Exploring biogeographical provinces with machine learning and Argo float data

Abstract:

Previous studies have identified biogeographical provinces, or regions with unique and coherent characteristics, using a range of techniques and datasets. The new availability of high-resolution vertical profiles of chlorophyll-a fluorescence, oxygen, and nitrate from profiling floats open up the possibility of identifying such regions using information about the vertical distribution of chlorophyll and nutrients as well as the surface concentrations or modeled data. Machine learning provides an opportunity to explore these data and identify biogeographical provinces without defining distinguishing characteristics a priori. Preliminary work completed applying a Gaussian Mixture Model (GMM) to vertical profiles of chlorophyll from biogeochemical profiling floats showed promise in identifying the dominant regional patterns and identifying profile clusters. Subtropical gyres in all basins showed similar patterns of vertical chlorophyll distribution, in terms of the location and magnitude of a deep chlorophyll maximum, resulting in a single cluster describing the majority of those data. Similar regional and seasonal patterns can be identified from the dominant profile clusters in other regions. There are many potential applications of these clusters, including input to Ocean General Circulation Models that have previously relied either on satellite data or coupled biogeochemical models for information about the expected chlorophyll distribution. Further work including temperature and salinity profiles and vertical nutrient distribution may allow for even more detailed investigation of biogeographical provinces that considers the full vertical behavior.