Complete Data Fusion of Multi-target Retrieval Products

Observations of the Earth's atmosphere for the vertical profiling of atmospheric variables are provided by many space-borne missions, airborne and ground-based campaigns, aiming at global and continuous measurements which can highlight trends in the atmospheric species and provide the input to the physical and chemical models that are used to predict the evolution of the atmospheric status. In the last two decades, there has been a strong focus on the development of innovative techniques to exploit all the available information from measurements of the same portion of the atmosphere to retrieve the best vertical profile estimate. In this framework, a new method of data fusion, referred to as Complete Data Fusion (CDF), was proposed as a-posteriori algorithm to combine independent measurements of the same profile into a single estimate for a comprehensive and concise description of the atmospheric state. This method uses standard retrieval products and requires a simple implementation. Multi-target retrievals are frequently applied to the analysis of remote sensing observations to determine simultaneously atmospheric constituents reducing the systematic error caused by interfering species. It is thus crucial to adapt the CDF algorithm to fuse profiles obtained from multi-target retrievals, in order to extend its application to a greater number of remote sensing data. In this work we present the results of the first application of the complete data fusion to multi-target retrieval products showing how the inputs of the CDF have to be modified to take into account that state vectors of the fusing measurements may contain only the same atmospheric variables or include different variables as well. We applied the method to simulated measurements in the thermal infrared and in the far infrared spectral ranges, considering the instrumental specifications and performances of IASI-NG and FORUM instruments, respectively. The results obtained demonstrate that the CDF can deal with state vectors from multi-target retrievals both when they contain the same variables and when they have only a subset of variables in common, providing outputs of improved quality with respect to the input data.