We consider the problem of area coverage in sensor networks with a realistic physical layer. Most of the previous work in this area has been done with an ideal environment. In this paper, we show that protocols developped in such an environment cannot provide satisfying results in a more realistic world. To do this, we analyze a well-spread localized protocol, from Tian and Georganas, considering the lognormal shadowing model. This study clearly demonstrates that the resulting area coverage is not sufficient, and that some modifications are needed to get back some good results. We thus propose an enhanced version of this protocol and experimentally show that it is very efficient, regardless of the considered environment.