Unleashing the power of old maps: Extracting symbology from nineteenth century maps using convolutional neural networks to quantify modern land use on historic wetlands

Abstract

Topographical maps from the nineteenth century hold significant historical and environmental value, providing insights into landscape changes over the past two centuries. These maps feature distinct symbols representing various land cover types, such as forests and wetlands, offering a unique historical perspective on land-use changes. For example, there has been a significant reduction in wetlands because of agricultural expansion and intensification which lead to biodiversity loss and increased greenhouse gas emissions globally. Our study uses U-net CNN to automatically segment wetland symbols from nineteenth century maps from hundreds of map sheets for an area of interest a large river catchment in Ireland. Extracted wetland extents were intersected with digital land cover datasets to estimate current land cover on former wetland (on both organic and mineral soils). Utilizing U-Net, we successfully automated the segmentation of wetland symbols from hundreds of nineteenth-century map sheets, focusing on a large river catchment area in Ireland. Our analysis achieved a very high F1 score of 98.2% and a Kappa of 89%. While it is challenging to verify the veracity of historical map content, the largely untapped information contained within these maps are important for understanding landscape change over time, and especially before the era of Earth observation & remote sensing. The data extracted from these sources can inform modern environmental management strategies, for example, in targeted rewetting of peatlands, or in habitat restoration.