Benchmarking data-driven rainfall–runoff models in  Great Britain: a comparison of long short-term memory (LSTM)-based models with four lumped conceptual models

Lees, Thomas; Buechel, Marcus; Anderson, Bailey; Slater, Louise; Reece, Steven; Coxon, Gemma; Dadson, Simon

doi:10.5194/hess-25-5517-2021

Cited by 127 publications

(95 citation statements)

References 58 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this study, we trained LSTM models using the same hyperparameters as those trained in Lees et al (2021). We offer a brief introduction to the state-space formulation of the LSTM (Kratzert et al, 2019b) because it offers a clear explanation for why we explore the cell state (c t ), since it reflects the state vector of the LSTM.…”

Section: Methodsmentioning

confidence: 99%

“…Following Lees et al (2021), we trained a single LSTM to predict runoff for 669 basins from the CAMELS-GB dataset (Coxon et al, 2020b). The input sequences are digested into the LSTM, each consisting of 1 year's worth of daily data (365 timesteps).…”

Section: Experimental Designmentioning

confidence: 99%

“…To our knowledge, we are the first to apply techniques developed in machine learning interpretability and natural language processing research (Hewitt and Liang, 2019) to hydrology. We carry out a comprehensive evaluation of the LSTM cell states across a sample of 669 catchments in Great Britain (Lees et al, 2021). This allows us to rigorously assess whether the LSTM has learned concepts that generalize over space.…”

mentioning

confidence: 99%

See 2 more Smart Citations

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

Lees

Reece

Kratzert³

et al. 2022

Hydrol. Earth Syst. Sci.

Self Cite

View full text Add to dashboard Cite

Abstract. Neural networks have been shown to be extremely effective rainfall-runoff models, where the river discharge is predicted from meteorological inputs. However, the question remains: what have these models learned? Is it possible to extract information about the learned relationships that map inputs to outputs, and do these mappings represent known hydrological concepts? Small-scale experiments have demonstrated that the internal states of long short-term memory networks (LSTMs), a particular neural network architecture predisposed to hydrological modelling, can be interpreted. By extracting the tensors which represent the learned translation from inputs (precipitation, temperature, and potential evapotranspiration) to outputs (discharge), this research seeks to understand what information the LSTM captures about the hydrological system. We assess the hypothesis that the LSTM replicates real-world processes and that we can extract information about these processes from the internal states of the LSTM. We examine the cell-state vector, which represents the memory of the LSTM, and explore the ways in which the LSTM learns to reproduce stores of water, such as soil moisture and snow cover. We use a simple regression approach to map the LSTM state vector to our target stores (soil moisture and snow). Good correlations (R2>0.8) between the probe outputs and the target variables of interest provide evidence that the LSTM contains information that reflects known hydrological processes comparable with the concept of variable-capacity soil moisture stores. The implications of this study are threefold: (1) LSTMs reproduce known hydrological processes. (2) While conceptual models have theoretical assumptions embedded in the model a priori, the LSTM derives these from the data. These learned representations are interpretable by scientists. (3) LSTMs can be used to gain an estimate of intermediate stores of water such as soil moisture. While machine learning interpretability is still a nascent field and our approach reflects a simple technique for exploring what the model has learned, the results are robust to different initial conditions and to a variety of benchmarking experiments. We therefore argue that deep learning approaches can be used to advance our scientific goals as well as our predictive goals.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Experimental Designmentioning

confidence: 99%

mentioning

confidence: 99%

See 1 more Smart Citation

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

Lees

Reece

Kratzert³

et al. 2022

Hydrol. Earth Syst. Sci.

Self Cite

View full text Add to dashboard Cite

show abstract

“…While this is a good start, it clearly leaves many questions unasked and unanswered. In particular, we have not yet conducted a comparison with a variety of physically/process‐based models—to cleanly perform such a comparison is nontrivial (Kratzert, Klotz, Shalev, et al., 2019; Lees et al., 2021) since different models may use different input information. However, this is certainly something that should be explored in future work.…”

Section: Conclusion Remarks and Outlookmentioning

confidence: 99%

Exploring the Potential of Long Short‐Term Memory Networks for Improving Understanding of Continental‐ and Regional‐Scale Snowpack Dynamics

Wang

Gupta

Zeng

et al. 2022

Water Resources Research

View full text Add to dashboard Cite

Accurate estimation of the spatio‐temporal distribution of snow water equivalent is essential given its global importance for understanding climate dynamics and climate change, and as a source of fresh water. Here, we explore the potential of using the Long Short‐Term Memory (LSTM) network for continental and regional scale modeling of daily snow accumulation and melt dynamics at 4‐km pixel resolution across the conterminous US (CONUS). To reduce training costs (data are available for ∼0.31 million snowy pixels), we combine spatial sampling with stagewise model development, whereby the network is first pretrained across the entire CONUS and then subjected to regional fine‐tuning. Accordingly, model evaluation is focused on out‐of‐sample predictive performance across space (analogous to the prediction in ungauged basins problem). We find that, given identical inputs (precipitation, temperature, and elevation), a single CONUS‐wide LSTM provides significantly better spatio‐temporal generalization than a regionally calibrated version of the physical‐conceptual temperature‐index‐based SNOW17 model. Adding more meteorological information (dew point temperature, vapor pressure deficit, longwave radiation, and shortwave radiation) further improves model performance, while rendering redundant the local information provided by elevation. Overall, the LSTM exhibits better transferability than SNOW17 to locations that were not included in the training data set, reinforcing the advantages of structure learning over parameter learning. Our results suggest that an LSTM‐based approach could be used to develop continental/global‐scale systems for modeling snow dynamics.

show abstract

“…With the increase of computing power and the development of novel algorithms, deep learning models have become powerful tools that have been widely utilized in hydrology studies (Fang et al, 2017;Orland et al, 2020;Liu et al, 2021;Feng et al, 2021a) in the past few years. The majority of studies focus on daily rainfall-runoff modeling and streamflow forecasting (Kratzert et al, 2018;Kratzert et al, 2019;Feng et al, 2020;Qian et al, 2020;Sarkar et al, 2020;Van et al, 2020;Lees et al, 2021). Most recent research has focused on improving neural network model accuracy with physical information (Feng et al 2020;Fang et al 2020;Rahmani et al 2021;Gauch et al 2021;Klotz et al 2021).…”

Section: Introductionmentioning

confidence: 99%

Real-Time Streamflow Forecasting Framework, Implementation and Post-Analysis Using Deep Learning

Xiang¹,

Demir²

2022

Preprint

View full text Add to dashboard Cite

Rainfall-runoff modeling and streamflow prediction using deep learning algorithms have been studied significantly in the last few years. The majority of these studies focus on the simulation and testing of historical datasets. Deployment and operation of a real-time streamflow forecast model using deep learning will face additional data and computational challenges such as inaccurate rainfall forecast data and real-time data assimilation with limited studies guiding on these difficulties. We proposed a real-time streamflow forecast framework that includes pre-event model training using deep learning, real-time data acquisition, and post-event analysis. We implemented the framework for 124 USGS gauged watersheds across Iowa to forecast 120-hour streamflow rates since April 2021. This is the first time deep learning models have been used to predict streamflow in real-time operational settings at a large scale, and we anticipate seeing more real-time implementations of deep learning models in the future.

show abstract

Benchmarking data-driven rainfall–runoff models in Great Britain: a comparison of long short-term memory (LSTM)-based models with four lumped conceptual models

Cited by 127 publications

References 58 publications

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

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

Exploring the Potential of Long Short‐Term Memory Networks for Improving Understanding of Continental‐ and Regional‐Scale Snowpack Dynamics

Real-Time Streamflow Forecasting Framework, Implementation and Post-Analysis Using Deep Learning

Contact Info

Product

Resources

About