A Systematic Review of Deep Learning Applications in Streamflow Data Augmentation and Forecasting

Sit, Muhammed; Demiray, Bekir Zahit; Demir, İbrahim

doi:10.31223/x5hm08

Cited by 6 publications

(4 citation statements)

References 145 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to add more features to the deep neural network model, a long short-term memory neural network (Hochreiter and Schmidhuber, 1997) extends the lifespan of the short-term memory. In hydrological forecasting tasks that require longer memory, such as flood forecasting and rainfall forecasting, LSTMs and their variations are frequently used (Sit and Demir, 2019;Sit et al, 2022a). A concise description of an LSTM node and its mathematical foundation can be found in Equations 4 through 9, and Figures 6 and 7.…”

Section: Long Short-term Memory (Lstm) Networkmentioning

confidence: 99%

“…Estimating streamflow for an ungauged basin or regionalizing an ungauged basin is often done by taking advantage of spatial proximity, physical similarity, regression-based methods, or hydrological signature methods. Besides these approaches, Guo et al (2021) proposed a machine learning approach for this task, albeit its utilization is quite limited in the literature (Sit et al, 2022a).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Spatial Downscaling of Streamflow Data with Attention Based Spatio-Temporal Graph Convolutional Networks

Sit¹,

Demiray²,

Demir³

2023

Preprint

View full text Add to dashboard Cite

Accurate streamflow data is vital for various climate modeling applications, including flood forecasting. However, many streams lack sufficient monitoring due to the high operational costs involved. To address this issue and promote enhanced disaster preparedness, management, and response, our study introduces a neural network-based method for estimating historical hourly streamflow in two spatial downscaling scenarios. The method targets two types of ungauged locations: (1) those without sensors in sparsely gauged river networks, and (2) those that previously had a streamflow sensor, but the gauge is no longer available. For both cases, we propose the ScaleGNN, a graph neural network based on Attention-Based Spatio-Temporal Graph Convolutional Networks (ASTGCN). We evaluate the performance of ScaleGNN against a Long Short-Term Memory (LSTM) baseline and spatial persistence in estimating discharge values over a 36-hour period. Our findings indicate that ScaleGNN surpasses spatial persistence in the first scenario, while both neural network approaches demonstrate their effectiveness compared to spatial persistence in the second scenario.

show abstract

Section: Long Short-term Memory (Lstm) Networkmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Spatial Downscaling of Streamflow Data with Attention Based Spatio-Temporal Graph Convolutional Networks

Sit¹,

Demiray²,

Demir³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Many classical machine-learning approaches have been used in streamflow forecasting and environmental studies (Bayar et al, 2009;Li and Demir, 2023) including Support Vector Machines (SVMs) and Linear Regression (LR) (Granata et al, 2016;Yan et al, 2018;Sharma and Machiwal, 2021). However, advancements in artificial intelligence (AI) coupled with the increasing capabilities of graphics processing units (GPUs) have opened up new possibilities and accelerated the progress of deep learning techniques, which has led to the widespread usage of these techniques in streamflow forecasting as well (Sit et al, 2022a). Out of various neural network architectures explored for streamflow forecasting (Sit et al, 2021a;Xiang and Demir, 2022b;Chen et al, 2023), Recurrent Neural Networks (RNNs), especially the Long Short-Term Memory (LSTM) neural network and Gated Recurrent Units (GRUs), have emerged as the most extensively studied and researched models in this domain.…”

Section: Introductionmentioning

confidence: 99%

Enhancing Hydrological Modeling with Transformers: A Case Study for 24-Hour Streamflow Prediction

Demiray,

Sit,

Mermer

et al. 2023

Preprint

View full text Add to dashboard Cite

In this paper, we address the critical task of 24-hour streamflow forecasting using advanced deep-learning models, with a primary focus on the Transformer architecture which has seen limited application in this specific task. We compare the performance of five different models, including Persistence, LSTM, Seq2Seq, GRU, and Transformer, across four distinct regions. The evaluation is based on three performance metrics: Nash-Sutcliffe Efficiency (NSE), Pearson’s r, and Normalized Root Mean Square Error (NRMSE). Additionally, we investigate the impact of two data extension methods: zero-padding and persistence, on the model's predictive capabilities. Our findings highlight the Transformer's superiority in capturing complex temporal dependencies and patterns in the streamflow data, outperforming all other models in terms of both accuracy and reliability. The study's insights emphasize the significance of leveraging advanced deep learning techniques, such as the Transformer, in hydrological modeling and streamflow forecasting for effective water resource management and flood prediction.

show abstract

“…Sit et al. (2021) proposed a model based on the combination of graph convolutional and GRU architectures, following the methods developed by Seo et al. (2018), to build a model for the prediction of a 36 hr of streamflow using data coming from different sensors in a river network.…”

Section: Introductionmentioning

confidence: 99%

Graph Convolutional Recurrent Neural Networks for Water Demand Forecasting

Zanfei

Brentan

Menapace

et al. 2022

Water Resources Research

View full text Add to dashboard Cite

Forecasting of urban water demand is essential today for managing water supply systems (WSSs). A reliable knowledge of the future water demand supports taking informed operational, tactical and strategical decisions. However, the stochastic nature of water demand makes the development of a robust forecasting model a challenging task (House-Peters & Chang, 2011). In the past years, machine learning methods gained considerable attention in the forecasting field, due to the increasing availability of computational resources and data (Ghalehkhondabi et al., 2017). This trend is likely to increase in the future, thanks to the new era of big data coming from smart meters (Nguyen et al., 2018). Many authors successfully developed powerful methods to forecast water demand, providing innovative solutions to water utilities. For instance, Herrera et al. ( 2010) considered multiple machine learning methods to develop a prediction model for the water demand in a city in Spain. Apart from exhibiting the superior performances of the support vector regression method on the case study, the authors also used the prediction output for a hydraulic model highlighting the importance of the forecasting application.Artificial neural networks (ANNs) have been widely used, especially for short-term forecasting (Donkor et al., 2014). Among all the different architectures, the feedforward ANN has been consistently adopted due to its powerful performance with a relatively easy implementation. This is the case of the works from Bougadis et al. (2005), Adamowski and Karapataki (2010), Adamowski et al. (2012), Romano and Kapelan (2014), and Pesantez et al. ( 2020). Among plenty of relevant studies on water demand forecasting based on ANNs, it highlights the work of Ghiassi et al. (2008), who developed a dynamic artificial neural network model to predict daily, weekly and monthly water demand. Besides, Ghiassi et al. (2008) applied the model to real water demand data, successfully obtaining a reliable forecast for the three different time horizons. Finally, the proposed dynamic artificial neural network was compared with more conventional methods resulting in an overall better performance. However, many different neural network architectures have been used to develop powerful prediction models. Among them, the recurrent ANN model showed significantly better results. For instance, Guo et al. (2018) proposed a deep learning model based on the gated recurrent unit (GRU) architecture to forecast the water demand data of two district metering areas (DMAs) with a 15-min time step. The results showed how the GRU-based model outperformed the conventional ANNs. Furthermore, Mu et al. (2020) developed a long short-term memory (LSTM) architecture to predict water demands of a WSS in China. The proposed model was

show abstract

A Systematic Review of Deep Learning Applications in Streamflow Data Augmentation and Forecasting

Cited by 6 publications

References 145 publications

Spatial Downscaling of Streamflow Data with Attention Based Spatio-Temporal Graph Convolutional Networks

Spatial Downscaling of Streamflow Data with Attention Based Spatio-Temporal Graph Convolutional Networks

Enhancing Hydrological Modeling with Transformers: A Case Study for 24-Hour Streamflow Prediction

Graph Convolutional Recurrent Neural Networks for Water Demand Forecasting

Contact Info

Product

Resources

About