An ANN Approach in Predicting Solar and Geophysical Indices from Ionospheric TEC Over Indore

TL;DR

This study develops an ANN model trained on TEC data to accurately predict solar and geophysical indices, achieving approximately 97% prediction accuracy, which is novel for reverse problem modeling in ionospheric studies.

Contribution

The paper introduces a novel ANN approach for predicting solar and ionospheric indices from TEC data, addressing the reverse problem for the first time.

Findings

Achieved ~97% prediction accuracy for Rz12, IG12, and F10.7 indices.

Developed a TEC-based ANN model using one year's hourly data.

Demonstrated the effectiveness of the model with low RMSEs.

Abstract

In this paper, preliminary results from the artificial neural network (ANN) based model developed at IIT Indore has been presented. One year's hourly total electron content (TEC) database has been created from the International Reference Ionosphere (IRI) 2016 model. For the first time, a reverse problem has been addressed, wherein the training has been performed for predicting the three indices: 13-month running sunspot number, ionospheric index, and daily solar radio flux also called targets to the network when hourly TEC values are the inputs. The root mean square errors (RMSEs) of these targets have been compared and minimized after several training of the dataset using different sets of combinations. Unknown data fed to the network yielded 0.99%, 3.12%, and 0.90% errors for Rz12, IG12, and F10.7 radio flux, respectively, thus signifying ~97% prediction accuracy of the model.