Forecasting hourly [formula omitted] over Sonmiani using an artificial neural network algorithm

Abstract

This article describes the efforts made to forecast daily hourly critical frequency foF2 of the ionosphere over a station at the crest of equatorial ionization anomaly (EIA) region by using artificial neural network (ANN) technique. The ANN based model is developed to forecast foF2 1-h in advance over Sonmiani (25.19°N, 66.74°E) by making use of ionosonde data of 2018–2020. The input parameters of the model include day of year (DOY), local time (Hour), 27- day running mean of solar radio flux (F10.7), running mean of 8 previous values of geomagnetic planetary Ap index (AP8) and 1-h prior value of ionosonde foF2 (foF2pObs). The model is comprising of one hidden layer with 27 neurons, tangent hyperbolic as activation function, adaptive movement estimation (Adam) as optimization algorithm and mean squared error (MSE) as cost function. The optimal configuration is obtained by training the model on 500 epochs using above-mentioned inputs and configuration. The model is then validated over unseen data of 2021. It is found that the developed ANN based model predicts 1-h ahead values of foF2 with an average RMSE of 1.0 MHz on comparing with ionosonde measurements. For the same unseen data set RMSE of 1.6 MHz is found between ionosonde measurements and IRI-Plas 2020 estimations. The developed space weather model may be further improved by incorporating long time series data in future to capture geomagnetically as well as atmospherically induced short-term variations.