Machine learning for groundwater pollution source identification and monitoring network optimization

The identification of the source in groundwater pollution is the only way to drastically deal with resulting environmental problems. This can only be achieved by an appropriate monitoring network, the optimization of which is prerequisite for the solution of the inverse modeling problem, i.e., identifying the source of the pollutant on the basis of measurements taken within the pollution field. For this reason, a theoretical confined aquifer with two pumping wells and six suspected sources is studied. Simulations of combinations of possible source locations, and hydraulic parameters, produce sets of measurement features for a 29 × 29 grid representing potential monitoring wells. Three sets of simulations are conducted to produce synthetic datasets, representing different groundwater pollution modeling methods. Features (input-X variables) coupled with respective sources (output-Y variables) are formulated in two different dataset formats (Types A, B) in order to train classification (random forests, multilayer perceptron) and computer vision (convolutional neural networks) algorithms, respectively, to solve the inverse modeling problem. In addition, appropriate feature selection and trial-and-error tests are employed for supporting the optimization of monitoring wells’ number, locations and sampling frequency. The methodology can successfully produce various sub-optimal monitoring strategies for various budgets. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s00521-022-07507-8.