[P1-17] Estimation of carbon flux in urban vegetation ecosystems using multi-satellite data based on machine learning
Keywords:Carbon flux, Urban, Machine learning
Achieving carbon neutrality in urban areas is critical for mitigating global climate change. Quantifying urban carbon exchange is essential for managing urban ecosystems and assessing the effects of green management practices. While anthropogenic emissions play an important role for urban carbon exchange, this study first focused on urban vegetation to accurately estimate natural carbon flux at high resolution using multi-satellite data and machine learning (ML). The recently developed ML-based carbon flux model uses various satellite and numerical model data from MODIS, Himawari-8, and LDAPS with 250 m spatial resolution, and evaluated for flux tower sites in South Korea. To extend the applicability of the model to urban areas, high resolution satellite data from Sentinel-2 and Landsat-8 were considered in the original model structure. Tree-based models as well as deep neural networks were further examined their performance for estimating natural carbon flux in urban areas. The modified model for urban areas is capable of reflecting urban vegetation signals on various urban green spaces that were neglected in the original coarse resolution model.