GNSS are widely used in localization applications. These systems have the advantage of being able to locate, at low-cost, receivers anywhere on the planet's surface without prior knowledge of the position. However, the accuracy of the provided position strongly depends on the quality of reception and on the number of signals received. And these depend on the environment around the receiver. Indeed, certain environments such as urban canyons or forests are known to be challenging because multiple phenomena such as multipath, Non-Light Of Sight (NLOS), or interference are common there. These phenomena induce an erroneous estimate of the position. In applications where the localization function does not constitute a critical function involving the safety of the overall system, this erroneous estimate only constitutes an inconvenience. But in applications for which the safety of goods or people is at stake, the tolerance of what is called the integrity risk is infinitesimal, not to say null. Moreover, in this type of application, it is not so much the position that matters but the ability to limit the unknown position error. For that, the literature proposes different possibilities of protection level calculation. The ideal protection level would be the one that limits the unknown error as closely as possible without ever underestimating it. This makes it possible to minimize hazardous operations as well as unavailability of the localization function. In this paper, we are interested in the impact of GNSS observation weighting models on the protection levels. We propose a model capable of carrying a multi-criteria and multi-parametric strategy allowing a better adaptivity to the navigation context. The encouraging experimental results show that a parametric and constrained modelling strategy of errors, from the design step, relaxes the requirements on the calculation of the protection levels. They also show the complementarity between the characterization of errors which must be concerned with the functional behaviour and a Fault Detection and Exclusion layer which is responsible for the dysfunctional behaviour.