Aboveground biomass estimation using multimodal remote sensing observations and machine learning in mixed temperate...

Area of Science:

• Forestry
• Remote Sensing
• Ecology

Background:

• Aboveground tree biomass (AGB) is vital for forest carbon storage assessment.
• Traditional Forest Inventory and Analysis (FIA) methods lack the sampling intensity for fine-resolution AGB estimates.
• Remote sensing (RS) offers a more efficient approach to forest carbon monitoring.

Purpose of the Study:

• To develop accurate predictive models for aboveground tree biomass (AGB) using the random forest (RF) algorithm.
• To assess the effectiveness of various remote sensing data sources in AGB estimation.
• To enhance the efficiency and accuracy of forest carbon monitoring for decision-making.

Main Methods:

• Utilized 67 explanatory variables derived from three remote sensing data sources (LiDAR, aerial imagery, satellite images).
• Developed nine random forest (RF) models for AGB prediction, each undergoing variable selection and hyperparameter tuning.
• Evaluated model performance using metrics such as Root Mean Square Error (RMSE) and R-squared (R²).

Main Results:

• The optimal RF model incorporated 28 explanatory variables, achieving an RMSE of 27.19 Mgha⁻¹ and an R² of 0.41.
• Combining LiDAR data with aerial and satellite image metrics significantly improved AGB prediction accuracy.
• The study demonstrated the potential of integrated RS data for large-area biomass mapping.

Conclusions:

• Remote sensing data, particularly when combined (e.g., LiDAR with imagery), significantly enhances the accuracy of aboveground tree biomass (AGB) estimation.
• The random forest algorithm provides a robust framework for developing predictive AGB models.
• Improved AGB estimation using RS is crucial for effective carbon stock assessment and informed climate change decision-making.