Extraction and use of noise models from transient electromagnetic data

Abstract

Measurements using the transient electromagnetic (TEM) method are unavoidably contaminated by noise from various sources. Knowledge of the local noise conditions during data acquisition can be useful for tuning the acquisition parameters in the field and for quantification of uncertainties in the interpretation phase. Normally, assessment of the noise conditions requires dedicated noise measurements with the TEM transmitter off. We have developed a frequency-domain method for estimating noise models, in the form of noise power spectra, from data acquired with the TEM system in normal operation with the transmitter on. The method is based on a well-known method for power spectral density estimation, combined with a windowing approach specifically tailored for the characteristics of TEM signals and interpolation in the frequency domain. Because this method does not require the survey to be interrupted, the noise model can be updated continuously, with no added survey cost or inconvenience, to reveal temporal and spatial variations in the noise conditions. The noise models can be used to compute reliable estimates of the noise levels of data after common linear processing steps such as averaging, e.g., for use as input to an inversion code. We exemplified the use of the method through semisynthetic examples, using field-measured noise combined with synthetic TEM signals. From the examples, we concluded that the method produces accurate and stable noise models, which in turn can be used to accurately predict the noise levels of processed data. In practical field applications of the method, several modeling parameters must be selected. We recommend using field noise measurements combined with synthetic TEM signals to obtain suitable parameters by tuning the modeling parameters to yield accurate noise models. Once calibrated for a given TEM system, the set of parameters should be reusable in future surveys.