Full-Spectrum Hyperspectral Modeling of Leaf Dry Matter Content Using a Stacked Ensemble Framework.

Area of Science:

• Plant physiology
• Remote sensing
• Machine learning

Background:

• Leaf dry matter content (LDMC) is a key plant functional trait.
• Accurate LDMC estimation is crucial for ecological and agricultural monitoring.
• Traditional spectral indices show limitations in predicting LDMC across diverse species.

Purpose of the Study:

• To evaluate the predictability of LDMC using hyperspectral reflectance data.
• To compare the performance of narrow-band spectral indices and full-spectrum machine learning models.
• To develop an enhanced LDMC estimation model using a stacked ensemble approach.

Main Methods:

• Collected hyperspectral reflectance data from diverse plant species under various conditions.
• Assessed narrow-band spectral indices for LDMC prediction.
• Trained and compared individual full-spectrum machine learning models.
• Integrated models into a stacked ensemble framework with a meta-learner.

Main Results:

• Narrow-band spectral indices demonstrated limited predictive performance for LDMC.
• Individual full-spectrum machine learning models showed moderate predictive ability.
• The stacked ensemble model achieved a high coefficient of determination (R2=0.896) on an independent test set.
• The ensemble approach significantly outperformed individual models.

Conclusions:

• Hyperspectral data combined with machine learning, particularly stacked ensembles, offers a robust method for LDMC estimation.
• The developed stacked ensemble model enhances accuracy and reliability in predicting leaf biochemical properties.
• This approach holds significant potential for large-scale vegetation monitoring and analysis.