New data for old trees

Researchers have been able to extract quantitative data on tree distributions from historical maps for the first time in Leeds and Edinburgh. These important new data are the result of new techniques, applying machine learning (ML) to help us understand more about how our cities have changed over time.

The research was published this week as a collaboration between the University of Leeds, University of Edinburgh, National Library of Scotland, and Forest Research and is free to access via Computers, Environment and Urban Systems. Interactive maps of the results are also available via the National Library of Scotland website.

The lead author, PhD researcher Elle Smith, highlights:

The tree symbol datasets produced in this work mean we can begin to analyse how urban forests have changed over time in Leeds and Edinburgh and examine how this has impacted the populations of the two cities, at a level of detail we haven’t been able to before.

 

 

Applying machine learning (ML) to historical maps is a relatively new method which allows us to gather information about the past over large areas and examine changes over time. This new research develops a method for generating training data and training a ML model to extract tree symbol information from historical Ordnance Survey (OS) maps of Edinburgh and Leeds which date from the 1890s. The maps are the most detailed created by OS, and cover more than 500 towns across England, Scotland and Wales at scale of 1:1056 and 1:500. As such, the model produced could be used to generate a very detailed assessment of historical trees in any of these locations. Researching past urban forests and how they have changed over time can help us understand the urban forests we have today. This understanding can help us in future tree planning to prevent mistakes being repeated and to maximise the benefits provided by urban trees.

 

 

The study estimates approximately 37 mapped tree symbols per hectare for Leeds (1888-90) and approximately 40 tree symbols per hectare for Edinburgh (1893-94) across the regions covered by the OS maps. This is the first time that quantitative data have been obtained for historical urban tree counts in these two cities.

Comparing the historical data to present day tree distributions indicates where the tree density appears to have changed. This comparison suggests that there has been a reduction in tree numbers in the northwest of the study area in Leeds. In Edinburgh, the spatial pattern of trees is similar to the present day but tree density appears to have reduced in the south of the study area. Both Leeds and Edinburgh have differing rates of development within the city which is likely to have impacted how their urban forests have evolved.

 

 

Along with tree symbol locations, the model determined a size category for each symbol and a species - coniferous or broadleaved. These results suggest that the proportion of conifer trees symbols detected in Edinburgh was greater than that of Leeds. These details can be used to inform historical canopy cover estimates as long as information about how the OS maps were created is taken into consideration.

 

Chris Fleet, Map Curator at the National Library of Scotland and coauthor of the study, said:

The NLS is keen to further the use of machine-learning to extract information from our historic maps. This ground-breaking research will make it possible to reconstruct precise historic urban tree densities and distributions across all towns and cities in Victorian Britain.

Dr Stuart King, another coauthor, based at the University of Edinburgh, added:

This project is a great example of the way that machine learning can be used to answer new questions on existing archives that are too difficult or costly to answer by hand.

 

The research was supported by the Natural Environment Research Council (NERC), including a SENSE CDT studentship. Full details of the research are available as Open Access via Computers, Environment and Urban Systems.