Abstract
Magnetotelluric (MT) signals are often contaminated with noise from natural or man-made processes that may not fit a normal distribution or are highly correlated. This may lead to serious errors in computed MT transfer functions and result in erroneous interpretation. A substantial improvement is possible when the time-series are presented as clean as possible for further processing. Cleaning of MT time-series is often done by manual editing. Editing of magnetotelluric time-series is subjective in nature and time consuming. Automation of such a process is difficult to achieve by statistical methods. Artificial neural networks (ANNs) are widely used to automate processes that require human intelligence. The objective here is to automate MT long-period time-series editing using ANN. A three-layer feed-forward artificial neural network (FANN) was adopted for the problem. As ANN-based techniques are computationally intensive, a novel approach was made, which involves editing of five simultaneously measured MT time-series that have been subdivided into stacks (a stack = 5 × 256 data points). Neural network training was done at two levels. Signal and noise patterns of individual channels were taught first. Five channel parameters along with interchannel correlation and amplitude ratios formed the input for a final network, which predicts the quality of a stack. A large database (5000 traces for pattern training and 900 vectors for interchannel training) was prepared to train the network. There were two error parameters to minimize while training: training error and testing error. Training was stopped when both errors were below an acceptable level. The sensitivity of the neural network to the signal-to-noise ratio and the relative significance of its inputs were tested to ensure that the training was correct. MT time-series from four stations with varying degrees of noise contamination were used to demonstrate the application of the network. The application brought out the ability of the network to pick out signals even in a high-noise environment. This suggest the possibility of automating the editing of MT time-series with artificial neural networks.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.