Semi-Automatic Classification of Weather Maps

In this paper we analyze weather maps to distinguish between the three main circulation forms which are essential factors for weather composition and are fundamental for weather forecasters. We propose a set of features specifically tailored for the classification of these circulation forms in General Weather Situations and use these to train a support vector machine for classification. Additionally, we propose a semi-automatic algorithm to extract the necessary data directly from the weather maps itself. This enables us to also analyze historic map material for which the original data is not available anymore. In order to reconstruct the weather data, we extract and analyze the isolines from the weather maps based on color and line thickness as well as symbolic and numerical features using template matching techniques. We reconstruct the dense wind alignment field and air velocity field from these sparse data and extract expressive feature vectors to classify the presented main circulation forms. Our algorithm shows an overall classification success rate of 61% for the three main circulation forms zonal, meridional and mixed.