The aim of this paper is to show the use of a multivariate autoregressive model for synthetic hydrologic series generation. Such model uses as input data the streamflows historical series but it also considers an exogenous variable for classes determination. This variable is the SST which is related, not just statistical but physically with the streamflows. As an example, the development of the model for three specific rivers of Colombian system is presented in a long term horizon, using the NOAA's SST medium term forecast.