Bayesian LSTM With Stochastic Variational Inference for Estimating Model Uncertainty in Process‐Based Hydrological Models

Significant attention has recently been paid to deep learning as a method for improved catchment modeling. Compared with process‐based models, deep learning is often criticized for its lack of interpretability. One solution is to combine a process‐based hydrological model with a residual error model...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Water resources research Jg. 57; H. 9
Hauptverfasser: Li, Dayang, Marshall, Lucy, Liang, Zhongmin, Sharma, Ashish, Zhou, Yan
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Washington John Wiley & Sons, Inc 01.09.2021
Schlagworte:
Artificial neural networks
Bayesian analysis
Bayesian inference
Bayesian theory
Catchment models
Catchments
Deep learning
Errors
Groundwater
Groundwater models
Hydrologic models
Hydrologic processes
hydrological model
Hydrology
Intervals
LSTM
Machine learning
Markov chain
Markov chains
Mathematical models
MIKE SHE
model uncertainty
Modelling
Natural language processing
Neural networks
Parameters
probabilistic prediction
Probability theory
regression analysis
Regression models
Reliability
Sharpness
Statistical analysis
Statistical inference
Statistical methods
stochastic variational inference
Uncertainty
Watersheds
ISSN:0043-1397, 1944-7973
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Significant attention has recently been paid to deep learning as a method for improved catchment modeling. Compared with process‐based models, deep learning is often criticized for its lack of interpretability. One solution is to combine a process‐based hydrological model with a residual error model based on deep learning to give full scope to their respective advantages. In classical residual error models, Bayesian inference via Markov chain Monte Carlo (MCMC) is commonly used to provide an estimation of the uncertainty. However, deep neural networks tend to have excessively large numbers of parameters, making MCMC an unsuitable approach. Here, we introduce an alternative to Bayesian MCMC sampling called stochastic variational inference (SVI) which has recently been developed for Bayesian deep learning in Natural Language Processing. We implement SVI in a Long Short‐Term Memory (LSTM) network and construct residual error models in process‐based hydrological models. This approach is examined in the contrasting geographical and climatic characteristics of two catchments from China, the Tangnaihai catchment and the Shiquan catchment. Compared with the Bayesian linear regression model, the Bayesian LSTM provides better uncertainty estimates. Specifically, the proposed method improves the Continuous Ranked Probability Score (CRPS) by over 10% in both two catchments. In the Tangnaihai catchment, it provides more than 10% narrower uncertainty intervals in terms of Sharpness with slightly superior Reliability. In the Shiquan catchment, it provides comparable uncertainty intervals with better Reliability. Further, our study highlights the scalability of SVI to high‐dimensional parameter spaces in hydrological applications (e.g., distributed hydrological models, groundwater models). Key Points Residual error models are constructed to take advantages of process‐based hydrological models and Bayesian deep learning Bayesian Long Short‐Term Memory provides better streamflow uncertainty estimates than a Bayesian linear regression model Stochastic variational inference as an alternative to Markov chain Monte Carlo is viable to estimate tens of thousands of parameters
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
content type line 23
ISSN:0043-1397
1944-7973
DOI:10.1029/2021WR029772