Performance Evaluation of Machine Learning Methods for Leaf Area Index Retrieval from Time-Series MODIS Reflectance...

Area of Science:

• Earth Observation
• Biophysical Parameter Retrieval
• Machine Learning Applications

Background:

• Leaf Area Index (LAI) is a critical biophysical parameter for ecological studies.
• Current remote sensing LAI retrieval methods suffer from spatial and temporal gaps due to clouds and aerosols.
• Generating spatially complete and temporally continuous LAI products is essential for practical applications.

Purpose of the Study:

• To evaluate the performance of four machine learning algorithms for retrieving LAI from time-series MODIS reflectance data.
• To compare the effectiveness of different spectral band combinations for LAI retrieval.
• To identify the most suitable machine learning algorithm for generating high-quality LAI products.

Main Methods:

• Utilized time-series Moderate Resolution Imaging Spectroradiometer (MODIS) reflectance data.
• Applied four machine learning algorithms: Back-Propagation Neural Network (BPNN), Radial Basis Function Networks (RBFNs), General Regression Neural Networks (GRNNs), and Multi-Output Support Vector Regression (MSVR).
• Evaluated algorithm performance based on sensitivity to training sample size, spectral band usage, and training time.

Main Results:

• GRNNs, RBFNs, and MSVR showed low sensitivity to training sample size, unlike BPNN.
• Using red, near-infrared (NIR), and short-wave infrared (SWIR) bands improved retrieval accuracy compared to using only red and NIR bands or single bands.
• GRNNs consistently outperformed the other three algorithms across various conditions.

Conclusions:

• Machine learning, particularly GRNNs, offers a robust approach for overcoming limitations in traditional LAI retrieval methods.
• The choice of spectral bands significantly impacts LAI retrieval accuracy.
• GRNNs provide a promising solution for generating reliable, gap-filled LAI products essential for regional and global applications.