Nearest Neighbor Search for Diagnosing Rain/Non-Rain Discrimination

Abstract

This study examines the rain occurrence by the passive microwave imagery during typhoons. The dataset consists of 53 typhoons affecting the watershed over 2001-2008. This study employs nearest neighbor search (NNS) classifier which is often used for diagnosing forecast problems. The multilayer perceptron (MLP) and logistic regressions (LR) are selected as the benchmarks. The results show that for the rain/non-rain discrimination, the best performing classifier is NNS according to the AUC measures. The results show that the use of NNS can effectively improve the AUC measures for diagnosing rain occurrence. Overall, the use of NNS is a relatively effective algorithm comparing to other classifiers.