Abstract. Emission inventories (EIs) are key-tools for air quality management. However, EIs are expensive, and they have uncertainties. A way to improve the accuracy of EIs is data assimilation. Multiple inverse methods have been used at various scales. However, typically, when applying these methods at the city scale, one encounters, in addition to problems related to the precision of the first guess, or the reliability and representativeness of the observations, or the shortcomings of the dispersion model, the problem of co-location of sources and observation sites. The latter problem results in spurious corrections to the a priori EI. Here we present a methodology to improve an EI of carbon monoxide over a city. We use a 3-D variational approach, in which a cost function that includes balanced terms addressing observation and emission errors is minimized to obtain an ameliorated EI. In addition to positivity, the method addresses the co-location of sources and observations by means of a factor that multiplies the emission error covariance matrix. The factor is chosen so that the reliability of the initial inventory is increased at the observation sites, reducing the local influence of the observations, avoiding spurious corrections to the EI and increasing the temporal and spatial extent of the corrections. The method is applied to Santiago de Chile. We find that the a posteriori inventory shows a decrease in total emissions of 8% with respect to the a priori inventory. Nevertheless, locally over 100% changes are found in the eastern area of Santiago during the morning hours.