Abstract: Uncertainty in the numerical simulation of groundwater DNAPLs migration is inevitable due to the complex and unknown hydrogeological conditions in the actual world. In order to overcome the computational burden caused by repetitive calls of the DNAPLs migration model in uncertainty analysis, this study is carried out, based on the traditional sparse grid (SG) surrogate model. In this paper, an improved SG method DA-LA-SG is proposed, which couples the local adaptive (LA) and dimensional adaptive (DA). The surrogate efficiency and accuracy of the proposed DA-LA-SG are verified through two analytical cases and a PCE migration laboratory experiment, and the proposed model is applied to the uncertainty analysis of indoor sandbox PCE migration simulation. The results show that in the early stage of establishing the surrogate model, the surrogate efficiency of LA-SG is better than or close to that of DA-SG and DA-LA-SG, but the efficiency of DA-SG and DA-LA-SG is gradually better than that of LA-SG with the increasing number of the interpolation nodes. DA-LA-SG performs the best. DA-LA-SG can be used to efficiently and accurately establish the likelihood function surrogate model of the PCE migration model, and analyze the uncertainty of parameters in PCE migration simulation. The posterior distribution of the model parameters is obtained, and it is found that the background medium permeability k1, the fitting coefficient n1 and the weak lens permeability k2 can be identified clearly, but the background medium porosity μ1, the weak lens porosity μ2 and the fitting coefficient n2 are basically evenly distributed, which are poorly recognized and insensitive to saturation observations.