Spatiotemporal analysis of extreme rainfall events using an object-based approach

This research presents a spatiotemporal object-based method for the identification and evaluation of extreme rainfall events at catchment scale using satellite products. Extreme rainfall events are defined as connected rainfall fields in space and time, with a relatively high statistical volume of water. This method provides a complete diagnosis of extreme events identifying the structure of rainfall events. A multidimensional connected-component labeling algorithm is used to associate (cluster) convective rainfall events, and with this create a 4D rainfall object. The main characteristics of rainfall extreme events are determined by rainfall objects extracted from the clusters, such as volume, area, duration, orientation, and speed, among others. The methodology is applied to the subtropical catchment of the Tiete River for the identification and classification of different types of extreme events during monsoon seasons and the verification of a near-real time satellite-based product. Results show the importance of spatial and temporal structures in the comparison of products and real-life events. This method also provides insights to better understand the rainfall concentration (location) of events and their dynamics over catchments.