In many environmental management problems, the construction of occurrence maps of species of interest is a prerequisite to their management. However, this task is difficult since observations are often imperfect and costly. There is a lack of methodological tools for designing efficient spatial sampling strategies. We developed original model-based static and adaptive sampling methods dedicated to the construction of occurrence maps. We adopt the classical image analysis probabilistic framework of Hidden Markov Random Fields and we express the sampling strategy as the solution of an optimisation problem. Resolution is intractable and we propose heuristic approaches to derive approximately optimal solutions both for static and adaptive sampling. An empirical comparison, on simulated data, of the adaptive and static HMRF-based sampling methods with random sampling and adaptive cluster sampling, shows that in most cases the adaptive sampling approach, when possible, provides a restored map of better accuracy than the other approaches.