Modern weather prediction is based on the data from the networks of automatic meteorological stations combined with the observations from meteorological satellites. When the measurement series are broken, the accuracy of a weather forecast is lowered. The research was aimed to develop a new technology of supplementing by means of artificial neural networks the incomplete series of air temperature measurements made by automatic meteorological stations. The temperature data were obtained from hourly measurements performed in 1984 at the Wrocław Swojec meteorological station located in Lower Silesia. Three perceptrons with a single hidden layer were developed and applied to complete 3-, 6- and 12-hour registration breaks. The results showed that neural networks may provide a useful tool for supplementing broken air temperature measurement series.

103-111