Tidal Flats Dataset Covers Coastal Region in North of 18°N Latitude of China (1989–2020)

Hu, Z. W.¹  Xu, Y.¹  Yin, Y. M.¹  Zhang, K. Y.¹  Wu, G. F.¹  Wang, C.^2*  Cui, L. J.^3*

1. MNR Key Laboratory for Geo-Environmental Monitoring of Great Bay Area, Shenzhen University, Shenzhen 518060, China;

2. Center for Satellite Application on Ecology and Environment, Ministry of Ecology and Environment, Beijing 100094, China;

3. Institute of Wetland Research, Chinese Academy of Forestry, Beijing Key Laboratory of Wetland Ecological Function and Restoration, Beijing 100091, China

Abstract: Coastal tidal flats are ecologically vulnerable regions, susceptible to climate change, sea level rises and human activities. Using time-series satellite images and field survey data, we developed a supervised classification method for mapping tidal flats on Google Earth Engine (GEE) and then obtained the Tidal flats dataset covers coastal region in north of 18°N latitude of China (1989–2020). This dataset is on an annual basis for over 30 years and has a spatial resolution of 30 m. It consists of 256 data files in the format of .shp with the data size of 318MB (Compressed to one single file with 201 MB).

Keywords: coastal zone; tidal flats; China; 1989–2020

DOI: https://doi.org/10.3974/geodp. 2022.01.17

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2022.01.17

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2021.10.06.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2021.10.06.V1.

1 Introduction

Coastal tidal flat wetlands are the transition zone between land and sea, maintaining biodiversity and productivity among various nature ecosystems^[1]. Due to an continuous increase in human activities to coastal areas over the years, substantial coastal tidal flats in China have been degraded and claimed^[2]. Mapping and understanding the spatio-temporal distribution of tidal flats are significant for their management, protection, restoration and sustainable development. Traditional field surveys are always time-consuming and laborious as the muddy condition and inaccessibility when flooding, while remote sensing technology could greatly improve the efficiency for large-scale tidal flat mapping. Also, repeated satellite observations over past decades help to represent the historical change of tidal flats^[3].

We deployed a supervised classification method with time-series satellite images at 30 m resolution provided in Google Earth Engine (GEE) to generate annual tidal flat map in the north of China at 18°N latitude during 1989–2020, and now we make this dataset available for public use.

2 Metadata of the Dataset

The metadata of Tidal flats dataset covers coastal region in north of 18°N latitude of China (1989–2020)^[4] is summarized in Table 1. It mainly includes full name, short name, authors, years, temporal resolution, spatial resolution, data format, data size, data files, data publisher and data sharing policy, etc.

3 Materials and Methods

3.1 Data and Methods

The multispectral images from Landsat 4/5 Thematic Mapper (TM), Landsat 7 Enhanced Thematic Mapper Plus (ETM+) and Landsat 8 Operational Land Imager (OLI) over the year of 1989 to 2020 were used as data sources, and training sample points were selected based on our field survey data. The spectral features of tidal flats and other land covers were firstly analyzed in terms of training sample points, and then several key characteristics were worked out from the time-series satellite images. Random forest (RF) classification algorithm with the aforementioned characteristics generated as predictor variables was implemented in the GEE platform to yield annual tidal flat map. Final dataset was obtained as each annual map was processed by a post-processing.

3.2 Objective and Study Area

3.2.1 Definition of Tidal Flat Wetland of the Dataset

Taking the reference in previous studies^[5], field samples and remote sensing images, the tidal flat wetlands in this dataset include intertidal barren areas and salt marshes. Specifically, the intertidal barren area refers to the area that is flooded by seawater during high tide and exposed as a sand or soil at low tide^[6]. The salt marsh area refers to undeveloped natural vegetations adjacent to the intertidal barren areas, such as mangroves, halophytic herbs and shrubs, etc.^[7].

3.2.2 Study Area

According to the China national surveying of coastal zones and tidal flats^[8], with topography and landscape across the coastal zones in China taken into account, our study area is defined as the coastal buffer formed from a 10 km buffer landward and a 40 km buffer seaward along the coastline (Figure 1).The coastline was derived from Global Self- consistent, Hierarchical, High-resolution Geography Database (GSHHG^[9]).

Table 1  Metadata summary of the Tidal flats dataset covers coastal region in north of 18°N latitude of China

3.3 Algorithms

3.3.1 Feature Extraction

The pixels of clouds, cloud shadows and snows were masked in terms of the Pixel_QA band^[10], and spectral indexes of the remaining pixels were then calculated (Table 2). The pixels within different tide levels show different periodic time-series characteristics^[11]. For examples, mNDWI value increases with the increase of tide level, while the bare soil index decreases. NDVI value changes seasonally with the growth of plants. In order to reflect the periodic changes of different land covers, the spectral index series were sorted ascendingly and then, the mean, standard deviation and 5^th, 25^th, 75^th and 95^th percentiles of annual spectral indexes were obtained. Finally, six statistical metrics for each spectral index were computed in each pixel position, and a total of 42 features were calculated for seven spectral indexes.

Figure 1  Location of study area (Tidal flat wetland distribution map in 2020)

Table 2  Spectral indexes used in this study

Note: ρ_Red, ρ_Green, ρ_Blue, ρ_Nir and ρ_Swir are the reflectance values of red, green, blue, near infrared and shortwave infrared channels.

3.3.2 Random Forest Algorithm

Random Forest is an ensemble algorithm based on classification and regression tree (CART)^[18] where overall structure is a tree structure, consisting of root nodes, decision nodes, decision tree branches and leaf nodes. The implementation of RF is as follows:

Input data: N training samples with M-dimensional feature vectors;

Parameters: The number of samples for every subset (n), the feature dimension to construct every decision tree (F) and the number of trees (N_Tree);

Step 1: A subset of the training set is obtained by randomly selecting n samples from training samples via the method of Bagging;

Step 2: F features (F≤M) are randomly selected from M features, and a classification tree is constructed using the randomly selected subset;

Step 3: The Step 1 and Step 2 are conducted repeatedly to construct N_Tree classification trees;

Step 4: An unlabeled object is predicated by all the N_Tree decision trees, and the category with the most votes is regarded as the final classified label.

3.4 Technical Workflow

All the image pre-processing and classification processes were conducted on the GEE platform (Figure 2).

Figure 2  Technique workflow of the classification algorithm

3.4.1 Training Sample Selection

We conducted field campaigns in Jiangsu, Guangxi, Guangdong and Shanghai, but more sample points were selected from Google Earth Pro (GEP). A total 12,704 sample points were selected in four years for training the RF classifier and validation (Table 3). We labeled each sample point in four land covers (Table 3) and some typical ones are presented in Figure 3.

Table 3  The descriptions of land covers and the number of training samples

Figure 3  Typical land covers: (a) intertidal barren flat, (b) mangrove forest, (c) agriculture field and (d) built-up area

3.4.2 Random Classification on GEE Platform

The GEE platform uses Google’s computing infrastructure to implement parallel geospatial data processing, and it greatly reduces the time for data processing. The platform integrates a variety of machine learning methods, among which the random forest algorithm is a good choice for large scale land cover mapping with high accuracy.

In this study, 42 spectral variables plus land topography and ocean bathymetry (ETOPO1) data were used to train the RF classifiers. The number of classification trees was set to 100, and 43 predictor variables were randomly combined to construct every single classification tree. The default settings of other parameters provided by GEE platform were applied in this study.

3.4.3 Post-Processing

The majority filter was adopted to eliminate salt-and-pepper noises of each classification map. Considering that the tidal flats and salt marshes are normally spatially adjacent, isolated salt marsh patches far from tidal flats were automatically removed by analyzing the adjacency relationship. Moreover, the digital elevation model provided by Shuttle Radar Topography Mission (SRTM) was utilized to remove the misclassified pixels with the threshold set at 35 m. Some remaining misclassification patches were further corrected manually, whilst some holes were filled. Finally, 32 annual tidal flats maps were obtained.

4 Data Results and Validation

4.1 Data Components

The dataset contains 32 vector maps in the format of .shp over 1989–2020 which were derived from 30 m resolution raster maps.

4.2 Data Products

A 30 m tidal flat map (Figure 4) covering the coastal region in North of 18°N latitude of China by using RF machine learning algorithm with Landsat time-series images via GEE. In 1989, the total area of coastal tidal wetlands is 13,653.7 km², including 12,840.7 km² bare tidal flats and 813.0 km² salt marsh wetlands. And in 2020, the total area of coastal tidal wetland in China is about 8,100.8 km², including 7,135.9 km² bare tidal wetlands and 964.9 km² salt marsh wetlands. From the comparison between 1989 and 2020, the area of bare tidal flats decreased by 44.4%, but salt marshes increased by 18.7%.

Tidal wetlands mainly occur in gently sloping coastal zones and estuaries. For example, total coastal wetlands in Jiangsu province accounts for about 21.0% across the study area in 2020, whilst the estuary wetlands of Yangtze River, Pearl River, Yellow River and Liaohe River totally represent about 22.3%.

Figure 4  Maps of tidal flat wetlands in three distinct regions: (a) Liaohe Estuary wetland, (b) Yellow River estuary wetland, (c) Coastal area of Yancheng, Jiangsu province

4.3 Data Validation

532 random points were randomly generated within the study area and imported into Google Earth Pro (GEP) software. The ground truths of these points were manually interpreted on their corresponding high spatial resolution images in GEP software, including 147 tidal flat points, 103 salt marsh points and 282 other points. Overall accuracies and Kappa coefficients were computed (Figure 5) by comparing the mapping results. All the overall accuracies are higher than 88%, and the Kappa coefficients are higher than 0.80. Average overall accuracy of the 32 annual maps is 90.84%, and average Kappa coefficient is 0.85. These accuracy metrics indicate the high accuracy of this dataset.

Figure 5  Overall accuracies and Kappa coefficients of the maps

5 Conclusion

The dataset is annual and has a spatial resolution of 30 m. The overall accuracies of all the 32 maps show their reliability. This dataset can be used in the spatio-temporal change analysis, evaluation of coastal ecosystem services and decision-making of sustainable developments for tidal flat wetlands in China.

Author Contributions

Cui, L. J., Wang, C. and Wu, G. F. made conceptual design and provided financial support. Xu, Y. and Zhang, K. Y. collected and processed sample data and images. Hu, Z. W. and Zhang, K. Y. designed algorithms. Yin, Y. M. contributed to dataset validation. Hu, Z. W. and Xu, Y. prepared the paper, and Cui, L. J., Wang, C. and Wu, G. F. revised the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]      Zhang, X., Li, P. Y., Li, P., et al. Present conditions and prospects of study on coastal wetlands in China [J]. Advances in Marine Science, 2005, 23(1): 87–95.

[2]      Yao, H. Characterizing landuse changes in 1990–2010 in the coastal zone of Nantong, Jiangsu province, China [J]. Ocean & Coastal Management, 2013, 71: 108–15.

[3]      Wen, Q., Zhang, Z., Xu, J., et al. Spatial and temporal change of wetlands in Bohai rim during 2000–2008: an analysis based on satellite images [J]. Journal of Remote Sensing, 2011, 15(1): 183–200.

[4]      Hu, Z. W., Xu, Y., Yin, Y. M., et al. Tidal flats dataset covers coastal region in north of 18°N latitude of China (1989–2020) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2021. https://doi.org/10.3974/geodb.2021.10.06.V1. https://cstr.escience.org.cn/CSTR:20146.11.2021.10.06.V1.

[5]      Wang, G. Definition of coastal tidal flats [J]. Chinese Fishery Economy, 2013(1): 99–104.

[6]      GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[7]      Fang, R, K. Environmental Dictionary [M]. Beijing: Science Press, 2003.

[8]      Peng, J., Wang, Y, L. Study of coastal tidalflats in China [J]. Journal of Peking University (Natural Science), 2000, 36(6): 832–839.

[9]      Su, S. J. Comprehensive survey of coastal zone and tideland resources in China in the past seven years [J]. Marine and Coastal Zone Development, 1988, (2): 30–32.

[10]   Wessel, P., Smith, W. H. F. A global, self-consistent, hierarchical, high-resolution shoreline database [J]. Journal of Geophysical Research Solid Earth, 1996, 101(B4): 8741–8743.

[11]   Foga, S., Scaramuzza, P. L., Guo, S., et al. Cloud detection algorithm comparison and validation for operational Landsat data products [J]. Remote Sensing of Environment, 2017, 194: 379–390.

[12]   Zhang, K. Y., Dong, X. Y., Liu, Z. G., et al. Mapping tidal flats with Landsat 8 images and Google Earth Engine: a case study of the China’s eastern coastal zone circa 2015 [J]. Remote Sensing, 2019, 11(8): 924.

[13]   Cheng, B., Liu, Y., Liu, X., et al. Research on extraction method of coastal aquaculture areas on high resolution remote sensing image based on multi-features fusion [J]. Remote Sensing Technology and Application, 2018, 33(2): 296–304.

[14]   Gong, C., Gang, D., Wang, D. Remote sensing monitoring water area of Dongting lake based on MNDWI [J]. Journal of Water Resources Research, 2015, 4(3): 234–239.

[15]      Dong, Z. Y., Wang, Z. M., Liu, D. W., et al. Mapping wetland areas using Landsat-derived NDVI and LSWI: a case study of west Songnen Plain, Northeast China [J]. Journal of the Indian Society of Remote Sensing, 2014, 42(3): 569–576.

[16]   Nguyen, C. T., Chidthaisong, A., Diem, P. K., et al. A Modified bare soil index to identify bare land Features during agricultural Fallow-Period in Southeast Asia using Landsat 8 [J]. Land, 2021, 10(3): 1–18.

[17]   Guo, B., Zang, W. Q., Zhang, R. Soil salizanation information in the Yellow River Delta based on feature surface models using Landsat 8 OLI Data [J]. Ieee Access, 2020, 11(1): 288–300.

[18]   Zhang, J. H., Feng, L. L., Yao, F. M. Improved maize cultivated area estimation over a large scale combining MODIS-EVI time series data and crop phenological information [J]. Isprs Journal of Photogrammetry and Remote Sensing, 2014, 94: 102–113.

[19]   Malik, M., Shukla, J. P., MishraI, S. Relationship of LST, NDBI and NDVI using Landsat-8 data in Kandai­himmat Watershed, Hoshangabad, India [J]. Indian Journal of Geo-Marine Sciences, 2019, 48(1): 25–31.

[20]   Breiman, L. Random forests [J]. Machine Learning, 2001, 45(1): 35–32.