International audienceThis article investigates the problem of initializing upper-level potential vorticity by using the detection of dry intrusions that can be seen in water vapor images. First, a satellite image processing technique has been developed for the identification and tracking of dry intrusions on geostationary satellite images. This technique can also be applied to images derived from model fields through a radiative transfer model. A linking algorithm automatically compares the trajectories of the dry intrusions in the model and in the satellite images. Differences of brightness temperatures are then converted to differences of tropopause height through a simple linear model, which is based on the correlation found in the background. As the scheme is likely to provide observations of the tropopause height, it also suggests that a space-alignment representation of the errors be used. A simple one-dimensional study provides a depiction of the background error covariance in alignment space, which is compared to the traditional approach of background error covariance in amplitude space. An approximate form of the Ertel potential vorticity operator is then used to incorporate pseudo-observations inside a global four-dimensional variational assimilation scheme. A case study of cyclogenesis highlights the difference between the amplitude-space assimilation of potential vorticity values and the alignment space assimilation of the height of the tropopaus