Most studies on semantic question answering (QA) are predominantly focused on encyclopedic knowledge graphs like DBpedia and Wikidata. These studies cover, if at all, the spatial and temporal characteristics of geospatial entities in isolation, not addressing them simultaneously. In this paper, we introduce a pipeline for creating question answering datasets for evaluating the reasoning capabilities of QA models in the context of geographic changes over time. This pipeline generates questions, GeoSPARQL queries, and corresponding answers by leveraging subgraph and query template extraction techniques. We exemplify this pipeline with the creation of the GeoChangesQA dataset with questions over a knowledge graph of US counties and states and their changes from 1629 to 2000. By evaluating GeoChangesQA using a Transformer-based model, we demonstrate that historical geospatial questions pose a substantial challenge for semantic question answering.