In the context of the global biodiversity crisis, human activities are the principal cause of natural habitat degradation, fragmentation, and destruction. Globally, the species extinction rate has reached an unprecedented level in human history and about one million species are nowadays threatened with extinction. Conservation biology is a multidisciplinary research area which attempts to address the current biodiversity crisis challenges. The development of practical approaches to promote conservation and reduce the research-implementation gap is one of its objectives. Last decades, systematic conservation planning (SCP) emerged in this direction as a framework relying on optimization and computer science research. Its main target is to provide decision support in the planning of conservation actions through the integration of ecological targets along with socio-economical constraints. In this PhD thesis, we introduced a formal approach for modelling and solving SCP problems based on constraint programming, a method from artificial intelligence based on automated reasoning. The main motivation of this approach was to provide more expressiveness into SCP (i.e. extend the breadth and variety of problems that users can represent and solve), notably through the integration of advanced spatial constraints and landscape indices. Formal approaches are often more demanding to implement and scale up than heuristic approaches. However, they provide satisfiability and optimality guarantees on the produced solutions. The insights offered by these guarantees can substantially improve the quality of decision support. We evaluated the methods developed in this thesis on real data from New Caledonian forests. As the smallest biodiversity hotspot in the world, New Caledonia has to struggle with many conservation challenges. Moreover, the developed, insular and low populated New Caledonian context allows high proximity between conservation stakeholders, which makes it an appropriate