Uncertainty is a frequent issue in high-dimensional data, and a sampling scheme for approximate Bayesian computation based on meta-sampling (MS) was previously proposed to address this issue. MS was organized to map directly the row data onto a Hilbert space of isolation kernel as a result it can improve the sampling scheme; however, it has not been sufficiently studied. Therefore, this study analyzes the dynamics of MS using cross-recurrence plots (CRPs) and illustrates the feasibility of employing cross-recurrence plots as a representation of the Hilbert space based on the isolation kernel. This mapping entails a partitioning approach based on the random forest algorithm, which leverages Voronoi diagrams. Specifically, the Hilbert space is constructed from the identifiers of the Voronoi sites associated with each random tree. Recurrence plots are defined based on the concurrence of site identifiers for both the maxima weighted of the kernel mean embedding and the points generated during meta-sampling. The MS algorithm is designed to generate new points within the parameter space and compute their similarity to an observation without requiring a simulation run. Through this approach, we demonstrate the consistent dynamics between CRPs, and their measures such as the recurrence rate, determinism, number of diagonal lines, their lengths and similarity. CRPs provide detailed insights and can be used as criteria for successful parameter estimation.