Atlantic tropical cyclones (TCs) affect the United States every year, and depending on their intensity, can cause coastal storm surge and inland inundation, leading to substantial damages in the east coast. Accurate TC risk assessment, however, is not easily quantified given the limited spatial and temporal historical records. This paper describes the development of a model framework for TC forecasts based on large ensemble generation. We illustrate the capability of the ensemble technique to improve TC forecasts by generating probabilistic forecasts of track and intensity. We show that the using this approach can generate tracks that are statistically similar to those of the underlying historical records, even TCs with long return periods. We began with historical compilations of the NOAA National Climatic Data Center (NCDC) tropical cyclone database. TCs reaching a hurricane strength and making landfall in or passing close to the United States were identified. The geographical area influenced by these hurricanes was discretized and the parameters of Markov chains and multivariate distributions were derived for each discretized area. Synthetic tracks were generated using repetitive random draws from the spatiotemporal distribution of historical genesis and storm motion, conditioned by Markov chains for each 6-hour displacement. Kolmogorov–Smirnov (K–S) tests revealed that the ensemble members and ’best track’ data are drawn from the same underlying population at the 95% significance level, with no test statistics exceeding the critical value of 0.13. The proposed algorithm is validated in both macro and micro scales. The results revealed that the general pattern of hurricane activities conforms well to the historical observations with high spatial correlations, over 90% for the Saffir–Simpson hurricane categories of 1 through 4 and near 80% for the category 5. The track generator also produces a history of potential wind and translational speeds for both of these scales.