Revisiting the quantification of power plant CO2 emissions in the United States and China from satellite: A comparative study using three top-down approaches

Top-down constraints of CO2 emissions from coal-fired power plants are critical to improving the accuracy of CO2 emission inventory and designing carbon reduction strategies. Different top-down models based on satellite observation have been proposed in previous studies, but discrepancies between these models and the underlying drivers are rarely explored, limiting the confidence of their application for monitoring point-source CO2 emissions from satellite. Here, we apply three top-down models to estimate CO2 emissions from individual coal-fired power plants in the United States (US) and China in 2014–2021 from Orbiting Carbon Observatory 2 (OCO-2) satellite observations. The first one applies the Gaussian plume model to optimize emissions by fitting modeled CO2 enhancement to the observation. The second and third methods apply the same inversion framework using the maximum likelihood estimation, but with WRF-Chem and WRF-FLEXPART as forward models, respectively. We evaluate consistency between the three methods in estimating emissions of 10 power plants in the US, using daily reported values from the US Environmental Protection Agency (EPA) as a benchmark, and then apply the three methods to quantify emissions of 13 power plants in China. Results show that the WRF-Chem and WRF-FLEXPART based inversion results are consistent with the EPA reported emission rates, with correlation coefficients (r) of 0.76 and 0.85 and mean biases (MB) of 4.06 and 3.97 ktCO2/d relative to EPA reports at all 10 power plants, respectively. The Gaussian plume model driven by wind fields from WRF-Chem model without the wind rotation shows comparable ability in reproducing the EPA reported emission rates at 7 power plants (r = 0.82, MB = 6.17), but is not applicable for the other three cases. We find that application of the high-resolution three-dimensional wind fields can better capture the shape of observed plumes, especially under complex wind conditions, compared to the Gaussian plume model which relies on wind field at a single point, and thus the Gaussian plume model has difficulty to optimize emissions under inhomogeneous wind fields or when observations are far away from the power plant. In general, using the WRF-FLEXPART model as the forward model in the inverse analysis shows advanced capability to simulate narrow-shape plumes in the absence of numerical diffusion which is inherent in Eulerian model such as WRF-Chem. Emissions estimated by the three top-town methods show a moderate consistency at 13 coal-fired power plant cases in China, with 8 of 13 cases showing differences of <30% between at least two methods. However, large differences emerge when wind fields are inhomogeneous and number of available observations is limited. Using different meteorological wind fields and OCO-2 data versions can also bring substantial differences to the posterior emissions for all three approaches. We find that the posterior CO2 emissions, though only reflecting instantaneous emission rates at satellite overpass time, are not proportional to the reported capacities of these power plants, indicating that attributing CO2 emissions simply based on the capacity of power plants in some bottom-up approaches may have significant discrepancies. Our study exposes the capability and limitation of different top-down approaches in quantifying point-source CO2 emissions, advancing their application for better serving increasing constellations of point-source imagers in the future.