Day 4

Detailed paper information

Back to list

Paper title DeepExtremes - Deploying Artificial Experiments on High-Resolution Data Cubes for explaining extreme event impacts
  1. Miguel Mahecha Universität Leipzig Speaker
  2. Fabian Gans Max Planck Institute for Biogeochemistry
  3. Gustau Camps-Valls University of Valencia
  4. Gunnar Brandt Brockmann Consult GmbH
  5. Guido Kraemer Remote Sensing Center for Earth System Research, Leipzig University
  6. Karin Mora Remote Sensing Centre for Earth System Research, Leipzig University
  7. Miguel-Ángel Fernández-Torres Image Processing Laboratory, University of Valencia, Valencia, Spain
  8. Christian Requena Biogeochemical Integration, Max-Planck-Institute for Biogeochemistry, Jena, Germany;
  9. Vitus Benson Biogeochemical Integration, Max-Planck-Institute for Biogeochemistry, Jena, Germany
  10. Markus Reichstein Max Planck Institute for Biogeochemistry
  11. Carsten Brockmann Brockmann Consult GmbH
  12. Michele Ronco Image Processing Laboratory, Universitat de València, València, Spain
Form of presentation Poster
  • C1. AI and Data Analytics
    • C1.07 ML4Earth: Machine Learning for Earth Sciences
Abstract text Droughts, heat-waves, and in particular their co-occurrences are among the most relevant climate extremes for both ecosystem functioning and human wellbeing. A deeper process understanding is needed to soon enable an early prediction of impacts of climate extremes. Earlier work has shown that vegetation responses to large-scale climate extreme events are highly heterogeneous, with critical thresholds varying according to vegetation type, event duration, pre-exposure, and ecosystem management. However, much of our current knowledge has been derived from coarse scale downstream data products and hence remains rather anecdotal. We do not yet have a global overview of high-resolution signatures of climate extreme impacts on ecosystems. However, obtaining these signatures is a nontrivial problem, as multiple challenges remain not only in the detection of extreme event impacts across environmental conditions, but also in explaining the exact impact pathways. Extreme events may happen clustered in time or space, and interact with local environmental factors such as soil conditions. Explainable artificial intelligence methods, applied to a wide collection of consistently sampled high-resolution satellite derived data cubes during extremes should enable us to address this challenge. In the new ESA funded project DeepExtremes we will work on this challenge and build on the data cube concepts developed in the Earth System Data Lab. The project adopts a nested approach of global extreme event detection and local impact exploration and prediction by comparing a wide range of XAI methods. Our aim is to explore how to shed light into the question how climate extremes affect ecosystems globally and in near-real time. In this presentation we describe the project implementation strategy, methodological challenges, and invite the remote sensing and XAI community to join us in addressing one of the most pressing environmental challenges of the future decades.