One of the main objectives of petroleum exploration consists of predicting reservoir location. Data collected in the basin are used to better understand the sedimentary architecture but are usually insufficient to accurately characterize this architecture. Three-dimensional stratigraphic forward modeling has brought new insights in the understanding of sediment distribution. It gives the opportunity to investigate several geological models and to tackle reservoir presence probability. However, simulation time is a strong limitation to properly taking the uncertainties into account during operational studies. Here, we propose a methodology based on metamodels (or surrogate models) to perform sensitivity and risk analyses. The objective is to reduce the simulation time necessary to quantify the regional impact of the input parameters and to estimate probability maps of reservoir presence. The approach consists of building functions that approximate the spatial outputs of the simulator (such as sediment thickness or net-to-gross distributions in the basin) and that are fast to evaluate. These functions are then called instead of the stratigraphic forward simulator for uncertainty quantification. The proposed methodology is applied to a three-dimensional synthetic case study, considering uncertainty on input parameters related to sediment transport, accommodation space, and sediment supply. The sensitivity analysis quantifies in each location the influence of the parameters on the sediment distribution, which can help to better understand the influence of each uncertain process on the basin architecture. In addition, probability maps of reservoir presence are estimated. The proposed approach is a promising trade- off between simulation time and information that can be inferred.