| dc.description.abstract | Coastal terrain is an important factor to understand natural and anthropogenic forcing contributed to the coastal processes. However, it has long been difficult to map the elevation in the intertidal and shallow water areas, due primarily to the limitation in vessel navigation. In this study, we test an algorithm of satellite derived bathymetry (SDB), aiming to broaden remote sensing applications in coastal areas. Since the physical principle of SDB based one Beer’s law had been well developed in the last two decades, we focus on the model transferability to see whether a multispectral model used to estimate water depth trained in one particular location can be adopted in another one. There are two application test for evaluate this technical question –the first is Application test in South China Sea is used to evaluate feasibility of model transfer in spatial domain, while Application test in Houwan (Kenting National Park) is to confirm model transferability in temporal domain. A total of 4 series of optical satellite missions are used in this study, including Sentinel-2 and WorldView-2 data for Dongsha and Taiping Island, and Landsat-7 and SPOT data for Houwan coastline.
To increase the accuracy of depth measurements, for Landsat-7 and Sentinel-2 images we average out multiple retrieving results to avoid random noises in each individual image. Besides, different categories of substrates are classified separately to create their own corresponded model, in prevention of estimation bias from the inherent albedo of various substrates.
Results in Application in South China Sea indicates that the retrieval depth is limited to ~10m with a root-mean-square-error (RMSE) between 1.5-3 m, as validated with ground-truth provided by air-borne LiDAR surveys. An operation of substrates classification can increase the accuracy of RMSE between 0.1-1 m. In Application in Houwan, coastal dynamic of Houwan area investigating temporal changes of coastline is achieved, where a RMSE of the digital surface model (DSM) is 1.6 m as overlapped with DSM built by a drone campaign. In conclusion, the test of 3 applications as “model transfer application”, “substrates classification” and “multiple images averaging conduct” show a great potential to improve the quality and application in SDB, whose can achieve RMSE at 0.1-1m in coastal areas.
| en_US |