Three-dimensional modelling is at the critical path to map the by-passed oil in multilayer fluvial systems in the San Jorge Basin. Integrated reservoir modelling teams dedicate an important amount of time to create these three-dimensional models to decrease risk pursuing chemical injection for enhanced oil recovery. Traditional static reservoir modelling requires an important effort from the geologist to construct the interwell correlation. The objective of this work is to show the implementation of two unsupervised algorithms to automate/assist integrated reservoir modelling. We create multiple possible three-dimensional models of real multilayer static reservoirs and accelerate simulation. The first part of the work obtains the stratigraphic representation of the entire reservoir structure. We use the available lithology well logs as spontaneous potential and gamma ray to identify automatically the permeable and shale rocks with unbiased interpretation by their deflection responses in each well for the entire target reservoirs. Then we construct a graph in which each of the deflections is represented by a node. The edges that join each pair of nodes have an assigned weight depending on the difference in depth and the distance in plan of the nodes. Then we use an adapted version of the Girvan-Newman algorithm to make a community detection by eliminating nongeological connections/features, to find the community with greater modularity. These communities represent the existing correlations between the deflections of the different wells. In the second part of this work, we obtain the facies distribution in the reservoir, using 3D Markov chains. We implemented Jaccard distance to measure the mismatch of geological features and objects between the true synthetic case and reconstructed model. Through modeling of two incomplete synthetic cross section cases using Markov chain propagation of a transition matrix the reconstruction reveal that we recover 90% of the original case even when we Input only 5% of the true data initially in the model. Then we tested a very fine real 3D model created by an experienced geologist. The analysis of the reconstructed model features reveals that the Jaccard distance is a reliable indicator of the geological features. Using the computational algorithms implemented, it is possible to obtain a stratigraphic model and the facies model in less than four hours that speed up definitely reservoir modelling.