Monitoring of geomagnetic field parameters and its variations is mainly carried out using ground-based magnetic observatories and variational stations. However, the imperfection of equipment used and the communication channels involved causes the presence of gaps in the time series of geomagnetic data, which, along with the spatial anisotropy of data sources, creates significant obstacles to their automated processing. In addition, the well-known methods for imputation of time series gaps provide the root-mean-square recovery error significantly exceeding the level acceptable for geophysical observations. Thus, the paper proposes a method for recovering geomagnetic data based on statistical methods for processing time series and machine learning principles using marked data and characterized by the fact that a pair of the time series fragments preceding and succeeding a missing fragment provide an indicative description of the time series fragment of interest, which together form a training sample to search for the missing fragment by a set of its attributes, followed by linear scaling to restore the original trend of an information signal. Analytical estimates of parameters of geomagnetic data time series are given, under which it is possible to apply the proposed method to recover both daily variations and several-minutes-long fragments.
Approach to the recovery of geomagnetic data by comparing daily fragments of a time series with equal geomagnetic activity
