[Temporal stage and method selection of tree species classification based on GF-2 remote sensing image]

Area of Science:

• Forestry
• Remote Sensing
• Ecology

Background:

• Accurate tree species identification is crucial for understanding forest ecosystems.
• Applying domestic high-resolution satellite data (Gaofen) for tree species classification requires methodological exploration.

Purpose of the Study:

• To evaluate the effectiveness of different image combinations, classification features, and classifiers for tree species classification using Gaofen-2 data.
• To investigate the impact of temporal data selection on classification accuracy.

Main Methods:

• Object-based image analysis using Gaofen-2 satellite data.
• Construction of single and multi-temporal datasets.
• Feature optimization using C5.0 algorithm.
• Classification using Support Vector Machine (SVM) and Random Forest (RF) algorithms.

Main Results:

• Overall classification accuracy ranged from 63.5% to 83.5%, with Kappa coefficients from 0.57 to 0.81.
• Multi-temporal data generally yielded better results than single-temporal data.
• Feature optimization positively impacted classification accuracy.
• Support Vector Machine (SVM) showed stability, while Random Forest (RF) performance depended on feature quality.

Conclusions:

• Multi-temporal Gaofen-2 data, optimized features, and appropriate classifiers enhance tree species classification.
• Temporal data selection significantly influences classification outcomes.
• SVM and RF offer distinct advantages depending on data characteristics and classification challenges.