| 摘要: | 大多數研究都是根據在標準網格位置或沿特定橫斷面測量的定量值來調查黑潮的。然而，本研究
將黑潮與其周邊水域區分開來，並利用衛星資料分析了 1993 年至 2020 年從單一網格到流域規模
的表面流趨勢。這種獨特的方法結合了分割和影像處理。當方差擬合優度大於 0.8 時，它比傳統
的閾值/直方圖技術更能有效地描繪出不均勻的黑潮範圍，對海洋學和氣候研究有所貢獻。
從劃定的黑潮來看，整體的流速呈現減弱的趨勢。為了分析其變化趨勢是均勻的還是隨區域變化
的，本研究根據水文特徵將已劃定的黑潮劃分為 5 個區域。使用三種統計方法、兩種頻率方法和
一種貝葉斯 (bayesian) 方法，以區域平均值和逐個像素為基礎分析其趨勢。值得注意的是呂宋島
和台灣沿岸的風力減弱。相反，吐噶喇海峽 (Tokara Strait) 至名古屋及名古屋-150°E 方向的風力
增強。有趣的是，黑潮南流-II 處存在著由減弱到加強的過渡區。在計算效率方面，檢測斷點和估
計趨勢段（DBEST）和小波轉換（WT）優於其他方法。集合經驗模態分解 (EEMD) 和季節性趨
勢 Loess (STL) 具有相似的效率水準。加性季節和趨勢斷點 (BFAST) 是為線性趨勢而設計的，而
EEMD 適用於一般趨勢。突變、季節性變化和趨勢的貝葉斯估計器 (BEAST) 結合了多種模型以
減少過度擬合，並從 GlobCurrent 和漂移數據中產生高度相關的趨勢。本研究探討結合空間濾波
演算法擷取海面流的適用性。因此，對光柵濾波器進行了 80-13,505 張每日影像的測試，以檢測
每週、季節和氣候尺度上的黑潮 (KC)。所選的柵格濾波器包括卷積、拉普拉斯、北梯度、銳化、
最小/最大、直方圖均衡化、標準差和自然斷點。此外，還採用了海面流、海面溫度（SST）、海
面高度（SSH）等常規資料集，以及總熱通量、表面密度（SSD）和鹽度（SSS）等非常規資料。
此外，由於很少有研究表明葉綠素-ff 可以作為夏季 SST 的替代物來提取 KC，因此納入了有爭議
的水色數據。有趣的是，只有結合演算法，過濾器的性能才是統一的，並且在季節和氣候尺度上
都能蓬勃發展. 與使用單獨的濾鏡和指定值譜來識別黑潮特徵的典型場景相反，本研究採用了不
同的方法。它研究了根據 SST、SSH、總熱通量、SSS、SSD、葉綠素-ff 和海面流計算出的黑潮中
心線之間的相關性。從台灣東北部經吐噶喇海峽到日本南部，黑潮中心線各段觀測到的熱通量、
葉綠素-ff 和表層鹽度資料有偏差，這很有啟發性。這種相關性凸顯了各種海洋數據的相互聯繫，
有助於更深入地了解黑潮系統。
;Most studies investigated Kuroshio on quantitative values measured at standard grid locations or
along a transect. However, this research delineates Kuroshio from its surrounding waters, and
surface current trends are analyzed from 1993 to 2020, extending from a single location to a basin
scale. This unique approach combines segmentation and image processing. When integrated with
the goodness of variance fit greater than 0.8, it proves more efficacious than conventional threshold/
histogram techniques in delineating patchy Kuroshio extent contributing to oceanography and
climate studies.
The generalized trend from delineated Kuroshio exhibited a systemwide weakening. To analyze
whether the trend is uniform or varies with Latitude, delineated Kuroshio is divided into five
sections based on the hydrological characteristics. The trend was analyzed using three statistical,
two frequency, and one Bayesian approach on a regional mean and a pixel-by-pixel basis. Weakening
in Luzon and along Taiwan is worth noting.
In contrast, a strengthening in the Tokara Strait to Nagoya and Nagoya-150°E. Interestingly, the
transition zone from weakening to strengthening in the Kuroshio South-II. Regarding computational efficiency, Detecting Breakpoints and Estimating Segments in Trend (DBEST) and Wavelet
Transform (WT) outperformed the other methods. Ensemble Empirical Mode Decomposition
(EEMD) and Seasonal Trend Loess (STL) have similar levels of efficiency. Breaks for Additive
Season and Trend (BFAST) is designed for a linear trend, while EEMD is suitable for a general
trend. The Bayesian Estimator of Abrupt, Seasonal Change and Trend (BEAST) combines multiple models to reduce overfitting and produces highly correlated trends from GlobCurrent and
drifter data.
This research also investigates the applicability of combining spatial filter＇s algorithm to extract
surface ocean current. Accordingly, the raster filters were tested on 80–13,505 daily images to
detect Kuroshio Current (KC) on weekly, seasonal, and climatological scales. The selected raster
filters are convolution, Laplacian, north gradient, sharpening, min/max, histogram equalization,
standard deviation, and natural break. In addition, conventional data sets of sea surface currents,
sea surface temperature (SST), sea surface height (SSH), and non-conventional data such as total
heat flux, surface density (SSD), and salinity (SSS) were employed. Moreover, controversial data on ocean color are included because very few studies revealed that chlorophyll-α is a proxy for
SST in the summer to extract KC. Interestingly, the performance of filters is uniform and thriving
for seasonal and on a climatological scale only by combining the algorithms.
Contrary to the typical scenario of identifying Kuroshio signatures using an individual filter and
by designating a value spectrum, this research takes a different approach. It investigates the correlation between Kuroshio＇s centerlines computed from SST, SSH, total heat flux, SSS, SSD,
chlorophyll-α, and sea surface currents. The deviations observed in the various segments of
Kuroshio＇s centerline extracted from heat flux, chlorophyll-α, and SSS flowing across Tokara
Strait from northeast Taiwan to south Japan are enlightening. This correlation highlights the interconnectedness of various oceanographic data, providing a deeper understanding of the Kuroshio
system. |
