| 摘要: | 沿岸地區被視為經濟及人文發展的重要地區，對於漁業、交通及觀光等發展都是相當重要的。此外，在生態方面沿岸的潮間帶地區更能用於調節氣候、過濾水質以及防止洪患等。然而，沿岸潮間帶地區具有易受到海水侵蝕及河川淤泥堆積的特性，地形變化快速，若能發展可以長期快速且大面積的監測方法，便能當作未來用來規劃發展該區域的重要資料來源。
本研究的研究區域為臺灣沿岸最大的沿岸沙洲-外傘頂洲，使用衛星影像來建置潮間帶的地形可達到快速且大範圍的要求。本研究使用Sentinel-2、Landsat 7/8自2014年至2017年的光學衛星影像，並計算每張影像的(改良的)常態差異水體指標，使用隨影像灰值分布不同而設定的閾值來辨識水體及陸域，再使用時間區間內的多張判識好水體的影像建置淹水機率圖。而後改善因為影像對應的潮汐高度採樣不均勻所造成的誤差，提出了Sampling Errors Reduction (SER)方法，使用DTU16全球海洋潮汐模型模擬每張影像對應的潮汐高度，並加以分析其分布情形後對淹水機率圖逐像素改進，再使用影像對應最高及最低潮汐高度對改善後淹水機率圖依比例賦予高程，最後潮間帶的數值高程模型即為高程位於平均較高高潮及平均較低低潮之間的區域。
本研究的結果與由單音束聲納所採集到的地形資料做驗證，我們所提出的方法建置的潮間帶數值高程模型，RMSD可以達到28.8公分的精度，經過SER方法有6.2%的改善幅度，且在影像張數越多的情況下可以有更好的精度及改善幅度，但改善幅度會受到影像對應的潮汐分布所影響
;Coastal zones serving as economic and cultural hubs, prominently feature activities such as fisheries, transportation, and tourism, underscoring their significance. Furthermore, ecologically, coastal intertidal zones play crucial roles in climate regulation, water filtration, and flood prevention. However, the coastal areas are easily affected by erosion and sediment deposition, coupled with the rapid morphological changes in intertidal zones, emphasizing the need to develop long-term, rapid, and large-scale monitoring methods. Such methods could serve as essential data sources for future planning and development of these regions.
The study area of this study is on the largest tidal flat along the coast of Taiwan, the Waisanding Tidal Flat. Traditionally, the topography of this shallow shoal has been surveyed using methods such as Single Beam Echo Sounder (SBES), airborne Light Detection and Ranging (LiDAR), or stereoscopic imagery captured by unmanned aerial vehicles. However, these methods are often time-consuming and resource-intensive.
Utilizing satellite imagery offers a solution to meet the requirements of rapid and large-scale terrain reconstruction in intertidal zones. Past studies have employed radar and optical imagery or both to enhance temporal resolution. This study adopts a methodology for automatic tidal flat reconstruction. It utilizes optical satellite imagery from Sentinel-2, Landsat 7/8, from 2014 to 2017. The (Modified) Normalized Difference Water Index ((M)NDWI) is calculated for each image, employing thresholding based on variations in pixel intensity to delineate water and land. Subsequently, multiple images within a time interval are used to construct flood probability maps. The study then focuses on addressing errors caused by uneven sampling of tide heights corresponding to the images by proposing the Sampling Errors Reduction (SER) method. This method incorporates tide heights simulated by the DTU16 global ocean tide model for each image, analyzing their distribution, and iteratively improves flood probability maps at a pixel level. Finally, elevation values in the intertidal zone are assigned proportionally based on the highest and lowest tide heights corresponding to the images, resulting in an intertidal Digital Elevation Model (DEM) situated between Mean Higher High Water (MHHW) and Mean Lower Low Water (MLLW).
Comparing the results with DEM collected from the SBES, the DEM of the intertidal zone constructed using the proposed method achieves a Root Mean Square Difference (RMSD) of 28.8 cm, with a 6.2% improvement through the SER method. Furthermore, higher accuracy and improvement rates are observed with more images, although the improvement rate is influenced by the distribution of tide heights corresponding to the images. |
