Prediction of zenith tropospheric delay by multi-layer perceptron

The aim of the present study was to use the artificial neural network approach and specifically the multi-layer perceptron algorithm in order to predict total zenith tropospheric delay (ZTD) for various time spans of 1, 3 and 6 hours. The test data was ZTD values derived from the analysis centers of the EUREF Permanent tracking Network. The prediction process was applied to six EUREF permanent GPS stations for using period data of 2006 and 2007. The results obtained show an agreement at the order of few centimetres (2-3 cm) with those derived from EPN. Comparisons were also made with ZTD values calculated by other methods like the radiosonde observations and Saastamoinen model using ground measurements in order to confirm the final results and the feasibility of the neural network methodology.