To facilitate the large-scale integration of distributed wind generation (DWG), the uncertainty of DWG outputs needs to be quantified, and the maximum DWG hosting capacity (DWGHC) of distribution systems must be assessed. However, the structure of the high-dimensional nonlinear dependencies and the abnormal marginal distributions observed in geographically dispersed DWG outputs lead to the increase of the complexity of the uncertainty analysis. To address this issue, this paper proposes a novel assessment model for DWGHC that considers the spatial correlations between distributed generation (DG) outputs. In our method, an advanced dependence modeling approach called vine copula is applied to capture the high-dimensional correlation between geographically dispersed DWG outputs and generate a sufficient number of correlated scenarios. To avoid an overly conservative hosting capacity in some extreme scenarios, a novel chance-constrained assessment model for DWGHC is developed to determine the optimal sizes and locations of DWG for a given DWG curtailment probability. To handle the computational challenges associated with large-scale scenarios, a bilinear variant of Benders decomposition (BD) is employed to solve the chance-constrained problem. The effectiveness of the proposed method is demonstrated using a typical 38-bus distribution system in eastern China.