Comparison of the Kriging and neural network methods for modeling foF2 maps over North China region

Abstract

The F2 layer critical frequency of the ionosphere (foF2) is one of the most significant parameters for studying the ionosphere. To investigate the large-scale characteristics of the ionosphere over particular regions, modeling foF2 is an effective method. In this paper, we use both the Kriging (KG) and neural network (NN) methods to reconstruct foF2 maps over North China. The neural network is trained by the genetic algorithm (GA) to avoid the ‘local minimum’ phenomenon in most NN applications. We then carry out a comparison between foF2 provided by both the KG and NN methods with vertical model operation of ionosonde data including Beijing, Qingdao, Suzhou, and Changchun. All of the foF2 data used in the comparison are obtained from the oblique and vertical mode operation of ionosonde from the China Ground-based Seismo-ionospheric Monitoring Network. To allow for a possible seasonal and diurnal variation, data obtained from summer, winter, and equinox months are applied in the present comparison. In addition, we make a comparison during a magnetic storm period. The results of our comparisons demonstrate that both the KG and NN methods are appropriate tools for modeling foF2 maps. However, when the data set is spare, the performance of the NN method is better than the KG method. On the other hand, the KG method is more robust than the NN method during a magnetic storm.