Hydrology and Earth System Sciences
Hydrology and Earth System Sciences
HESS
Articles | Volume 26, issue 12
Articles | Volume 26, issue 12
https://doi.org/10.5194/hess-26-3079-2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/hess-26-3079-2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.
Articles | Volume 26, issue 12
Research article
 | 
20 Jun 2022
Research article |  | 20 Jun 2022

Hydrological concept formation inside long short-term memory (LSTM) networks

Thomas Lees, Steven Reece, Frederik Kratzert, Daniel Klotz, Martin Gauch, Jens De Bruijn, Reetik Kumar Sahu, Peter Greve, Louise Slater, and Simon J. Dadson

Data sets

Hydrological Concept Formation inside Long Short-Term Memory (LSTM) networks T. Lees https://doi.org/10.5281/zenodo.5600851

Catchment attributes and hydro-meteorological timeseries for 671 catchments across Great Britain (CAMELS-GB) G. Coxon, N. Addor, J. Bloomfield, J. Freer, M. Fry, M., J. Hannaford, N. Howden, R. Lane, M. Lewis, E. Robinson, T. Wagener, and R. Woods https://doi.org/10.5285/8344e4f3-d2ea-44f5-8afa-86d2987543a9

Model code and software

NeuralHydrology Github Branch (pixel) T. Lees, S. Dadson, S. Reece, L. Slater, D. Klotz, M. Gauch, F. Kratzert, R. Sahu-Kumar, J. De Bruijn, and P. Greve https://doi.org/10.5281/zenodo.5541446

Lane_et_al_Benchmark_FUSE_GB R. Lane, G. Coxon, J.Freer, and T. Wagener https://doi.org/10.5523/bris.3ma509dlakcf720aw8x82aq4tm

neuralhydrology/neuralhydrology F. Kratzert https://github.com/neuralhydrology/neuralhydrology

tommylees112/neuralhydrology T. Lees https://github.com/tommylees112/neuralhydrology/tree/pixel

Download
Short summary
Despite the accuracy of deep learning rainfall-runoff models, we are currently uncertain of what these models have learned. In this study we explore the internals of one deep learning architecture and demonstrate that the model learns about intermediate hydrological stores of soil moisture and snow water, despite never having seen data about these processes during training. Therefore, we find evidence that the deep learning approach learns a physically realistic mapping from inputs to outputs.
Read more