Abstract:Archival aerial photographs are a unique but underused and potentially game-changing source to study twentieth-century environmental and climate change dynamics. While satellite imagery with comparable high resolution appeared only in the early twenty-first century, archival aerial imagery with native sub-1-meter resolution became ubiquitous in the 1940s. Archival aerial photography therefore quadruples the time depth of high-resolution analysis to eighty years, allowing for a more reliable identification of structural trends. Moreover, the greater time-depth brings into focus the Great Acceleration that started in the 1940s, and virtually in real time. The article uses a human manual analysis of a sample from two time series (1943 and 1971) of archival photographs of the Oshikango area of Namibia (see Figure 1) to demonstrate how aerial photography complements conventional datasets. Namibia was one of the first places in colonial Africa where what subsequently became the standard protocol for "aerial mapping" was used and for which the imagery and the "flight plans" have survived. The standard protocol makes the imagery compatible with any archival aerial photography from the 1940s to 1990s and the flight plans contain key information to identify, interpret, and combine the individual photographs into orthomosaics. Although the use of manual analysis of aerial photography is not new, unlocking the full explanatory potential of high-resolution mass data requires machine reading and analysis. Current machine reading methods, however, are based on the pixel method, which identifies such features as farms, water holes, and trees only as low-resolution pixel aggregates. In contrast, the object method of machine analysis, combined with Geographical Information Systems (GIS) technology to unlock the sub-1-meter native resolution of historical aerial photography, renders visible individual trees and other features, including their precise location and size, allowing for the dimensions of trees and other features to be measured between different time series of images. The interrelationships between different features in the environment can thus be assessed more precisely in space and over time, for example comparing tree growth and surface water sources. A major challenge is that the object method used for high resolution geospatial imagery cannot be easily applied to monochromatic archival aerial photography because it has been designed for analyzing multispectral satellite imagery. As discussed in the article, using the manual sample as a training data set for an experimental machine-learning protocol demonstrates proof of concept for automatically extracting such features as farms, water holes and trees as individual objects from archival aerial photography. This increases the time depth of available high-resolution land use, environmental, and climate data from 2000 back to the 1940s and provides a base line for the Great Acceleration and brings the massive changes from the 1940s through the 1990s in focus as captured in aerial photography.

Assessing the accuracy of georeferenced landcover data derived from oblique imagery using machine learning

Mapping of land cover with open‐source software and ultra‐high‐resolution imagery acquired with unmanned aerial vehicles

AI-supported Framework of Semi-Automatic Monoplotting for Monocular Oblique Visual Data Analysis

Land cover classification in an era of big and open data: Optimizing localized implementation and training data selection to improve mapping outcomes

How Well Do Deep Learning-Based Methods for Land Cover Classification and Object Detection Perform on High Resolution Remote Sensing Imagery?

Fine-Scale Mapping of Natural Ecological Communities Using Machine Learning Approaches

A Comprehensive Approach Towards Enhancing Land Use Land Cover Classification Through Machine Learning and Object-Based Image Analysis

Archival Aerial Photographs of Africa: Present Potential and Imagining a Machine-Learning Future

Object based classification of a riparian environment using ultra-high resolution imagery, hierarchical landcover structures, and image texture

Sub-alpine shrub classification using UAV images: Performance of human observers vs DL classifiers

A review of supervised object-based land-cover image classification

Automating Ground Control Point Detection in Drone Imagery: From Computer Vision to Deep Learning

Propagating observation errors to enable scalable and rigorous enumeration of plant population abundance with aerial imagery

Land use/land cover mapping from airborne hyperspectral images with machine learning algorithms and contextual information

Integrating Google Earth imagery with Landsat data to improve 30-m resolution land cover mapping

Measuring Detailed Urban Vegetation with Multisource High-Resolution Remote Sensing Imagery for Environmental Design and Planning

Mapping the scientific research on natural landscape change with rephotography

Recognizing protected and anthropogenic patterns in landscapes using interpretable machine learning and satellite imagery

Spatial analysis of remote sensing image classification accuracy

A New and Automated Method for Improving Georeferencing in Nighttime Thermal ECOSTRESS Imagery

Anthropogenic Object Localization: Evaluation of Broad-Area High-Resolution Imagery Scans Using Deep Learning in Overhead Imagery