An Algorithm to Detect Rainy Cloud Pixels Based on GOES Visible and Near Infrared Reflectances

It has been shown that cloud microphysical parameters exhibit some potential skill in detecting and estimating rainfall for warm-top clouds (defined as warmer than 235 K in the 11 μm window).  The cloud microphysical parameters extracted from MODIS are related to GOES visible and near infrared reflectances.  The microphysical parameters studied are the cloud water path, droplet effective radius, and the optical thickness.  During the daytime the visible and near infrared reflectances were used to derive a detection algorithm.  The introduced detection algorithm is based on the calibration of a preselected set of potential rainy pixels.  A rainfall event includes rain and no-rain pixels; however, in most of the cases the areas are dominated by no-rain pixels, and consequently the algorithm may have more skill for detecting no-rain pixels than rain pixels.  The proposed algorithm is calibrated with pixels that are most likely to be rainy pixels and also including similar amount of no-rain pixels, with the purpose that the algorithm will easily detect the pixels that exhibit the microphysical properties associated with rain.  During the daytime the parameters associated with microphysical properties are the ones that exhibit low values of albedo (near infrared reflectance) and high values of visible reflectance.  The optimization problem consists of identifying the appropriate threshold values of albedo  and visible reflectance ( to determine the set of pixels that are more likely to be rain pixels.  Rain/no-rain regression equations are developed using albedo, visible reflectance, and brightness temperature from the GOES infrared channels as predictors.  Forward selection and multicollinearity algorithms are used to determine the parsimonious models with the best predictors that provide the optimal thresholds  and  to discriminate whether or not a given pixel is a rain or a no rain pixel.  To ensure that the algorithm has an appropriate detection skill the Heike Skill Score (HSS) was selected as the objective function.  In summary, a searching pattern optimization algorithm was used to determine the values of ,  and  such that the HSS was maximized.

Puerto Rico (PR), one of the Greater Antilles Islands, was used as a test bed since in PR warm rain is associated with weak tropical waves, and easterly winds from the eastern Atlantic bring sea breezes into the island and stimulate orographic rainfall over the mountains characterized by warm IR brightness temperatures.  The algorithm was calibrated using four storms--two with warm and two with cold clouds--and was validated with independent data set that includes two cold and one warm rain case.  A warm rain cloud is the one dominated by a large amount of warm rain pixels and a cold rainy cloud is dominated by cold rain pixels.  NEXRAD data were used for calibration and validation and results show that the proposed algorithm has some potential to improve the detection skill of the SCaMPR algorithm, which will be the operational GOES-R rainfall rate algorithm.  During the nighttime the microphysical parameters are potentially associated with brightness temperature differences and consequently an expansion of the proposed algorithm can be developed in the near future.