Carbon dioxide (CO₂) is the most important anthropogenic greenhouse gas and its concentration in atmosphere has been increasing rapidly due to the increase of anthropogenic CO₂ emissions. Quantifying anthropogenic CO₂ emissions is essential to evaluate the measures for mitigating climate change. Satellite-based measurements of greenhouse gases greatly advance the way of monitoring atmospheric CO₂ concentration. In this study, we propose an approach for estimating anthropogenic CO₂ emissions by an artificial neural network using column-average dry air mole fraction of CO₂ (XCO₂) derived from observations of Greenhouse gases Observing SATellite (GOSAT) in China. First, we use annual XCO₂ anomalies (dXCO₂) derived from XCO₂ and anthropogenic emission data during 2010–2014 as the training dataset to build a General Regression Neural Network (GRNN) model. Second, applying the built model to annual dXCO₂ in 2015, we estimate the corresponding emission and verify them using ODIAC emission. As a results, the estimated emissions significantly demonstrate positive correlation with that of ODIAC CO₂ emissions especially in the areas with high anthropogenic CO₂ emissions. Our results indicate that XCO₂ data from satellite observations can be applied in estimating anthropogenic CO₂ emissions at regional scale by the machine learning. This developed method can estimate carbon emission inventory in a data-driven way. In particular, it is expected that the estimation accuracy can be further improved when combined with other data sources, related CO₂ uptake and emissions, from satellite observations.

Atmospheric CO₂ concentrations are sensitive to the effects of climate extremes on carbon sources and sinks of the land biosphere. Therefore, extreme changes of atmospheric CO₂ can be used to identify anomalous sources and sinks of carbon. In this study, we develop a spatiotemporal extreme change detection method for atmospheric CO₂ concentrations using column-averaged CO₂ dry air mole fraction (XCO₂) retrieved from the Greenhouse gases Observing SATellite (GOSAT) from 1 June 2009 to 31 May 2016. For extreme events identified, we attributed the main drivers using surface environmental parameters, including surface skin temperature, self-calibrating Palmer drought severity index, burned area, and gross primary production (GPP). We also tested the sensitivity of XCO₂ response to changing surface CO₂ fluxes using model simulations and Goddard Earth Observing System (GEOS)-Chem atmospheric transport. Several extreme high XCO₂ events are detected around mid-2010 over Eurasia and in early 2016 in the tropics. The magnitudes of extreme XCO₂ increases are around 1.5–1.8 ppm in the Northern Hemisphere and 1.2–1.4 ppm in Southern Hemisphere. The spatiotemporal pattern of detected high XCO₂ events are similar to patterns of local surface environmental parameter extremes. The extreme high XCO₂ events often occurred during periods of increased temperature, severe drought, increased wildfire or reduced GPP. Our sensitivity tests show that the magnitude of detectable anomalies varies with location, for example 25% or larger anomalies in local CO₂ emission fluxes are detectable in tropical forest, whereas anomalies must be half again as large in mid-latitudes (~37.5%). In conclusion, we present a method for extreme high XCO₂ detection, and large changes in land CO₂ fluxes. This provides another tool to monitor large-scale changes in the land carbon sink and potential feedbacks on the climate system.

Using measurements of the column-averaged CO₂ dry air mole fraction (XCO₂) from GOSAT and biosphere parameters, including normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), leaf area index (LAI), gross primary production (GPP), and land surface temperature (LST) from MODIS, this study proposes a data-driven approach to assess the impacts of terrestrial biosphere activities on the seasonal cycle pattern of XCO₂. A unique global land mapping dataset of XCO₂ with a resolution of 1° by 1° in space, and three days in time, from June 2009 to May 2014, which facilitates the assessment at a fine scale, is first produced from GOSAT XCO₂ retrievals. We then conduct a statistical fitting method to obtain the global map of seasonal cycle amplitudes (SCA) of XCO₂ and NDVI, and implement correlation analyses of seasonal variation between XCO₂ and the vegetation parameters. As a result, the spatial distribution of XCO₂ SCA decreases globally with latitude from north to south, which is in good agreement with that of simulated XCO₂ from CarbonTracker. The spatial pattern of XCO₂ SCA corresponds well to the vegetation seasonal activity revealed by NDVI, with a strong correlation coefficient of 0.74 in the northern hemisphere (NH). Some hotspots in the subtropical areas, including Northern India (with SCA of 8.68 ± 0.49 ppm on average) and Central Africa (with SCA of 8.33 ± 0.25 ppm on average), shown by satellite measurements, but missed by model simulations, demonstrate the advantage of satellites in observing the biosphere–atmosphere interactions at local scales. Results from correlation analyses between XCO₂ and NDVI, EVI, LAI, or GPP show a consistent spatial distribution, and NDVI and EVI have stronger negative correlations over all latitudes. This may suggest that NDVI and EVI can be better vegetation parameters in characterizing the seasonal variations of XCO₂ and its driving terrestrial biosphere activities. We, furthermore, present the global distribution of phase lags of XCO₂ compared to NDVI in seasonal variation, which, to our knowledge, is the first such map derived from a completely data-driven approach using satellite observations. The impact of retrieval error of GOSAT data on the mapping data, especially over high-latitude areas, is further discussed. Results from this study provide reference for better understanding the distribution of the strength of carbon sink by terrestrial ecosystems and utilizing remote sensing data in assessing the impact of biosphere–atmosphere interactions on the seasonal cycle pattern of atmospheric CO₂ columns.