Neural Networks for identification of neutron monitor faulty tube-counts

Danislav T. Sapundjiev, S. M. Stankov, Jean-Claude Jodogne
Royal Meteorological Institute of Belgium

Neutron monitor data quality and control is fundamental for successful and reliable application of this instrument to space weather forecasting. The majority of the operating neutron monitors were built in the years 1965-1975. Despite all care and efforts to maintain their operation, noise and spurious peaks in one or more individual detector tubes are still observed in the output data. This requires data control and verification by an operator which is not suitable for real-time applications. The usual algorithm is to check the individual tube counts against the ratios with the remaining tubes Sapundjiev et al., 2012. In some cases, more than half of the individual tubes do not pass the tube-ratio-test and the measurement cannot be accepted. In this work we are investigating the applicability and the advantages of neural networks (NNs) to detect faulty and good tube counts and the potency to recover erroneous data from as little as a single correct measurement. In order to evaluate this method, besides the real operational data for the training, we also used complex tailored data with as little as one good detector measurement. Finally we are testing the NNs for real time data-control and correction complimentary to the tube ratios method.