Multi Source Data Fusion for Forest Above Ground Biomass Estimation using Deep Learning

This project leverages deep learning, specifically U-Net-based architectures (U-Net3+, TransU-Net, SwinU-Net, and Attention U-Net), to estimate forest Above-Ground Biomass (AGB) using multi-source remote sensing data. The framework integrates Sentinel-1 radar, Sentinel-2 and Landsat-8 multispectral imagery, climatic variables, and LiDAR-derived topographic parameters under three fusion scenarios. Compared to conventional single-source methods, this approach provides a scalable and accurate pathway for biomass estimation, supporting carbon accounting, climate change monitoring, and sustainable forest management. Explainable AI (XAI) techniques were applied to assess feature importance and interpret model predictions.

Study Area

The study was conducted across heterogeneous forest landscapes with diverse vegetation structures and topographic conditions in the Northwest of the US. The selected sites include regions with varying canopy densities, climatic gradients, and terrain complexity in the Rocky Mountains. These areas provided suitable conditions for testing multi-source data fusion and evaluating the robustness of deep learning models across different environments.

Forests within the study regions are subject to structural heterogeneity and environmental variability. Such characteristics amplify the challenges of biomass estimation and highlight the importance of data fusion approaches that combine spectral, radar, climatic, and topographic information.

Dataset

Inputs:

• Sentinel-1 C-band SAR backscatter

• Sentinel-2 MSI multispectral bands

• Landsat-8 OLI multispectral imagery

• Climatic variables (temperature, precipitation, etc.)

• LiDAR-derived topographic parameters (elevation, slope, aspect)

Labels:

• Above-Ground Biomass (AGB) reference maps derived from ground-truth and validated biomass datasets.

All data are pre-processed into GeoTIFF format and harmonized spatially and temporally for deep learning

https://github.com/sinax9696/Multi-Source-Data-Fusion-for-Forest-Above-Ground-Biomass-Estimation-using-Deep-Learning.git