5.53 The Role of A Priori Information in the Retrieval of CO Profiles from Terra-MOPITT Measurements Shu-peng Ho, John C. Gille, David P. Edwards, Jean-Luc Attie, Merritt N. Deeter, Juying Warner, Gene L. Francis and Daniel Ziskin National Center for Atmospheric Research, Boulder, Colorado 1. INTRODUCTION Anthropogenic activities have significant impacts on biogeochemical cycles. To better understand the changes occurring in the atmosphere and to clearly distinguish natural from anthropogenic influences, it is extremely important to monitor the temporal and spatial distributions of gases and identify their sources and sinks. Carbon monoxide (CO) is one of the key tropospheric trace species. With roughly a 2 month lifetime, and with diverse sources, both natural and anthropogenic (CH 4 oxidation, NMHC oxidation, biomass burning, fossil fuel burning etc.), CO can serve as a useful tracer of atmospheric transport. CO also affects the concentration of the hydroxyl radical (OH), which is involved in much of the chemistry in the troposphere. However, OH has an extremely short lifetime and is difficult to measure. Therefore, the ability to continuously monitor CO from space should provide an important window on tropospheric chemistry. To measure the spatial and temporal variation of the CO profile and total column amount in the troposphere, the Measurements of Pollution In The Troposphere (MOPITT) instrument was launched in 1999 on board the NASA Terra satellite. MOPITT is an eight-channel gas correlation radiometer; each channel generates an average (A) signal and a difference (D) signal (Drummond, 1992). The A signals are sensitive to the background emissions, while the D signals are sensitive to the target gas vertical distribution. MOPITT measurements can resolve the vertical distribution of tropospheric CO in 3-4 layers with a 22X22 km horizontal resolution. The MOPITT operational retrieval is based on the Maximum Likelihood (ML) method (Pan et al., 1998; Rodgers, 2000). The ML retrieval algorithm seeks the statistically most likely CO profile consistent with both the observed radiances and a priori information. The role of the a priori mean profile and covariance matrix is to constrain the retrieved profile to fall within the range of physically realistic solutions (based on variability ____________________________________________ Corresponding author address : Dr. Shu-Peng Ho, Atmospheric Chemistry Division, NCAR, PO BOX 3000 Boulder, Colorado, 80307-3000. E-mail address: spho@ucar.edu, Tel: (303)497-2922 statistics of a selected set of observed in-situ profiles). The relative weighting of the a priori information and information from the measured radiances in the retrieved profile is controlled directly by the a priori covariance matrix and measurement error covariance matrix. The operational MOPITT CO retrieval currently uses a fixed global a priori. This approach was adopted initially to ensure that observed geographical variations in the retrieval results are due to information in the measured radiances rather than features of the a priori. Studies (Hansen et al.,1995; Pan, et al., 1998) have confirmed that the choice of a priori affects the accuracy of the retrievals. There is some question as to whether the use of a global a priori is an adequate representation of the seasonal variation in the CO profiles for diverse locations. Furthermore, only limited regional surface observations and aircraft measurements from field experiments are available for the construction of the MOPITT global a priori. The purpose of this paper is to quantify the sensitivity of the retrieval to the use of a fixed global a priori in the ML method. We conduct simulation experiments to explore the impact of using a fixed global a priori on the MOPITT CO retrievals. This is further illustrated using an alternative criterion to dynamically choose an a priori error covariance matrix (the DAP method) which constrains the CO retrievals using primarily measurement errors. The method will be described in section 2. The simulated CO profiles retrieved from the DAP method are compared to CO profiles retrieved from the ML method in section 3. 2. DATA and the DYNAMIC A PRIORI Method MOPITT has four CO thermal channels, two CO solar reflectance channels and two CH 4 solar reflectance channels. Characteristics of the MOPITT channels are described in Drummond (1992). The radiative transfer equation (RTE) for the upwelling MOPITT A and D signals for each channel is described in Pan et al., (1998). The MOPITT transmittance model (Edwards, et al., 1999) is used in this study to simulate MOPITT radiances while the mixing ratio of target and interfering gases, viewing geometry, vertical thermal profiles and surface emissivity and temperature are