IFAC PapersOnLine 50-1 (2017) 12829–12834 ScienceDirect ScienceDirect Available online at www.sciencedirect.com 2405-8963 © 2017, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. Peer review under responsibility of International Federation of Automatic Control. 10.1016/j.ifacol.2017.08.1932 © 2017, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. TEC modelling via neural network using observations from the first GLONASS R&D data network in Brazil and the RBMC Arthur A. Ferreira * Renato A. Borges * Claudia Paparini ** Sandro M. Radicella ** * Dep. of Electrical Engineering, University of Brasilia (UnB), 70910-900, Brasília, DF, Brazil (e-mail: arthur.ferreira@aluno.unb.br; raborges@ene.unb.br). ** The Abdus Salam International Centre for Theoretical Physics (ICTP), strada Costiera 11, 34151, Trieste, Italy (e-mail: {paparini;rsandro}@ictp.it). Abstract: This work presents a result on the use of neural networks (NNs) model to estimate Total Electron Content (TEC) behavior based on Global Navigation Satellite Systems (GNSS) measurements in the Brazilian equatorial and low latitude sectors. The main goal of the proposed NN is to estimate GPS (Global Positioning System) TEC values at locations without a GNSS receiver that may be used, for instance, as background models in regional TEC mapping procedures. The proposed approach is useful especially for single frequency users that rely on corrections of ionospheric range errors by TEC models. The data used was collected on the first GLONASS (Globalnaya Navigatsionnaya Sputnikovaya Sistema) network for research and development (GLONASS R&D network), recently inaugurated in Brazil, and also on the Brazilian Network for Continuous Monitoring of the GNSS Systems (RBMC), with a temporal interval of 15s or 30s and a spatial resolution of about 300 km over an area corresponding to a longitudinal extension of 650 km. The input parameters for the NN used in this work are the latitude, longitude, day of the year (doy), time of the day, the global geomagnetic storm index (Kp-index), and the solar radio flux at 10.7 cm, and the output the vertical TEC (vTEC e ). The vTEC used for training the NN is calculated with the GPS-TEC Analysis Application, version 2.9.3. Future work considers applying the vTEC calculated with the ICTP method in the training process which allows the use of both GPS and GLONASS TEC. Information on the new GLONASS R&D network, future research possibilities and collaborations are also provided. Keywords: TEC modeling, Neural Network, GLONASS data network. 1. INTRODUCTION It is well know that an electromagnetic wave is not affected when traveling through a static electric or magnetic fields in a linear medium such as a vacuum. However, when traveling in a dispersive medium, such as the atmosphere, different as- pects cause variation on the propagation speed, polarization and signal power (Borre and Strang (2012); Hoque and Jakowski (2015)). In the context of GNSS L band signals, the ionospheric refraction, proportional to the TEC value, introduces most of the delay that may cause range errors in the positioning system of up to 100 m (Jakowski et al. (2011)). This fact is even worst in the equatorial and low-latitude regions, since in these areas the TEC presents strong temporal and spatial variation due to mainly three different dynamic process: the equatorial ion- ization anomaly, post-sunset plasma enhancement and evening plasma bubbles (Takahashi et al. (2014)). Some benefits of knowing the correct TEC value within a good spatial resolution is related with the improvement of accuracy 1 At the time of this work Claudia Paparini was working at ICTP in Tri- este. Now she works at ESSP-SAS (European Satellite System Provider) in Toulouse. in global positioning systems, especially for single frequency users, as well as a better understanding of the different param- eters that affect it, such as solar and magnetic activities, and the ability for monitoring and forecast space weather events (Denardini et al. (2016b); Denardini et al. (2016a)). In this con- text, the use of NNs have provided good results in applications for regional TEC modelling being capable of recovering TEC values with good performance (Leandro and Santos (2007); Habarulema et al. (2009)). This fact is related with the abilities of a NN to learn, generalize and adapt to different patterns of input/output sets with nonlinear behavior (Haykin (1999)). In this framework, this work aims to use NNs to estimate TEC behavior based on GNSS measurements in the Brazilian equatorial and low latitude sectors. The idea is based on the work of Leandro and Santos (2007), but considering different activation functions for the NNs, more input parameters similar to the approach presented in Habarulema et al. (2009) and using data from the new GLONASS R&D network. Specifically, the input parameters are the latitude, longitude, day of year (doy), time of the day, Kp-index, and the solar radio flux at 10.7 cm, and the output the vTEC e . The vTEC used for training the NN