The understanding of the real in-service behavior of complex structures is the first step before modeling. That is still a challenge that requires structural monitoring in conjunction with structural modeling. Moreover, structural reassessment based on reliability updating during in-service conditions is needed for the identification and updating of boundary conditions of the structural model. Polynomial chaos allows providing an automatic treatment of data for monitored structures since it leads, on the one hand, to the decomposition of random variables and their efficient representation when considering maximum likelihood estimate and, on the other hand, to provide a format suitable for direct stochastic finite element analysis. In this paper, we illustrate a reliability updating of two monitored port structures to highlight similarities and specificities: we have chosen two on-piles quays instrumented by the GeM in the west coast of France. One of these quays involves an imperfect mechanical connection. One of the novelties of this paper is to identify the probabilistic properties of this defect. These structures are subjected to complex loads and among them we can underline wind actions. In fact, due to global warming, speed of wind is considerably increasing in some areas and will change the reliability level of some infrastructures. We present at first, from measurements of trajectories of stochastic fields of loads along the quay, an updating of basic variables representing random mechanical parameters of sensible components of the structure. Another novelty of the paper concerns similarities in the distribution of basic random variables which allow us to generalize the probabilistic modeling for other non-instrumented and similar quays, i.e., with the same global design and anchorage technology. Then, a mechanical finite element model of the quays is used as a transfer function of the random variable inputs to model the in-service behavior of the structure during severe conditions of wind action on a lift-crane. Non-intrusive stochastic finite element method is finally selected to evaluate the probability of failure for several scenarios.