Surrogate optimization of deep neural networks for groundwater predictions

Müller, Juliane; Park, Jangho; Sahu, Reetik Kumar; Varadharajan, Charuleka; Arora, Bhavna; Faybishenko, Boris; Agarwal, D.

doi:10.1007/s10898-020-00912-0

Cited by 51 publications

(47 citation statements)

References 66 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, data‐driven approaches have gained momentum in watershed modeling because of their computational efficiency and agility to incorporate diverse and multiscale data that are often difficult to incorporate into current process‐based models (Shen, 2018). The value of ML for watershed modeling has been illustrated by applications focused on streamflow prediction (Kratzert, Klotz, Brenner, Schulz, & Herrnegger, 2018), early warning of droughts and floods (Mosavi, Ozturk, & Chau, 2018; Park, Im, Jang, & Rhee, 2016), groundwater level fluctuations (Müller et al, 2019), and chemical equilibrium calculations (Leal, Kulik, & Saar, 2017). However, as data‐driven models are developed directly from observations, their effectiveness is limited when data are sparse.…”

Section: Emerging Technologies Poised To Advance Watershed Hydrobiogementioning

confidence: 99%

Emerging technologies and radical collaboration to advance predictive understanding of watershed hydrobiogeochemistry

Hubbard

Varadharajan

et al. 2020

Hydrological Processes

Self Cite

View full text Add to dashboard Cite

Increasing population and resource-intensive lifestyles are driving enhanced demands for clean water, food, and energy. In parallel, landuse change, climate change, and perturbations-including drought, floods, fires, and early snowmelt-are significantly reshaping interactions within watersheds throughout the world. While watersheds are the Earth's key functional unit for assessing and managing water resources, hydrological processes in watersheds also mediate biogeochemical interactions that support terrestrial life on Earth (Kaushal, Gold, Bernal, & Tank, 2018; National Research Council, 2012). Although society is dependent upon clean water availability, tractable prediction of watershed hydrobiogeochemical behavior, including watershed response to perturbations, remains a challenge. Central to the challenge are complex, multiscale interactions between plants, microorganisms, organic matter, minerals, dissolved constituents, and migrating fluids, which occur within and across bedrock-to-canopy compartments and along extensive lateral gradients of a watershed. Several recent community reports have synthesized formidable challenges associated with watershed science and technology (AGU, 2018;Blöschl et al., 2019).Here, we discuss emerging technologies and collaboration modes that are critical for developing generalizable insights about and predictive understanding of complex watershed hydrobiogeochemical behavior, which are important for underpinning optimized natural resource management.Recent developments in field observatories and open-science principles provide foundational pillars for advancing predictive understanding of watershed hydrobiogeochemistry using emerging technologies. Field observatories have fostered crossdisciplinary collaboration and provided platforms for quantifying hydrological, biological, geological, geochemical, and atmospheric processes and their couplings (Bogena, White, Bour, Li, & Jensen, 2018). Observatory networks in the United States include the Critical Zone

show abstract

Section: Emerging Technologies Poised To Advance Watershed Hydrobiogementioning

confidence: 99%

Emerging technologies and radical collaboration to advance predictive understanding of watershed hydrobiogeochemistry

Hubbard

Varadharajan

et al. 2020

Hydrological Processes

Self Cite

View full text Add to dashboard Cite

show abstract

“…The MLP is a feedforward type of neural network with different hyperparameters that need to be adjusted before its training. The MLP was chosen as it was the best performing model in terms of accuracy and compute time, based on comparison with CNN, RNN, LSTM neural networks (Müller et al, 2020).…”

Section: Neural Network Model and Hyperparameter Tuningmentioning

confidence: 99%

“…Based on the outcome for l, the optimizer at the upper-level selects the next set of hyperparameters for which the lower-level problem is solved, and so on until convergence at the upper-level is achieved. For solving the upper-level optimization problem, we use a derivative-free optimization algorithm that uses radial basis function surrogate models, see Müller et al (2020) for further details. Since a stochastic optimizer is used to solve the lower-level problem (Equation 3), the performance of the MLP for a given architecture θ depends on the random number seed of the stochastic optimizer.…”

Section: Neural Network Model and Hyperparameter Tuningmentioning

confidence: 99%

“…DL techniques can utilize the climate and hydrogeology data to capture the relationships between groundwater levels and other dependent features such as nearby river flow, precipitation and temperature. Recent advances in ML have enabled making groundwater predictions by using purely data-driven models (Taormina et al, 2012;Moosavi et al, 2013;Sahoo et al, 2017;Müller et al, 2020). ML techniques have been used for both prediction and optimization purposes including modeling of groundwater levels and or quality, optimization of groundwater well design, pumping rate, and location (Banerjee et al, 2011;Gaur et al, 2013).…”

Section: Introductionmentioning

confidence: 99%

“…The DL model could predict groundwater levels up to 4 weeks of lead time with a prediction error of about 8% of the annual groundwater-level change. Our previous study (Müller et al, 2020) compared results from a variety of DL methods including multilayer perceptron (MLP), RNN, long short term memory (LSTM), and 1D-convolutional neural network (CNN) designed with our hyperparameter optimization approach for both single-and multi-well groundwater level predictions in California, and were able to attain prediction accuracies of 6-20%, depending on the DL model. Each of the referenced applications utilize different ML models and architecture under different scenarios such as multipoint vs. single-point sites, with data of varying temporal resolutions (hourly, daily, weekly, and monthly).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Impact of Input Feature Selection on Groundwater Level Prediction From a Multi-Layer Perceptron Neural Network

et al. 2020

Self Cite

View full text Add to dashboard Cite

With the growing use of machine learning (ML) techniques in hydrological applications, there is a need to analyze the robustness, performance, and reliability of predictions made with these ML models. In this paper we analyze the accuracy and variability of groundwater level predictions obtained from a Multilayer Perceptron (MLP) model with optimized hyperparameters for different amounts and types of available training data. The MLP model is trained on point observations of features like groundwater levels, temperature, precipitation, and river flow in various combinations, for different periods and temporal resolutions. We analyze the sensitivity of the MLP predictions at three different test locations in California, United States and derive recommendations for training features to obtain accurate predictions. We show that the use of all available features and data for training the MLP does not necessarily ensure the best predictive performance at all locations. More specifically, river flow and precipitation data are important training features for some, but not all locations. However, we find that predictions made with MLPs that are trained solely on temperature and historical groundwater level measurements as features, without additional hydrological information, are unreliable at all locations.

show abstract

An efficient underground water prediction using optimal deep neural network

Sureshkumar

Rajasomashekar²,

Sarala³

2022

Concurrency and Computation

View full text Add to dashboard Cite

For the humanity to the whole and all the creatures of this world, the underground water is the greatest resource to rely upon that highly forms an indispensable factor toward augmented livelihood. In spite of the lack of detailed knowledge, global warming is found to profoundly influence underground water resources through changes in underground water recharge. Prediction of the underground water under a changing climate is essential to living beings. In this article, underground water prediction using optimal deep neural networks (optimal DNN) has been attempted. Initially, the features of temperature and rainfall among the input data have been selected and after which, the chosen data have been fed to the DNN to predict the underground water. In DNN, weight parameters are optimally selected with the help of fish swarm optimization (FSO). The implementation has been done on MATLAB. The simulation results found that the proposed FSO-DNN prediction approach outperforms the existing prediction approaches by 78.9% accuracy, 83% sensitivity, 88% specificity, 95.8% positive predictive values, 52.3% negative predictive values, and 95.8% F-measure.

show abstract

Surrogate optimization of deep neural networks for groundwater predictions

Cited by 51 publications

References 66 publications

Emerging technologies and radical collaboration to advance predictive understanding of watershed hydrobiogeochemistry

Emerging technologies and radical collaboration to advance predictive understanding of watershed hydrobiogeochemistry

Impact of Input Feature Selection on Groundwater Level Prediction From a Multi-Layer Perceptron Neural Network

An efficient underground water prediction using optimal deep neural network

Contact Info

Product

Resources

About