Random Forest classification offers an effective, low-cost way to detect changes in urban extent and density over time. We built a classifier that achieves 76% accuracy in a four-way classification task: distinguishing water, dense urban, light urban, and non-urban land. The method produces a time-series of classified areas which can be used to reality check city planners' understanding of their city's urban form: which areas grew fastest, and where density increased. Extending this method through additional input layers, or through engineered input layers such as NDVI and corner detection, could further enhance performance. The researched was possible due to the existence of detailed land use maps created by the Texas GIS services; without such reference data, the method is still feasible but may require creating hand-labeled reference images based on property tax or zoning data. Such methods may be particularly valuable in fast-growing US cities - many concentrated in the Sunbelt - where urban planners are confronting rapid change with spatial patterns that are hard to apprehend; and in fast-growing cities of the developing world where urban planning data is sparse.