Storm Identification and Tracking for Hydrologic Modeling Using Hourly Accumulated NEXRAD Precipitation Data

Watershed responses to storm events are strongly affected by the spatial and temporal patterns of rainfall; that is, the spatial distribution of the precipitation intensity and its evolution over time. Although real storms are moving entities with non-uniform intensities in both space and time, hydrological applications often synthesize these attributes by assuming storms that are uniformly distributed and have variable intensity according to a pre-defined hyetograph shape. As one considers watersheds of greater size, the non-uniformity of rainfall becomes more important, because a storm may not cover the watershed's entire area and may not stay in the watershed for its full duration. In order to incorporate parameters such as storm area, propagation velocity and direction, and intensity distribution in the definition of synthetic storms, it is necessary to determine these storm characteristics from spatially distributed precipitation data. To date, most algorithms for identifying and tracking storms have been applied to short time-step radar reflectivity data (i.e., 15 minutes or less), where storm features are captured in an effectively synoptic manner. For the entire United States, however, the most reliable distributed precipitation data are the one-hour accumulated 4 km × 4 km gridded NEXRAD data of the U.S. National Weather Service (NWS) (NWS 2005. The one-hour aggregation level of the data, though, makes it more difficult to identify and track storms than when using sequences of synoptic radar reflectivity data, because storms can traverse over a number of NEXRAD cells and change size and shape appreciably between consecutive data maps. In this paper, we present a methodology to overcome the identification and tracking difficulties and to extract the characteristics of moving storms (e.g. size, propagation velocity and direction, and intensity distribution) from one-hour accumulated distributed rainfall data. The algorithm uses Gaussian Mixture Models (GMM) for storm identification and image processing for storm tracking. The method has been successfully applied to Brazos County in Texas using the 2003 Multi-sensor Precipitation Estimator (MPE) NEXRAD rainfall data.