Estimation of terrestrial carbon fluxes over east Asia through satellite remote sensing and AsiaFlux data

Abstract

Understanding how carbon fluxes between the land and atmosphere change spatio-temporally is critical for researches on global carbon cycle and climate change. In this study, we quantified the exchanges of terrestrial carbon fluxes over East Asia using various machine learning techniques with multiple satellite-derived products and AsiaFlux in-situ data. Net Ecosystem Exchange (NEE) as a carbon flux estimate was calculated from AsiaFlux data. Various satellite-derived products that might be related to carbon fluxes were used, including LST, NDVI, EVI, FPAR, LAI, GPP, land cover and rainfall. Machine learning techniques used in this study were random forest, support vector regression and Cubist. The machine learning models were compared in terms of performance and importance of input variables was also examined.