Dramatic changes have occurred over the past century on many parts of the planet due to natural factors and intensifying human activities. Understanding these changes is critical for quantifying long-term environmental trends and modeling future conditions. A vast underexploited resource for such analysis lies in historical aerial and satellite stereo imagery captured with analog cameras from the early 1900s to the early 2000s. Originally acquired for military and mapping purposes, the stereoscopic nature of historical images offers unique potential to reconstruct 3D Earth surface changes across the 20th century. Recent algorithmic advances in photogrammetry and computer vision have greatly enhanced this potential. Despite the early recognition of their value, these datasets remain underexploited due to challenges related to 1) fragmented and inaccessible archives, 2) digitization and associated costs, and 3) a lack of scalable automated processing solutions. This review article addresses these challenges by analyzing 198 studies that digitally process historical aerial and satellite stereo imagery scanned from film. As the primary focus of this article is to review the processing of historical imagery, the greater emphasis on 3D reconstruction, such as the generation of digital elevation models (DEMs), and on vertical accuracy reflects both the focus of most published studies and the types of metrics commonly reported in literature. However, we also review work on 2D products, such as orthophotos, when photogrammetric processing is involved. We provide an overview of the accessed archives, processing strategies, and software pipelines. We discuss emerging tools and advances in image matching algorithms and georeferencing solutions, and we highlight how historical imagery can support a wide range of geoscientific applications, from climate change to urban development. Finally, we emphasize the urgent need to unlock these archives and develop efficient reproducible workflows to preserve and exploit this irreplaceable remote sensing dataset before physical degradation or institutional neglect make it inaccessible.
Piermattei et al. (Thu,) studied this question.