In this paper, three different methodologies are employed for the prediction of atmospheric attenuation for the performance evaluation of High Throughput Satellite Communication systems. The first one is based on numerical weather predictions and in particular the ECMWF forecasts that uses high resolution deterministic forecast and the probabilistic forecasts the perturbated products. The second methodology is based machine learning algorithms, which are advanced statistical methods. The two algorithms tested in this study are the random forest and the gradient boosting, both based on regression trees. Finally, the last method that is employed is the recurrent neural networks and in particular the Long Short Term Memory. These neural networks are used for the prediction of time series using memory blocks. All the algorithms are tested using data from the ALPHASAT experiment at Chilbolton and Chilton, UK. The obtained results are very encouraging.