Translator Disclaimer
19 October 2020 Combing Sentinel-1 and Sentinel-2 image time series for invasive Spartina alterniflora mapping on Google Earth Engine: a case study in Zhangjiang Estuary
Author Affiliations +
Abstract

As one of the most threatening invasive alien species to mangroves in China, Spartina alterniflora (S. alterniflora) has broadly existed along the Chinese tropical and subtropical coasts. Monitoring S. alterniflora with remote sensing is urgent and requisite for scientific invasive plant control and management. However, given the spectral similarity between S. alterniflora and other wetland types, such as mud covered by algae and the optical image coverage gaps due to cloud and tidal inundation in coastal areas, accurate and timely mapping of S. alterniflora is challenging. Using the extended Jeffries–Matusita distance (JBh), we first explored the best time window for detecting S. alterniflora with satellite data in Zhangjiang Estuary, Fujian, China. Then we presented a hierarchical classification framework to alleviate the spectral confusion problem, combining cost-free Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 multispectral image time series on the Google Earth Engine platform. Specifically, we integrated the inundation frequency map derived from the SAR time series, elevation, and slope criteria to calculate the potential areas of S. alterniflora and mangroves, then used the random-forest classification algorithm to identify S. alterniflora, and finally refined the classified map with a yearlong water mask. The optimal time windows of one month, two months, and three months identified by JBh were January, November and January, and November, January, and August, respectively; we got the high classification accuracies with corresponding overall accuracies of 99.35%, 99.63%, and 99.63%, respectively. The results suggested classification accuracy could be improved with a wider temporal window, but would saturate with 3-month imagery. The generated 10-m mangrove and S. alterniflora maps of 2017 and 2018 clearly showed the relatively stable spatial pattern of mangroves and the rapid expansion of S. alterniflora. The thriving S. alterniflora in Zhangjiang Estuary suggests the necessity of high-frequency and large-scale monitoring of invasive species along the coastal estuaries of China.

1.

Introduction

Biological invasions pose a primary threat to global biodiversity and the economy.1 Invaders can entirely alter a species composition and ecosystem function through outcompeting native species. Invasive species also exert important socio-economic impacts, with economic losses estimated at more than 200 billion Chinese Yuan per year in China.2 Spartina alterniflora (S. alterniflora), one of the most threatening invasive alien species to mangroves in China, was the only coastal saltmarsh vegetation among the 16 most serious invasive alien species released by the State Environmental Protection Administration of China in 2003. Researchers suggested that, without intervention, S. alterniflora could gradually replace mangroves in China’s coastlines of moderate salinity.3 Methods used to control invasive S. alterniflora include biological, chemical, and mechanical controls, and they all require the invaders’ distribution information as the precondition.4 Accurate and timely mapping of S. alterniflora is urgent and requisite for the scientific invasive plant control and management.

Satellite sensors provide repeated and cost-effective observations over inaccessible areas, which supports efficient and reliable information for monitoring invasive S. alterniflora and complements traditional time-consuming and labor-intensive field surveys.5 Medium spatial resolution (10 to 30 m) optical satellite imagery has been extensively used to monitor S. alterniflora at national,69 regional, and local10 levels due to easy-access, low-cost, and long-time historical data archives. But it is still challenging to use optical satellite imagery for routine S. alterniflora monitoring. First, the commonly used Landsat satellite imagery has a moderate spatial resolution (30 m), which results in mixed pixel problems.11 Second, commercial high-resolution satellite imagery has a comparatively high cost and is generally acquired on demand rather than at regular temporal intervals.12 Third, persistent cloud cover and tidal inundation in tropical and subtropical coastal regions prevents the acquisition of high-quality optical data.11,13,14 Finally, the spectral similarity between S. alterniflora and mud covered by biofilm or algae leads to S. alterniflora misclassification.1517

Investigating the optimal temporal window for invasive plant detection and image selection is an efficient way to lessen the number of images required while maintaining accurate detection.18,19 Studies have indicated that the optimal temporal window for S. alterniflora detection varies from region to region. For instance, Ouyang et al. (2013)18 pointed out that the germination and early growth stage, as well as the flowering stage, were the best time windows to identify S. alterniflora through measuring the field spectra of S. alterniflora and neighboring saltmarsh plants in Shanghai Chongming Dongtan National Nature Reserve. Sun et al. (2016)19 classified saltmarsh vegetation species with high temporal Chinese HJ-1 satellite data (30 m) in Xinyang Estuary, Jiangsu, China, and suggested that November was a good time for the identification of S. alterniflora there. Liu et al. (2018)7 pointed out that early spring and winter are good times for S. alterniflora identification in the south of China. Assuming different phenological features of S. alterniflora in different latitudes,20 optimal timing assessment with local time-series data is necessary. Sun et al. (2016) used the C5.0 decision tree classifier and the importance ratio of variables method to identify the best period for S. alterniflora identification. Evangelista et al. (2009)21 and Liu et al. (2017)22 applied the Maxent model to study the important periods for invasive plant detection. Rajah et al. (2018) used classification accuracy as the criteria to determine the optimal season for detecting invasive American Bramble.23 Furthermore, different methods of optimal temporal window analysis, which have been applied for crop mapping,2427 such as the extension of Jeffries–Matusita (JM) distance26 and classification accuracy-based methods,24,25,28 could also be applied for invasive plant identification.

Although the combination of optical and synthetic aperture radar (SAR) imagery offers great potential for invasive plant remote sensing,23 few studies have integrated the use of time-series imagery to map invasive plants in complex coastal wetlands.10,23,29 Invasive plants usually have distinct phenological characteristics from native species in order to occupy ecological niches better.23 Reflectance variations due to phenological fluctuations in leaf pigmentation, water content, and structure provide opportunities for mapping invasive plants with optical image time series.23 SAR is a powerful tool for inundation extent mapping, providing valid measurements for cloudy and rainy days.30 SAR image time series can be used to estimate inundation frequency, an important factor determining saltmarsh zonation patterns.31 S. alterniflora is a perennial deciduous grass, while the biofilm or algae on mud is ephemeral. Thus SAR-based inundation frequency data could contribute to the discrimination of S. alterniflora from ephemeral biological structures on mud, potentially addressing the misclassification problem of S. alterniflora. Rajah et al. (2018)23 fused Landsat-8, Sentinel-2, and Sentinel-1 imagery at the feature level and detected a harmful invasive alien plant, American Bramble, with support vector machine algorithm. But S. alterniflora usually lives in intertidal wetlands, which are heterogeneous and dynamic. The difficulty of simultaneously collecting SAR and optical satellite imagery at low tide limits the application of the method by Rajah et al. (2018)23 in identifying S. alterniflora. Image fusion at the decision level,32 a hierarchical classification method integrating optical, SAR and other ancillary data, is required to regularly map S. alterniflora in a cost-effective manner.33

In recent years, the cloud-based platform Google Earth Engine (GEE), providing a multipetabyte catalog of remote sensing data and machine learning algorithms, offers a new choice for researchers interested in geospatial big remotely sensed data analysis.34,35 GEE has gained broad applications, and enabled time series analysis of wetlands3537 and mangroves,38,39 facilitating our ability to monitor S. alterniflora.

In this study, we choose Zhangjiangkou National Mangrove Nature Reserve (ZNMNR), a representative mangrove and S. alterniflora ecotone, as the study region to identify the invasive S. alterniflora using remote sensing imagery on the GEE. The objectives of this paper are: (1) to investigate the optimal temporal window to identify S. alterniflora with Sentinel-2 optical image time series; (2) to apply a hierarchical classification framework through integrating Sentinel-1 SAR and Sentinel-2 multispectral image time series to map mangroves and S. alterniflora; and (3) to monitor S. alterniflora and mangroves in Zhangjiang Estuary from 2017 to 2018.

2.

Materials and Methods

2.1.

Study Area

The ZNMNR is located at Zhangjiang Estuary, Fujian province, China (117°24′07″E∼117°30′00″E, 23°53′45″N∼23°56′00″N, Fig. 1). It has the northernmost and largest concentrated natural mangrove forests in China and has been included in the List of Wetlands of International Importance since 2008. S. alterniflora is distributed on the tidal flats in both the eastern and western parts of the reserve. The elevation range of the estuary is 6 to 8 m above the mean sea level. The climate is a subtropical maritime monsoon climate with an annual average temperature of 21.2°C. The maximum (minimum) temperature is 38.1°C (0.2°C). The annual average rainfall is 1714.5 mm, with the rainy season from April to September. The annual average relative humidity is 79%. The annual average sunshine hours are 2125.1 h.

Fig. 1

(a) The Sentinel-2A false-color image of the study area on January 12, 2018, the spatial distribution of the sample data (colored polygons), and the black rectangle denotes the study area (the band combination of the Sentinel-2A image is the NIR band, red band, and green band); (b) the location of Fujian province; and (c) the location of ZNMNR (the red polygon highlights the location of the study region).

JARS_14_4_044504_f001.png

2.2.

Data

2.2.1.

Satellite data

Cloud-free Sentinel-2 Level-1C top of atmosphere reflectance data40 in the GEE (ImageCollection ID: COPERNICUS/S2) with the highest quality and lowest tide level in each month from 2017 to 2018 were selected by visual inspection (Table 1). We used data from December 2017 to November 2018 for optimal temporal window selection analysis. The wide-swath multispectral imager of the Sentinel-2A/2B satellites has 13 spectral bands with 4 bands at 10 m, 6 bands at 20 m, and 3 bands at 60 m spatial resolution. This study used blue, green, red, near-infrared (NIR), and short-wave infrared (SWIR) bands. The spatial resolution of blue, green, red, and NIR bands is 10 m. The SWIR band was resampled at a 10-m pixel size.

Table 1

Information of the Sentinel-2 imagery used in this study.

SensorDateSensorDate
Sentinel2-B MSI2017-12-10Sentinel2-B MSI2018-08-17
Sentinel2-B MSI2018-01-12Sentinel2-B MSI2018-09-29
Sentinel2-B MSI2018-02-18Sentinel2-A MSI2018-10-01
Sentinel2-B MSI2018-03-23Sentinel2-A MSI2018-11-23
Sentinel2-B MSI2018-04-04Sentinel2-A MSI2017-01-02
Sentinel2-B MSI2018-05-19Sentinel2-B MSI2017-11-10
Sentinel2-B MSI2018-06-28
Sentinel2-B MSI2018-07-28

We also used Sentinel-1 C-band Ground Range Detected SAR data41 from GEE (ImageCollection ID: COPERNICUS/S1) and collected a total of 29 and 28 scenes covering the study region in 2017 and 2018, respectively. The data are of the Interferometric Wide swath mode with dual-band vertical transmit/horizontal receive (VH) polarization and a pixel size of 10 m.

2.2.2.

Digital elevation model and vector maps

We used the 30-m SRTM digital elevation model (DEM) data and the derived variable (slope) to mask out regions of high elevation or steep slope where S. alterniflora and mangroves are not likely to exist.38 We also applied the boundary vector data of the reserve to mask out regions outside of the reserve.

2.2.3.

Sample data

We used in situ data derived from the Global Geo-Referenced Field Photo Library38 and Google Earth very high-spatial resolution satellite images from 2017 to 2018 as the background reference to create regions of interests (ROIs) for algorithm training and accuracy assessment. Sample polygon ROIs of five land covers, namely mangroves, S. alterniflora, water, mud flats, and other vegetation, were delineated by visual interpretation. Note that other vegetation here corresponded to other vegetation species except S. alterniflora and mangroves in the ZNMNR, mainly distributed around aquaculture ponds. The number of sample polygons for training and validation of the random forest classifier in 2017 and 2018 are listed in Table 2. For the optimal temporal window analysis of different vegetation species in the study area, we independently collected another ROI dataset of mangroves, S. alterniflora, and other vegetation to limit the tidal inundation influence.

Table 2

The number of samples for random forest classifier in this study.

TypeMangrovesS. alternifloraWaterMud flatsOther veg.
2017 and 201840 (10.13)44 (1.64)15 (22.06)24 (5.68)22 (0.66)
Note: The first number is the polygon number, the second number in the bracket is the area of the sample (unit: hm2).

2.3.

Optimal Time Window Selection

We used the JM distance to measure the period-by-period separability for each pair of vegetation species and applied the extension of JM distance (JBh) to calculate the optimal time window when considering the separability of multiclasses, refer to Hao et al. (2014)26 for details. The JM distance was widely used to quantify the separability of different bands,18 different vegetation categories,19 different crops,26,42,43 etc. Hao et al. (2014)26 applied JM distance and JBh to learn how separability of different crops changes over the growing season and to calculate the optimal time period combination order for crop classification. As crops are also plants, the method from Hao et al. (2014)26 applies to the situation of detecting S. alterniflora from mangroves and other vegetation here. Normalized difference vegetation index (NDVI) is a measure of vegetation density and condition and is widely used for vegetation phenology studies.44 We used NDVI for optimal time window analysis26 here. We first calculated the JM distance of the collected ROI dataset every month for all pair-wise vegetation species and selected the single month with the highest JM distance. Then by comparing any one of the other months to the selected month, we calculated the JBh distance of all possible combinations and selected the combination with the largest JBh as the optimal time period. By repeating the process, we obtained the optimal month combination order, which can guide image selection for vegetation classification.

The phenological patterns over the course of a year are of particular interest for annual vegetation mapping in this study. To verify that NDVI values derived from 1-year imagery (Table 1) could represent the phenological trends of different vegetation species, we used 4-year Sentinel-2 imagery to produce phenological data for comparison. Based on image time series from 2016 to 2019, we calculated NDVI values of S. alterniflora, mangroves, and other vegetation. Then we removed the outliers due to factors such as clouds, cloud shadows, and tidal inundation by visual inspection of each image. Finally, we adopted the harmonic analysis of time series (HANTs) algorithm to get high-quality phenological data.39

2.4.

S. Alterniflora and Mangrove Detection

The proposed hierarchical classification framework mainly comprises three steps: (1) calculate the potential areas of mangroves and S. alterniflora; (2) apply random forest classifier to detect mangroves and S. alterniflora; and (3) use the yearlong water mask and the criterion of interaction with the sea to refine the distribution of mangroves and S. alterniflora (Fig. 2).

Fig. 2

Flowchart of mangrove and S. alterniflora detection.

JARS_14_4_044504_f002.png

2.4.1.

Detecting potential areas of S. alterniflora and mangroves

Features including tidal inundation and topography (elevation and slope) were used to calculate potential areas of S. alterniflora and mangroves, which mainly existed near coastlines inundated by tides. We used the Sentinel-1 SAR image time series in each year to produce the inundation frequency map FVH<1938 (e.g., Fig. 3 based on imagery in 2018). FVH<19<90% was applied to create the inundation mask where S. alterniflora and mangroves possibly exist. We used the criteria DEM<10  m and slope<10  deg to mask out uplands and steep-slope regions,38 and the criterion intersection with the inundation mask was applied to produce the potential areas of mangroves and S. alterniflora.

Fig. 3

The Sentinel-1 VH backscatter of S. alterniflora, mangroves, mud flats, and water in the study region based on all Sentinel-1 data in 2018. The dashed line represents 19  dB VH backscatter.

JARS_14_4_044504_f003.png

2.4.2.

Random forest classification

Up to now, the random forest classifier has been widely applied in different remote sensing research fields, such as land use and land cover, wetlands, and forests.15,37,4551 We used both spectral and texture features as classification inputs. The spectral features include the red, green, blue, NIR, and SWIR bands, normalized difference water index (NDWI),52 modified normalized difference water index (mNDWI), NDVI, enhance vegetation index (EVI),53 and land surface water index (LSWI).54 The texture feature is the standard deviation of NDVI in a box with a radius of 5 pixels. Note that before running the random forest classifier, the potential areas of mangroves and S. alterniflora were used to mask the spectral and texture features. As to the parameters for the random forest classifier, the number of trees was 100, and the number of variables per split was the square root of the number of variables. The sample data were randomly divided into training (70%) and validation (30%). We quantified classification accuracy using overall accuracy, confusion matrix, user accuracy, producer accuracy, and kappa coefficient.

To further investigate the optimal month combination order in the ZNMNR, we tested the 1-month (January), 2-month (January and November), and 3-month (January, November, and August) imagery in 2018 with the random forest classifier. When the optimal time window was greater than 1 month, the inputs of the random forest classifier were all Sentinel-2 images in the calculated optimal time window.

2.4.3.

Refining the S. alterniflora and mangrove map with the intersection-with-sea criterion

We assume vegetation that does not interact with the sea is not S. alterniflora or mangroves to further refine the classification results. Impervious surfaces were removed with the reserve boundary vector data and no sand beaches existed in the study region, thus we used frequency map FVH<19>80% to calculate the yearlong water body mask.38 We deleted mangrove patches and S. alterniflora patches with pixel numbers smaller than 1 to lessen problems such as too fragmented vegetation patches.

3.

Results

3.1.

Separability Analysis and Optimal Time Window Selection

Based on the mean and standard deviation of the monthly NDVI data (Fig. 4), we can see that the NDVI seasonal variation of mangroves was not as distinct as that of S. alterniflora and other vegetation, corresponding to the fact that mangroves are evergreen species. The NDVI value of mangroves was larger than those of S. alterniflora and other vegetation in almost every month, suggesting that the separation of mangroves from other two categories was much easier than the identification of S. alterniflora and other vegetation. We could infer from Fig. 4 that January and November were the optimal periods for S. alterniflora identification and separation from other vegetation in the ZNMNR. Comparing Figs. 4 and 5, NDVI time series derived from 12-month imagery here can approximately reflect the phenological curves of different plants, especially S. alterniflora, based on 4-year dataset. Our results also showed a similar phenological spectral behavior of S. alterniflora to that of Li et al. (2019).39 They demonstrated that months in winter and spring, such as January, were the most important months for the classification of mangrove species and S. alterniflora in Zhangjiang Estuary.39 Our results were in line with Li et al. (2019),39 suggesting that the workflow utilized here was reliable and could be applied to other regions.

Fig. 4

NDVI curves for the three categories (mangroves, S. alterniflora, and other vegetation). The nodes denote the mean values and error bars represent the standard deviations. (“Dec-17” means December in 2017).

JARS_14_4_044504_f004.png

Fig. 5

NDVI profiles of mangrove, S. alterniflora, and other vegetation established by HANTs based on Sentinel-2 imagery from 2016 to 2019. DOY represents day of the year.

JARS_14_4_044504_f005.png

We used the JM distance to quantify the differences between the three vegetation categories in the study region (Fig. 6). The JM distance ranged from 0 to 2, with excellent separability for JM distance values >1.9. The JM distance values between mangroves and S. alterniflora in every month from December 2017 to November 2018 were mostly >1.99 [Fig. 6(a)], suggesting that one scene in any month was sufficient for the discrimination of mangroves from S. alterniflora. The JM distance values between mangroves and other vegetation were >1.90 from January to May and from November to December, and deteriorated from June to October. These indicate the discrimination of evergreen mangroves from other vegetation is also feasible with one scene, especially in spring and winter in the ZNMNR. The JM distance values between S. alterniflora and other vegetation were >1.5 only in November and January and increased with NDVI data of more months [Figs. 6(a)6(c)]. Every vegetation class showed good separability at some time, indicating the possibility of high overall accuracy using multitemporal classification. This has been confirmed by previous studies about crop discrimination.42

Fig. 6

JM distance values of all pair-wise vegetation comparisons for (a) each month; (b) two months, one of which is January, and (c) three months, two of which are January and November using the ROI dataset; and (d) JBh distance values of one month, two months (one is January), and three months (two are January and November). (“Dec-17” means December in 2017). The blue line representing the JM distance values of Mangrove—S. alterniflora pair, overlapped with the red line representing the JM distance values of Mangrove—Other vegetation pair in (b) and (c).

JARS_14_4_044504_f006.png

As to the separability of multiclasses, the first two maximum JBh distance values of each month were 2.54 in January and 2.51 in November [Fig. 6(d)]. By comparing image data in January with other months, we found that the maximum JBh distance value was 2.66 in November. By combing image data in January and November with other months, we learned that the maximum JBh distance value was 2.70 in August. The maximum (minimum) JBh distance values were 2.54 (2.19) with 1-month data, 2.66 (2.60) with 2-month data, 2.70 (2.67) with 3-month data, respectively. The JBh variations with 2-month and 3-month data were much smaller compared with that of 1-month data. Thus, January was the optimal month; January and November were the optimal 2-month combination; January, November, and August were the optimal 3-month combination for the detection of S. alterniflora and mangroves in the ZNMNR. We did not calculate the optimal 4-month combination, because the separability did not increase much when adding other images (JBh increased by <0.1 from 2-month to 3-month combination).

3.2.

Classification Accuracy Assessment

The overall accuracies of 1-month (January), 2-month (January and November), and 3-month (January, November, and August) imagery in 2018 were 99.35%, 99.63%, 99.63%, respectively, suggesting that classification accuracy could be improved with a wider temporal window, but would saturate with 3-month imagery. We focused on the classification results with 1-month and 2-month data (Table 3). The producer accuracy and user accuracy of mangrove, mud flat, and water were >90%, whereas the user accuracies of other vegetation were lower than 90% with either 1-month or 2-month imagery. The classification accuracy of other vegetation was lower than the remaining land covers. The producer accuracies of mangrove and mud flats were improved from 1-month imagery to 2-month imagery, suggesting the identification of mangrove and mud flats benefits from adding seasonal signals with more images from different months.

Table 3

Accuracy assessment with 1-month image on January 2, 2-month images on January 2 and November 23 in 2018; and 2-month images on January 2 and November 10 in 2017 for the study region.

YearTimeTypeMangroveSAMud flatsWaterOVPA (%)UA (%)
2018January 2Mangrove264000498.51100
SAa049000100100
Mud flat001080298.1899.08
Water0006310100100
OV00101794.4473.91
Overall accuracy=99.35%, kappa coefficient=0.989
January 2 and November 23Mangrove267000199.63100
SAa04810097.96100
Mud flat001090199.0998.20
Water0006310100100
OV00101794.4489.47
Overall accuracy=99.63%, kappa coefficient=0.994
2017January 2 and November 10Mangrove2680000100100
SAa04810097.96100
Mud flat001080298.1890.76
Water0010621098.4299.84
OV00011794.4489.47
Overall accuracy=98.70%, kappa coefficient=0.978

aSA and OV represent S. alterniflora and other vegetation, respectively.

3.3.

Importance of Features

We calculated the variable importance scores for all features with a random forest classifier. The top 10 features of each image combination in 2018 were presented in Fig. 7. Implementing a random forest classifier with one image in January or November 2018, the two most important features were texture feature and NDVI. When two images in January and November 2018 were used in the model, the five most important features were January NDVI, January NIR, January NDWI, January red, and November texture bands, respectively. Using three images in January, August, and November 2018 in the model, the five most important features were January NDWI, January NDVI, August texture feature, January NIR, and January blue bands, respectively. Although we used the 11 features as inputs, the features with high importance scores changed when using imagery of different periods, especially when multiperiod imagery was used. The variable importance score also decreased when the number of features increased, which was due to how the importance was measured.27

Fig. 7

Feature importance scores of the 10 most important features for each image combination. (a) January in 2018; (b) November in 2018; (c) January and November in 2018; and (d) January, August, and November in 2018.

JARS_14_4_044504_f007.png

3.4.

Monitoring S. alterniflora and Mangroves in Zhangjiang Estuary in 2018 and 2017

Figures 8(a)8(d) present the inundation frequency map, inundation frequency mask, random forest classification map based on 3-month imagery, and yearlong sea water map in the study region in 2018, corresponding to the main steps of the proposed hierarchical classification framework. The mangrove and S. alterniflora maps based on 1-month, 2-month and 3-month imagery in 2018 are displayed in Fig. 9. Obviously, mangroves were distributed in the southwest of the study areas, whereas S. alterniflora were distributed on both sides of the north and south of the Zhangjiang River. The spatial distribution patterns of mangroves and S. alterniflora are similar based on data of different months in 2018 (Fig. 9). But comparing the enlarged views of the mangrove and S. alterniflora maps, we can see that classification results derived from 1-month imagery missed some small patches of S. alterniflora (the red circle in Fig. 9) compared with those derived from 2-month or 3-month imagery. Similarly, the areas of S. alterniflora and mangroves based on 1-month data in 2018 were 114.07 and 58.63  hm2, respectively, clearly underestimated in comparison with results from 2-month data and 3-month data (Table 4).

Fig. 8

(a) The inundation frequency map; (b) the inundation frequency mask; (c) the classified map using random forest with three-month imagery; and (d) the yearlong sea water mask in the study region in 2018.

JARS_14_4_044504_f008.png

Fig. 9

The distribution maps of S. alterniflora and mangroves based on 1-month, 2-month, and 3-month Sentinel-2 satellite imagery in 2018 and 2-month imagery in 2017. (a), (d), (g), and (j) are the overviews, whereas (b), (c), (e), (f), (h), (i), (k), and (l) are the enlarged views of the red rectangles on the left.

JARS_14_4_044504_f009.png

Table 4

Areas of mangroves and S. alterniflora in 2018 and 2017.

Year20182017
Data1-month2-months2-months
Mangroves58.6858.5258.61
S. alterniflora140.77139.60128.70

Comparing results based on 2-month imagery from 2017 to 2018 (Table 4), we can see that the mangrove area was relatively stable, whereas S. alterniflora area increased quickly. Verified in Figs. 9(d) and 9(j), S. alterniflora in the southwest of the study region occupied the whole mud flat area in 2017 and further enlarged in 2018. The red rectangles in Figs. 9(f) and 9(l) clearly presented the expansion of the S. alterniflora extent from 2017 to 2018.

4.

Discussion

4.1.

Potential of Time Series Sentinel-1 and Sentinel-2 Imagery to Monitor S. alterniflora and Mangroves

The areas of mangroves and S. alterniflora are not always in consensus with literature due to differences in classification methods and satellite images including the spatial, spectral, and temporal resolutions. Researchers presented that mangrove and S. alterniflora areas were 116.11  hm2 (Ref. 55) and 62.19  hm2 (Ref. 38) in 2015, and 117 and 60  hm2 in 2016.56 The magnitude of S. alterniflora and mangrove area in our study (Table 4) was comparable with previous research, and the invasion of S. alterniflora also contributed to the discrepancies of S. alterniflora area. By comparison with other studies, the expansion of S. alterniflora in both the southwestern and northwestern parts of the study area from 2015 to 2018 was obvious.55 The distribution of S. alterniflora was more scattered in comparison to the research of Chen et al. (2019)56 in 2016, probably because S. alterniflora expanded as time went on, and the spatial resolution of Sentinel-2 imagery is higher than that of Landsat imagery, contributing to the detection of small-area S. alterniflora clumps. S. alterniflora can reproduce both sexually and clonally by seed, rhizome, or vegetative fragmentation. Its seeds are dispersed by currents and tides, which can be transported long distances. We suggest that the clonal growth of S. alterniflora mainly contributed to its area increase57 in this study [red rectangles in Figs. 9(f) and 9(l)]. In Sec. 3.4, we found the area of the detected S. alterniflora based on 1-month data in 2018 was smaller than those derived from 2-month and 3-month imagery. The reasons for this might be that (1) January is not the growth month for S. alterniflora in the ZNMNR when biomass and leaf greenness is relatively low,58 leading to confusion between S. alterniflora and mud flat; (2) by adding seasonal signals with 2-month or 3-month imagery, the random-forest algorithm can identify S. alterniflora better; and (3) in the period of several months, S. alterniflora expanded.

Although different methods and datasets have been used to map S. alterniflora in the ZNMNR,39,55 our study outperforms them in terms of higher image spectral resolution and temporal resolution. Liu et al. (2017) used Google Earth RGB imagery, which might degrade the performance of the classification method. Their overall accuracies were no larger than 90%, and the S. alterniflora producer and user accuracies were smaller than 92%.55 Li et al. (2019) processed 24-month Sentinel-2 NDVI time series from 2017 to 2018 as a whole to map mangrove species and S. alterniflora,39 which might introduce S. alterniflora distribution bias due to its dynamic invasion process. The overall accuracy of their method was 84%, whereas the classification accuracy of S. alterniflora was 77%.39 Our study produced S. alterniflora maps with higher classification accuracy and shorter time intervals, contributing to the timely monitoring of invasive plants. As to the optimal time window analysis, period-by-period JM distance values efficiently demonstrate class separability through time.42 Different from other studies that used the Maxent model21 or classification accuracy criteria,24,25,28 the JM distance and JBh method can effectively provide the optimal time window results with less computation resources26,27 and is suitable for routine large-scale S. alterniflora monitoring.

4.2.

Inundation Frequency Map for Discriminating S. alterniflora from Ephemeral Biological Structures on Mud

Our study demonstrated the inundation frequency map derived from SAR time series could be used to effectively reduce the spectral confusion between S. alterniflora and ephemeral biological structures on mud flats. For instance, Fig. 10(a) presents a Sentinel-2 optical image on February 11, 2017, in which large patches of green ephemeral biological structures were visible. Figure 10(b) shows the corresponding inundation frequency mask. The red circles in Figs. 10(c)10(f) highlighted the nonpersistent biological structures. The inundation frequency mask effectively deleted areas where ephemeral structures existed and retained areas where S. alterniflora and other land covers persisted. In addition, from Fig. 10(g), the nonpersistent biological structures of S. alterniflora cannot be discriminated by NDVI, and the NDVI confusion between mud flats and S. alterniflora is evident.

Fig. 10

(a) The overall view of the original Sentinel-2 image; (b) inundation frequency mask; (c)–(f) the corresponding enlarged views of the two red rectangles in the overview figure; and (g) NDVI of different land covers, including S. alterniflora, mangroves, mud flats, water, nonpersistent biological structures, and other vegetation (OV) in the study region on February 11, 2017.

JARS_14_4_044504_f010.png

It should be noted that the mixed pixel problem, the spectral confusion between S. alterniflora, mud flats, and other vegetation is still a challenge. For example, the boundary between mangroves and S. alterniflora in Fig. 8(c) was mixed with mud flat and other vegetation. It is possible to solve the mixed pixel problem by detecting S. alterniflora with very-high-resolution multispectral remote sensed imagery, thereby contributing to identifying invasive S. alterniflora in the early invasion stage.

4.3.

Factors Affecting the Mapping Accuracy

Multisource images have uncertainties such as different spatial resolutions, scanning systems, and angular effects, affecting the detection accuracies of S. alterniflora. Our study used Sentinel-1 SAR imagery to calculate an inundation frequency map and yearlong water body, and used Sentinel-2 imagery for vegetation classification, which could limit the multisensor related uncertainties to some degree. The algorithm was also used on the GEE, which was efficient, reproducible, and recommendable. We do not need to do extensive visual interpretation of the entire satellite image except for the sample data collection.

The parameters of the method, such as features used in the random forest classifier, land cover types, thresholds, and data sources, can affect the mapping accuracy of S. alterniflora and mangroves. Löw et al. (2013) set the number of trees to a relatively high number of 500, and pointed out that a higher number of trees did not improve the performance of the random forest classifier.59 We tested the importance of this parameter by changing it from 20 to 50, 100, and 500, whereas other parameters were set the same in our study area. The overall classification accuracies based on 2-month imagery with the number of trees set as 20, 50, 100, and 500 were 99.35%, 99.63%, 99.63% and 99.63%, respectively, consistent with the viewpoints from Löw et al. (2013).59 In addition, we used 11 features including red, green, blue, NIR, and SWIR bands, NDWI, mNDWI, NDVI, EVI, LSWI, and one texture feature as inputs for the random forest classifier in this study. We did not apply feature selection, whereas most researchers tried to select the most important features from tens to hundreds of features.59 We tested our method with subsets of the feature set and 2-month imagery in 2018. According to the overall accuracies with different feature sets listed in Table 5, the 20-m SWIR band could degrade the classification accuracy to some degree in our relatively small test area. In addition, the water indices mNDWI and LSWI were related to the SWIR band. The mNDWI, LSWI, and SWIR bands were not among the four most important features (Fig. 7). On the other hand, researchers highlighted the contribution of the SWIR band in vegetation species discrimination,12,60 and mNDWI ranked fifth out of top 10 most important features based on imagery in November 2018 here. More features could not further improve the accuracy of the classification model in this study (Table 5). In the last step of the algorithm, we deleted mangrove and S. alterniflora patches with pixel number <1; researchers interested in smaller vegetation patches can ignore these criteria.

Table 5

The overall classification accuracies with different feature sets and 2-month imagery in 2018.

Feature setBlue, green, red, NIR, and SWIRBlue, green, red, and NIRNDVINDVI, NDWI, and texture feature
Overall accuracy (%)99.1699.6397.7798.88

It is worth noting that the premise of using our algorithm is that mangroves and S. alterniflora are affected by the tidal flood, and vegetation that is not connected to seawater is not considered as mangroves or S. alterniflora. However, in a few areas of the world, such as China’s Hainan Province, Australia, and Bangladesh, some mangroves live on land without a connection to the ocean, so our algorithm may not be applicable there.

5.

Conclusions

In this study, we estimated the optimal time window using the extended JM distance and mapped mangroves and S. alterniflora integrating Sentinel-1 and Sentinel-2 image time series with a hierarchical classification framework. Our main conclusions are summarized as follows. (1) Mangroves and S. alterniflora show clear differences on the monthly NDVI time series curves due to different phenology and coverages. January, November and January, and November, January, and August were the optimal time windows for the identification of S. alterniflora and mangroves in the ZNMNR. The 2-month imagery (January and November) could provide a very high classification accuracy, and additional images could not significantly improve the overall accuracy. (2) The proposed hierarchical classification framework produced annual S. alterniflora and mangrove maps with high overall accuracies of 98.70% and 99.63% based on 2-month imagery in 2017 and 2018, respectively. (3) The generated 10-m mangrove and S. alterniflora maps of 2017 and 2018 clearly showed the relatively stable distribution pattern of mangroves and the rapid expansion of S. alterniflora.

S. alterniflora thriving in the ZNMNR suggests the necessity of high-frequency and large-scale monitoring of estuaries along the coast of China. The proposed workflow and generated mangrove and S. alterniflora datasets here may provide resource managers, government officials, and ecologists guidelines for remote sensing data collection and processing for S. alterniflora monitoring.

Acknowledgments

This research was funded by the Natural Science Foundation of Guangdong Province of China (Grant No. 2018A030310032); Open fund of the Key Laboratory for National Geography State Monitoring (Ministry of Natural Resources of the People’s Republic of China) (Grant No. 2017NGCM08); Guangdong Key Lab of Ocean Remote Sensing (South China Sea Institute of Oceanology Chinese Academy of Sciences) (Grant No. 2017B030301005-LORS1806); the National Natural Science Foundation of China (NSFC) (Grant No. 41876205, 41906026); and the Key Special Project for Introduced Talents Team of Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou) (Grant No. GML2019ZD0302). The authors are grateful to the colleagues who participated in the data collection. Thanks go to Fangfang YAO for their feedback on the earlier version of the manuscript and to Yuming Tang for discussions.

References

1. 

W. V. Reid et al., Ecosystems and Human Well-Being-Synthesis: A Report of the Millennium Ecosystem Assessment, Island Press, Washington DC (2005). Google Scholar

2. 

F. Wan, X. Zheng and J.-Y. Guo, Biology and Management of Invasive Alien Species in Agriculture and Forestry, Science Press, Beijing (2005). Google Scholar

3. 

Y. Zhang et al., “Interactions between mangroves and exotic Spartina in an anthropogenically disturbed estuary in southern China,” Ecology, 93 (3), 588 –597 (2012). https://doi.org/10.1890/11-1302.1 ECGYAQ 0094-6621 Google Scholar

4. 

D. J. Emry, H. M. Alexander and M. K. Tourtellot, “Modelling the local spread of invasive plants: importance of including spatial distribution and detectability in management plans,” J. Appl. Ecol., 48 (6), 1391 –1400 (2011). https://doi.org/10.1111/j.1365-2664.2011.02050.x Google Scholar

5. 

B. Satyanarayana et al., “Assessment of mangrove vegetation based on remote sensing and ground-truth measurements at Tumpat, Kelantan Delta, east coast of peninsular Malaysia,” Int. J. Remote Sens., 32 (6), 1635 –1650 (2011). https://doi.org/10.1080/01431160903586781 IJSEDK 0143-1161 Google Scholar

6. 

P. Zuo et al., “Distribution of Spartina spp. along China’s coast,” Ecol. Eng., 40 160 –166 (2012). https://doi.org/10.1016/j.ecoleng.2011.12.014 Google Scholar

7. 

M. Liu et al., “Rapid invasion of Spartina alterniflora in the coastal zone of mainland China: new observations from Landsat OLI Images,” Remote Sens., 10 (12), 1933 (2018). https://doi.org/10.3390/rs10121933 Google Scholar

8. 

D. Zhang et al., “Introduction and spread of an exotic plant, Spartina alterniflora, along coastal marshes of China,” Wetlands, 37 (6), 1181 –1193 (2017). https://doi.org/10.1007/s13157-017-0950-0 Google Scholar

9. 

D. Mao et al., “Rapid invasion of Spartina Alterniflora in the coastal zone of mainland China: spatiotemporal patterns and human prevention,” Sensors, 19 (10), 2308 (2019). https://doi.org/10.3390/s19102308 SNSRES 0746-9462 Google Scholar

10. 

J. Ai et al., “Integrating pan-sharpening and classifier ensemble techniques to map an invasive plant (Spartina alterniflora) in an estuarine wetland using Landsat 8 imagery,” J. Appl. Remote Sens., 10 (2), 026001 (2016). https://doi.org/10.1117/1.JRS.10.026001 Google Scholar

11. 

X. Zhu et al., “Tidal and meteorological influences on the growth of invasive Spartina alterniflora: evidence from UAV remote sensing,” Remote Sens., 11 (10), 1208 (2019). https://doi.org/10.3390/rs11101208 Google Scholar

12. 

M. Karlson et al., “Assessing the potential of multi-seasonal WorldView-2 imagery for mapping West African agroforestry tree species,” Int. J. Appl. Earth Obs. Geoinf., 50 80 –88 (2016). https://doi.org/10.1016/j.jag.2016.03.004 Google Scholar

13. 

C. Tarantino et al., “Ailanthus altissima mapping from multi-temporal very high resolution satellite images,” ISPRS J. Photogramm. Remote Sens., 147 90 –103 (2019). https://doi.org/10.1016/j.isprsjprs.2018.11.013 IRSEE9 0924-2716 Google Scholar

14. 

A. D. Campbell and Y. Wang, “Salt marsh monitoring along the mid-Atlantic coast by Google Earth Engine enabled time series,” PLoS One, 15 (2), e0229605 (2020). https://doi.org/10.1371/journal.pone.0229605 POLNCL 1932-6203 Google Scholar

15. 

B. Proença et al., “Potential of high-resolution Pléiades imagery to monitor salt marsh evolution after Spartina invasion,” Remote Sens., 11 (8), 968 (2019). https://doi.org/10.3390/rs11080968 Google Scholar

16. 

A. Thomson et al., “The use of airborne remote sensing for extensive mapping of intertidal sediments and saltmarshes in eastern England,” Int. J. Remote Sens., 24 (13), 2717 –2737 (2003). https://doi.org/10.1080/0143116031000066918 IJSEDK 0143-1161 Google Scholar

17. 

C. Hladik, J. Schalles and M. Alber, “Salt marsh elevation and habitat mapping using hyperspectral and LIDAR data,” Remote Sens. Environ., 139 318 –330 (2013). https://doi.org/10.1016/j.rse.2013.08.003 Google Scholar

18. 

Z.-T. Ouyang et al., “Spectral discrimination of the invasive plant Spartina alterniflora at multiple phenological stages in a saltmarsh wetland,” PLoS One, 8 (6), e67315 (2013). https://doi.org/10.1371/journal.pone.0067315 POLNCL 1932-6203 Google Scholar

19. 

C. Sun et al., “Classification mapping and species identification of salt marshes based on a short-time interval NDVI time-series from HJ-1 optical imagery,” Int. J. Appl. Earth Obs. Geoinf., 45 27 –41 (2016). https://doi.org/10.1016/j.jag.2015.10.008 Google Scholar

20. 

M. L. Kirwan, G. R. Guntenspergen and J. T. Morris, “Latitudinal trends in Spartina alterniflora productivity and the response of coastal marshes to global change,” Global Change Biol., 15 (8), 1982 –1989 (2009). https://doi.org/10.1111/j.1365-2486.2008.01834.x Google Scholar

21. 

P. H. Evangelista et al., “Mapping invasive tamarisk (Tamarix): a comparison of single-scene and time-series analyses of remotely sensed data,” Remote Sens., 1 (3), 519 –533 (2009). https://doi.org/10.3390/rs1030519 Google Scholar

22. 

X. Liu et al., “Appling the one-class classification method of maxent to detect an invasive plant Spartina alterniflora with time-series analysis,” Remote Sens., 9 (11), 1120 (2017). https://doi.org/10.3390/rs9111120 Google Scholar

23. 

P. Rajah, J. Odindi and O. Mutanga, “Feature level image fusion of optical imagery and synthetic aperture radar (SAR) for invasive alien plant species detection and mapping,” Remote Sens. Appl., 10 198 –208 (2018). https://doi.org/10.1016/j.rsase.2018.04.007 Google Scholar

24. 

S. Meng et al., “Optimal temporal window selection for winter wheat and rapeseed mapping with Sentinel-2 images: a case study of Zhongxiang in China,” Remote Sens., 12 (2), 226 (2020). https://doi.org/10.3390/rs12020226 Google Scholar

25. 

J. E. Böhler, M. E. Schaepman and M. Kneubühler, “Optimal timing assessment for crop separation using multispectral unmanned aerial vehicle (UAV) data and textural features,” Remote Sens., 11 (15), 1780 (2019). https://doi.org/10.3390/rs11151780 Google Scholar

26. 

P. Hao et al., “The potential of time series merged from Landsat-5 TM and HJ-1 CCD for crop classification: a case study for Bole and Manas Counties in Xinjiang, China,” Remote Sens., 6 (8), 7610 –7631 (2014). https://doi.org/10.3390/rs6087610 Google Scholar

27. 

P. Hao, L. Wang and Z. Niu, “Potential of multitemporal Gaofen-1 panchromatic/multispectral images for crop classification: case study in Xinjiang Uygur Autonomous Region, China,” J. Appl. Remote Sens., 9 (1), 096035 (2015). https://doi.org/10.1117/1.JRS.9.096035 Google Scholar

28. 

T. G. Van Niel and T. R. McVicar, “Determining temporal windows for crop discrimination with remote sensing: a case study in south-eastern Australia,” Comput. Electron. Agric., 45 (1–3), 91 –108 (2004). https://doi.org/10.1016/j.compag.2004.06.003 CEAGE6 0168-1699 Google Scholar

29. 

R. Ismail, O. Mutanga and K. Peerbhay, “The identification and remote detection of alien invasive plants in commercial forests: an overview,” South African J. Geomatics, 5 (1), 49 –67 (2016). https://doi.org/10.4314/sajg.v5i1.4 Google Scholar

30. 

X. Shen et al., “Inundation extent mapping by synthetic aperture radar: a review,” Remote Sens., 11 (7), 879 (2019). https://doi.org/10.3390/rs11070879 Google Scholar

31. 

T. Holt et al., “Impact of storm tides and inundation frequency on water table salinity and vegetation on a Juvenile Barrier Island,” J. Hydrol., 554 666 –679 (2017). https://doi.org/10.1016/j.jhydrol.2017.09.014 JHYDA7 0022-1694 Google Scholar

32. 

C. Pohl and J. L. Van Genderen, “Review article multisensor image fusion in remote sensing: concepts, methods and applications,” Int. J. Remote Sens., 19 (5), 823 –854 (1998). https://doi.org/10.1080/014311698215748 IJSEDK 0143-1161 Google Scholar

33. 

L. Jiao et al., “A hierarchical classification framework of satellite multispectral/hyperspectral images for mapping coastal wetlands,” Remote Sens., 11 (19), 2238 (2019). https://doi.org/10.3390/rs11192238 Google Scholar

34. 

N. Gorelick et al., “Google Earth Engine: planetary-scale geospatial analysis for everyone,” Remote Sens. Environ., 202 18 –27 (2017). https://doi.org/10.1016/j.rse.2017.06.031 Google Scholar

35. 

M. Mahdianpari et al., “The first wetland inventory map of newfoundland at a spatial resolution of 10 m using Sentinel-1 and Sentinel-2 data on the Google Earth Engine cloud computing platform,” Remote Sens., 11 (1), 43 (2019). https://doi.org/10.3390/rs11010043 Google Scholar

36. 

J. N. Hird et al., “Google Earth Engine, open-access satellite data, and machine learning in support of large-area probabilistic wetland mapping,” Remote Sens., 9 (12), 1315 (2017). https://doi.org/10.3390/rs9121315 Google Scholar

37. 

E. R. DeLancey et al., “Comparing deep learning and shallow learning for large-scale wetland classification in Alberta, Canada,” Remote Sens., 12 (1), 2 (2020). https://doi.org/10.3390/rs12010002 Google Scholar

38. 

B. Chen et al., “A mangrove forest map of China in 2015: analysis of time series Landsat 7/8 and Sentinel-1A imagery in Google Earth Engine cloud computing platform,” ISPRS J. Photogramm. Remote Sens., 131 104 –120 (2017). https://doi.org/10.1016/j.isprsjprs.2017.07.011 IRSEE9 0924-2716 Google Scholar

39. 

H. Li et al., “Incorporating the plant phenological trajectory into mangrove species mapping with dense time series Sentinel-2 imagery and the Google Earth Engine platform,” Remote Sens., 11 (21), 2479 (2019). https://doi.org/10.3390/rs11212479 Google Scholar

40. 

SUHET, Sentinel-2 User Handbook, European Space Agency(2015). https://sentinel.esa.int/documents/247904/685211/Sentinel-2_User_Handbook Google Scholar

41. 

SUHET, Sentinel-1 User Handbook, European Space Agency(2013). https://sentinel.esa.int/documents/247904/685163/Sentinel-1_User_Handbook Google Scholar

42. 

T. G. Van Niel, T. R. McVicar and B. Datt, “On the relationship between training sample size and data dimensionality: Monte Carlo analysis of broadband multi-temporal classification,” Remote Sens. Environ., 98 (4), 468 –480 (2005). https://doi.org/10.1016/j.rse.2005.08.011 Google Scholar

43. 

T. Murakami et al., “Crop discrimination with multitemporal SPOT/HRV data in the Saga Plains, Japan,” Int. J. Remote Sens., 22 (7), 1335 –1348 (2001). https://doi.org/10.1080/01431160151144378 IJSEDK 0143-1161 Google Scholar

44. 

M. Forkel et al., “Trend change detection in NDVI time series: effects of inter-annual variability and methodology,” Remote Sens., 5 (5), 2113 –2144 (2013). https://doi.org/10.3390/rs5052113 Google Scholar

45. 

K. Millard and M. Richardson, “Wetland mapping with LiDAR derivatives, SAR polarimetric decompositions, and LiDAR–SAR fusion using a random forest classifier,” Can. J. Remote Sens., 39 (4), 290 –307 (2013). https://doi.org/10.5589/m13-038 CJRSDP 0703-8992 Google Scholar

46. 

S. Banks et al., “Wetland classification with multi-angle/temporal SAR using random forests,” Remote Sens., 11 (6), 670 (2019). https://doi.org/10.3390/rs11060670 Google Scholar

47. 

M. Mahdianpari et al., “Random forest wetland classification using ALOS-2 L-band, RADARSAT-2 C-band, and TerraSAR-X imagery,” ISPRS J. Photogramm. Remote Sens., 130 13 –31 (2017). https://doi.org/10.1016/j.isprsjprs.2017.05.010 IRSEE9 0924-2716 Google Scholar

48. 

C. Wang et al., “Decadal trend in agricultural abandonment and woodland expansion in an agro-pastoral transition band in Northern China,” PLoS One, 10 (11), e0142113 (2015). https://doi.org/10.1371/journal.pone.0142113 POLNCL 1932-6203 Google Scholar

49. 

C. Wang et al., “Spatially differentiated trends in urbanization, agricultural land abandonment and reclamation, and woodland recovery in Northern China,” Sci. Rep., 6 37658 (2016). https://doi.org/10.1038/srep37658 SRCEC3 2045-2322 Google Scholar

50. 

C. Wang, Q. Gao and M. Yu, “Quantifying trends of land change in Qinghai-Tibet Plateau during 2001–2015,” Remote Sens., 11 (20), 2435 (2019). https://doi.org/10.3390/rs11202435 Google Scholar

51. 

E. R. DeLancey et al., “The synergistic use of RADARSAT-2 ascending and descending images to improve surface water detection accuracy in Alberta, Canada,” Can. J. Remote Sens., 45 759 –769 (2019). https://doi.org/10.1080/07038992.2019.1691516 CJRSDP 0703-8992 Google Scholar

52. 

C. J. Tucker, “Red and photographic infrared linear combinations for monitoring vegetation,” Remote Sens. Environ., 8 (2), 127 –150 (1979). https://doi.org/10.1016/0034-4257(79)90013-0 Google Scholar

53. 

A. Huete et al., “A comparison of vegetation indices over a global set of TM images for EOS-MODIS,” Remote Sens. Environ., 59 (3), 440 –451 (1997). https://doi.org/10.1016/S0034-4257(96)00112-5 Google Scholar

54. 

X. Xiao et al., “Characterization of forest types in northeastern China, using multi-temporal SPOT-4 VEGETATION sensor data,” Remote Sens. Environ., 82 (2–3), 335 –348 (2002). https://doi.org/10.1016/S0034-4257(02)00051-2 Google Scholar

55. 

M. Liu et al., “Monitoring the invasion of Spartina alterniflora using multi-source high-resolution imagery in the Zhangjiang Estuary, China,” Remote Sens., 9 (6), 539 (2017). https://doi.org/10.3390/rs9060539 Google Scholar

56. 

Y. Chen et al., “Remote sensing monitoring of Zhangjiang Estuary Wetland,” J. For. Environ., 39 (1), 61 –69 (2019). https://doi.org/10.13324/j.cnki.jfcf.2019.01.011 Google Scholar

57. 

Y. Xiao et al., “Effects of salinity and clonal integration on growth and sexual reproduction of the invasive grass Spartina alterniflora,” Flora-Morphol. Distrib. Funct. Ecol. Plants, 206 (8), 736 –741 (2011). https://doi.org/10.1016/j.flora.2010.12.003 Google Scholar

58. 

G.-B. Ren et al., “Monitoring the invasion of smooth cordgrass Spartina alterniflora within the modern Yellow River Delta using remote sensing,” J. Coastal Res., 90 (sp1), 135 –145 (2019). https://doi.org/10.2112/SI90-017.1 JCRSEK 0749-0208 Google Scholar

59. 

F. Löw et al., “Impact of feature selection on the accuracy and spatial uncertainty of per-field crop classification using support vector machines,” ISPRS J. Photogramm. Remote Sens., 85 102 –119 (2013). https://doi.org/10.1016/j.isprsjprs.2013.08.007 IRSEE9 0924-2716 Google Scholar

60. 

J.-B. Féret and G. P. Asner, “Spectroscopic classification of tropical forest species using radiative transfer modeling,” Remote Sens. Environ., 115 (9), 2415 –2422 (2011). https://doi.org/10.1016/j.rse.2011.05.004 Google Scholar

Biography

Di Dong received her BS degree from Wuhan University in 2012 and her PhD in cartography and geography information system from the Institute of Remote Sensing and Digital Earth of UCAS in 2017. She received the CAS-DAAD scholarship and studied at Helmholtz Centre for Ocean Research Kiel from 2015 to 2017. She is a researcher at South China Sea Institute of Planning and Environmental Research, SOA. Her research interests include ocean remote sensing, mesoscale eddies, and internal waves.

Chao Wang received his BS and master’s degrees from Anhui Normal University in 2012 and 2015, respectively, and his PhD in environmental sciences from University of Puerto Rico at Rio Piedras in 2019. His research interests lie within the interdisciplinary research fields of ecology and hydrology, which bridge the disciplinary perspectives in understanding the ecohydrological processes and bi-directional interactions and feedbacks at different spatiotemporal scales through integrating multi-source observations and model simulations.

Jinhui Yan is a professor at South China Sea Institute of Planning and Environmental Research, South China Sea Bureau of Ministry of Natural Resources. His research interests include physical oceanography and remote sensing.

Qingyou He received his BS degree in marine science from Ocean University of China in 2011 and his PhD in physical oceanography from University of Chinese Academy of Sciences in 2017. He is a researcher at the State Key Laboratory of Tropical Oceanography, South China Sea Institute of Oceanology, Chinese Academy of Sciences. His research interests include remote sensing and oceanic mesoscale eddies.

Jisheng Zeng is a senior researcher at South China Sea Institute of Planning and Environmental Research, South China Sea Bureau of Ministry of Natural Resources. His research interests include surveying and mapping, remote sensing.

Zheng Wei is a senior researcher at South China Sea Institute of Planning and Environmental Research, South China Sea Bureau of Ministry of Natural Resources. He received his BS, MS degrees, and PhD in remote sensing from Wuhan University in 2006, 2008 and 2012, respectively. His research interests include remote sensing in coastal zones with LiDAR.

© The Authors. Published by SPIE under a Creative Commons Attribution 4.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.
Di Dong, Chao Wang, Jinhui Yan, Qingyou He, Jisheng Zeng, and Zheng Wei "Combing Sentinel-1 and Sentinel-2 image time series for invasive Spartina alterniflora mapping on Google Earth Engine: a case study in Zhangjiang Estuary," Journal of Applied Remote Sensing 14(4), 044504 (19 October 2020). https://doi.org/10.1117/1.JRS.14.044504
Received: 3 April 2020; Accepted: 29 September 2020; Published: 19 October 2020
JOURNAL ARTICLE
18 PAGES


SHARE
Advertisement
Advertisement
Back to Top