Improving the simultaneous application of the DSN-PC and NOAA GFS datasets

Our surface-based sensor network, called Distributed Sensor Network for Prediction Calculations (DSN-PC) obviously has limitations in terms of ver-tical atmospheric data. While efforts are being made to approximate these upper-air parameters from surface-level, as a first step it was necessary to test the network’s capability of making distributed computations by applying a hybrid approach. We accessed public databases like NOAA Global Forecast System (GFS) and the initial values for the 2-dimensional computational grid were produced by using both DSN-PC measurements and NOAA GFS data for each grid point. However, though the latter consists of assim-ilated and initialized (smoothed) data the stations of the DSN-PC network provide raw measurements which can cause numerical instability due to measurement errors or local weather phenomena. Previously we simultaneously interpolated both DSN-PC and GFS data. As a step forward, we wanted for our network to have a more significant role in the production of the initial values. Therefore it was necessary to apply 2D smoothing algorithms on the initial conditions. We found significant difference regarding numerical stability between calculating with raw and smoothed initial data. Applying the smoothing algorithms greatly improved the prediction reliability compared to the cases when raw data were used. The size of the grid portion used for smoothing has a significant impact on the goodness of the forecasts and it’s worth further investigation. We could verify the viability of direct integration of DSN-PC data since it provided forecast errors similar to the previous approach. In this paper we present one simple method for smoothing our initial data and the results of the weather prediction calculations.