About NeuralHydrology
Everything started with the idea of using LSTMs as a general rainfall-runoff model, back in 2016. Daniel and myself (Frederik) were self-studying machine learning and trying to keep up with the fast-paced developments of that field. At that time, both of us were still working at the Institute for Hydrology and Watermangement (former Institute of Water Management, Hydrology and Hydraulic Engineering), Daniel as PhD-student and I as a student assistant. A lot of our free-time went into designing and conducting first experiments, to see if the LSTM is able to model the rainfall-runoff relationship at all. By then, we jokingly referred to this idea as neural hydrology, not knowing that Bob Abrahart (1999) coined a similar term (NeuroHydrology) already 20 years ago. In his paper, Bob Abrahart advocates the use of neural networks in hydrology, while still referring to classical multi layer perceptrons and not recurrent neural networks.
Starting in late 2017, we did first experiments using the CAMELS data set. At this time we became aware of the potential that lies in LSTM-based rainfall-runoff modeling and we put more and more effort into designing and writing our first paper.
At the same time, I got the offer to join the Institute for Machine Learning (at this time it was still named Institute for Bioinformatics) as a PhD student under the supervision of Sepp Hochreiter, the inventor of the LSTM. Obviously I didn’t hesitate and Daniel should follow me later the same year.
In December 2018 things really started to take off, when we met Sella Nevo at NeurIPS 2018 and one week later I met Grey Nearing at the AGU Fall Meeting. Sella Nevo is the team leader of the Google Flood Initiative, which we first met at EGU 2016, when he was starting to build his team. Now, we work closely together with his team at Google Research, Tel Aviv, and they are even providing funding for my PhD. Grey Nearing, became a very close collaborator and almost anything we did since then has been a collaboration of all of use together. Outcomes of this collaboration are for example our latest HESS paper entitled “Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets” or the follow-up work, published in WRR, with the title “Towards Improved Predictions in Ungauged Basins: Exploiting the Power of Machine Learning”.
Currently, we are working on a bunch of new ideas, all settled around LSTM-based modeling for hydrological applications. All updates will be posted at this homepage, so stay check for updates from time to time.
About Frederik Kratzert
I did my Bachelor in Civil Engineering at the Technical University in Berlin, Germany, followed by a Master in Environmental Engineering at the University of Life Sciences and Natural Resources in Vienna (BOKU), Austria. After working one year as research assistant at the Institute for Hydrology and Watermangement at the BOKU, I joined the Institute for Machine Learning at the Johannes Kepler University in 2018 as a PhD student. Soon after I got a faculty research award from Google, which allowed me to concentrate to 100 percent on my research about deep learning models in hydrology. In summer 2021 I finished my PhD and finally joined the Flood Forecasting team at Google Research.
Personal Blog: https://kratzert.github.io/
GitHub: https://github.com/kratzert
Twitter: @fkratzert
About Daniel Klotz
Doodling, errors, environmental sciences, and machine learning. I have a weird set of interest which led me to a rather unusual path. After fiddling around with a broad range of studies (dipping into architecture, germanistic, landscape planning and natural resource management), I settled upon a study in ecological engineering. Back then I figured that it will provide strong ethical and designing principles combined with technical depth. Anyhow, this choice led me to work at the Institute for Hydrology and Watermangement at BOKU, Vienna, which brought the opportunity to pursue a PHD in hydrology there (and, as mentioned above, it is where Frederik and I started to plot our research program). As of now, I joined the Institute for Machine Learning at the Johannes Kepler University, where I try to propel the application of neural networks to hydrological applications.
Personal Website: danklotz
Twitter: @ido87
Flickr: jeanklotz
GitHub: danklotz
About Martin Gauch
After finishing my Bachelor’s and starting my Master’s degree in computer science at Karlsruhe Institute of Technology, Germany, I did an exchange year in Canada at the University of Waterloo. That’s where I got into the topic of deep learning for hydrology—and I liked it so much that I stayed and completed my Master’s in Waterloo. Currently, I’m a Research Associate at the University of Waterloo, and in January 2021 I’ll officially start my PhD at the Institute for Machine Learning at Johannes Kepler University.
Personal Website: https://gauchm.github.io/
GitHub: https://github.com/gauchm
About Grey Nearing
Grey Nearing is a professor in the Land, Air, and Water Resources Department at the University of California, Davis, and a visiting researcher at Google Research. He previously worked with hydrological model development teams at NASA and NCAR.
Google Scholar: link