Where were all the trees? Computer vision meets landscape history to unlock the potential of historical Ordnance Survey maps

Abstract The first edition of the Ordnance Survey's 1:2500 County Series is marked with treed areas, such as woodlands and orchards, and, remarkably, freestanding trees. Consequently, it is an unparalleled source of information on the distribution of trees across Britain in the mid‐late nineteenth century, when the maps were first surveyed. In order to make use of this data, tree symbols need to be digitised, and we need to better understand to what extent the mapped treescape accurately represents the treed landscape of the past. This paper presents a non‐technical introduction to our new Deep Learning Computer Vision method for rapidly identifying and georeferencing tree symbols on the County Series maps: You Only Look for a Symbol Once (YOLSO). We report on archival research to elucidate the Ordnance Survey's decision‐making process around the mapping of trees, finding that they aimed to plot all large and conspicuous freestanding trees, but only if they did not confuse or obscure other map detail. We assess the accuracy of the Ordnance Survey's tree mapping by comparing numbers of tree symbols on the 1:2500 maps with larger scale plans and against historical photographs, plotted or taken at or around the same time. We demonstrate the usefulness of a digital historic tree dataset by examining the density of freestanding trees in relation to the density of settlement in three cities and their hinterlands: York, Cardiff and Milton Keynes. Synthesis and applications . The work presented here envisages a future National Historic Tree Map of Great Britain, which could be used to examine how the distribution of trees characterises different types of landscape, and as a baseline for treescape change throughout the twentieth century and beyond. Read the free Plain Language Summary for this article on the Journal blog.