We consider an integrated biomass inventory and energy production problem that arises from scenario analysis in national energy planning. It addresses together two decision stages that have previously been kept distinct: the decisions on the purchase of biomass and the decisions on the production of electricity and heat of each power plant. We model the problem as a stochastic mixed 0-1 integer linear programming problem. Since practical instances of this problem are very large, we experimentally assess a relaxed formulation to obtain near-optimal solutions. In addition, we study a Benders decomposition that exploits the problem structure of the relaxed formulation. In a distributed computing architecture this decomposition allows to increase the number of included scenarios and thus to better address the uncertainty of the data. Computational results indicate that at parity of information the approximation provided by the relaxed model is good. However, by allowing to increase the amount of information treated it can provide more accurate predictions. On a multicore computing architecture a state-of-the-art MIP solver operating on the undecomposed model is sufficient to achieve similar performance as the Benders decomposition. However, the use of the solver in a distributed computing environment is not obvious and the Benders decomposition is a more easily implementable and scalable approach.