Karst vegetation coverage detection using UAV multispectral vegetation indices and machine learning algorithm.

Area of Science:

• Ecological remote sensing
• Geospatial analysis
• Environmental monitoring

Background:

• Karst vegetation is crucial for ecological restoration in karst landscapes.
• Vegetation Indices (VIs) help assess ecological restoration status.
• Remote sensing and machine learning are advancing karst vegetation surveys.

Purpose of the Study:

• To evaluate Unmanned Aerial Vehicle (UAV) multispectral data for karst vegetation detection.
• To compare machine learning models for vegetation coverage accuracy.
• To identify important spectral features for vegetation detection.

Main Methods:

• Collected UAV multispectral data at 100m, 200m, and 400m altitudes.
• Compared Random Forest, Support Vector Machine, Gradient Boosting Machine, and Deep Learning models.
• Analyzed spectral values and Vegetation Indices (VIs) for variable importance.

Main Results:

• Gradient Boosting Machine (GBM) model achieved 95.66% overall accuracy using all data.
• Modified Soil Adjusted Vegetation Index (MSAVI) was significantly correlated with vegetation detection.
• The best model accurately predicted vegetation and other land types.

Conclusions:

• The GBM_all model is feasible for accurate karst vegetation detection using UAV data.
• UAV imaging at various altitudes provides a methodological reference for karst vegetation monitoring.
• The study confirmed the effectiveness of remote sensing and machine learning for ecological assessments.