top of page
Search

A New Frontier in Urban Tree Sensing

Developing an urban tree data and analytics platform in the city of Amsterdam



Project Mission

Space4Good, together with consortium partners LucidMinds and UNL, successfully finished the MIT-AI research and development project supported by the RVO called ReTreevAIble. The project aimed to provide policymakers and other stakeholders with a highly detailed individual urban tree data and analytics platform by using state-of-the-art technologies. This platform supports making decisions in a data-driven way and avoids the need for extensive fieldwork.


Pilot Study Amsterdam Results

The Municipality of Amsterdam served as the pilot location for the project due to their existing database with geolocated trees to train the artificial intelligence (AI) models. For some Amsterdam-based Space4Good members, it meant the models could even be ground-validated from their living room windows!


Space4Good’s important contributions to the RetreevAIble project included: 1) Responsibility for research and the acquisition of necessary input datasets that contributed to developing an extensive tree database. 2) Developed remote sensing-based urban tree crown detection algorithms 3) Created AI models for individual tree species identification. 4) Finally, we were able to support above-ground biomass estimations from LiDAR data.

.

A core result of this project is the development of a remote sensing pipeline that fuses data from multiple sources on the individual tree level. Highlights were utilising the open-source remote sensing data to extract over 8000 3D reconstructions of trees (Fig. 1). Each of these was autonomously given a species label based on the geolocation in the Amsterdam tree database. Ultimately, this dataset was used to test and train our models for tree species identification. The results of these models were greatly promising. We were able to achieve differentiation of up to 7 urban tree species or categories with an overall accuracy above 90%!




Figure 1. Individual trees extracted from the remote sensing data.

From left to right: Linden, Poplar, Acacia & Chestnut.


Extracting Digital Twins of Urban Trees

To give insights into how we were able to segment these digital twins of urban trees, this section describes the methodology steps. The approach solely uses open-source remote sensing data, namely the nationwide Dutch AHN4 LiDAR point cloud and the high-resolution Superview multispectral data made available by the Netherlands Space Office Satellietdataportaal. An algorithm was developed to delineate the tree crowns from the data in high detail (Fig 2a). For each unique digital tree, variables were extracted using the 3D geometry from LiDAR and satellite images from different seasons. Examples derived from LiDAR are geometric variables such as tree height and crown diameter. Superview images are utilised to calculate several vegetation indices over time that encapsulate phenology features such as seasonal variations in leaf colour and texture. The rich tree database from the municipality of Amsterdam enabled us to label each tree with a ground-truth species name.


The final labelled database with digital trees is fed to train our state-of-the-art AI models. The predictions on an independent dataset proved to be highly promising and exemplify how this can be used for urban tree mapping (Fig. 2b). In addition to this, the developments within the ReTreevAIble project can be applied outside of the urban scope, better enabling agroforestry carbon assessments and biodiversity indications, for example.




Figure 2. a) Tree crown detections-left b) Predicted tree species mapping-right


Consortium Partners

We are thankful to have partnered within this highly skilled AI consortium and to work alongside LucidMinds and UNL to the overall outcomes of ReTreevAIble. To LucidMinds, we thank you for your leading role in project management and your contributions to biomass estimations and scenario analysis for individual urban trees. To UNL, thank you for your major role in the development of the visualisations on the dashboard. We look forward to what the ReTreevAIble project has enabled and unlocked for urban and nature-based solutions worldwide! 


Would you like more information or are you interested in collaborating with Space4Good? Visit our website or contact us via hello@space4good.com.



258 views0 comments

Recent Posts

See All

Comments


Anchor 1
bottom of page