Rain-Area Identification Using TRMM/TMI Data by Data Mining Approach

Abstract

Rain-area identification distinguishes between rainy and non-rainy areas, which is the first step in some critical real-world problems, such as rain intensity identification and rain-rate estimation. We develop a data mining approach for oceanic rain-area identification during typhoon season, using microwave data from the Tropical Rainfall Measuring Mission (TRMM) satellite. Three schemes tailored for the problem are developed, namely (1) association rule analysis for uncovering the set of potential attributes relevant to the problem, (2) three-phase outlier removal for cleaning data and (3) the neural committee classifier (NCC) for achieving more accurate results. We created classification models from 1998-2004 TRMM Microwave Imager (TRMM-TMI) satellite data and used Automatic Rainfall and Meteorological Telemetry System (ARMTS) rain gauge data measurements to evaluate the model. Experimental results show that our approach achieves high accuracy for the rain-area identification problem. The classification accuracy of our approach, 96%, outperforms the 78.6%, 77.3%, 83.3% obtained by the scattering index, threshold check, and rain flag methods, respectively.