Optimization of Global CO2 Fluxes at High Resolution using the Coupled MLEF-PCTM Model
Ravi
Lokupitiya, Atmospheric Science, Colorado State University, ravi@atmos.colostate.edu
Dusanka
Zupanski, Cooperative Institute for Research in teh Atmosphere, CSU, zupanski@cira.colostate.edu
(Presenting)
Scott
Denning, Atmospheric Science, Colorado State University, denning@atmos.colostate.edu
Nick
Parazoo, Atmospheric Science, Colorado State University, nparazoo@atmos.colostate.edu
Randall
Kawa, NASA GSFC, kawa@maia.gsfc.nasa.gov
Zhengxin
Zhu, NASA GSFC, zhu@code916.gsfc.nasa.gov
Increasing CO2 concentrations in the atmosphere believed to be a significant factor of the global warming. About half of the CO2 emitted by anthropogenic activities is taken up by the sink processes on land and ocean. There is a missing sink, which is hard to locate among the land and ocean processes. Hence the study of spatial and temporal variability of CO2 sources and sinks on the surface is important.
Inverse modeling is widely used to optimize surface CO2 fluxes using the observed concentrations in the atmosphere. Traditionally used Batch Mode inversions solve the problem by dividing the globe into several large regions. However this technique is lack of understanding the smaller scale variations of the fluxes and considering larger regions may leads to aggregation errors. In this study, we attempt to solve the fluxes in much finer scale compared to the batch mode inversions.
We introduced Maximum Likelihood Ensemble Filter (MLEF), coupled with Parameterized Chemistry Transport Model (PCTM) as an observation operator to optimize the surface CO2 fluxes. We conducted two experiments using synthetic data: (1) a very large problem, with observations defined in every grid cell, and (2) a more realistic problem with a network of 85 weekly observations. Our results show coupled MLEF-PCTM model can efficiently process very large observation vectors, and is thus suited for flux estimation using continuous-sampling towers and global satellite retrievals. Severely under constrained nature of the flux estimation problem with current observing system requires aggressive covariance localization and smoothing to obtain reasonable results.