TY  - GEN
ID  - heidok27372
KW  - velocity field estimation
KW  -  convolutional neural network
KW  -  density?driven active solute transport
KW  -  Hele?Shaw cell experiment
AV  - public
CY  - Heidelberg
TI  - Flow Field Estimation of Active Solute Transport ? Information Transfer from Synthetic Data
to Hele-Shaw Cell Experiments
Using Convolutional Neural Networks
Y1  - 2019///
N2  - Variable density groundwater flow associated with active solute transport is understood reasonably well. Nevertheless, predictions are operationally still difficult due to joint effects of nonlinear processes and uncertain boundary conditions. Gaining deeper insight into the dynamics of these groundwater systems therefore relies on the availability of accurate and dense measurements of the complete system state and parameters. Often, such measurements are hard to come by, hence our information is incomplete. Recent deep learning methods in conjunction with numerical simulation of the physical processes to create large training datasets enable the information transfer to real world problems. To demonstrate this, I chose a laboratory experiment on density-driven active solute transport observed in a Hele-Shaw cell, where high resolution measurements of the solute concentration distribution are available. With the use of deep convolutional neural networks I was able to estimate the otherwise inaccessible flow fields and to identify the influence of background flow for this experiment without explicit knowledge of the boundary conditions. The situation of missing data, as encountered here, is typical also for other hydrological systems, from soil-vegetation-atmosphere interactions to catchment dynamics and groundwater recharge. Hence, I believe that the approach has wide applicability.
A1  - Kreyenberg, Philipp Johannes
UR  - https://archiv.ub.uni-heidelberg.de/volltextserver/27372/
ER  -