Systematic underestimation of temporal trends in lower tropospheric CO over coastal cities from MOPITT Level 3 data

MOPITT data are valuable for the analysis of temporal trends in CO concentrations, owing to the nearly unbroken record of observations since the start of the 21st century. The information content of MOPITT's CO profile retrievals is however sensitive to a range of scene-specific factors such as thermal contrast in the lower troposphere. To maximise the information content of MOPITT data and minimize the influence of the a priori, users are advised to restrict analysis to daytime observations over land during the summer season, since this is when thermal contrast conditions are greatest, thus maximizing the instrument's ability to sense CO in the lowermost layers of the troposphere. While such filtering is possible with Level 2 ('L2') data, where each individual retrieval (22 x 22 km spatial resolution) is available for analysis, Level 3 ('L3') data are a 1^ox 1^oarea-averaged version of the L2 retrievals bounded by an L3 gridbox. In the case of L3 gridboxes that straddle the coastline ('coastal L3 gridbox'), this often leads to the averaging together of retrievals over both land and water, which usually have very different lower tropospheric sensitivity characteristics, especially in the summer season. This results in a loss of information in the L3 data and, subsequently, a significant bias towards the a priori profile in the lower troposphere - an a priori that is derived from a multi-year model climatology and features no temporal trend. This is of great importance for analyses of temporal trends in lower tropospheric CO concentrations. We show, for the case study city of Halifax, Canada, which is located within a coastal L3 gridbox, that the decrease in summertime CO detected using L3 data is 3 times slower than that detected with the underlying L2 retrievals filtered so that only the profiles retrieved over land remain (-1.16 ppbv y^-1vs -3.28 ppbv y^-1). Finally, given that approximately half of the world's 100 largest cities by population are found in coastal L3 gridboxes, we document how many reveal a similar discrepancy when they are the target of temporal trend analyses using MOPITT data.