Fast Neural Network Surrogates For Complex Groundwater Flow Models

Surrogate modeling approach has been adopted in the study to replace computationally expensive physical-based numerical flow and transport model. Two approximate surrogate models namely, Artificial Neural Network (ANN) and Gaussian Process Model (GPM) are developed individually using a scenario database generated from the density dependent numerical flow and transport model OpenGeoSys (OGS). The state-space surrogates have the flexibility to move freely from one point to another within a time frame of decades and also to allow for moderate extrapolation in the case of extreme abstractions. The performance of the GPM was better in many cases with a little compromise on the computational time. A comparison of the performances of both the surrogates shows that they are reliable enough to be used in the management frameworks for decision making