Cluster analysis on spatial-temporal patterns of SST and SSS: enhancing the explainability of K-means clustering results

In this paper, we present seasonal and interannual differences of sea surface temperature (SST) and sea surface salinity (SSS) in different ocean regions, using cluster analysis in model data. This is important because thermohaline features reveal inherent ocean patterns and is also related to atmosphere. The volume of SST and SSS data is massive, which contributes to more obvious characteristics after clustering. The goal of our research is to explore thermohaline patterns of oceans both spatially and temporally. Specially, it first classifies single ocean area into diverse groups based on SST and SSS characteristics and then compares between different regions, enhancing the explainability of cluster analysis by introducing several influencing factors. We collect 1951-2014 SST and SSS data from the NASA GISS CMIP6 model and apply K-means clustering based on python to analyze SST and SSS datasets. In our research, we explore the possibility of optimizing parameters of K-means clustering, which will potentially reduce computational cost and achieve reproducibility of coding experiments. Moreover, we introduce three influencing factors: precipitation, evaporation and heat flux, to have a comprehensive view of regional changes in SST and SSS both seasonally and interannually. The results of our research show West Pacific Ocean is divided into three seasonal patterns and a transfer towards warmer and saltier seawater is seen in both this region and the North Atlantic Ocean. Future research might include more factors such as currents, rivers and ice melting in the analysis, rather than simply consider rainfall flux, water evaporation flux and heat flux.