Central Wellington; individual trees outlined in yellow
The Wellington Urban Tree Explorer contains a map layer of individual tree crowns across Wellington. The data were produced by a deep learning model called Mask R-CNN (Region-based Convolutional Neural Network), which identifies individual tree objects from top-view aerial imagery.
Dr Jan Schindler leads a joint New Zealand-Singapore Data Science research programme that uses data science, remote sensing and 3D modelling for extracting tree species information from multi-resolution, remotely sensed data.
The aims of the programme are to advance our current methods for measuring individual trees by developing and applying novel deep learning algorithms and to model their interactions with the human and physical environment.
“As the world becomes increasingly urbanised, urban trees and forests become increasingly important for overall human well-being,” says Jan. “There are plenty of studies showing the benefits of urban trees and forests from a number of perspectives – health, climate and ecology.”
However, sustaining and enhancing biodiversity and healthy living environments requires careful management of trees in urban areas and forests. But these decisions are currently limited by the quality of available data, tools, and techniques.
“With the information we gather using these new deep-learning techniques, we can detect trees and identify species in greater detail than ever before, moving towards analysing the socio-economic impacts of trees in cities,” says Jan. “The methods are transferable to non-urban areas and are already being successfully applied to a number of tree mapping projects in pastoral hill country and native forests.”
Features of the website include sliders for layer transparency, coloured maps showing tree heights and diameters, selecting data based on attribute ranges, and being able to zoom in to get information on individual trees, such as their crown dimensions and height.
The project team consists of researchers from Manaaki Whenua, Scion, the University of Canterbury, Victoria University of Wellington, the Institute of High Performance Computing Singapore and Nanyang Technological University.
