TreeNet goes AI

Reconstructing radial stem size changes of trees with machine learning

by Mirko Luković (EMPA), Roman Zweifel (WSL), Guillaume Thiry (ETHZ), Ce Zhang (ETHZ) and Mark Schubert (EMPA)

Publication in the Journal of the Royal Society Interface

We already know from image processing programs like Photoshop that we can fix an image almost as if by magic if it has white spots, for example. Behind this are machine learning methods, also known as artificial intelligence (AI). This work has adopted these AI methods to be applied to time series data, specifically time series of dendrometer data collected in TreeNet. And just like with photos, it works with TreeNet data.
The work is a collaboration between WSL, EMPA and ETHZ and was supported by the WSL internal project deepT

Paper abstract:

Like many scientists, ecologists depend heavily on continuous uninterrupted data in order to understand better the object of their study. Although this might be straightforward to achieve under controlled laboratory conditions, the situation is easily complicated under field conditions where sensors and data transmission are affected by harsh weather, living organisms, changes in atmospheric conditions etc. This often results in parts of the data being corrupted or missing altogether. We propose the use of the most recent machine-learning techniques to reverse such data losses in multi-channel time series. In particular, we focus on tree stem growth data obtained from the TreeNet project, which monitors the changes in stem radius and environmental conditions of a few hundred trees across Switzerland. In the first part of the study, we test the performance of five architectures based on encoders, and recurrent and convolutional neural networks and we show that a deep neural network combining long short-term memory with one-dimensional convolutional layers performs the best. In the second part, we adopt this model to reconstruct the original TreeNet dataset, which we then use in a separate classification problem to show the effect of the proposed gap-filling procedure.