Conventional simulation of fractured carbonate reservoirs is computationally expensive because multi-scale heterogeneities and fracture-matrix transfer must be taken into account. The computational requirement increases exponentially when multiple simulation runs are required for sensitivity analysis and quantification of the uncertainty range. This can be prohibitive, especially for giant carbonate reservoirs. Yet, sensitivity analysis and uncertainty quantification are particularly important to analyse, determine and rank the impact of geological and engineering parameters on the economics and sustainability of different EOR techniques. We use experimental design to set up multiple screened simulations of a high-resolution model of a Jurassic Carbonate ramp, which is an analogue for the highly prolific reservoirs of the Arab D formation in Qatar. We consider CO2 water-alternating-gas (WAG) injection, which has been shown to be a successful EOR method for carbonate reservoirs. The simulations are used as a basis for generating polynomial response surfaces for prediction of hydrocarbon recovery and net gas utilisation. We compare response surfaces from polynomial regression to response surfaces generated with polynomial chaos expansion (PCE). PCE allows for non-linear mapping of parameter uncertainty to the predicted results. In the current work, the geological parameter uncertainties affecting WAG modelling in fractured carbonates are evaluated. These include end-member fault transmissibility configurations, wettability scenarios, hysteresis models and fracture intensity. Effective fracture permeabilities are computed using discrete fracture networks (DFN) for sparsely distributed regional fractures. The results enable us to adequately explore the parameter space, quantify and rank the interrelated effect uncertain model parameters on CO2 WAG efficiency in fractured carbonate reservoirs. The results highlight the first order impact of fracture intensity, wettability and hysteresis on hydrocarbon recovery and net gas utilisation. Furthermore, reduced order (i.e. proxy) models enable us to calculate quick estimates of the probabilistic uncertainty range and achieve significant computational speed-up compared with the conventional Monte Carlo framework.@en