Harnessing Spectral Libraries From AVIRIS-NG Data for Precise PFT Classification: A Deep Learning Approach.

Area of Science:

• Remote Sensing
• Ecology
• Computer Science

Background:

• Traditional land cover classification methods lack the detail to capture subtle variations in plant physiology and biochemistry.
• Hyperspectral sensing offers detailed spectral signatures for improved forest classification and differentiation of plant species and plant functional types (PFTs).

Purpose of the Study:

• To advance the classification and monitoring of PFTs in Shoolpaneshwar wildlife sanctuary, Gujarat, India.
• To develop and utilize a comprehensive spectral library for precise PFT classification using hyperspectral data and machine learning.

Main Methods:

• Acquisition of hyperspectral data using Airborne Visible/Infrared Imaging Spectrometer-Next Generation (AVIRIS-NG) and ASD Handheld Spectroradiometer (400-1600 nm).
• Development of a spectral library for 130 plant species and grouping into five PFTs using Fuzzy C-means clustering.
• Identification of key spectral features using ISODATA clustering and Jeffries-Matusita (JM) distance analysis.
• Evaluation of machine learning classifiers: Parzen Window (PW), Gradient Boosted Machine (GBM), and Stochastic Gradient Descent (SGD).

Main Results:

• The Gradient Boosted Machine (GBM) classifier demonstrated the highest performance.
• GBM achieved an overall accuracy of 0.94 and a Kappa coefficient of 0.93 for PFT classification.
• Effective feature selection was achieved through spectral analysis, enhancing classification accuracy.

Conclusions:

• Hyperspectral sensing, coupled with machine learning, is a critical tool for accurate PFT classification and monitoring.
• The developed spectral library and feature selection methods significantly improve the precision of PFT differentiation.
• The study highlights the potential of remote sensing for ecological assessments in biodiversity-rich areas.