Substitution for lost one-hour means of the geomagnetic elements for the first half of the 20-th century at the Hurbanovo Geomagnetic Observatory by means of neural networks

Abstract

The existence of long-acting observatories by itself does not guarantee that
their historical magnetograms are available or complete. In the archive of the Hurbanovo
Geomagnetic Observatory (acronym HRB; geographical coordinates 47.86° N, 18.19° E),
records of the geomagnetic field made on photo paper covering the period between the
two World Wars were found for which the values of the baselines are unknown. We
studied if a feed-forward neural network with one hidden layer can be used to supplement
one-hour means of the geomagnetic elements D, H and Z of observatory HRB, using for
this purpose the geomagnetic data of observatories Potsdam, Seddin and Niemegk (all of
them being referenced to Niemegk). We focused our interest on the first half of the 20-th
century. The neural-network model for element D proved to be applicable to substitute
for the lost data of the magnetic declination at observatory HRB; however, the usability
of the model for both elements H and Z turned out to be limited to a few years close to
beginning or end of data gaps. Further we supplemented the time series of annual means
of geomagnetic elements D, H and Z at observatory HRB with the model data.