Tomlinson, L.A. 1995 A comparison of automated and hand-scaled K-indices at Eyrewell Geomagnetic Observatory . Lower Hutt: Institute of Geological & Nuclear Sciences Institute of Geological & Nuclear Sciences science report 95/40 19 p.
Abstract: Three of the four digital algorithms acknowledged by IAGA for obtaining K-indices directly from digital data were implemented and used to obtain K-indices from the Eyrewell digital data for 1994. The results were compared to those obtained from hand-scaling the indices. The Linear-phase, robust, non-linear smoothing (LRNS) method developed by Hattingh et al. at the Hermanus observatory in South Africa gave the best agreement with hand scaling provided the appropriate control parameter was selected. A slightly better agreement was obtained when the algorithm was applied to blocks of data beginning and ending at local midnight. The level of agreement was similar when the algorithm was applied to data from 1992 and 1993. There are two other criteria that can be used in comparing digital algorithms to hand scaled data. Firstly there should be very few instances where the algorithm and hand-scaling give differences greater than one k unit and secondly that there should be no consistent tendency for the digital algorithm to give higher or lower values than hand-scaling, ie the statistical distribution of K-indices should remain unaltered. It is well known that different people scaling the same record will frequently give indices differing by one unit, especially on borderline cases and the same is to be expected when comparing a digital algorithm with hand-scaled results. Discrepancies of more than one k unit is an indication of differences in the adopted SR curves. The LRNS gave the lowest incidence of differences of two K units (about 30 instances per year) and differences of more than 2 K units were very rare (less than 5 instances per year). No check was made of the records in these cases to see if there had been a gross error in the hand-scaled index or if there had been noise on the digital record to account for the discrepencies. The LRNS algorithm had a slight tendency to give higher K-indices during magnetically quiet periods, and a slight tendency to give lower values during periods of moderate disturbance. These effects are commented on in the text and are considered to be the result of the difficulty of observing small disturbances by eye and to be the result of the discrete sampling of the digital record not recording peak excursions. The other two algorithms had definite tendencies to give higher (FMI method) or lower (AS method) results. The statistical distribution of indices for the LRNS method was closest to that obtained from hand scaling. (auth)