Machine learning for modeling forest canopy height and cover from multi-sensor data in Northwestern Ethiopia

Area of Science:

• Remote Sensing
• Forestry Science
• Geospatial Analysis

Background:

• Accurate forest height and canopy cover mapping is crucial for biomass estimation, degradation, and restoration monitoring.
• Light Detection and Ranging (LiDAR) sensors provide detailed forest composition data over large areas.
• Global Ecosystem Dynamics Investigation (GEDI) LiDAR offers valuable forest structure information.

Purpose of the Study:

• To predict forest canopy cover and height in tropical regions.
• To integrate GEDI LiDAR data with multisensor imagery and machine learning for enhanced forest mapping.
• To assess the accuracy of combined data sources for forest parameter estimation.

Main Methods:

• Utilized Global Ecosystem Dynamics Investigation (GEDI) LiDAR data.
• Incorporated predictor variables from Shuttle Radar Topography Mission (SRTM) DEM, Sentinel-2, and Sentinel-1 SAR data.
• Employed random forest regression for modeling and validated results using GEDI Level 2A and 2B data.

Main Results:

• Achieved high accuracy in predicting forest canopy height (R²=0.86, RMSE=3.65m) and canopy cover (R²=0.87, RMSE=0.15) at 30m resolution for 2022.
• Demonstrated that combining multiple data sources (Sentinel-1, Sentinel-2, SRTM, GEDI) significantly improves prediction accuracy.
• Generated 30m resolution maps of forest canopy height and cover.

Conclusions:

• The integration of GEDI LiDAR, Sentinel, and SRTM data provides a robust approach for mapping tropical forest canopy characteristics.
• The findings support the development of effective forest management plans for sustainable resource utilization.
• Multi-source data fusion enhances the reliability and accuracy of remote sensing-based forest monitoring.