Pattern formation processes on single-crystal catalytic surfaces involve transitions between alternative surface phases coupled with oscillatory reaction dynamics. A qualitative Monte Carlo model has been developed in order to describe dynamics of Pt(1 1 0) surface reconstruction under reactive conditions on atomic scale. We use kinetics of a phenomenological time-dependent Ginzburg-Landau (TDGL) type, which is known to give a correct qualitative picture of oscillations. However, the lattice dynamics model involves realistic effects of nearest neighbor interactions dependent on the surface coverage. The surface phase distribution oscillates chaotically together with adsorbate coverage, and, in addition, undergoes a slow roughening process, resulting in the growth of island size and an increase of the oscillation period.