Prediction of chloride ingress in reinforced concrete structures under complex environmental conditions is essential for assessing and maintaining structural durability. However, conventional computational approaches are often time-consuming. Our recent work developed a data-driven computational framework based on Dynamic Mode Decomposition (DMD for predicting the evolution of chloride ingress and demonstrated its accuracy and computational efficiency. Nevertheless, the robustness of this framework remains to be investigated, particularly with respect to variations in boundary conditions and differences in the data representation of concentration fields. This gap provides the motivation for the present study, in which chloride ingress is treated as a benchmark diffusion problem. First, the predictive performance of the DMD-based framework is examined under various environmental exposure scenarios, including single-surface ingress, multi-surface ingress, and internal chlorine-source ingress. Second, prediction results are compared for two types of input data: Finite Element-based nodal field data defined on unstructured meshes and spatially grid-interpolated field data defined on equivalent Cartesian meshes. This comparison is motivated by the fact that data obtained from experimental measurements or finite element simulations are generally not available on regularly spaced grids. The presented analyses contribute to a better understanding of the robustness of the DMD-based computational framework for predicting chloride ingress in concrete and provide practical insights for further methodological development and data preparation strategies aimed at improving predictive performance.