Observing changes in Earth surface topography is crucial for many Earth science disciplines. Documenting these changes over several decades at regional to global scale remains a challenge due to the limited availability of suitable satellite data before the year 2000. Declassified analog satellite images from the American reconnaissance program Hexagon (KH-9), which surveyed nearly all land surfaces from 1972 to 1986 at meter to sub-meter resolutions, provide a unique opportunity to fill the gap in observations. However, large-scale processing of analog imagery remains challenging. We developed an automated workflow to generate Digital Elevation Models and orthophotos from scanned KH-9 mapping camera stereo images. The workflow includes a preprocessing step to correct for film and scanning distortions and crop the scanned images, and a stereo reconstruction step using the open-source NASA Ames Stereo Pipeline. The processing of several hundreds of image pairs enabled us to estimate reliable camera parameters for each KH-9 mission, thereby correcting elevation biases of several tens of meters. The resulting DEMs were validated against various reference elevation data, including snow-covered glaciers with limited image texture. Pixel-scale elevation uncertainty was estimated as 5 m at the 68% confidence level, and less than 15 m at the 95% level. We evaluated the uncertainty of spatially averaged elevation change and volume change, both from an empirical and analytical approach, and we raise particular attention to large-scale correlated biases that may impact volume change estimates from such DEMs. Finally, we present a case study of long-term glacier elevation change in the European Alps. Our results show the suitability of these historical images to quantitatively study global surface change over the past 40–50 years.

观察地球表面地形的变化对于许多地球科学学科而言至关重要。由于在2000年之前合适的卫星数据有限，要记录几十年来从区域到全球范围内的这些变化仍然是一个挑战。来自美国侦察计划Hexagon（KH-9）的解密的模拟卫星图像已解密，该计划对几乎所有地面进行了勘测从1972年到1986年的米到亚米分辨率，为填补观测的空白提供了独特的机会。但是，模拟图像的大规模处理仍然具有挑战性。我们开发了一种自动化的工作流程，可以从扫描的KH-9制图相机立体图像生成数字高程模型和正射照片。该工作流程包括一个预处理步骤，以纠正胶片和扫描的失真并裁剪扫描的图像，以及使用开源NASA Ames立体声管道的立体声重建步骤。数百个图像对的处理使我们能够估计每个KH-9任务的可靠相机参数，从而校正几十米的仰角偏差。针对各种参考海拔数据验证了生成的DEM，包括具有有限图像纹理的冰雪覆盖的冰川。在68％的置信度下，像素级高程不确定性估计为5 m，而在95％的置信度下，小于15 m。我们通过经验和分析方法评估了空间平均海拔变化和体积变化的不确定性，并且我们特别关注可能影响此类DEM体积变化估计值的大规模相关偏差。最后，我们提出了一个长期研究欧洲阿尔卑斯山冰川海拔变化的案例。我们的结果表明，这些历史图像适用于定量研究过去40-50年间的全球表面变化。