HESS Opinions: Never train a Long Short-Term Memory (LSTM) network on a single basin

Kratzert, Frederik; Gauch, Martin; Klotz, Daniel; Nearing, Grey

doi:https://doi.org/10.5194/hess-28-4187-2024

Articles | Volume 28, issue 17

https://doi.org/10.5194/hess-28-4187-2024

© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Special issue:

HESS Opinions 2023

https://doi.org/10.5194/hess-28-4187-2024

© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 28, issue 17

Opinion article

|

12 Sep 2024

Opinion article |

| 12 Sep 2024

HESS Opinions: Never train a Long Short-Term Memory (LSTM) network on a single basin

Frederik Kratzert, Martin Gauch, Daniel Klotz, and Grey Nearing

Data sets

Results and experimental data Frederik Kratzert, Martin Gauch, Daniel Klotz, and Grey Nearning https://doi.org/10.5281/zenodo.11247607

``Never train an LSTM on a single basin'' Frederik Kratzert https://doi.org/10.5281/zenodo.10139248

``Never train a Long Short-Term Memory (LSTM) network on a single basin'' Frederik Kratzert https://doi.org/10.4211/hs.474ecc37e7db45baa425cdb4fc1b61e1

Model code and software

``Never train a Long Short-Term Memory (LSTM) network on a single basin'' Frederik Kratzert https://doi.org/10.5281/zenodo.13691802

Short summary

Recently, a special type of neural-network architecture became increasingly popular in hydrology literature. However, in most applications, this model was applied as a one-to-one replacement for hydrology models without adapting or rethinking the experimental setup. In this opinion paper, we show how this is almost always a bad decision and how using these kinds of models requires the use of large-sample hydrology data sets.