Machine learning algorithms for modeling groundwater level changes in agricultural regions of the U.S.

Sahoo, Sasmita; Russo, T. A.; Elliott, Joshua; Foster, Ian

doi:10.1002/2016wr019933

Cited by 322 publications

(143 citation statements)

References 89 publications

Supporting

Mentioning

139

Contrasting

Unclassified

Order By: Relevance

“…Popular black box models include Box-Jenkins type models (e.g. Changnon et al 1988;Gehrels et al 1994;Van Geer and Zuur 1997), which originate from economics (e.g., Box and Jenkins 1970), artificial neural networks (e.g., Daliakopoulos et al 2005), and other artificial intelligence approaches (e.g., Sahoo et al 2017;Wunsch et al 2018). An increasingly popular gray box model is a method that makes use of predefined, physically realistic response functions (e.g.…”

Section: Data-driven Modelingmentioning

confidence: 99%

Solving Groundwater Flow Problems with Time Series Analysis: You May Not Even Need Another Model

Bakker¹,

Schaars²

2019

Groundwater

View full text Add to dashboard Cite

Time series analysis is a data‐driven approach to analyze time series of heads measured in an observation well. Time series models are commonly much simpler and give much better fits than regular groundwater models. Time series analysis with response functions gives insight into why heads vary, while such insight is difficult to gain with black box models out of the artificial intelligence world. An important application is to quantify the contributions to the head variation of different stresses on the aquifer, such as rainfall and evaporation, pumping, and surface water levels. Time series analysis may be applied to answer many groundwater questions without the need for a regular groundwater model, such as what is the drawdown caused by a pumping station? Or, how long will it take before groundwater levels recover after a period of drought? Even when a regular groundwater model is needed to solve a groundwater problem, time series analysis can be of great value. It can be used to clean up the data, identify the major stresses on the aquifer, determine the most important processes that affect flow in the aquifer, and give an indication of the fit that can be expected. In addition, it can be used to determine calibration targets for steady‐state models, and it can provide several alternative calibration methods for transient models. In summary, the overarching message of this paper is that it would be wise to do time series analysis for any application that uses measured groundwater heads.

show abstract

Section: Data-driven Modelingmentioning

confidence: 99%

Solving Groundwater Flow Problems with Time Series Analysis: You May Not Even Need Another Model

Bakker¹,

Schaars²

2019

Groundwater

View full text Add to dashboard Cite

show abstract

“…Of the possible statistical methods available for this approach, the artificial neural networks technique was selected. This technique has been widely used for statistical downscaling in the hydrosciences [39][40][41], in spatial data analysis [42][43][44], in studies for groundwater management [45], for predicting groundwater levels [46][47][48][49][50][51][52], as well as for predicting groundwater levels with GRACE data [53]. Neural network studies have also illustrated the method's ability to simulate complex hydrological characteristics across various regions and time periods [54,55].…”

Section: Downscaling Grace Datamentioning

confidence: 99%

Downscaling GRACE Remote Sensing Datasets to High-Resolution Groundwater Storage Change Maps of California’s Central Valley

2018

View full text Add to dashboard Cite

NASA's Gravity Recovery and Climate Experiment (GRACE) has already proven to be a powerful data source for regional groundwater assessments in many areas around the world. However, the applicability of GRACE data products to more localized studies and their utility to water management authorities have been constrained by their limited spatial resolution (~200,000 km 2 ). Researchers have begun to address these shortcomings with data assimilation approaches that integrate GRACE-derived total water storage estimates into complex regional models, producing higher-resolution estimates of hydrologic variables (~2500 km 2 ). Here we take those approaches one step further by developing an empirically based model capable of downscaling GRACE data to a high-resolution (~16 km 2 ) dataset of groundwater storage changes over a portion of California's Central Valley. The model utilizes an artificial neural network to generate a series of high-resolution maps of groundwater storage change from 2002 to 2010 using GRACE estimates of variations in total water storage and a series of widely available hydrologic variables (PRISM precipitation and temperature data, digital elevation model (DEM)-derived slope, and Natural Resources Conservation Service (NRCS) soil type). The neural network downscaling model is able to accurately reproduce local groundwater behavior, with acceptable Nash-Sutcliffe efficiency (NSE) values for calibration and validation (ranging from 0.2445 to 0.9577 and 0.0391 to 0.7511, respectively). Ultimately, the model generates maps of local groundwater storage change at a 100-fold higher resolution than GRACE gridded data products without the use of computationally intensive physical models. The model's simulated maps have the potential for application to local groundwater management initiatives in the region.

show abstract

“…Oikonomou et al (2018) use an exogenous seasonal autoregressive integrated moving average stochastic model and ensemble smoother for predicting water table levels and filling in data gaps. Others studies model groundwater levels and deal with missing data by using higher spatial and/or temporally more frequent data sets, such as remotely sensed data from the GRACE satellites (Mukherjee & Ramachandran, 2018;Sun, 2013), impulse response functions to relate precipitation to groundwater levels (Marchant & Bloomfield, 2018;von Asmuth et al, 2002), machine learning using artificial neural networks (Daliakopoulos et al, 2005;Sahoo et al, 2017), and interpolation methods that use secondary variables to improve estimation in sparsely sampled areas (Desbarats et al, 2002;Passarella et al, 2017;Peterson et al, 2011). Interpolation techniques are commonly used to transform point measurements at groundwater wells to groundwater surfaces across an aquifer.…”

Section: Introductionmentioning

confidence: 99%

Evaluation of Groundwater Levels in the Arapahoe Aquifer Using Spatiotemporal Regression Kriging

Ruybal

Hogue

McCray

2019

Water Resources Research

View full text Add to dashboard Cite

Groundwater monitoring is fundamental to understanding system dynamics, trends in storage, and the long‐term sustainability of an aquifer. Water‐level data are the key source of information used to understand the response. However, groundwater‐level data are often irregularly sampled, leading to temporal gaps in the record, and are not adequately distributed spatially across an aquifer. This presents challenges when spatially interpolating potentiometric surfaces and creating groundwater maps due to data availability. We present a spatiotemporal kriging methodology to improve spatial and temporal confidence in groundwater‐level predictions at unsampled locations. The space–time data set consists of a trend and residual component modeled with a linear regression and utilize a sum‐metric model to represent spatiotemporal covariances. The Arapahoe aquifer is used as a case study to demonstrate the benefits of spatiotemporal kriging over spatial kriging across a sparsely gauged and irregularly sampled aquifer. The Arapahoe aquifer is a major source of water for many residents along the Rocky Mountain Front Range in Colorado. The results show superior performance of spatiotemporal kriging to predict groundwater levels over the traditional spatial kriging. Spatiotemporal kriging represents realistic temporal and spatial changes in water levels and avoids some of the problems inherent to spatial kriging. This study demonstrates the power of spatiotemporal kriging to help inform system dynamics in irregularly sampled aquifers.

show abstract

Machine learning algorithms for modeling groundwater level changes in agricultural regions of the U.S.

Cited by 322 publications

References 89 publications

Solving Groundwater Flow Problems with Time Series Analysis: You May Not Even Need Another Model

Solving Groundwater Flow Problems with Time Series Analysis: You May Not Even Need Another Model

Downscaling GRACE Remote Sensing Datasets to High-Resolution Groundwater Storage Change Maps of California’s Central Valley

Evaluation of Groundwater Levels in the Arapahoe Aquifer Using Spatiotemporal Regression Kriging

Contact Info

Product

Resources

About