Machine learning for postprocessing ensemble streamflow forecasts

Sharma, Sanjib; Ghimire, Ganesh R.; Siddique, Ridwan

doi:10.2166/hydro.2022.114

Cited by 11 publications

(6 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…LSTMs trained across a large number of watersheds have been shown to outperform process‐based hydrologic models by a substantial margin (Kratzert et al., 2018), even at hourly timescales (Gauch, Kratzert, et al., 2021) and for watersheds treated as unseen by the LSTM (Kratzert, Klotz, Herrnegger, et al., 2019). Given their high degree of performance, LSTMs are being considered for a range of hydrologic applications, including forecasting (Cheng et al., 2020; Sharma et al., 2021), streamflow estimation for ungauged sites (Yin et al., 2021), and post‐processing of process‐based hydrologic models (Frame, Kratzert, Raney, et al., 2021).…”

Section: Introductionmentioning

confidence: 99%

Assessing the Physical Realism of Deep Learning Hydrologic Model Projections Under Climate Change

2022

Water Resources Research

View full text Add to dashboard Cite

Deep learning (DL) models currently represent the state-of-the-art in hydrologic prediction (Nearing et al., 2021;Shen et al., 2021), as evidenced by their unmatched accuracy in predicting streamflow (Liu et al., 2020), soil moisture (Li et al., 2021), evapotranspiration (Ahmed et al., 2021), stream temperature (Rahmani et al., 2021, and water quality indicators (Aldhyani et al., 2020;Zhi et al., 2021). For streamflow prediction, long short-term msemory networks (LSTMs, Hochreiter & Schmidhuber, 1997) have proven particularly effective due to their strong inductive bias toward storing information over time (Hoedt et al., 2021). This property is well suited for capturing hydrologic dynamics driven by multi-scale memory effects within a watershed, such as the persistence and release of water from soil moisture and snowpack. LSTMs trained across a large number of watersheds have been shown to outperform process-based hydrologic models by a substantial margin (Kratzert et al., 2018), even at hourly timescales (Gauch, Kratzert, et al., 2021) and for watersheds treated as unseen by the LSTM (Kratzert, Klotz, Herrnegger, et al., 2019). Given their high degree of performance, LSTMs are being considered for a range of hydrologic applications, including forecasting (Cheng et al., 2020;Sharma et al., 2021), streamflow estimation

show abstract

Section: Introductionmentioning

confidence: 99%

Assessing the Physical Realism of Deep Learning Hydrologic Model Projections Under Climate Change

2022

Water Resources Research

View full text Add to dashboard Cite

show abstract

“…In constructing the hybrid error model used in this work, we emphasized interpretability and parsimony over complexity. Future work could explore more advanced error correction procedures, potentially drawing on the forecast post‐processing literature (Seo et al., 2006; Sharma et al., 2023; Siqueira et al., 2021), more complex optimization schemes for the dynamic residual model, or non‐linear relationships to state variables in the dynamic residual model. Furthermore, this study accomplished only a limited exploration of the spatial generalizability of the approach and future work should examine in more detail how performance varies by hydroclimate regime.…”

Section: Discussionmentioning

confidence: 99%

“…There are two main components of the hybrid model. The first is an initial model updating step referred to as “error correction” (Shen et al., 2022a, 2022b), which is analogous to hydrologic post‐processing (Seo et al., 2006; Sharma et al., 2023; Siqueira et al., 2021; Zha et al., 2020). The error correction model f creates a mapping between the process model state variables ( θ t , SV ) and the raw errors ( e t ), including autocorrelation in the errors through lagged error terms ( e t ‐1: t ‐ p ) out to lag p :

{e}_{t}=f\left({\theta }_{t,SV},{e}_{t-1:t-p}\right)+{\varepsilon }_{t}

…”

Section: Methodsmentioning

confidence: 99%

A Hybrid, Non‐Stationary Stochastic Watershed Model (SWM) for Uncertain Hydrologic Simulations Under Climate Change

Brodeur,

Wi,

Shabestanipour

et al. 2024

Water Resources Research

View full text Add to dashboard Cite

Stochastic Watershed Models (SWMs) are emerging tools in hydrologic modeling used to propagate uncertainty into model predictions by adding samples of model error to deterministic simulations. One of the most promising uses of SWMs is uncertainty propagation for hydrologic simulations under climate change. However, a core challenge is that the historical predictive uncertainty may not correctly characterize the error distribution under future climate. For example, the frequency of physical processes (e.g., snow accumulation and melt) may change under climate change, and so too may the frequency of errors associated with those processes. In this work, we explore for the first time non‐stationarity in hydrologic model errors under climate change in an idealized experimental design. We fit one hydrologic model to historical observations, and then fit a second model to the simulations of the first, treating the first model as the true hydrologic system. We then force both models with climate change impacted meteorology and investigate changes to the error distribution between the models. We develop a hybrid machine learning method that maps model state variables to predictive errors, allowing for non‐stationary error distributions based on changes in the frequency of model states. We find that this procedure provides an internally consistent methodology to overcome stationarity assumptions in error modeling and offers an important advance for implementing SWMs under climate change. We test this method on three hydrologically distinct watersheds in California (Feather River, Sacramento River, Calaveras River), finding that the hybrid model performs best in larger and less flashy basins.

show abstract

“…Ensemble predictions are collections of a limited number of diverse models, offering greater variety than single-model prediction. For example, Sharma et al [8] incorporated two separate models: a weather forecasting model and a simulation hydrology model. This integration resulted in an ensemble forecasting typically performing better with mediumrange time-frames than with shorter lead periods.…”

Section: Literature Reviewmentioning

confidence: 99%

“…Streamflow modeling is a vital technique for managing water resources, especially in the early detection of flood dangers [7,8,9]. Several types of advanced models can operate across the range of climate zones.…”

Section: Introductionmentioning

confidence: 99%

Comparative Analysis of Artificial Intelligence Methods for Streamflow Forecasting

Wei,

Hashim,

Lai

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Deep learning excels at managing spatial and temporal time series with variable patterns for streamflow forecasting, but traditional machine learning algorithms may struggle with complicated data, including non-linear and multidimensional complexity. Empirical heterogeneity within watersheds and limitations inherent to each estimation methodology pose challenges in effectively measuring and appraising hydrological statistical frameworks of spatial and temporal variables. This study emphasizes streamflow forecasting in the region of Johor, a coastal state in Peninsular Malaysia, utilizing a 28-year streamflow-pattern dataset from Malaysia's Department of Irrigation and Drainage for the Johor River and its tropical rainforest environment. For this dataset, wavelet transformation significantly improves the resolution of lag noise when historical streamflow data are used as lagged input variables, producing a 6% reduction in the root-mean-square error. A comparative analysis of convolutional neural networks and artificial neural networks reveals these models' distinct behavioral patterns. Convolutional neural networks exhibit lower stochasticity than artificial neural networks when dealing with complex time series data and with data transformed into a format suitable for modeling. However, convolutional neural networks may suffer from overfitting, particularly in cases in which the structure of the time series is overly simplified.Using Bayesian neural networks, we modeled network weights and biases as probability distributions to assess aleatoric and epistemic variability, employing Markov chain Monte Carlo and bootstrap resampling techniques. This modeling allowed us to quantify uncertainty, providing confidence intervals and metrics for a robust quantitative assessment of model prediction variability.

show abstract

Machine learning for postprocessing ensemble streamflow forecasts

Cited by 11 publications

References 63 publications

Assessing the Physical Realism of Deep Learning Hydrologic Model Projections Under Climate Change

Assessing the Physical Realism of Deep Learning Hydrologic Model Projections Under Climate Change

A Hybrid, Non‐Stationary Stochastic Watershed Model (SWM) for Uncertain Hydrologic Simulations Under Climate Change

Comparative Analysis of Artificial Intelligence Methods for Streamflow Forecasting

Contact Info

Product

Resources

About