| 摘要: | 海底地形資訊是改善底棲環境分類結果的關鍵因素。然而，獲取水深資訊並不容易且費用高，尤其是在偏遠島嶼。 ICESat-2 是一個衛載 LiDAR ，可提供含有地理位置的高度資訊，進而解決近岸地區測深的挑戰。在這項研究中，我們提出以ICESat-2 數據與光學影像結合的演算法，以擴展衛星測深的應用。在我們的實驗中選擇南中國海的五個島嶼地區，利用光譜特徵根據目標島嶼底棲物質的異質性分類進行並分析評估。結果顯示ICESat-2 和 Sentinel-2 影像的結合，判定係數可以達到0.75-0.95，且均方根誤差在 0.66-1.87 m，這五個區域的最深估計到 19-32 m。 本研究也提出使用PlanetScope 衛星影像和ICESat-2 數據同時進行水深估計和底棲分類。 利用深度不變指數（Depth Invariant Index, DII）和海底反射指數（Bottom Reflectance Index, BRI）來減少水體的影響，再採用機器學習算法，包括隨機森林 (Random Forest, RF)、支持向量機 (Support Vector Machine, SVM) 和卷積神經網絡 (Convolutional Neural Network, CNN) 來分類底棲物質。 此三個分類法的總體準確率於BRI 分別為 79.02%、83.05% 和 86.49%，而 DII分別為 79%、82.75%、84.2%。 成果清楚地展示水深資訊在分類中的重要性，此外CNN方法可以得到最佳的底棲分類結果。;Bathymetry information is a critical factor in improving the classification results of benthic habitats. However, obtaining the bathymetry data is not always easy and affordable, especially in remote islands. ICESat-2 is a space-borne LiDAR satellite that provides geolocated photon height that can resolve the challenges in bathymetric mapping in the nearshore region. In this study, the combination of ICESat-2 data with optical images is developed to extend the application of satellite-derived bathymetry (SDB). Five islands in different parts of the South China Sea are selected, analyzed, and evaluated in our proposed model. The clustering step is used to address the heterogeneity of benthic habitats by dividing the target islands into groups based on spectral characteristics. The results show that an integration of ICESat-2 and Sentinel-2 imageries can achieve R2 at 0.75–0.95 and RMSE at 0.66–1.87 m with the deepest pixels at 19–32 m across these five islands.
This study also proposed a completed scheme for bathymetry estimation and integration of benthic classification using the PlanetScope satellite image and ICESat-2 data for the coastal region. Depth invariant index (DII) and bottom reflectance index (BRI) were utilized to reduce the water column′s effect. Moreover, two conventional machine learning algorithms including Random Forest (RF), Support Vector Machine (SVM) and a current deep convolutional neural network (CNN) were employed to address the benthic features. The overall accuracy of the three classifiers are 79.02%, 83.05%, and 86.49% with BRI compared to 79%, 82.75%, 84,2% of DII, respectively. These results clearly emphasize the importance of bathymetry information in the classification procedure. Moreover, the CNN approach can maximize the improvement in the benthic classification results in the coastal region. |
