Aims. We use four observational data sets, mainly from the Rosetta mission, to constrain the activity pattern of the nucleus of comet 67P/Churyumov-Gerasimenko (67P). Methods. We developed a numerical model that computes the production rate and non-gravitational acceleration of the nucleus of comet 67P as a function of time, taking into account its complex shape with a shape model reconstructed from OSIRIS imagery. We used this model to fit three observational data sets: the trajectory data from flight dynamics; the rotation state as reconstructed from OSIRIS imagery; and the water production measurements from ROSINA of 67P. The two key parameters of our model, adjusted to fit the three data sets all together, are the activity pattern and the momentum transfer efficiency (i.e., the so-called η parameter of the non-gravitational forces). Results. We find an activity pattern that can successfully reproduce the three data sets simultaneously. The fitted activity pattern exhibits two main features: a higher effective active fraction in two southern super-regions (~10%) outside perihelion compared to the northern regions (<4%), and a drastic rise in effective active fraction of the southern regions (~25−35%) around perihelion. We interpret the time-varying southern effective active fraction by cyclic formation and removal of a dust mantle in these regions. Our analysis supports moderate values of the momentum transfer coefficient η in the range 0.6-0.7; values η ≤ 0.5 or η ≥ 0.8 significantly degrade the fit to the three data sets. Our conclusions reinforce the idea that seasonal effects linked to the orientation of the spin axis play a key role in the formation and evolution of dust mantles, and in turn, they largely control the temporal variations of the gas flux.