Comparing WWII aerial photographs to Sentinel-2 data

Abstract

During World War II West Papua was the scene of war between the Axis and the Allies. In this case, it was the Japanese on one side Australian and US forces on the other. After the war a lot of equipment was left in the jungle and because of the low population density on the island some material can still be found to this day. Planes and vehicles that were abandoned far enough from villages were able to withstand the test of time and are still hidden under the trees. In November 2018 an expedition force is heading to the island to find some of these planes. This assignment was meant as a preparation for that expedition and it was my job to find places that could be interesting to investigate closer in the field. The research question that had to be answered was as follows: is it possible to classify and link historical data to modern data. To do this I looked at old aerial photographs of the airports, and at modern Sentinel-2 and Landsat multispectral images of the area. The aerial photographs are stitched into a panorama and geolocated so that the coordinates were known. Next features like the average and gradient of surrounding pixels are computed so the image could be classified. Classifications were then performed on the panoramas and on the Landsat 8 dataset in order to create a map of the region of interest and possibly point out potential war equipment locations. The Landsat data was also used to create a heatmap to show the warmest spots in the region. These spots could potentially point to material that is warmer than its surroundings, e.g. a metal car in a grass field that has been standing in the sun for some time. The classification of the panoramas proved to be a difficult task and the final result leaves much to be desired. The classification of the Landsat 8 dataset went better and the results show a useable map. However the most interesting class in this case was the war equipment class, which cannot be validated. The heatmap could mostly be used to highlight the roads, but there are certain spots that light up and cannot be explained without field data. These spots might be interesting to check out during the expedition.