The increasing need for resources, the expansion of settlements, and Climate Change have put humanity in a difficult relationship with the environment.
The Earth Science and Climate Change research group at the Institute of Machine Learning focuses on building deep learning models to better describe environmental systems. Our current research revolves mainly around water, the most vital natural resource. Within this field, we introduced a new approach to predict the amount of water in river systems. Unlike traditional simulations with coarse, hand-crafted abstractions, we use Machine Learning to generate fully data-driven predictions. Our approach has proven to outperform traditional hydrological models --- the outcomes of decades of expertise and model development --- by a large margin. Ultimately, these new models will allow us to provide early flood warnings, to predict future droughts, to operate hydropower plants more efficiently, to improve access to drinking water, and more.
Modelling these processes is an innately spatio-temporal problem. Our models thus make use of a wide range of different inputs, ranging from weather data to land-surface characteristics to satellite observations. In the future, we are also looking forward to extending the application and scope of our research to other areas of geosciences.
recent publications in AI 4 Earth:
- HESSDUncertainty Estimation with Deep Learning for Rainfall-Runoff ModellingHydrology and Earth System Sciences Discussions 2021
- HESSA note on leveraging synergy in multiple meteorological datasets with deep learning for rainfall-runoff modelingHydrology and Earth System Sciences 2021
- HESSRainfall–runoff prediction at multiple timescales with a single Long Short-Term Memory networkHydrology and Earth System Sciences 2021
- WRRWhat Role Does Hydrological Science Play in the Age of Machine Learning?Water Resources Research 2020
- WRRToward Improved Predictions in Ungauged Basins: Exploiting the Power of Machine Learning2019
- HESSTowards Learning Universal, Regional, and Local Hydrological Behaviors via Machine Learning Applied to Large-Sample Datasets2019
- SpringerNeuralHydrology – Interpreting LSTMs in Hydrology2019