 |
Center for Nonlinear Studies |
Understanding the fundamental causes of regional variability in extreme precipitation events is one of the most important insights needed in water resource planning to provide society with a reliable source of usable water and to minimize the impacts of extreme precipitation events. The statistics of populations of extreme precipitation events can provide very specific and informative information that will be vital to water resource planning by providing enhanced information regarding the similarities of the precipitation events, as well as the their characteristics. We demonstrate how viewing extreme precipitation events as “objects”, calculating simple physical and environmental based characteristics for each object, and applying machine learning algorithms can provide new information and knowledge on the variability of extreme precipitation events. To do this, we apply an object-oriented connectivity algorithm developed at the Center for Hydrometeorology and Remote Sensing (CHRS) at the University of California Irvine, which segments gridded near global satellite precipitation data into 4D objects (longitude, latitude, time and intensity) and calculates characteristics for each object. These characteristics are stored in an “Nxd” or “design” matrix, where N is the number of observations and d is the number of dimensions. We then apply an unsupervised machine-learning algorithm, the Finite Mixture Model, to the object characteristic dataset. The results of the mixture models are analyzed and discussed in terms of the fundamental factors that impact the event population characteristics and demonstrate how to detect changes in precipitation characteristics that correspond to climate phenomena.
Host: Brian Munsky