Meteorological models depend on high quality meteorological measurements. However, the presence of missing measurements presents a major challenge in handling weather data. The aim of this thesis was to present and test various methods for imputing missing values in cases where the entire weather station experiences an outage. In such scenarios, there are no internal station measurements available that would help us impute the missing data, which forces us to rely on data from surrounding weather stations. We achieved the best results using Random Forest which achieved mean absolute error of 0.6 °C with a standard deviation of 0.1 °C. We also demonstrated that the accuracy of imputations was dependent on the meteorological characteristics of the location of the meteorological station, and the weather patterns that vary on a monthly basis.
|