Grünblick: an AI Software Toolkit of Above the Ground Biomass Estimation

Abstract

Estimating forest biomass from satellite data is crucial for monitoring global carbon stocks and understanding forests’ role in mitigating climate change. Remote sensing provides large-scale, consistent, and frequent observations, making it more efficient than traditional field measurements. Above the Ground Biomass (AGB) estimates help assess deforestation, forest degradation, and land-use changes, which are vital for conservation efforts and policymaking. Accurate biomass data support the design of accurate climate models and carbon credit programs, ensuring better management of natural resources. Additionally, satellite-based monitoring enhances early detection of illegal logging and forest loss, aiding enforcement and sustainable land management. To support applications that benefit from global-scale biomass monitoring, we present a toolbox for Above-Ground Biomass (AGB) estimation, specifically targeting regions in the Northern Hemisphere. This toolbox features a suite of pixel-wise regression models based on a modern deep learning framework (U-Net). The model is optimized using a variety of backbone architectures, including ResNet and EfficientNet, to enhance performance. Additionally, the toolbox incorporates multiple data modalities, such as the fusion of Sentinel imagery, to improve estimation accuracy.