Estimating biophysical parameters of rice with remote sensing data using support vector machines

Area of Science:

• Agricultural Remote Sensing
• Plant Physiology
• Spectroscopy

Background:

• Accurate monitoring of rice crop health is crucial for optimizing agricultural yields.
• Leaf Area Index (LAI) and Green Leaf Chlorophyll Density (GLCD) are key biophysical parameters indicating plant health and nitrogen status.

Purpose of the Study:

• To evaluate the efficacy of hyperspectral reflectance transformations in predicting rice LAI and GLCD.
• To compare the predictive performance of Support Vector Machines (SVMs) against Stepwise Multiple Regression (SMR) models.

Main Methods:

• Hyperspectral reflectance measurements (350-2500 nm) were collected from experimental rice fields under varying nitrogen levels.
• Four data transformations (reflectance, first-order derivative, second-order derivative, logarithm) were applied to the spectral data.
• Stepwise Multiple Regression (SMR) and Support Vector Machines (SVMs) were employed for predictive modeling.

Main Results:

• The polynomial kernel SVM using raw reflectance (R) achieved the best prediction for LAI (RMSE = 1.0496).
• The analysis of variance kernel SVM using logarithm-transformed reflectance (LOG) was optimal for GLCD prediction (RMSE = 523.0741).
• SVM models demonstrated superior performance compared to SMR models for both parameters.

Conclusions:

• Hyperspectral data, particularly when transformed and analyzed with SVMs, is a powerful tool for non-destructively assessing rice biophysical parameters.
• SVMs offer a robust and versatile approach for analyzing spectral data in precision agriculture applications.
• This study highlights the potential of advanced machine learning techniques for improving crop monitoring and management.