Building an AI dataset to Estimate Vegetation Carbon Using a QGIS-based Annotation Tool

Recently, interest in vegetation biomass has been increasing worldwide due to carbon dioxide regulations. Accordingly, various studies are being conducted to identify local carbon sinks for carbon reduction. In particular, many studies are being conducted to estimate the absorption of forests, which are known as representative sinks, by utilizing remote sensing data to estimate vegetation structural features (tree type, crown density, tree height, understory vegetation, etc.). In the field of remote sensing, spaceborne satellites that can measure biomass include optical-based LANDSAT and Sentinel series and SAR- based ALOS series satellites. Recently, biomass measurement methods using LiDAR data have been attracting attention. In particular, GEDI (Global Ecosystem Dynamics Investigation) data is recently being utilized as data that can measure tree height and biomass with high accuracy through a high- resolution relative height matrix (Qi & Dubayah, 2016). In previous studies, GEDI Lv2 source data was used to analyze the vertical structure of vegetation in tropical rainforests and savanna regions and to estimate biomass(Silva et. al, 2018). Accordingly, this study proposes a method to construct artificial intelligence learning data that can estimate vegetation carbon capture using aerial photographs and satellite images, and to perform deep learning model training such as Quantized U-net, DeepLap v3+, and HRNet using the constructed data, and to verify the predicted capture amount using Spaceborne LiDAR, i.e. GEDI data