Canopy height models (CHM) provide detailed environmental vertical structure information and are an important indicator and input for ecological and geospatial applications. These models are often spatiotemporally inconsistent, necessitating additional modeling to scale them in space and time. Yet, such scaling is hindered by a lack of spatially diverse data. To address this, we use United States Geological Survey 3D Elevation Program lidar data to produce 22,796,764 one meter resolution CHM chips, stratified across the dominant land covers of the conterminous United States. For each CHM, we pair a matching time-aligned aerial image from the United States Department of Agriculture National Agriculture Imagery Program. This dataset can be used to train models for large scale CHM production.