| 摘要: | 摘要
遙測的演進逐步提高了衛星影像的應用性，例如城市規劃、精準農業、空氣污染監測、和林業管理。同時，許多應用需要更高的時空解析度的衛星影像，以進一步加強其對更大尺度變化的監視能力。然而，沒有單個衛星可以產生具有高空間和時間解析度的影像。當面臨於需要高空間和時間解析度的應用時，如地表水監測，這種類型的問題將是一個巨大的問題。
由於自然或人為因素，地表水覆蓋面會動態地收縮或膨脹。自1970年代以來，各種遙測技術已用於監測地表水。但是，沒有單個衛星可以提供具有高空間和時間解析度的監測能力。為了克服此一眾所周知的問題，有兩種可能的策略，影像模擬和影像融合。影像模擬透過學習多時相影像並模擬應有的影像內容，來補足不可直接獲得的遙測影像。而影像融合透過整合來自不同感測器的影像來獲取更多訊息。
隨著最新技術的發展，我們(1）有了更多的衛星影像來了解水位高度與地表水覆蓋率之間的關係，並且(2）地球同步衛星(如Himawari-8)可以提供超高時間解析度的影像來監測水覆蓋的動態。因此，本研究試圖測試最近的遙測技術如何能夠幫助水覆蓋率的監測。
具體而言，本研究旨在通過結合Landsat OLI和Advance Himawari Imager(AHI)的影像來獲得時空修正的正規化差異水指數（mNDWI）影像。對於影像模擬，我們尋找兩張具有最接近水位高度的影像，並透過線性內插的推估特定水位的影像。而對於影像融合，使用稱為Spatial and Temporal Adaptive Fusion Model (STARFM)方法來推估特定水位高的影像。
本研究測試了六種方法，分別是A1、A2、A3、A4，A5和A6。其中A1和A2是影像模擬方法，A3和A4是使用模擬參考影像的影像融合方法，而A5和A6是使用最接近水位高的參考影像之影像融合方法。此外，A1、A3和A5根據潮汐模型估算水位，而A2、A4和A6則根據AHI影像估算水位高進行影像模擬。
我們對正常情況和異常情況檢查了這些方法。這兩種情況之間的區別基於水位高度是否符合潮汐模型。實驗結果表明，在正常情況下，A2、A4和A6可以獲得更好的結果，相關係數通常大於0.90，異常情況下的整體精度大於0.95。綜上所述，本研究得出以下結論。隨著遙測技術的最新發展，影像模擬和影像融合都可以解決數據不足的問題。超高時間解析度的AHI影像可以幫助追蹤潮間帶水域的動態變化，從而實現更高的時空解析度之水域監測能力。
關鍵字：水覆蓋監測、影像模擬、時空影像融合
;ABSTRACT
Recent improvements of remote sensing largely increase the applicability of satellite images, which has been utilized in many applications such as urban planning, precision agriculture, air pollution monitoring, and forestry management. In the meantime, higher spatial and temporal resolutions of satellite imagery are desired to further strengthen its monitoring capability for changes at finer scales. However, no single satellite can produce images with both high spatial and temporal resolutions. This type of problem will be an enormous issue while sensors are applied in applications that needs these both high resolutions, among which is the surface water coverage monitoring.
The surface water coverage dynamically changes as it can shrink or expand due to natural or human factors. Various remote sensors had been applied in detecting and monitoring surface water since the 1970s, such as Moderate Resolution Imaging Spectroradiometer (MODIS), and Suomi National Polar-orbiting Partnership - Visible Infrared Imaging Radiometer Suite (NPP-VIIRS). However, no single satellite can provide imagery with both high spatial and temporal resolutions. To overcome this well-known issue in remote sensing, there are two possible strategies, image simulation and image fusion. As the image simulation becomes one of the solutions to solve the lack of remotely-sensed data by generating multitemporal images, the image fusion is a remote sensing technique that has been applied to obtain more information by integrating images from different sensors.
With recent advancement, we have (1) more satellite images to understand the relationship between water height and surface water coverage and (2) geostationary satellites such as Himawari-8 to provide ultra-high temporal resolution images to monitor the dynamics of water coverage. Therefore, this research tries to examine how recent remote sensing advancement can help water coverage monitoring.
To be specific, this research aims to derive spatial and temporal modified Normalized Difference Water Index (mNDWI) images by combining imagery from Landsat Operational Land Imager (OLI) and Advance Himawari Imager (AHI). For image simulation, we generate an image of a certain water height via linear interpolation with two images that have nearest water heights. For the image fusion, an existing model called Spatial and Temporal Adaptive Fusion Model (STARFM) is applied to generate an image of a certain water height.
This research proposes six approaches, which are A1, A2, A3, A4, A5, and A6, where A1 and A2 are image simulation approaches, A3 and A4 are image fusion approaches with simulated reference data, and A5 and A6 are image fusion approaches with reference images that have closest water heights. The A1, A3, and A5 estimate water heights from a tide model while A2, A4, and A6 estimate water heights based on AHI imagery.
We examine these approaches with normal cases and anomaly cases. The difference between these two types of cases is based on whether water heights follow the tide model or not. The experimental result shows that A2, A4, and A6 can achieve better results for normal case with correlation coefficient generally greater than 0.90 and overall accuracy more than 0.95 for anomaly case. In summary, this research has the following conclusions. With the current advance in remote sensing technologies, both the image simulation and image fusion can solve the problem of data lacking. The ultra-high temporal resolution AHI imagery can help capture the dynamics of water coverage changes in the intertidal area, with which higher spatial and temporal resolution water coverage monitoring capability can be achieved.
Keywords: Water coverage monitoring; Image simulation; Spatial and temporal image fusion |
