Author(s): Kaitong Qin; Songjie Wu; Chen Chen; Tiejian Li; Jiaye Li
Linked Author(s): Chen Chen
Keywords: Carbon emissions; GRNN; Remote sensing; XCO2
Abstract: Global climate warming is advancing at a unprecedent pace, primarily fueled by the rapid increase in carbon emissions. Unlike traditional carbon emission inventories, the satellite-based observations of column carbon dioxide concentration (XCO2) provide a new perspective for estimating carbon emissions. However, a key challenge lies in defining the XCO2 background, essential for determining XCO2 enhancement, and establishing relationship between XCO2 (anomaly) and carbon emissions. This study comparatively analyzes various methods for defining XCO2 background areas, including national, latitudinal ranges, or non-emission areas approaches. Then a novel method for carbon emission estimation is proposed, leveraging K-means clustering and Generalized Regression Neural Network (GRNN). Using XCO2 (anomaly), net primary productivity (NPP), and population distribution as inputs, and the Open-source Data Inventory for Anthropogenic CO2 (ODIAC) as output, the GRNN model is trained on data from 2014 to 2019 and then used to estimate the spatial distribution of national carbon emissions in 2020. A relatively well-performing model attains a determination coefficient (R2) of 0.969 and Mean Absolute Error (MAE) of 0.186 gC/m2/day, demonstrating a significant improvement compared to existing research. The findings can provide recommendations for XCO2 enhancement and improve our understanding of satellite-based carbon estimation.
DOI: https://doi.org/10.3850/iahr-hic2483430201-217
Year: 2024