Remote sensing-based biomass estimation of dry deciduous tropical forest using machine learning and ensemble analysis

Area of Science:

• Forestry
• Remote Sensing
• Ecology
• Machine Learning Applications

Background:

• Forests are critical for global carbon balance, but face threats from climate change and deforestation.
• Accurate monitoring of forest biomass is essential for tracking carbon stocks and emissions.
• Developing nations face challenges in monitoring forest carbon due to high costs of traditional field measurements.

Purpose of the Study:

• To develop and validate a framework for monitoring above-ground biomass (AGB) at finer scales using open-source satellite data.
• To address the limitations of field-based biomass estimation in developing countries like India.
• To integrate machine learning techniques with satellite data for continuous spatial AGB estimates.

Main Methods:

• Integrated four machine learning (ML) techniques with field surveys and satellite data.
• Applied the framework to a dry deciduous tropical forest in India as a case study.
• Utilized Sentinel-2 satellite data for wet and dry seasons to estimate AGB.

Main Results:

• Random Forest (adjusted R² = 0.91) and Artificial Neural Network (adjusted R² = 0.77) models showed high accuracy for AGB estimation using wet season Sentinel-2 data.
• ML models performed poorly in estimating AGB using dry season satellite data (adjusted R² between -0.05 - 0.43).
• Ensemble analysis improved prediction reliability and quantified spatial uncertainty.

Conclusions:

• The proposed framework provides a scalable and cost-effective method for monitoring forest AGB using open-source satellite data.
• Wet season satellite data and specific ML models (Random Forest, ANN) are more suitable for AGB estimation in the study area.
• Ensemble modeling and uncertainty quantification enhance the robustness of AGB estimates for forest carbon management.