Dynamics in mangroves assessed by high-resolution and multi-temporal satellite data : a case study in Zhanjiang Mangrove National Nature Reserve ( ZMNNR ) , P . R . China

Mangrove forests are declining across the globe, mainly because of human intervention, and therefore require an evaluation of their past and present status (e.g. areal extent, species-level distribution, etc.) to implement better conservation and management strategies. In this paper, mangrove cover dynamics at Gaoqiao (P. R. China) were assessed through time using 1967, 2000 and 2009 satellite imagery (sensors Corona KH-4B, Landsat ETM+, GeoEye-1 respectively). Firstly, multi-temporal analysis of satellite data was undertaken, and secondly biotic and abiotic differences were analysed between the different mangrove stands, assessed through a supervised classification of a high-resolution satellite image. A major decline in mangrove cover ( −36 %) was observed between 1967 and 2009 due to rice cultivation and aquaculture practices. Moreover, dike construction has prevented mangroves from expanding landward. Although a small increase of mangrove area was observed between 2000 and 2009 (+24 %), the ratio mangrove / aquaculture kept decreasing due to increased aquaculture at the expense of rice cultivation in the vicinity. From the land-use/cover map based on ground-truth data (5 × m plot-based tree measurements) (August–September, 2009) as well as spectral reflectance values (obtained from pansharpened GeoEye1), bothBruguiera gymnorrhiza nd smallAegi eras corniculatum are distinguishable at 73–100 % accuracy, whereas tall A. corniculatumwas correctly classified at only 53 % due to its mixed vegetation stands with B. gymnorrhiza(overall classification accuracy: 85 %). In the case of sediments, sand proportion was significantly different between the three mangrove classes. Overall, the advantage of very high resolution satellite images like GeoEye-1 (0.5 m) for mangrove spatial heterogeneity assessment and/or species-level discrimination was well demonstrated, along with the complexity to provide a precise classification for non-dominant species (e.g. Kandelia obovata) at Gaoqiao. Despite limitations such as geometric distortion and single panchromatic band, the 42 yr old Published by Copernicus Publications on behalf of the European Geosciences Union. 5682 K. Leempoel et al.: Dynamics in mangroves assessed by satellite data in Gaoqiao, China Corona declassified images are invaluable for land-use/cover change detections when compared to recent satellite data sets.


Introduction
Mangroves provide a wide array of ecological and economic benefits (Dahdouh-Guebas and Koedam, 2006a;Dahdouh-Guebas et ab, 2006a;Nagelkerken et al., 2008;Walters et al., 2008).However, they are presently considered as one of the most threatened ecosystems in the world (Duke et al., 2007).In fact, mangroves once fringing ~7 5 % of tropical coasts (Chapman, 1976) are reduced to about 25% mainly due to human intervention (Rönnbäck, 1999); recent esti mates on global mangrove cover indicate between 137 760 and 152 000 km2 of mangrove forest remaining in 123 coun tries and territories (Giri et al., 2011;Spalding et al., 2010).In general, mangrove denudation at the rate of 2.1 % annu ally is much higher than the loss of tropical forests and coral reefs (Valiela et al., 2001).
This constant pressure on mangrove ecosystems under lines the demand for mangrove biogeographical data and vegetation maps (produced at species level and areal extent) that can ultimately be used by local authorities for better con servation and management practices (Masso i Aleman et al., 2010).With the difficulty of conventional monitoring tech niques in mangrove environments, the data obtained from aerial and/or satellite remote sensing sensors became essen tial, particularly when analysing vegetation history and dy namics (Dahdouh-Guebas and Koedam, 2008).The coarse resolution of most remote sensors (e.g.Landsat, SPOT, etc.) provides enough information to discriminate mangroves at the species level but only for large and homogeneous stands.In mangrove research, both spatial and spectral resolution need to be high to enable delineation of small patch size of certain species.In this context, modern sensors like IKONOS (1 m), QuickBird (2.4 m) and GeoEye-1 (0.5 m) provide very high resolution (VHR) images that enable spotting volumi nous single trees on the ground (Dahdouh-Guebas et al., 2004).Overall, remote sensing, combined with ground-truth observations in a GIS environment, remains time-saving as well as cost-effective for qualitative and quantitative assess ment of the mangrove vegetation (Dahdouh-Guebas, 2002;Dahdouh-Guebas et al., 2006b;Satyanarayana et al., 2011;Green et al., 2000).
The main objective of the present study was to analyse the spatial heterogeneity of different mangrove species at Gaoqiao (Leizhou Peninsula, China) using ground-truth and high-resolution satellite imagery (GeoEye-1 data).Firstly, a diachronic observation of three satellite images (obtained from Corona KH-4B, Landsat ETM+ and GeoEye-1) en abled the assessment of land cover change and mangrove extent over a 42yr period (1967-2000-2009).Secondly, abi otic and biotic factors influencing local mangrove distribu tion were compared between the different mangrove stands, assessed through a supervised classification of the very high resolution GeoEye-1 satellite image.
The study area is influenced by a northern tropical climate with the mean annual temperature of 23 °C and precipitation of 1500mm (Gao et al., 2009).The coldest and hottest days are during January and July, respectively.The rainfall, with uneven rates of precipitation, lasts from May to September.The tides are diurnal with a salinity between 20 and 23 %o, and pH from 7.6 to 7.8 (Liang and Dong, 2004).

Fieldwork
Data on mangrove vegetation, pore-water salinity and sedi ment characteristics were collected along nine transects (run ning perpendicular to the coast/creek line) in 5 m x 5 m area plots at 50m intervals (Fig. 1).A total number of 70 plots, located in the main stands of the high tide area, were studied between 6 August and 8 September, 2009.The geographi cal coordinates of all plots were obtained using a handheld global positioning system (Garmin, GPS 60 , USA) (accu racy < 6 m).In each plot, the vegetation data consisted of tree identification, height (m) and girth G 130 measurements (the girth at 130 cm height along the tree, which was subsequently converted into the diameter -D 130).In the case of trees smaller than 130 cm, girth was measured at 10 cm height (Goio)-This kind of situation was encountered mostly for small A. corniculatum, which formed pure stands and thus is not problematic for the class assignment method used.In addition, most A. corniculatum individuals were constituted of several stems deeply anchored in the mud, making the dis tinction between individual trees and stems almost impos sible.It was therefore decided to measure all stems instead in each plot.Based on these measurements, different tree structural parameters such as density (stems m-2 ), basal area (m2), relative density (%), dominance (%) (i.e.relative basal area), and relative frequency (%) were estimated using stan dard formulae (Cottam and Curtis, 1956;Cintron and Scha effer Novelli, 1984;Dahdouh-Guebas and Koedam, 2006b).The pore-water salinity (from a 20 cm deep hole dug into the ground) was measured at three different areas (randomly chosen) within each plot using a hand refractometer (Atago®, MASTER-S/Mill(Y, Japan).A soil sample (~ 250 g) was col lected from each plot and analysed for its textural content through hydrometer method (Bouyoucos 1962).Soils sam ples were analysed at the Soil Laboratory of the Agriculture College in Guangdong Ocean University.

Remote sensing
Three satellite images (namely from panchromatic Corona KH-4B, a recently declassified US military programme (dated 17 December 1967), spatial resolution: 1.8 m; the pan sharpened Landsat ETM+ (30 October 2000), spatial res olution: 15 m; and the pansharpened GeoEye-1 (16 Octo ber 2009), spatial resolution: 0.5 m) were used to identify changes in mangrove and adjacent land-use/cover (e.g.aqua culture) patterns.All images were projected under the same projection system (WGS_1984_UTM_Zone_49N).The panchromatic Corona image was obtained from the United States Geological Survey (USGS), and georeferenced using ground control points (GCPs) against the GeoEye-1 (ArcMap v. 9.3.1).Since it contains only single-band in formation, the supervised and/or un-supervised classification techniques were not supported.Instead, the mangrove area in 1967 was extracted by on-screen digitization.As the aqua culture ponds were not easily recognizable in 1967, the man grove/aquaculture ratio was calculated only for 2 0 0 0 and 2009.
Both GeoEye-1 (satellite image produced by GeoEye) and Landsat ETM+ (courtesy of the U.S. Geological Survey) images were subjected to pansharpening (applying Brovey transformation and nearest-neighbour resampling technique in ERDAS IMAGINE® v. 8.5) for clear identification of the features.In the case of recent ( 2009) data, the pansharp ened NDVI (normalized difference vegetation index) layer of GeoEye-1 was added to the pansharpened multispectral image of GeoEye-1 as a band.Supervised classifications of mangrove and aquaculture cover were carried out on both images using maximum likelihood classification to achieve better accuracy (Green et al., 2000).
Discrimination of current mangrove stands was only per formed on GeoEye-1 image.All aforementioned species were sampled during fieldwork; however, only B. gymnor rhiza and A. corniculatum formed large pure stands.Within mangroves, three different classes were defined by five man grove plots each through the training sets of the signature ed itor (Fig. 1) (Neukermans et al., 2008;Satyanarayana et al., 2011).Bruguiera gymnorrhiza training sites were defined by the five highest values of dominance while the two classes of A. corniculatum (i.e.tali A. corniculatum and small A. cor niculatum) were defined by high dominance as well as av erage tree height (five highest and five smallest respectively) (Table 1).Other classes in the supervised classification in cluded water, sand, open area (i.e.grassland), and aquacul ture ponds (in total seven classes including mangroves).Fi nally, the accuracy assessment of the land-use/cover map was carried out using confusion matrix with 113 GCPs scattered across mangrove and non-mangrove areas.

Correlation between dominant vegetation and abiotic factors
Information on soil texture and pore-water salinity for each plot was regrouped according to the mangrove class of the plot.One-way analysis of variance (ANOVA) was applied to test whether the means of the groups were equal, while Student's t tests (with Bonferroni correction) were used to compare two groups.For some data sets, however, conditions of application for using ANOVA were not observed.In such cases, the Kruskal-Wallis test substituted the ANOVA, and the Wilcoxon rank sum test replaced the Student's t test.All statistical tests were computed using R (R Development Core Team, 2010).3 Results

Mangrove cover dynamics
Between 1967 and 2009, mangrove degradation occurred along the landward sides of Gaoqiao (Fig. 2).The most per ceptible change was mainly observed on its southeast corner where a large mangrove stand has been converted into rice culture and aquaculture ponds, delimited by a new dike.Sim ilarly, mangroves in the north, adjacent to main water chan nels, were cleared for the same reason.Changes in the river course pattern were also evident at some places in the north (Fig. 2).Time-series data revealed that certain rice fields were replaced by aquaculture ponds between 2000 and 2009.
In this context, the dike limits remained the same with only those aquaculture developments on its landward side (Fig. 2).At the same time (i.e. between 2000 and 2009), the man grove / aquaculture ratio at Gaoqiao decreased despite a small increase (192 ha) of mangrove cover (Table 2).

Land-use/cover classification
Spectral reflectance values of the seven land-use/cover classes were separable (Fig. 3).Aquaculture and water classes were distinguished primarily by the blue and NDVI bands.In contrast, differences between open areas and sand classes were found to be small.For the mangrove classes, al though values were similar for the blue and green bands, their differentiation in the red, near infrared (NIR), and NDVI bands was considerably high.The highest value for NDVI was found for B. gymnorrhiza, followed by tali A. cornicula tum and small A. corniculatum (Fig. 3).

Mangrove distribution
In agreement with the aforementioned spectral characteris tics, the supervised classification (  3).
The sediments at Gaoqiao were predominantly of a clayeysilt nature (Table 4).Sand content was high at small A. cor niculatum sites, and it also revealed significant differences between the mangrove classes.In contrast, clay and silt con tent as well as salinity varied less and remained insignificant among classes.

Mangrove cover changes
Mangrove degradation has been reported substantially since increasing pressure of rice cultivation is exercised.Twothirds of the original mangrove coverage disappeared in the last 50yr, especially between 1960 and 1970s, due to defor estation, land reclamation for aquaculture or tourist resorts and urbanization activities (Li and Lee, 1997;Chen et al., 2009;Chen and Ye, 2011;Spalding et al., 2010;Ren et al., 2008).At Gaoqiao, 39.5% of mangrove cover was lost in the past four decades  for rice cultivation and aquaculture practices for shrimps and crabs (Fig. 2).After acquiring the Ramsar site of international importance, the dike which was constructed previously did not change, but the type of land use inside the dike area was shifted from rice cultivation to aquaculture ponds.Nevertheless, the neg ative impacts of aquaculture on mangrove vegetation and the booming of this industry during the 1980-1990s for eco nomic reasons are well known (Rönnbäck, 1999;Dahdouh-Guebas et al., 2002;Primavera, 1998;Hamilton et al., 1989).
In fact, in order to filter nitrogen and phosphorus loads within a mangrove ecosystem, several authors suggest a minimum sustainable ratio for mangrove and aquaculture areas pro posed at 2 -2 2 ha of forest per 1 ha of aquaculture, depend ing on the study (Robertson and Phillips, 1995;Costa-Pierce, 2002;Primavera et al., 2007).In the case of Gaoqiao, the mangrove/aquaculture ratio is < 1 and continued to decrease in 2009 (Table 2) due to aquaculture expansion.In addition, the actual mangrove cover (i.e.775 ha) estimated using 2009 GeoEye-1 satellite imagery was found to be much lower than the area (i.e.2000 ha) projected by Ramsar (2009).Therefore a revision of the area statistics should be considered.

Land-use/cover classification
In addition to B. gymnorrhiza and both tali and small A. cor niculatum, we also tried to provide supervised classes for S. apetala, K. obovata and the mixed mangrove stands of A. corniculatum-A.marina sampled on the southern transect, in addition to B. gymnorrhiza and both tali and small A. cor niculatum stands.However, due to its lower dominance and limited coverage, the accuracy of the classified map was poor with unrealistic species' distribution and therefore ignored.
Overall, the importance of NIR and NDVI bands for (dom inant) mangrove species discrimination at Gaoqiao was evi dent (Fig. 3).Despite of the same species, a clear-cut differ ence in the spectral reflectance of tali and small A. cornicula tum sites was remarkable.This could be due to both variation in their greenness/biomass and the dominance of B. gymnor rhiza in tali A. corniculatum class (18 %) compared to small A. corniculatum class (3 %).This latter hypothesis may also explain the misclassified plots of tali A. corniculatum in the B. gymnorrhiza class (Table 3).In addition to earlier records from mangrove regions close to Hong Kong, Yingluo Bay and Leizhou Bay (Tam et al., 1997;Ye et al., 2005;He et al, 2007;Ren et al, 2008;Gao et al., 2009;Ramsar, 2009), the present study highlights the dominance of A. corniculatum.However, previous studies do not report dominance of B. gymnorrhiza, which could be ex plained by different areas (even in the same region) support ing different species in relation to their size and geographic location (sea or landward), freshwater runoff and the extent of inundation in Gaoqiao.
For example, species like B. gymnorrhiza are usually char acteristic of interior sites away from main flooding channels (Ye et al., 2003;Satyanarayana et al, 2010), and this might be the reason for its abundance inside the forest (Fig. 4).
Kandelia obovata being more tolerant to water logging than B. gymnorrhiza (Ye et al., 2003) was found along small per manent watercourses and waterlogged areas.In fact, water channels in the study zone are often boarded by the tail A. corniculatum class, and it is in these areas that most K. obovata individuals exist.Open areas in the upper inter tidal and supralittoral zones were the last to be inundated, as was observed in situ, and this may explain the very low den sity of propagules and the absence of mangroves observed in these areas, indicating their limited transport by water.
The intriguing pattern of tail A. corniculatum and small A. corniculatum distribution observed along the convex and concave creek sides, respectively (Fig. 4), is likely to corre spond to accretion and erosion zones (Fig. 2 Bay.Therefore, a pioneer mangrove species like A. marina, which occupies the frontier edge of the tidal flat (i.e.low tidal flats) in China (Ye et al., 2005; Ren et al., 20 1 1 ), was found essentially in the southernmost transect.However, the salin ity in mangroves close to Gaoqiao was reported to be 23 %o (Liang and Dong, 2004).This important difference of mea surement could explain why no significant correlation was found with salinity.An important point is the observation of old A. marina sites in the northern part of the study area, which may reflect an ancient disjunct zonation of this species restrained by parasitism or by man.
Mangrove-associated species such as Cerbera manghas L. and Ipomoea pes-caprae L. have mostly occupied the ele vated grounds where tidal inundation is less frequent.They also colonize, mixed with some mangrove species (A.Cor niculatum, E. Agallocha), along the banks of rivers and small streams inland that supply the rice fields.This abrupt transi tion in zonation along the land-sea gradient inside the man groves and the presence of landward mangrove species both testify a restricted zonation in light of the artificial barrier (i.e.dike construction) at Gaoqiao.

Conclusions
The applicability of very high resolution imagery such as GeoEye-1 for mangrove spatial heterogeneity assessment and species-level discrimination, along with its difficulty to provide a precise classification for non-dominant species, has been demonstrated in the present study.If the mangrove stand size is considerable, it is possible to identify the same species (e.g. A. corniculatum) with different heights using GeoEye-1 NIR imagery even when present with different tree heights.In addition, the use of 42yr old Corona satellite im agery, compared to newly derived satellite data (e.g.GeoEye-1), allowed studying mangrove and other land-use/cover dy namics.While mangrove destruction between 1967 and 2000 was associated with the land reclamation for agriculture and aquaculture practices, conversion of agricultural fields into aquaculture ponds, between 2000 and 2009, was also respon sible for the decrease of mangrove/aquaculture ratio.The mangrove extent over 775 ha at Gaoqiao appeared to be de termined largely by its geographic (sea or landward) location.
The overwhelming dominance of A. corniculatum (which is known as "river mangrove") coincides with strong freshwa ter input in the vicinity.However, no environmental factors measured were able to discriminate B. gymnorrhiza and tali A. corniculatum stands.
Knowing the current status of mangrove distribution at Gaoqiao, we suggest that further studies (involving both re mote sensing and ground-truth assessments) should be fo cused on other mangrove patches in the south of the present study zone (Yingluo Bay) and other mangrove areas in ZMNNR (i.e.Beitan, Techeng, Taiping, Fucheng, Qishui, He'an and Tuii) for a complete monitoring and more efficient management.
Fig. 1. (A) Mangrove areas found in Leizhou Peninsula of southern China, and (B) pansharpened multispectral GeoEye-1 satellite (2009) imagery showing Gaoqiao mangrove cover facing the Gulf of Tonkin.The ground inventory has been carried out in transects (blue points) running perpendicular to the coast/creek line.Training sites used for the supervised classification of the three mangroves classes are indicated in yellow, red and green points.

Fig. 3 .
Fig. 3. Spectral reflectance (intensity of each band) of the landuse/cover classes used for supervised classification of Gaoqiao mangrove and adjacent areas.
Fig. 4) provided satisfac tory results.The interior position of B. gymnorrhiza could be noticed as a single patch with a few smaller ones nearby.Tali A. corniculatum was present in the northern part of the study zone and along the major water channels, whereas small A. corniculatum was mostly distributed in the peripheral margins and close to open areas.While aquaculture, open area and small A. corniculatum classes were represented by higher accuracy, others such as water and dike/sand, in cluding two mangrove classes (i.e.tali A.corniculatum and B. gymnorrhiza), were less well delineated.Tali A. cornicu latum had a low accuracy (53 %) owing to the fact that nearly seven plots were misclassified from B. gymnorrhiza to tali A. corniculatum (Table Fig. 4. Land-use/cover supervised classification of Gaoqiao based on the pansharpened multispectral GeoEye-1 (2009) image, and the fieldwork transects shown in Fig. 1.
), suggesting an important role of currents on the dynamics of habitats and vegetation regeneration.The low average salinity (8.5 %o) in the high tide area of Gaoqiao could be due to the present www.biogeosciences.net/10/5681/2013/Biogeosciences, 10, 5681-5689, 2013 sampling in the rainy season and also to constant freshwa ter input from Ximi and Qaoqiao rivers flowing into Yingluo

Table 1 .
Relative dominance and average height of A. corniculatum and B. gymnorrhiza in sample plots used to define mangrove classes for supervised classification (maximum likelihood classification) in ZMNNR, Gaoqiao, China.

Table 3 .
Confusion matrix showing accuracy assessment of the supervised land-use/cover classification for Gaoqiao, China.

Table 4 .
Sediment and salinity characteristics at three mangrove classes in ZMNNR, Gaoqiao, China.The average and standard deviation values as well as their significance level are indicated (a = 0.05 with Bonferroni correction for multiple test = 0.017).Significant results are bold.