Obtaining spectrum and various transfer functions via noise elimination studies on time domain electromagnetic data

The data obtained via observing physical events changing with time are defined as time series (TS). TS has a complex structure as summed signals containing noise components in addition to the information related to that physical events. Therefore, studies targeting to definition and reduction of the noise in TS are required in order to understand the physical meaning of the observed events. For this purpose, the applications are carried out by using different methods under the so-called both pre-processing and evaluation of data. The advantage and disadvanage of the methods that depends on the noise types together with their fundamental premises affect the success of the studies on reducing the noise components.The study uses the magnetotelluric (MT) method data based on measuring the changes in the natural electromagnetic field components of ground. The noise effects in TS cause deviations in auto- and cross-power spectrums of MT components and consequently systematic errors in the Earth response function (ERF) used for defining geoelectrical structure. Thus, the methods on noise reduction have been developed to estimate, classify eliminate, and weight the noisy sub-time segments before Earth response function (ERF) calculations. The studies are carried out by the assumption of an average error range of 30% exist in data sets. According to this criterion, the noise elimination studies can be summarized as 1- classifying and eliminating the noisy segments which impairing the stability of MT components in auto power spectral density (PSD classification); 2- classifying and eliminating the noisy segments by evaluating the noise effects which impairing the TS pattern and stability of the segments together with the change in 1 (IEG classification); 3- weighing the noise segments defined with IEG classification and reducing its effect on the solution and 4- estimating and eliminating the noise segments with artificial neural network (ANN) depending to the time and frequency characteristics of the data.The methods have been tested on synthetic data and then applied to two different field data sets. As a result of the classification studies, it is seen that the noise components, which cause contrast between the segments and where the conventional methods are insufficient, can be reduced. With IEG weighed robust approach, successful results have been obtained as an alternative to the conventional robust method. Additionally, the speed and superiority over automation of ANN and the applicability of MT data in noise elimination studies have been observed.