Mapping large-area tidal flats without the dependence on tidal elevations: A case study of Southern China
Introduction
Tidal flat is an important component of the intertidal zone that connects marine, terrestrial, and freshwater ecosystems, provides services such as shoreline stabilization, storm and wave buffering, and breeding and nursing ground (Barbier et al., 2011, Zohary and Gasith, 2014). However, tidal flats are under increased pressure from activities such as land reclamation, coastal development, sea-level rise, and river sediment reduction (Blum and Roberts, 2009, French, 2002, Syvitski et al., 2005), which result from anthropogenic impacts and climate change (Kirwan and Megonigal, 2013, Murray et al., 2014, Nicholls and Cazenave, 2010). For example, coastline hardening to protect human residential areas, results in rapid degradation of tidal flats. Many human-made structures (such as dams and reservoirs) constructed to control floods and saltwater intrusion not only reduce the sediment delivery to tidal flats, but also change the interactions between rivers and seas (Blum and Roberts, 2009, Kirwan and Megonigal, 2013). In areas experiencing rapid coastal development, the changes can be quick and wide-ranging. The sea-level rise worsens the situation (Barros et al., 2014). It might lead to conversion of tidal flats to open water (Craft et al., 2009) and result in increased saltwater intrusion into the fresh surface water (Bhuiyan and Dutta, 2012, Hong and Shen, 2012). Thus, a high-resolution and fast-updating tidal flat map that covers large areas is necessary for managing, protecting, and restoring coastal ecosystems.
To map the tidal flats, studies have been conducted based on traditional tidal stations or remote sensing (RS) imagery. The tidal stations provide valuable information on tidal datum at points, which could be interpolated to the coastline for mapping the tidal flats. However, for large areas of coastline, which are spatially undulated and varied (Maloney and Ausness, 1974, Milton and Farvar, 1968), the spatial sparsity of tidal stations due to high expenditure, as well as the difficulty in accessing such data, limit the availability of such information.
The development of RS techniques provides a large amount of imagery potentially available for mapping the intertidal zone (or tidal flats particularly) (Green et al., 1996). Currently many RS-based approaches for mapping the intertidal zone or tidal flats have been proposed. Existing RS-based approaches for mapping tidal flats can be classified into three types: terrain-based approaches, approaches based on simulated tidal elevations and RS images, and approaches based solely on RS images.
The terrain-based approaches mainly depend on digital elevation model (DEM). Such DEMs have been proposed to be built from waterlines that are extracted from images and tidal elevations at the precise time of captured images (Liu et al., 2012, Mason et al., 1995, Ryu et al., 2008), or to be built by Light Detection and Ranging (LIDAR) and Synthetic Aperture Radar (SAR) at low tides (Bell et al., 2016, Cracknell, 1999, Gade et al., 2008). Due to the scarcity of coastal DEMs with high resolution in spatial and temporal domain, the spatial variation of water levels, rapidly changing coastal topography, and high cost of airborne acquisitions (Liu et al., 2012, Murray et al., 2012), the terrain-based approaches are not applicable to large-area tidal flat mapping.
For mapping large areas of intertidal zones or tidal flats, approaches based on simulated tidal elevations and RS images were proposed. These approaches assign a simulated tidal height to each of the images covering large areas and maps of the intertidal zones or tidal flats were made by composing those images within a certain simulated tidal interval to classify them (Murray et al., 2014, Murray et al., 2012, Sagar et al., 2017). Certainly, the performance of these approaches is highly dependent on the simulation of tidal elevations over offshore regions. However, the tidal elevation simulation has uncertainties in accuracy over offshore regions (Turner et al., 2013), especially in large areas where anthropogenic impacts (such as aquacultures, harbor constructions, dams, and roads) have changed the tidal conditions. Meanwhile, in these approaches the tidal information of each image is represented by only a single value, which ignores the complexity in both tidal patterns and geographic heterogeneity. Therefore, these approaches suffer high uncertainty caused by the difference in simulated tidal elevations in large areas, especially in rapidly changing regions.
To avoid dependence on the modelled tidal elevations at the time of image acquisition, approaches based solely on RS images were proposed. Chen et al. (2016) manually selected the images with the highest or lowest shoreline to map the tidal flats of the Yangtze Estuary. The method of manually selecting images with the highest or lowest tides is laborious to be applied to large areas. Wang et al. (2018) classified each pixel from the satellite imagery during a period as water and non-water pixel, by setting a global threshold on its derived water indices and then differentiated the intertidal zone by using another global threshold on inundation frequency. The tidal flats were identified by applying a third global threshold of vegetation frequency to each pixel in intertidal zone over the period. This approach relies highly on the global thresholds, which ignores the complexity in geographic heterogeneity. In a state-of-the-art study based solely on RS images, Murray et al. (2019) mapped the tidal flats around the world using a machine learning classifier mainly based on features of statistics of the Landsat-derived water indices on each individual pixel location. The features were calculated from percentiles and interval means of Landsat observation values within a two-year observation period. Murray et al. (2019)'s approach depends on large training samples and numerous statistics on water indices. The large training samples mean a laborious work; meanwhile these numerous statistics are comparatively redundant in describing the water information. Therefore, existing ways of avoiding the dependence on the modelled tidal elevations in current approaches based solely on RS images might not work effectively for large areas.
Meanwhile, mapping tidal flats in estuaries is still comparatively ignored in the existing approaches for mapping tidal flats, although globally, tidal flats are present in estuaries and are key areas where intertidal zone connects freshwater ecosystems with the sea. Few station observation in these river-sea interaction areas is available because these areas are hardly monitored for accurate vertical tidal datum, due to the complexity of the tidal influence interfered by fresh water runoff (Maloney and Ausness, 1974) and thus are normally excluded from building tidal stations. In the existing RS-based tidal flats mapping approaches, estuaries are considered using a global threshold of coast distance. This method is over-simplified and would cause large errors in estimating the tidal flats in estuaries. In fact, the saltwater induced by tides can penetrate upstream to a few to tens of kilometers from the estuarine mouth (Garvine, 1975, Meade, 1966). Meanwhile, tidal flats in estuaries could vary because the river-sea interactions are highly dynamic due to the influences from sea-level rise, dams, river channel morphology change, etc. (Hoitink and Jay, 2016).
To overcome the above-mentioned problems in existing RS-based approaches for mapping tidal flats on a large spatial scale, we proposed an approach for mapping tidal flats over large area. For removing the dependence on tidal elevations efficiently, the proposed approach adopts a new way, which is different with those of relying on either global thresholds (Wang et al., 2018), or large training samples and numerous statistic values (Murray et al., 2019). The issue of separating tidal flats along estuaries is also specifically considered in the proposed approach. Section 2 presents the basic idea of the proposed approach, including composition of the high and low tidal datum based solely on RS data, the water and land classification by an adaptive binary classifier without needing large sample set, and the separation of tidal flats along estuaries by considering the intrusion of saltwater into fresh surface water in detail. We applied the proposed approach to mapping the tidal flats of southern China for the year 2017, based on Sentinel-1 SAR data (Section 3). Section 4 shows the results of the proposed approach and compares them with the state-of-the-art tidal flat maps produced by Murray et al., 2019, Wang et al., 2018. We also analyze the results of the proposed approach, based on the distance of saltwater intrusion into fresh surface water along estuaries. Section 5 discusses the effectiveness of the proposed approach in estimating tidal datum based solely on satellite imagery, as well as the limitations and strengths of the proposed approach compared with the two state-of-the-art approaches. Conclusions are presented in Section 6.
Section snippets
Basic idea
Tides are affected by the phases of the moon and meteorological conditions, thus they are time-variant and perform inconsistently even at any fixed time of each day. The instantaneous tidal state could be observed by satellites with a certain revisit time and multiple instantaneous observations could be used in estimation of tides and tide-related tidal flats. Li and Gong (2016) regarded each observation from satellite imagery during multiple revisits for the same location as a random sample
Study area
The study area is located in south China, covering about half of China’s coastline (Fig. 2). It includes Zhejiang, Fujian, Guangdong, Guangxi Zhuang Autonomous, Hong Kong, Macao, Taiwan, and Hainan. The region has complex tidal patterns (including diurnal tides, semidiurnal tides and mixed tides), which are calculated by the data from Oregon State University (OSU) TOPEX/POSEIDON Crossover database (TPXO9) (Egbert and Erofeeva, 2010) according to Reeve et al. (2012). The abundance in
Tidal flat map and its evaluation
The tidal flat map integrating tidal salinity influence, of southern China for 2017 is shown in Fig. 7.
According to the randomly generated 1377 validation samples, the tidal flat map has an overall accuracy value of 92.4% (Table 1) and the 95% quantile for confidence interval of overall accuracy is 90.8–93.7%. The user accuracies for the non-tidal and the tidal flats are 88.3% and 100.0%, respectively. The lower user accuracy for the non-tidal areas on the resulted map might have resulted from
Area statistics from different tidal flat maps of the study area
The areas of tidal flats reported in existing studies covering the southern China vary largely. In a country-scale study, Wang et al. (2018) reported a tidal flat area of 292,282 ha in southern China in 2016, while Murray et al. (2019)’s global tidal flat map records an area double the size of the former one, i.e., a tidal flat area of 581,082 ha for the same area in 2014–2016. Both the studies used the same satellite data source. Using a different data source, the proposed approach in this
Conclusions
In this paper, we proposed a large-area tidal flat mapping approach with a new way to avoid the dependence on tidal elevations. The approach synthesizes Sentinel-1 SAR data over a two-year time period (to acquire high and low tidal datums) and classifies land and water under each tidal datum using an adaptive binary classifier. Thus, tidal flats could be identified as the areas being water during high tides and being land during low tides. Furthermore, the proposed approach uses time series of
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This research was funded by the Science and Technology Basic Resources Investigation Program of China (No. 2017FY100706). The authors thank Dr. Mingming Jia of Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences, for her kindly providing the mangrove map of China for 2010, which was produced by her workgroup.
References (66)
- et al.
Analysis of SMOS sea surface salinity data using DINEOF
Rem. Sens. Environ.
(2016) - et al.
Analysis of high frequency geostationary ocean colour data using DINEOF
Estuar. Coast. Shelf Sci.
(2015) Chapter 10 Siliciclastic Tidal Flats
- et al.
A temporal waterline approach to mapping intertidal areas using X-band marine radar
Coast. Eng.
(2016) - et al.
Assessing impacts of sea level rise on river salinity in the Gorai river network, Bangladesh
Estuar. Coast. Shelf Sci.
(2012) - et al.
Classification of sediments on exposed tidal flats in the German Bight using multi-frequency radar data
Remote Sens. Environ.
(2008) - et al.
The conservative and non-conservative behavior of chromophoric dissolved organic matter in Chinese estuarine waters
Mar. Chem.
(2007) - et al.
Responses of estuarine salinity and transport processes to potential future sea-level rise in the Chesapeake Bay
Estuar. Coast. Shelf Sci.
(2012) - et al.
Continuous monitoring of coastline dynamics in western Florida with a 30-year time series of Landsat imagery
Remote Sens. Environ.
(2016) Effects of waterlogging and salinity on plant-water relations and on the accumulation of solutes in three mangrove species
Aquat. Bot.
(1985)