Spatial datasets (i.e., location-based social media, crime incident reports, and demographic data) often exhibit varied distribution patterns at multiple spatial scales. Examining these patterns across different scales enhances the understanding from global to local perspectives and offers new insights into the nature of various spatial phenomena. Conventional navigation techniques in such multi-scale data-rich spaces are often inefficient, require users to choose between an overview or detailed information, and do not support identifying spatial patterns at varying scales. In this work, we present a context-preserving visual analytics technique that aggregates spatial datasets into hierarchical clusters and visualizes the multi-scale aggregates in a single visual space. We design a boundary distortion algorithm to minimize the visual clutter caused by overlapping aggregates and explore visual encoding strategies including color, transparency, shading, and shapes, in order to illustrate the hierarchical and statistical patterns of the multi-scale aggregates. We also propose a transparency-based technique that maintains a smooth visual transition as the users navigate across adjacent scales. To further support effective semantic exploration in the multi-scale space, we design a set of text-based encoding and layout methods that draw textual labels along the boundary or filled within the aggregates. The text itself not only summarizes the semantics at each scale, but also indicates the spatial coverage of the aggregates and their hierarchical relationships. We demonstrate the effectiveness of the proposed approaches through real-world application examples and user studies.