Patterns of the loop current system and regions of sea surface height variability in the eastern Gulf of Mexico revealed by the self‐organizing maps

Abstract

The Self‐Organizing Map (SOM), an unsupervised learning neural network, is employed to extract patterns evinced by the Loop Current (LC) system and to identify regions of sea surface height (SSH) variability in the eastern Gulf of Mexico (GoM) from 23 years (1993–2015) of altimetry data. Spatial patterns are characterized as different LC extensions and different stages in the process of LC eddy shedding. The temporal evolutions and the frequency of occurrences of these patterns are obtained, and the typical trajectories of the LC system progression on the SOM grid are investigated. For an elongated, northwest‐extended, or west‐positioned LC, it is common for the LC anticyclonic eddy (LCE) to separate and propagate into the western GoM, while an initially separated LCE in close proximity to the west Florida continental slope often reattaches to the LC and develops into an elongated LC, or reduces intensity locally before moving westward as a smaller eddy. Regions of differing SSH variations are also identified using the joint SOM‐wavelet analysis. Along the general axis of the LC, SSH exhibits strong variability on time scales of 3 months to 2 years, also with energetic intraseasonal variations, which is consistent with the joint Empirical Orthogonal Function (EOF)‐wavelet analysis. In the more peripheral regions, the SSH has a dominant seasonal variation that also projects across the coastal ocean. The SOM, when applied to both space and time domains of the same data, provides a powerful tool for diagnosing ocean processes from such different perspectives.

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

“…On the other hand, a larger map size will provide more details on different patterns but may be difficult to visualize, which is the case in Weisberg and Liu (), where they had to show 50 patterns of the LC evolution in two separate figures. For SOM applications in the eastern GoM, Liu et al (Liu et al, ; Liu, Weisberg, Vignudelli, & Mitchum, ) used 16 and 20 patterns. For our purpose, we employ a 7 × 7 map size (49 patterns) for the spatial analysis and a 2 × 3 map size (6 patterns) for the temporal variability.…”

“…As the LC passes through the Yucatan Channel into the GoM, it stretches to the north and turns to outflow through the Florida Strait, where it becomes the Gulf Stream that travels along the northwestern Atlantic Ocean (Oey et al, ). During this process, the LC sheds eddies on an irregular basis (Lugo‐Fernandez, ) and produces Loop Current eddies (LCEs), which are anticyclonic rings that drift westward and influence the entire GoM (e.g., Alvera‐Azcárate et al, ; Liu et al, ; Oey, ; Weisberg & Liu, ). Elliot () describes LCEs as warm salty bodies (Caribbean water properties) with an estimated ring lifespan of one year traveling to the west at a mean speed of 2.1 km/day.…”

“…According to Liu and Weisberg (2005), SOM patterns are more accurate than the EOF leading modes, and they found asymmetric features between upwelling and downwelling current patterns revealed by SOMs that were not detected by EOFs. Furthermore, Liu et al (2006) showed how a SOM extracted complex patterns from multiple sets while an EOF failed to do so.…”

Temporal evolution of near‐surface chlorophyll over cyclonic eddy lifecycles in the southeastern Pacific

“…A MATLAB Toolbox of the SOM is freely provided by the Laboratory of Information and Computer Science in the Helsinki University of Technology (http://www.cis.hut.fi/somtoolbox/). Refer to Liu et al [2016] for details about the dual (both spatial and temporal) SOM analysis.…”

Enhanced Compositional Mapping through Integrated Full-Range Spectral Analysis

“…Evaluation of alternative approaches to clustering and classification for the data integration may reveal benefits for spatially limited compositions, potentially warranting any added complexity. For example, Self-Organizing Maps [68], already widely used in the geosciences e.g., [69,70], may overcome some of the limitations of ISODATA cluster analysis [71,72] and have shown greater success in detecting unique but spatially small data clusters, e.g., [73].…”