Characterization of and correction for cultural noise

Surveys of time varying electromagnetic fields result in time series consisting of signals and noise, the latter defined as that part of the data which cannot be explained by a theory. Man-made contributions to noise can be subdivided into active and passive sources and are complex in character. As Szarka has treated this topic extensively in a recent review paper (Szarka, 1988), only a few further examples are presented here. Following discussion of noise correction in transient electromagnetic investigations which consists mainly of sophisticated stacking and filter procedures, several aspects of its correction in magnetotelluric and geomagnetic depth sounding data are considered. These include:The methods of treatment of single time series in the presence of visible noise—its detection, removal and sometimes replacement by data predicted from undisturbed intervals.The investigation of time series interrelations. This is mainly coherence based and—if possible—takes advantage of remote reference techniques.The examination of the statistical properties of the time series by regression analysis. This leads to the weighting of time segments of data in order to achieve unbiased and minimum variance estimates based on identically and independently Gaussian distributed residuals.The application of constraints. These can further improve the estimates' quality.The use of simultaneously recorded multistation data. This can contribute remarkably to noise suppression as well as to the treatment of non-uniform source fields.Leveraging and confidence limits. Problems relating to the former have not yet been solved satisfactorily while the Jacknife method seems to be an easy way of determining the latter. The methods of treatment of single time series in the presence of visible noise—its detection, removal and sometimes replacement by data predicted from undisturbed intervals. The investigation of time series interrelations. This is mainly coherence based and—if possible—takes advantage of remote reference techniques. The examination of the statistical properties of the time series by regression analysis. This leads to the weighting of time segments of data in order to achieve unbiased and minimum variance estimates based on identically and independently Gaussian distributed residuals. The application of constraints. These can further improve the estimates' quality. The use of simultaneously recorded multistation data. This can contribute remarkably to noise suppression as well as to the treatment of non-uniform source fields. Leveraging and confidence limits. Problems relating to the former have not yet been solved satisfactorily while the Jacknife method seems to be an easy way of determining the latter. Thanks to the modern processing techniques reviewed in this paper it should be possible to obtain a rather dense net of high quality data in spite of the world-wide increasing noise level. As most processing codes are widely accessible current problems are more related to availability of instruments, carrying out the measurements and reserving enough time for thorough data processing.