A Rainfall Nowcasting Algorithm Based on Hydro-Estimator Data

A nowcasting algorithm based on satellite (Hydro-Estimator) data is introduced in this work. The algorithm predicts the spatial and temporal distribution of rainfall rate. The most likely raining areas are predicted and then the expected intensity of rainfall rate in each rain pixel is forecasted. The Otsu's method based on discriminant analysis was adopted to identify contiguous rain pixels. A tracking algorithm is introduced in this work and maximizes persistent pixels and minimizes the size-difference of a single rainfall cell observed at two instant of times. The cloud motion vector is used to predict the most likely rainfall areas. The predicted rainfall areas are divided into smaller regions to best represent the spatial and temporal rainfall variability. An exponential regression model in time and spatial domain is developed for each region to predict the growth and decay of rainfall intensity. It is assumed that the potential predictors are the two previous observations of reflectivity located in a neighborhood region with center on a predicted pixel. An iterative forward selection algorithm is used to eliminate irrelevant pixels and determine the best predictors for each region; and finally, the intensity of rainfall rate is forecasted at pixel level. Nine storms that occurred in Puerto Rico during 2003-2013 were used to validate the proposed algorithm. The Eulerian and Lagrangian algorithms were compared with our algorithm and results show that the performances of Eulerian and Lagrangian algorithms were slightly better than our algorithm.