2021
DOI: 10.5194/hess-2021-423
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Deep learning rainfall-runoff predictions of extreme events

Abstract: Abstract. The most accurate rainfall-runoff predictions are currently based on deep learning. There is a concern among hydrologists that data-driven models based on deep learning may not be reliable in extrapolation or for predicting extreme events. This study tests that hypothesis using Long Short-Term Memory networks (LSTMs) and an LSTM variant that is architecturally constrained to conserve mass. The LSTM (and the mass-conserving LSTM variant) remained relatively accurate in predicting extreme (high return-… Show more

Help me understand this report
View published versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
3
1

Citation Types

0
23
0

Year Published

2021
2021
2023
2023

Publication Types

Select...
6
1

Relationship

0
7

Authors

Journals

citations
Cited by 18 publications
(23 citation statements)
references
References 23 publications
0
23
0
Order By: Relevance
“…They found that various hydrometeorological and geological characteristics impacted the simulation accuracy and recommended continued work to increase trust in the use of ML technology in hydrology. Frame and colleagues (2021) aimed to understand whether process-based models are more reliable for predicting extreme events (in this case, streamflow) as compared to data-driven models [8]. This work demonstrated that under these testing conditions the data-driven models (including physics-informed ML model and purely ML model) performed better than process-based models at predicting peak flows under a variety of conditions, including extreme events.…”
Section: Introductionmentioning
confidence: 99%
“…They found that various hydrometeorological and geological characteristics impacted the simulation accuracy and recommended continued work to increase trust in the use of ML technology in hydrology. Frame and colleagues (2021) aimed to understand whether process-based models are more reliable for predicting extreme events (in this case, streamflow) as compared to data-driven models [8]. This work demonstrated that under these testing conditions the data-driven models (including physics-informed ML model and purely ML model) performed better than process-based models at predicting peak flows under a variety of conditions, including extreme events.…”
Section: Introductionmentioning
confidence: 99%
“…Machine learning methods provide great versatility (Shen, 2018;Shen et al, 2018;Reichstein et al, 2019) and have demonstrated unprecedented accuracy in various modelling tasks like predictions in un-gauged basins (PUB, e.g. Kratzert et al, 2019b;Prieto et al, 2019), in transfer learning to data-scarce regions (Ma et al, 2021) or flood forecasting (Frame et al, 2021;Nevo et al, 2021).…”
Section: Machine Learning In Hydrologymentioning
confidence: 99%
“…Therefore, it becomes more and more inaccurate to label machine learning methods as black boxes since techniques exist that shed light on the internals of machine learning methods (see also Nearing et al, 2021;Frame et al, 2021) -turning them toward so-called grey box models. Yet, internal investigation of machine learning models relies on additional methods that come with their own assumptions and caveats, and the current straight-forward interpretability of conceptual models serves as benchmark in the hydrologic community.…”
Section: Machine Learning In Hydrologymentioning
confidence: 99%
See 1 more Smart Citation
“…The method can be applied to either physically based, process-based and data-based deterministic 10.1029/2021WR031215 3 of 20 prediction/simulation models. It can also be applied in conjunction with prediction models based on deep learning, which are gaining increasing popularity for hydrological predictions (see, for instance, Frame et al, 2021).…”
mentioning
confidence: 99%