Mapping Ocean Salinity Data Using Gaussian Mixture Modeling

Abstract Intensification of the water cycle due to climate change has a significant impact on human societies and ecosystems. Globally, 77% of precipitation and 85% of evaporation occur over the oceans. Precipitation, river runoff, and ice melt dilute and evaporation and ice formation concentrate salinity in seawater, making ocean salinity a crucial indicator for quantifying water cycle change. However, ocean salinity is relatively underobserved, compared to ocean temperature, making long-term variations in salinity challenging to quantify. This study focuses on the development of a method to create two-dimensional maps of ocean salinity and its trends on pressure surfaces from sparse observations. To achieve this, we employ an unsupervised classification technique called Gaussian mixture modeling (GMM). GMM is used to identify coherent regions where temperature and salinity are tightly related at constant pressure. Within each region, sparse observations of salinity are aggregated and means and trends are quantified. The method is assessed using temperature and salinity data from the climate model ACCESS-CM2, subsampled to mimic historical observational programs. We demonstrate that the method has skill in predicting historical salinity trends from 1970 to 2014. In the South Atlantic, at a pressure of 539 dbar, the root-mean-square error of salinity and of the salinity linear trend is 0.040 g kg −1 and 2.1 10 −3 g kg −1 yr −1 , respectively. Applying this methodology at all depth levels and across all oceans could help fill in missing salinity observations and thus improve our understanding of the intensification of the global water cycle in response to climate change. Significance Statement Understanding ocean salinity is crucial for tracking changes in Earth’s water cycle and climate. However, collecting accurate salinity data has been challenging due to limited observations, especially in certain regions. This study used advanced techniques to address this issue. By grouping similar ocean regions using a method called Gaussian mixture modeling, we are able to predict missing salinity data and fill gaps in historical salinity records from 1970 to 2014. The results show that this approach effectively estimates past salinity data. The method can now be used to improve our understanding of long-term changes in the ocean’s salt content and help us better study how the water cycle and climate are evolving over time.