Machine Learning Based Statistical Characterization of a Turbulence Dissipation Rate Array: A Revisitation

Comprehension of energy dissipation rates in coastal waters is crucial to understanding such environmental fluid dynamical processes as coastal sediment transport, pollution dispersal, and heat and mass exchange across the air-sea interface.Traditional methods for understanding sub-surface turbulent velocity energetics have focused on the sampling of the turbulent velocity field using acoustic Doppler velocimeters (ADVs) and analyses directed towards the quantification of velocity array variability modeled as statistically independent frequency modes experiencing weak spectral energy transfer.These statistical methods while enlightening have not provided comprehensive insight into spatial ADV array structural dynamics especially in the area of probe system characterization.Understanding the structure of an ADV array as an information system is addressed and accomplished here via the analytical revisitation of three-dimensional velocity data obtained from a four-probe array deployed during the 2001-2003 Coupled Boundary Layers and Air-Sea Transfer (CBLAST) Low Program.The research objective was to show how machine learning algorithms can provide an alternative perspective for the characterization of coastal turbulent velocity information with respect to three important areasmultivariate turbulent kinetic energy level segmentation, nonlinear turbulent velocity modal analysis, and statistical modelling of probe relationships.The Coupled Boundary Layer and Air-Sea Transfer Low program (CBLAST-LOW) was a field experiment whose goal was to improve understanding of the parameterization of the marine boundary layer and air-sea interaction processes during low winds.During the experiment four ADVs were mounted on a submerged steel beam in a linear array approximately 3.5 meters below the water surface in the Martha's Vineyard Sound.Turbulent kinetic energy (tke) dissipation rates were estimated from power spectra of the vertical velocity component and through the use of the frozen turbulence field hypothesis from data acquired from September 22-23, 2003 [1].Each dissipation rate value was estimated from 20 minutes of data at 20 minute intervals over a time period characterized by two high and two low tides per day.Dissipation rates were elevated in general due to energy contamination by surface wave fluctuations and possessed local maxima due to the semidiurnal tidal component which inundated the sampling region.Gaussian mixture modelling (GMM) [2] using two spatially distant probes, probes 2 and 4, showed a linear proportionality covariance structure at high dissipation rates.A second cluster mode exists at low dissipation rates where low values for probe 2 were associated with a large spread of dissipation rate values at probe 4.This is thought to be due to a turbulence wake effect where a large uniform turbulence system sat on top of all the probes at high tide, causing linearly proportional dissipation rates for the two probes.At low tide, dissipation rates at probe 2 were extremely low but a residual medium local turbulence level still existed at probe 4.Generative topographic mapping (GTM) [3] is a non-linear latent variable model which furnishes a two-dimensional organized representation of noisy, nonlinear dissipation rate data exhibiting data clusters using latent variables constructed under the assumption of an underlying manifold data structure.Latent space is filled with a regular square array of feature nodes where the four-dimensional data space points lying on a manifold are images of the latent space under a local but nonlinear kernel function mapping.Latent space exhibits a segmentation of data points above and below a root mean

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.