Ocean Circulation in the Western Gulf of Mexico Using Self‐Organizing Maps

Abstract

The ocean circulation in the western Gulf of Mexico is influenced by Loop Current eddies (LCEs), which affect the thermohaline structure and distribution of environmental variables as well as the design and management of offshore structures and maritime activities. In this study, we assess the ocean circulation and its variability by applying dual self‐organizing maps (SOMs) to a 20‐year data set (1993–2012) of salinity and currents at a depth of 200 m from a Hybrid Coordinate Ocean Model. The geospatial variability is characterized via a 7 × 7 SOM, which is further grouped into seven patterns, while the temporal variability is characterized as six distinct regions using a 2 × 3 SOM. The influence of LCEs on the upper slope occurs along the 200‐m isobath, a zone under constant pressure by strong currents from impacting eddies. The temporal pattern variability also reveals a clear zone of LCE impact on the continental slope (between latitudes 22°N and 25°N) and a dominant offshore transport pattern around latitudes 25°N to 26°N. A rectified wavelet power spectral analysis reveals a frequency of LCE impact oscillating around four to five months with an estimated ring lifespan of 1 year.

Earth Observation and Machine Learning Reveal the Dynamics of Productive Upwelling Regimes on the Agulhas Bank

“…Although monthly climatology (long-term statistical mean) can be used to describe the overall spatial features found in an oceanic region from satellite data, it comes with a smoothing effect which can mask important variability which is captured here by the SOM analysis. Our SOM approach consisted of a 1-step classification which is recommended (Liu et al, 2006a) and widely used (e.g., Meza-Padilla et al, 2019) for oceanographic applications. It can be argued that taking a 2-step classification using a larger-sized SOM then reducing the number of clusters using a HAC following Ward's criterion (Ward's method) (e.g., Jouini et al, 2016) can allow to further explore the topological organization of the data.…”

“…QE indicates the average distance between data vectors and the BMU, while TE represents the data vectors' percentage where the first and second BMU are not neighbouring neurons (Liu et al, 2006a). In other words, QE measures how accurate is the representation of the input data while TE measures the topology preservation (Hernandezcarrasco and Orfila, 2018;Meza-Padilla et al, 2019). The lower QE and TE, the better is the mapping quality (Liu et al, 2006a).…”

“…The lower QE and TE, the better is the mapping quality (Liu et al, 2006a). However, QE and TE decrease with increasing and decreasing map sizes, respectively, leading to a careful trade-off between mapping accuracy and preservation of the topology (Mihanovicé t al., 2011;Meza-Padilla et al, 2019). The SOM-Toolbox returns also the BMU timeseries using the function som_bmus, which allows the identification of how the dominant neurons (patterns) vary/occur over time.…”

“…The lower QE and TE, the better is the mapping quality (Liu et al, 2006a). Although QE decreases with the number of nodes, TE increases and the interpretation of the SOM patterns becomes more complex (e.g., Meza-Padilla et al, 2019). The elbow criterion helps in determining an "optimum" (best guess) number of clusters by finding the limit at which the addition of clusters fails to add a significant amount of information (Schuenemann et al, 2009).…”

Inversion of Phytoplankton Pigment Vertical Profiles from Satellite Data Using Machine Learning

“…They have the advantage of having high interpretability and make it possible to find relationships between the distribution of data on the map and the main explanatory variables. This is particularly useful in the case of complex and noisy data-as it is the case with climatology/oceanography data where they have been used in a large variety of studies [50,51].…”