VU researchers identify more than 2.4 million trees in the Veluwe region

Earth scientists at VU Amsterdam have mapped and identified the species of more than 2.4 million trees in the Veluwe region. It is not often that so much data is obtained at the level of individual trees; it was made possible by the highly detailed Dutch Digital Elevation Model in combination with satellite images from the European Space Agency (ESA).

11/19/2020 | 3:46 PM

The study, which was published in the scientific journal Remote Sensing, combined field data on 2,197 trees in a forested area measuring 140 square kilometres in the Veluwe region with digital models and satellite data. The study distinguished between eight common tree species, including deciduous trees, such as the common oak and silver birch, and coniferous trees, such as the Scots pine and Japanese larch.

Identifying individual trees
The locations of these trees were determined by researchers in the field and subsequently linked to a digital crown model containing information on the height and crown surface areas of the trees, along with various other parameters. This digital crown model was derived from the detailed model The Current Dutch Elevation map (Actueel Hoogtebestand Nederland, AHN). The digital crown model was then linked to satellite data from the European Space Agency(ESA) throughout the various seasons. This made the colour differences between the tree species visible, season by season.

Algorithm recognizes patterns
Based on the crown structure and colour, the researchers developed an algorithm that classified all the trees in the area covered by the study with a high degree of accuracy. ‘The idea is that every species of tree is characterized by a unique structure and colour palette. A machine learning algorithm recognizes these patterns and can be applied over a large area,’ says Veerle Plakman, who carried out the study for her Master’s in Earth Sciences at VU Amsterdam.

From the back garden to the Veluwe
The concept underlying this study was thought up by co-author Thomas Janssen, a PhD student at VU Amsterdam, in his own back garden. ‘The Dutch Elevation Model is very detailed, with pixels of 50 x 50 cm. The elevation model around my own home enabled me to recognize the individual trees in my back garden very easily. It seemed to me that combining this with information on colours from satellite images would provide me with a mine of information for classifying individual trees by species,’ according to Janssen.

For the whole of the Netherlands
The automated recognition and classification of tree species developed in this study can easily be scaled up for use for the whole of the Netherlands. Because of the detail in the crown model, however, scaling up to national level would require a large amount of computer storage and processing capacity. Thomas Janssen thinks that national application is feasible: ‘My dream is a publicly accessible map of tree species for the Netherlands. Users would be able to give feedback on the map and add data, if required. This would mean that the map could be continually improved and kept up to date.’

Fire hazard and carbon
‘This study has massive potential,’ says Sander Veraverbeke, Assistant Professor of Earth Observation at VU Amsterdam and senior author of the article. ‘Forest managers need spatial information on the occurrence of deciduous and coniferous trees to, for instance, be able to assess the risk of spread of forest fires. Fires spread more easily in coniferous forests, whereas deciduous trees can form a natural barrier to fire because their foliage generally contains more moisture than that of coniferous trees. Our study provides this information at the individual tree level and this can lead to better-informed policy decisions.’

Quantity of stored carbon
The VU Amsterdam team collaborated with the Institute for Safety (Instituut Fysieke Veiligheid, IFV) in applying the results in practice. ‘We can, furthermore, link the structural information at the individual tree level to what we call allometric comparisons. These existing comparisons express the relationship between the structure of trees, species and biomass and, in turn, the quantity of carbon stored in these trees, too. This could bring us closer to a detailed assessment of the quantity of carbon stored in Dutch forests,’ says Veraverbeke in conclusion.