SST clustering for winter precipitation prediction in southeast of Iran: Comparison between modified K-means and genetic algorithm-based clustering methods

In this paper, two innovative approaches for temporal clustering of sea surface temperature (SST) data using genetic algorithm (GA) and K-means clustering methods are introduced. In both methods, new approaches have been developed to consider the relationship between variations of the variable being clustered (SST in this study) and another climatic variable (precipitation in this study) in selection of clusters. In the case study, these models are used for clustering SST in selected geographical zones in Gulf of Oman, Arabian Sea, and the northern part of Indian Ocean with respect to the precipitation variations in the selected rain gauges in Sistan and Baluchestan Province in southeast of Iran. For this purpose, fitness function of the GA model and Euclidean distance in the modified K-means method have been formulated to minimize the variance of the precipitation data associated with each selected cluster. Application of these modified clustering methods in the case study has resulted in temporal classification of SST data which also represent below and above normal precipitation seasons in the selected rain gauges. The results of the two clustering techniques have been used for development of seasonal precipitation prediction guidelines based on the SST variations in the selected geographical zones. The results have also shown that these models can be effectively used for prediction of below and above normal precipitation seasons in the study area. Comparison between the results of this study with the official forecasts of the Islamic Republic of Iran Meteorological Organization (IRIMET) has shown significant improvements.