1 Introduction

The Liaohe Delta coastal wetland is located at the junction of the Liaohe and the Daliaohe Estuary. It is one of four major estuary deltas in China and is also China’s main oil and grain production base [1]. Due to geographical location, the Liaohe Delta wetland was affected by rivers and oceans. With the acceleration of urbanization and reclamation activities, some coastal wetlands in this area were seriously degraded [2, 3]. By analyzing the distribution of coastal wetlands, we can understand the causes, mechanisms and evolution of coastal wetlands, and provide a scientific basis for sustainable development of coastal wetland resources. The Liaohe Delta area has become a hotspot research area in recent years, due to the concentrated distribution of various types of coastal wetlands, and strong human disturbance. Some scholars have studied the status and distribution characteristics of wetland resources in the Liaohe Delta with the help of remote sensing and geographic information system (GIS) technology, and established the Liaohe Delta wetland resource classification system [4]. In 1997, Huang Guilin used 3S technology to classify Liaohe Delta wetland types and areas, as well as the changes to and causes of the Shuangtaizi River Nature Reserve and wetland types [5]. Landsat thematic mapper (TM) satellite imagery from 1986, 1995, 2000 and 2006 was adopted to monitor changes to wetlands in Panjin city through visualization interpretation, computer automatic identification, and field verification, and the causes of the changes were analyzed [6]. However, use of Tiangong-2 data to reveal the distribution of wetland in the Liaohe Delta is still rare. The Tiangong-2 wide-band imaging spectrometer is a new generation optical remote sensor with wide-band and wide-field coverage, as well as image-spectrum merging. This is the first opportunity for large-scale measurement of visible light, short-wave infrared, and thermal infrared multi-spectra on a single instrument, as well as field full-push scanning imaging with integrated functions. The wide-band imaging spectrometer has 14 channels in the visible and near-infrared, and the spectral range is 0.4 μm to 1.0 μm. It has multiple bands, narrow spectral range and continuous banding, which are conducive to accurate interpretation of wetland information.

In this study, the suitability of remote sensing data from wide-band imaging spectrometer of Tiangong-2 for monitoring coastal wetland have been explored. Tiangong-2 remote sensing data and GIS technology were used to analyze Panjin wetland types, area and distribution. This information will be beneficial to ensure the background data of the wetland ecosystem and its biodiversity in the Liaohe Delta, and will lay a foundation for the protection and rational use of wetland resources.

2 Study Area and Data

2.1 Study Area

The Liaohe Delta is located in southwestern Liaoning Province, at the top of Liaodong Bay in the Bohai Sea. It is an alluvial marine plain formed by the Liaohe, Daliaohe, and Dalinghe rivers, including parts of Panjin and Yingkou cities. The geographical position is 121°25′ to 112°55′ E, and 40°40′ to 41°25′ N, in the concentrated wetland area southern part of northeastern China with extensive development of agriculture, petroleum industry, shallow sea breeding industry, and reed planting industry. Panjin City lies on the alluvial marine plain that is the main core of the Liaohe Delta. Landform include tidal plain and tidal flat. Panjin has a warm temperate continental semi-humid monsoon climate, and regional climate differentiation is not obvious. The annual average temperature is 8.3 °C to 8.4 °C, the frost-free period is 167 d to 174 d, the annual average rainfall is 611.6 mm to 640.0 mm, and the annual average evaporation is 1 669.6 mm. The annual sunshine hours are 2786.0 h [7]. There are more than 20 rivers in Panjin, which gather in the Shuangtaizi Estuary and the Daliaohe River before emptying into the sea. Panjin city has a complete system of administrative divisions and statistical data, so we selected this city in this study as an area of focus to analyze the applicability of Tiangong-2 remote sensing data in coastal wetland monitoring (Fig. 1).

Fig. 1.
figure 1

Location of the study area (Combination of 5/8/10 channel visible)

Full size image

2.2 Data

The Tiangong-2 Space Laboratory was launched on September 15, 2016, and was in orbit on September 23, 2016. Data services were provided after June 2017. Tiangong-2 is equipped with a new space application load device. Instruments include a wide-band imaging spectrometer, three-dimensional imaging microwave altimeters and ultraviolet edge imaging spectrometers for earth observation and space geoscience. The wide-band imaging spectrometer is the first in the world to be packaged as a single instrument. It integrates visible, short-wave infrared and thermal infrared multi-spectral large field of view full-push scan imaging. During orbit, it is mainly used for medium (higher) ground resolution; large-scale ground object monitoring, suitable for inland lakes, land and atmospheric exploration, as well as observation of ocean and coastal water color and temperature [8]. This study uses visible and near-infrared imagery from the Tiangong-2 wide-band imaging spectrometer. Because river wetlands, reservoirs and aquaculture farms are water bodies that change with the seasons, wet season data were used. Furthermore, the chlorophyll content in the summer vegetation growth period is high, which will be conducive to identifying marsh wetlands. Cloud cover should be less than 10% to ensure the quality of remote sensing imagery. Therefore, the remote sensing imagery data on June 1st, 2018 was selected. System radiation and geometric corrected image data were downloaded from the official manned space applications data sharing and distribution website (http://www.msadc.cn/), with a spatial resolution of 100 m. Lands at 8 operational land imager (OLI) remote sensing image of Panjin on May 5th, 2018 were used to evaluated the accuracy of Tiangong-2 data for wetland extraction. Image download from OLI official website (http://glovis.usgs.gov/) [9].

3 Methods

3.1 Remote Sensing Image Preprocessing

Atmospheric correction.

The images acquired by sensors on satellites need to be processed before use. To obtain the true reflectivity of the surface, radiation correction is needed to reduce the effects of sensors, atmospheric conditions, and solar radiation. The radiation correction in this study was done with ENVI 5.1 software. Lelvel-2 grade Tiangong-2 data was used in the study. There are the radiance products after system radiation correction and geometric correction. The DN value is radiance amplified by a certain magnification; the magnification parameter is included in the companion XML file. Then, atmospheric correction was done using the fast line-site atmospheric analysis of hypercubes (FlAASH) atmospheric correction method. Because there is no basic parameter of the Tiangong-2 in the ENVI software, it is necessary to edit the header file including the center wavelength and the band width corresponding to the 14 bands (Table 1). In addition, the spectral response function of the Tiangong-2 wide-band imaging spectrometer is required to complete the parameter setting of FlAASH atmospheric correction.

Table 1. Visible and near-infrared spectra from the wide-band imaging spectrometer (Unit: nm).
Full size table

To check the effects of atmospheric correction, spectral profiles of the same pixel, sampled from vegetation, before and after atmospheric correction are shown in Fig. 2. Figure 2a is the original DN value before atmospheric correction, and Fig. 2b is the reflectivity magnified 10,000 times after atmospheric correction. Figure 2 shows that atmospheric correction removed the influence of part of the atmosphere.

Fig. 2.
figure 2

Vegetation spectrum curve comparison before (2a) and after atmospheric correction (2b)

Full size image

Geometric correction.

Tiangong-2 data was geometrically corrected by Landsat OLI imagery with the right geometric information for the same time period. Quadratic polynomial and bilinear interpolation were selected as the parameters, and the above image was corrected for geometric accuracy. The root mean square error was controlled within 0.5 pixels. Due to bias between images, the corrected image projection and coordinates need to be converted to a common system, which ensures accuracy verification of wetland classification and allows a comparison. Then, the images were cut and enhanced using the boundaries of the study area.

3.2 Determination of Wetland Classification System

Current wetland classifications according to Ramsar Convention and the National Technical Regulations for Wetland Resources Survey are widely used [10]. Human production activities in the delta zone of this study area are severe. In the wetland classification system, wetland formation dynamics are the dominant factor, and the hydrological, ecological and plant dominant communities of the wetland are comprehensively considered. A wetland classification system of two levels was adopted in the study. In the first level, wetland was classified as natural or artificial. Natural wetlands indicate that biotic and abiotic elements in the landscape were not affected or interfered with by human activities. The relationship between energy flow and material flow in the landscape is still a natural ecosystem, or human beings just transform and manage the natural ecosystem slightly, without changing its system or landscape type, for instance, by artificially managing a reed field. Artificial wetlands are landscapes that were strongly disturbed by human activities. In the artificial wetland, many landscape elements have been artificially changed, and humans have formed new and different ecosystems. Humans need to invest a lot of energy into the system and take out products such as paddy fields and reservoirs. In the second level, wetlands were classified by hydrological status (water accumulation) and landscape type. In combination with the actual situation of wetlands in Panjin city and the marks that can be interpreted by Tiangong-2 remote sensing images, a wetland classification system suitable for the study area was established (Table 2, Fig. 3).

Table 2. Wetland classification system in Panjin city.
Full size table
Fig. 3.
figure 3

Wetland classification system in Panjin city (Combination of 5/8/10 channel visible)

Full size image

3.3 Wetland Information Extraction

According to the established wetland classification system and the characteristics of the study area, spectral characteristics of ground objects were analyzed. Spectral differences of water bodies, vegetation, and mudflats are obvious (Fig. 4), making them easier to classified. Using partition extraction [11], the study area is initially classified into water bodies, vegetation, and mudflats, then these three types are further classified separately.

Fig. 4.
figure 4

Spectral curves of three kinds of ground objects

Full size image

According to Fig. 4, water bodies have a much lower reflection than the other two classes in band 6 (central wavelength 750 nm); this feature can be used to extract water bodies in the study area [12]. Vegetation has a distinct absorption in the red band and strong reflectance in the near-infrared band. The vegetation index is designed by combining the reflectance of the red band and the near-infrared band. The Normalized Different Vegetation index (NDVI) is widely used, and its formula NDVI = (NIR-R)/(NIR + R) is calculated by using band 4 in the near-infrared bands and the red band 8 of the Tiangong-2 data. NDVI can effectively separate plant growing areas, mudflat, and buildings. Based on the threshold of band 6 and NDVI, the study area was classified into three major areas. A detailed classification of the three types was done separately according to the technical flowchart in Fig. 5.

Fig. 5.
figure 5

Technology roadmap

Full size image

Water body information extraction.

Rivers, aquaculture farms, and reservoirs in the study area are the subjects of water classification. The spectral differences of water bodies are not significant, but their shapes are. For example, small rivers are curved strips and large lakes have regular geometric shapes. According to these features, the secondary classification of water can be carried out. The shape index [13] is an effective expressions of patch-shape features. The shape index indicates the degree of regularity of the patch. The more irregular the shape, the smaller the shape index is. The patch area of a water body can also effectively identify the type. For example, the patch area of some offshore waters is significantly larger than that of a pond. Therefore, confusing water body can be distinguished based on both shape index and patch area. Aquaculture farms can be extracted according to regular textures. Some rivers are discontinuous and cannot be identified by the shape index. In this regard, with the help of GIS tools, the preliminary classification of water body raster data is converted into vector data, and these water bodies are distinguished by visualization interpretation to improve the classification accuracy.

Extraction of vegetation information.

The vegetation information includes coastal wetlands (including reeds, Suaeda salsa, and paddy fields) and some green spaces in the city center. The spectral characteristics of typical vegetation can be used for secondary classification. The spectral characteristic curves of Suaeda salsa and reeds are shown in Fig. 6. The two types of ground objects can be identified by using the peak value of reeds in the green band (band 10) and the valley value of Suaeda salsa in this band, as opposed to their peak value in the red band (band 8). Due to water information, the NDVI can be combined with the low reflectance feature of band 6 for paddy field extraction.

Fig. 6.
figure 6

Spectral curves of reed and Suaeda salsa

Full size image

Beach information extraction.

Since the spectral characteristics of mudflats and buildings are very similar, when the mudflat are extracted above a certain threshold of NDVI, the extracted information mixed with a lot of information of the building, and a small amount of water information. Mudflats in the study area are mainly distributed near the estuary, and have obvious spatial distribution characteristics. Combining these features, manual visual interpretation is performed to improve classification accuracy.

4 Classification Results and Accuracy Verification

4.1 Area and Spatial Distribution of Various Types of Wetlands

In 2018, wetlands occupied 3184.89 km2 in the study area. As described in Table 3, paddy fields occupy the largest area, accounting for 55.29% of the total. In order of decreasing area, the others are reed swamp, mudflat wetland, Suaeda salsa swamp, aquaculture farm, reservoir/pond and river wetland.

Table 3. Area and proportion of various types of wetlands in Panjin city in 2018.
Full size table

Figure 7 also shows that Panjin city has a wide range of wetlands, accounting for about 80% of the entire study area. The regional resource development in Panjin city is centered on paddy field, reed fields, and aquaculture farms. Paddy fields are the most widely distributed, largely in the eastern part of the research area. Reed swamps are the main type of the wetland on the north shore of Liaodong Bay, distributed west of the Shuangtaizi River, with a small amount on the east side. The development of reed fields in this region is in line with ecological rules and has economic value. Suaeda salsa swamp are mainly distributed on both sides of river in the center of the study area. Mudflat wetlands are mainly distributed near the mouth of the Liaohe River. There are aquaculture farms along the coast, mainly because the estuary is suitable for this purpose. River wetlands mainly concentrated along the Shuangtaizi River in the center and the Daliao River in the southeastern part of the study area.

Fig. 7.
figure 7

Distribution of wetland types in Panjin city in 2018

Full size image

4.2 Classification Accuracy

Because Landsat remote sensing data were widely used in the landuse/landcover classification, as well as with higher spatial resolution, wetland data imaged by Landsat 8 on May 5th, 2018 were used to evaluated the accuracy of Tiangong-2 data for wetland extraction, using the confusion matrix method [14]. The confusion matrix was established to calculate overall classification accuracy. In addition, the Kappa coefficient of classification accuracy was calculated for various types of wetlands. The size of the Kappa coefficient indicates the consistency of classification results for different wetland types and the spatial distribution of real objects. When the Kappa coefficient is less than 0.4, the consistency degree is not ideal. When the Kappa coefficient is between 0.4 and 0.75, the degree of agreement between the two is ordinary. When the Kappa coefficient is greater than 0.75, it indicates a good consistency between the two [15].

Table 4 indicates that Tiangong-2 data are highly consistent with Landsat classification data. The accuracies of producer and user are mostly over 90%, indicating that the classification results are good. Among these, the extraction precision for paddy field is very high.

Table 4. Classification accuracy of all types of wetlands.
Full size table

5 Conclusion

In this study, the application of the Tiangong-2 wide-band imaging spectrometer to wetland monitoring was explored. Monitoring results show that in 2018, Panjin city has a wide distribution of wetlands, with paddy fields mainly in the east, and large areas of reed swamp distributed in the western of the study area. Overall wetland classification accuracy using Tiangong-2 data reached 89.88%, and the kappa coefficient was 0.8748. Compared with the traditional classification method using Landsat data, the partition classification method improves the classification accuracy to a certain extent, and achieves remote sensing extraction of wetland information in Panjin City.