RAINFALL ESTIMATION OVER PUERTO RICO USING A CLASSIFICATION SYSTEM AND ANN

摘要

Precise remote detection and estimation of rainfall has become critical for protecting human lives and infrastructure. Researchers have developed diverse algorithms for deriving rainfall rates from instruments on geostationary satellite platforms such as the Geostationary Operational Environmental Satellites (GOES) due to its relatively high spatial and temporal resolution and uniform spatial coverage.rnValidations of the operational NOAA/NESDIS Hydro-Estimator (HE) algorithm conducted over Puerto Rico (PR) at a satellite pixel and island-wide scale showed that the algorithm has a low probability of detection. The poor performance of the HE over PR may in part be due to the fact that the algorithm was designed to operate over the continental United States and conditions over PR are considerably different. In order to achieve greater accuracy of detection and estimation over PR, a new rainfall algorithm is under development. The algorithm utilizes data from multiple bands of GOES-12 to extract diverse features from clouds (e.g., Brightness Temperature, Visual Reflectance, and Albedo). These features are utilized to perform a supervised classification of the image pixels into 4 previously defined classes. The characterized classes will only provide rainfall detection information. After the classification is completed, two artificial neural networks will be utilized to find a feature-rain rate relationship for each class. Preliminary results in terms of rainfall detection show that the algorithm's classification system has great potential for outperforming the HE over PR.