Abstract: Analysis of social media and traditional media provides significant information to first responders in times of natural disasters. Sentiment analysis, particularly of social media originating from the affected population, forms an integral part of multifaceted media analysis. The current paper extends an existing methodology to the domain of natural disasters, broadens the support of multiple languages and introduces a new manner of classification. The performance of the approach is evaluated on a recently collected dataset manually annotated by three human annotators as a reference. The experiments show a high agreement rate between the approach taken and the annotators. Furthermore, the paper presents the initial application of the resulting technology and models to sentiment analysis of social media data in German, covering data collected during the Central European floods of 2013.