Spatial and temporal variability of floods in Indonesia based on governmental data, Twitter messages and paper reports

Indonesia, with its tropical and monsoonal climate, is exposed to heavy precipitation and enormous rainfall accumulation which results in weather-driven hazards, including extreme rainfall events and floods. There are several conventional sources of data to estimate potential of anomalously high precipitation in Indonesia, including rain gauge data, satellite data and meteorological reanalysis. Even though they allow assessment of precipitation variability, their usefulness is limited by biases and data gaps. Furthermore, assessment of a variability in precipitation patterns is not the same as identification of their adverse societal effects, such as floods. Due to the proliferation of social media, these conventional data sets can be supplemented with crowd-sourced information that can potentially provide longer-term, accurate records and cover a larger area. In this study, we demonstrated that Twitter is a useful source for flood detection and created a flood database. Twitter-based flood database is derived for subregions of major islands within Indonesia: Java, Sumatra, Borneo and Sulawesi, and validated against data from governmental reports and local paper articles. Results show that Twitter-based retrieval performs well in comparison with other sources, but only in regions characterized by sufficiently large pool of active users. Flood events and extreme rainfall events (defined using in-situ and satellite data) were compared in terms of their spatial and temporal distribution, as well as their meteorological drivers. In general, on each of the island, there is a seasonal cycle: a wet season during boreal winter, when the Southeast Asian monsoon provides an environment supportive of rain events, and a dry season during boreal summer. On intraseasonal scale, Madden-Julian Oscillation (MJO) creates the conditions favorable for weather extremes. MJO activity causes an increase in the local rainfall rate, with a significant increase in a chance of observing extreme precipitation during favorable MJO phase.