The K-means rainfall nowcasting algorithm developed at the National Severe Storms Laboratory (NSSL) has been used for severe weather analysis, warnings and forecasting. This algorithm was selected by the GOES-R Algorithm Working Group (AWG) Hydrology Algorithm Team for producing nowcasts of 0-3 hour rainfall potential from current and previous GOES-R Rainfall Rate imagery. However, direct application of the current K-Means algorithm to full-disk satellite-based rainfall estimates has been challenging because the domain is much larger and contains various rain types at different scales.

This work outlines several modifications to the K-Means algorithm and their impacts. For instance, instead of weighting all precipitation features equally when creating an objectively analyzed motion field between two images, features were weighted based on the common area between the same features in two successive images in order to reduce the effects of feature changes on the apparent motion vector. In addition, a new approach to estimating feature growth and decay has been developed based on changes in the size and intensity of each rainfall feature between successive images. Preliminary results show enhanced skill with respect to the original K-Means algorithm.