Mapping of hydraulic transmissivity field from inversion of tracer test data using convolutional neural networks. CNN-2T

This paper introduces a new concept for mapping hydraulic transmissivity from temporal concentration data collected in multiple tracer tests. Based on convolutional neural network, the principle uses an encoder-decoder architecture with layers to establish relationship between and field. is established two phases networks. The first network designated trained reconstruct field using single test. To improve reconstruction quality, second then performs joint interpretation tests, which reprocesses all resulted each individual Both networks are by synthetic data, where models generated Gaussian variogram its properties considered as prior information aquifer heterogeneity. Tracer tests derived numerically solving forward problem obtain corresponding that feed training. accurately map fields, of accuracy relies volume nature heterogeneities training models, well number piezometers used monitor changes. Reconstruction other hand, less influenced noise. Effective requires large dataset, but time required dataset generation only order Gauss-Newton algorithm conventional inversion, while inference instantly.