Featured by the transitional zone between land and sea (Gao et al., 2021), tidal wetlands act as an important component of coastal ecosystems (Zhang et al., 2019b). And they provide significance service value not only for animals but for humans, for instance, providing habitats for animals (Chang et al., 2022), being as the buffer zone to protect coastal areas from severe natural disasters (Wang et al., 2020b), and providing land resources for urban development (Wang et al., 2019). However, tidal wetlands are also vulnerable to environmental changes. In recent years, reclamation (Feng et al., 2012), sea-level rise, seawater erosion and other threats from land or sea have led to the rapid disappearance of tidal wetlands (Chen et al., 2016), which has brought potential damage to coastal ecosystems, and this trend will continue in the future because of global climate change. Therefore, in order to get a better understanding of tidal wetlands’ dynamic changes in coastal ecosystems and provide conservation suggestions of tidal wetlands for authorities, it is of great significance to monitor tidal wetlands.
Generally, there are two methods of monitoring tidal wetlands, including traditional monitoring based on field survey and monitoring based on remote sensing technology. Traditional monitoring based on field survey usually conducts field measurements with the help of surveying and mapping instruments, which is tough to be applied to monitor large-scale tidal wetlands because it is time-consuming, laborious, costly and easily affected by tide (Zhang et al., 2019a ; Ghosh et al., 2016).
In recent years, remote sensing technology has played a huge role in monitoring tidal wetlands. Compared to the traditional monitoring method based on field survey, remote sensing technology can achieve simultaneous monitoring of large-scale tidal wetlands and this technology is economical (Zhang et al., 2014). Remote sensing technology achieves monitoring by extracting tidal wetlands from remote sensing images. However, extracting tidal wetlands by remote sensing technology is a little different from other ground objects, that is, the spatial distribution of tidal wetlands in a single remote sensing image can not reflect the real extent of tidal wetlands due to the influence of tidal cycle (Murray et al., 2012), which makes it hard to conduct extraction of tidal wetlands merely on the basis of a single remote sensing image. To cope with this challenge, scholars have developed many methods, mainly including two types: methods based on combining tidal height data with remote sensing images and methods merely based on remote sensing images (Zhao et al., 2020). The former usually attributes tidal height to every remote sensing image, and then find out two remote sensing images corresponding to the lowest and highest tide respectively, finally, tidal wetlands between the lowest and highest tide are considered as the real extent of tidal wetlands. Based on this method, some studies were conducted, for example, (Yan et al., 2021) used tidal height data to determine waterlines corresponding to the lowest and highest tide respectively and then used determined waterlines to modify the extent of tidal wetlands extracted from machine learning algorithms; (Sharma et al., 2021) sequenced remote sensing images according to tidal height data and then extract the tidal wetland by setting threshold; (Murray et al., 2014) selected Landsat images within 10% of high tide and low tide according to tide height data, and used the selected Landsat images to extract tidal wetlands in the Yellow Sea.
Compared to the former method, methods merely based on remote sensing images avoid using tidal height data, and usually use intensive time series remote sensing images to extract tidal wetlands by calculating water inundation frequency and setting the inundation frequency threshold. For instance, (Lopes et al., 2020) extracted tidal wetlands by using normalized water index frequency image and vegetation frequency image, and found that extraction method of combining two images could achieve better result; by using Google Earth Engine and method merely based on remote sensing images, (Wang et al., 2020b) extracted tidal wetlands of China. In addition, some scholars have proposed to extract tidal wetlands by using waterlines extracted from remote sensing images, for example, (Liu et al., 2016) extracted waterlines of Bohai Rim from 1794 HJ-A/B images and used the outermost waterline as the boundary of tidal wetlands.
Accurate and efficient monitoring of tidal wetlands can help decision-makers to make conservative policies of coastal ecosystem in time. However, there are some limitations in above-mentioned methods of monitoring tidal wetlands by remote sensing technology despite the desired results achieved. On the one hand, Considering the influence of remote sensing images’ temporal resolution and the cloudy weather, intensive time series remote sensing images or the remote sensing image corresponding to the lowest tide is usually unavailable, which will make the boundaries of extracted tidal wetlands inaccurate; for example, Landsat provides images with a temporal resolution of 16 days, which makes it hard to capture the spatial distribution of tidal wetlands during the lowest tide. On the other hand, in terms of methods merely based on remote sensing images, intensive time series images with the volume of more than one hundred gigabytes need to be processed, which will result in low efficiency and not suitable for large-scale, long time series extraction of tidal wetlands.
Therefore, in this study, the main objectives were: (1) to resolve the above-mentioned limitations by proposing to extract tidal wetlands from remote sensing image by the means of spatio-temporal data fusion algorithm; and (2) to select the optimal spatio-temporal data fusion algorithm suitable for the extraction of large-scale and long time series tidal wetlands, which should meet the requirements of less input data, moderate computation and higher precision. On the whole, we proposed to extract tidal wetlands by using spatio-temporal data fusion algorithm which fused a 250m MODIS image at time t0, a 30m Landsat 8 image at time t0 with a 250m MODIS image at the lowest tide time to generate 30m fusion image (Landsat-like image) at the lowest tide time; and we compared the performance of three kinds of spatio-temporal data fusion algorithms——Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) (Gao et al., 2006), Flexible Spatiotemporal DAta Fusion model (FSDAF) (Zhu et al., 2016) and GAN-based Spatio-Temporal Fusion Model(GANSTFM) (Tan et al., 2022) ——in the extraction of tidal wetland, and selected the optimal algorithm suitable for the extraction of large-scale and long time series tidal wetland.