Time-Traveling Trees: Machine-reading the Past, Present, and Future of Climate Change Adaptation from Archival Aerial Photography

This project begins with a “book:” eight historical case studies detailing environmental change will serve to produce 8 machine learning training data sets to develop a methodology for the automatic analysis of quantitative and visual environmental data from historical aerial photographs. This will allow for a new approach to study the 1940s-1990s Great Acceleration, which caused unprecedented environmental change. Becoming ubiquitous in the 1940s, aerial photography has a sub-1-meter resolution which is key to identifying non-linear, non-monumental, and pre-industrial rural environmental infrastructure (e.g. trees, waterholes, fields) and how it has changed over time. Since satellite imagery only gained a comparable high resolution in the early 2000s, the ability to machine-read aerial photographs will quadruple the availability of high-resolution geospatial data, thereby allowing a more robust and nuanced analysis of environmental and climate change. The project goals are: (1) to develop a historical ground truthing method to enhance the interpretation of historical aerial photographs to imagine historical landscapes and how they changed; (2) to produce machine deep learning training data sets by deploying historical ground truthing to identify and extract a set of features from sample areas in the study regions; (3) to use a more systematic and more historically informed object-based methodology to machine read archival aerial photography that is deployable at different spatial (local, regional, global) and temporal scales and make it available as an Open Access online software to read historical aerial photography world-wide. By unlocking the mass data contained in archival aerial photography 3TMAAP opens a new path of research by systematically integrating historical ground truthing into the automatization of geospatial analysis, dramatically deepening the time depth of high-resolution mass data for historical analysis and geospatial and climate modelling.