Doctoral researcher with the Global Diagnostic Modelling Research Group
Ecosystem and Earth System dynamics are often hard to model and predict and today, still many machine-learning approaches fail in encoding temporal dependencies that typically characterize dynamical natural systems. Yet, deep learning approaches in the time domain (e.g., Recurrent Neural Networks) are theoretically prepared to address such issues. However, such models are not physically interpretable, limiting the scientific insights they provide. The aim of my current work is to test the potential of temporal deep-learning approaches to model ecosystem processes and to find and explore possible pathways to gather scientific insights despite the models' black-box character; for example by combining machine learning and physically-based modeling (so-called hybrid modeling).
Kraft, B., Jung, M., Körner, M., Requena Mesa, C., Cortés, J., and Reichstein, M., (2019). Identifying Dynamic Memory Effects on Vegetation State Using Recurrent Neural Networks. Frontiers in Big Data, 2, p.31.
Reichstein, M., Besnard, S., Carvalhais, N., Gans, F., Jung, M., Kraft, B., and Mahecha, M. (2018). Modelling Landsurface Time-Series with Recurrent Neural Nets. In IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium (pp. 7640-7643). IEEE.
Requena-Mesa, C., Reichstein, M., Mahecha, M., Kraft, B., and Denzler, J. (2018). Predicting landscapes as seen from space from environmental conditions. In IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium (pp. 1768-1771). IEEE.