A comparison of satellite-derived sea surface temperature fronts using two edge detection algorithms

Abstract

Satellite-derived sea surface temperature (SST) fronts provide a valuable resource for the study of oceanic fronts. Two edge detection algorithms designed specifically to detect fronts in satellite-derived SST fields are compared: the histogram-based algorithm of Cayula and Cornillon (1992, 1995) and the entropy-based algorithm of Shimada et al. (2005). The algorithms were applied to four months (July and August for both 1995 and 1996) of SST fields and the results are compared with SST data taken by the M.V. Oleander, a container ship that makes weekly transits between New York and Bermuda. There is no significant difference in front pixels found with the Cayula-Cornillon algorithm and those found in the in situ (Oleander) data. Furthermore, for strong fronts, with gradients greater than 0.2 K/km, the distribution of fronts found with the Shimada et al. algorithm is quite similar to that of fronts found with the Cayula-Cornillon algorithm. However, there are significant differences in the number of weak fronts found. This is seen clearly in waters south of the Gulf Stream where the gradient magnitude of fronts found is less than 0.1 K/km. In this region, the probability that the Shimada et al. algorithm detects a front rarely falls below 4% while the other two algorithms find fronts less than 1% of the time. These results raise the question of exactly what qualifies as an SST front, a classic problem in edge detection.

Diurnal Variability of Turbidity Fronts Observed by Geostationary Satellite Ocean Color Remote Sensing

“…The JSD method does not only diminish the influence of impulsive noise but also discerns the finer-scale curvilinear frontal features in coastal regions due to its independence from temporal or spatial variations of geophysical parameters [11]. Its effectiveness was already proven by Shimada et al [11], and Chang and Cornillon [56] through comparison with other traditional methods (e.g., gradient magnitude and histogram edge detection algorithms) using the SST images, and the results showed that the entropy-based algorithm had better performance, especially for detecting short and weak fronts. Here, the JSD was calculated from four JSD matrices that were estimated in each of two 5ˆ5 pixel TSM subwindows with four different directions (horizontal, vertical, and two diagonals, shown in [11]).…”

On the Seasonal Cycle of the Statistical Properties of Sea Surface Temperature

“…Singularity exponents characterize the scaling at small scales, are independent of the gradient magnitude, and measure the degree of continuity of the field. Moreover, singularity exponents have the advantage over other popular approaches for detecting fronts, such as those based on histograms or gradient filters like the Sobel filter (Chang & Cornillon, 2015; Kirches et al., 2016), that they can be easily connected to statistical quantities that are central to turbulence theories (Section 2). Indeed, the singularity spectrum, which gives the fractal dimension of those points with the same exponent, emerges as a fundamental property of the ocean providing the link between anomalous scaling and the intensity of fronts.…”

Lagrangian Reconstruction to Extract Small‐Scale Salinity Variability From SMAP Observations

“…In addition, it should be pointed out that previous studies in this region have applied different detection methods to identify fronts in satellite SST observations. To evaluate those methods, Ullman and Cornillon (2000) and Chang and Cornillon (2015) applied an along‐track gradient algorithm to detect fronts in Oleander temperature observations similar to the method used here to detect fronts in salinity. In future work, some of the methods already developed to detect fronts in SST could be adapted to detect fronts in the 2D reconstructed SSS fields.…”