MIT Center for Global Change Science
| Report Series | Contact |
November 1992
Submitted to the Dept. of Earth, Atmospheric and Planetary Sciences on October 13, 1992 in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Atmospheric Chemistry
To obtain a copy of this Report in paper format, use the request form
or send an e-mail to
cgcs@mit.edu, specify Report 17, and provide a complete return postal address.
We investigate the feasibility of using an inverse method based on a linear Kalman filter to determine regional surface fluxes through comparisons between observations and predictions in a 3-D atmospheric transport model. The ability of the present ALE/GAGE observation sites to quantify the regional fluxes of anthropogenic trace gases is also studied. These investigations are done using CFCl3 as the test tracer since its sources are relatively well known. The first of these investigations is done in the low resolution spectral model of Golombek and Prinn (1986) to enable many feasibility tests to be affordably run. Once convinced that the Kalman filter can deduce regional sources, we test the resolving capabilities of a higher resolution model which would be more promising for actually deducing unknown surface fluxes. For this purpose we used the National Center for Atmospheric Research's (NCAR) Community Climate Model (CCM2).
The tests in the low resolution model are first done with the transport model being perfect in the sense that the "observations" were produced by running the model with the CFCl3 emissions derived from industry data. The inverse method used is capable in this case of accurately determining regional surface fluxes using the present ALE/GAGE sites and to converge to the correct solution within a year or two even using initial emission guesses very different from the final solution. We also investigate how well the Kalman filter approach works with a less than perfect chemistry-circulation model by using the ALE/GAGE observations of CFCl3 for the inversion. The success of this inversion depends largely on the ability of the model circulation to predict observed concentrations of CFCl3 since its chemistry is reasonably well understood. The larger the difference between the model and the observed values using the real (industry) emissions then the larger the bias will be in the estimated emissions. Such studies can help to understand the inherent biases in the model when used in an inverse scheme before trying to use the model to estimate unknown surface fluxes such as those for methane, nitrous oxide and carbon dioxide. We also investigate where additional observational stations could be placed to enhance the capability of the present ALE/GAGE network for determining regional net fluxes. It appears that Hateruma (24N, 123E) and to a lesser extent Kamchatka (51N, 156E) are very promising locations for new stations and that Hateruma is superior to the ALE/GAGE Oregon station in providing information about Asian sources. This type of analysis can aid the process of choosing observation sites by addressing how well each site contributes to the different goals for use of the data.
These tests have shown the Kalman filter can deduce regional sources provided the model is accurate enough. We have shown the Kalman filter can deduce regional sources provided the model is accurate enough. We have chosen NCAR's CCM2 as a promising candidate. The ability of this model to simulate atmospheric CFCl3 at the ALE/GAGE observing sites is tested. The model resolves the high frequency events seen in the data in most cases. However, the phasing is not always correct. Next we perform a series of test inversions using CFCl3 in the CCM2. We find that due to the high resolution and non-linearities of the CCM2 there are new considerations for posing inverse problems in such models. Despite the needed changes for the posing of the Kalman filter the results do suggest that the regional source strengths cannot be constrained using the ALE/GAGE observations in this high resolution transport model using calculated dynamics. This may put into question the use of transport models using calculated dynamics to deduce unknowns from observations of tracers. For such questions, a transport model based on observed dynamics may be better suited. Furthermore, it is also apparent that more and better located observation sites could improve our resolving capabilities.
