conclusions:

• Gaussian process regression (GPR) is a highly effective method for retrieving leaf anthocyanin content.
• Simple two-wavelength models using GPR, selected from the green peak and red edge regions, can achieve high accuracy (R² > 0.90).
• This research advances remote sensing capabilities for monitoring plant physiological status through anthocyanin estimation.