Comparison of different methods for smoothing initial 2D data of the DSN-PC system's weather prediction algorithm

Our Distributed Sensor Network for Prediction Calculations (DSN-PC) is a surface-based observational and computational network which currently calculates the change of an upper-air atmospheric parameter. The number of currently installed stations is limited thence we include data from the NOAA GFS database to create the 2D field of initial values for the prediction calculations. This hybrid application leads to numerical instability because some grid points get values from DSN-PC stations and the others are set by NOAA GFS data. Previously we applied a smoothing method based on moving average calculations. As presented in this paper, we tried two other methods. Applying any of the algorithms improved the numerical stability and prediction reliability. The type of the algorithm and the window size had a significant impact on the goodness of the forecasts. Below we compare these new results along with the previous, moving average method and the raw, unsmoothed case.