All material
View session timetable
Poster
License: Creative Commons Attribution 4.0 International (CC-BY-4.0)

Community members
View details

Automatic low-dimension explainable feature extraction of climatic drivers leading to forest mortality

Anand, Mohit1 , Camps-Valls, G.2 , Zscheischler, J.1
  1. Helmholtz Centre for Environmental Research
  2. Image Processing Laboratory (IPL), Universitat de València
<p> Forest mortality is a complex phenomenon because of the interaction of multiple drivers over <br/> a long period. Understanding these interactions and their relevant time scale is important for <br/> forest management. Unlike climate data which are continuous (daily or hourly resolutions), <br/> forest mortality related observations are discrete (e.g. the number of trees, mortality fraction) <br/> with lower frequencies (once/twice a year). Forests also have persistent memory with a large <br/> buffering capacity. All the above-mentioned reasons make the analysis of forest mortality <br/> difficult with the conventional tools. Deep learning is well suited for modelling multivariate <br/> time series with persistent non-linear interactions. In this study, we generate 200,000 years of <br/> hourly climate data using a weather generator (AWE-GEN). We aggregate the hourly data to <br/> daily values and feed it to a process based forest model (FORMIND). The forest model gives <br/> us mortality fractions per year, in line with the forest mortality related observations. For the <br/> method development phase, we use these simulated data. First, we use a variationalautoencoder <br/> to extract climatic features and use that for the prediction of forest mortality. In <br/> the second stage, we do the prediction of forest mortality and feature extraction together and <br/> illustrate the difference between the extracted features. Further different approach to <br/> disentangle the extracted feature are tested. Finally, we present the analysis of performance <br/> versus explainability for different approaches. </p>