Comparison between the Local Ensemble Transform Kalman Filter (LETKF) and 4D‐Var in atmospheric CO2 flux inversion with the Goddard Earth Observing System‐Chem model and the observation impact diagnostics from the LETKF. Issue 21 (5th November 2016)

• Record Type: Journal Article
• Authors: Liu, Junjie; Bowman, Kevin W.; Lee, Meemong
• Is Part Of: Journal of geophysical research. Volume 121:Issue 21(2016)
• Page Start: 13, 066
• Page End: 13, 087
• Publication Date: 2016-11-05
• Subjects: LETKF -- 4D‐Var -- GEOS‐Chem -- GOSAT -- observation impact; Atmospheric physics -- Periodicals; Geophysics -- Periodicals; 551.5
• DOI: 10.1002/2016JD025100
• Languages: English
• ISSNs: 2169-897X

Abstract

Abstract: Ensemble Kalman filter (EnKF) and 4D‐Variational (4D‐Var) are two advanced data assimilation methods that are the basis of numerical weather prediction and have been extensively used in trace gas assimilation and inverse modeling. In this study, we compare 4D‐Var and the Local Ensemble Transform Kalman Filter (LETKF), one type of EnKF, in estimating CO2 fluxes with both simulated and real satellite data from Greenhouse gases Observing Satellite (GOSAT) and propose a method to calculate flux changes and flux error reductions from assimilating each observation within the LETKF. The results show that the mean posterior flux accuracy across 11 land regions defined by the Atmospheric Tracer Transport Model Intercomparison Project is comparable between 4D‐Var and the LETKF, as shown in the Observing System Simulation Experiment, but the differences between the LETKF and 4D‐Var are relatively larger over data sparse regions. We show that this is most likely due to the fact that the observations from a much broader region have impact on flux estimation in 4D‐Var than in the LETKF. As a result, the posterior fluxes from 4D‐Var are more consistent with the atmospheric CO2 growth rate. We find that the inversion results are less dependent on inversion methods with the increase of observations. With real GOSAT observations, we show that the posterior flux changes in 2011 relative to 2010 are more consistent between these two methods than the absolute estimates. Key Points: The inversion results are less dependent on inversion methods with the increase of observations The observations from much broader region influence flux estimation in 4D‐Var than the LETKF over data sparse region The posterior flux changes in 2011 relative to 2010 are more consistent between these two methods than the absolute estimates