Coastal Mangrove Response to Marine Erosion: Evaluating the Impacts of Spatial Distribution and Vegetation Growth in Bangkok Bay from 1987 to 2017

: Long time-series monitoring of mangroves to marine erosion in the Bay of Bangkok, using Landsat data from 1987 to 2017, shows responses including landward retreat and seaward extension. Quantitative assessment of these responses with respect to spatial distribution and vegetation growth shows differing relationships depending on mangrove growth stage. Using transects perpendicular to the shoreline, we calculated the cross-shore mangrove extent (width) to represent spatial distribution, and the normalized difference vegetation index (NDVI) was used to represent vegetation growth. Correlations were then compared between mangrove seaside changes and the two parameters — mangrove width and NDVI — at yearly and 10-year scales. Both spatial distribution and vegetation growth display positive impacts on mangrove ecosystem stability: At early growth stages, mangrove stability is positively related to spatial distribution, whereas at mature growth the impact of vegetation growth is greater. Thus, we conclude that at early growth stages, planting width and area are more critical for stability, whereas for mature mangroves, management activities should focus on sustaining vegetation health and density. This study provides new rapid insights into monitoring and managing mangroves, based on analyses of parameters from historical satellite-derived information, which succinctly capture the net effect of complex environmental and human disturbances.


Introduction
Mangroves, together with their associated environments, such as rivers, deltas, offshore mudflats, and sandy habitats, provide important ecosystem services to support feeding, breeding, and nursery areas for migratory shorebirds and fish species [1,2]. A key ecological and human service from mangroves is shoreline protection through eliminating or minimizing the erosion impacts of wind and wave forces [3,4]. However, with increasing pressure from climate change and the impact of human activities, coastal mangroves are also affected by marine erosion and show different spatial and temporal responses to it [5][6][7][8][9][10]. These responses to erosion accumulate over time to affect the

Study Area
The Bay of Bangkok (study area shown in Figure 1) is part of the Gulf of Thailand and its coastline comprises intertidal mudflats intersected by deltas and estuaries associated with the Chao Phraya River, the Tachin River, the Meckl River, and the Mamba River. Over time, dense coverages of mangroves along its coast have been destroyed by human settlements, aquaculture, and salt pans, so that the current distribution is a narrow strip of fragile mangrove ecosystems at the outermost edge of the land [34].
Through two interviews with the fish pond owner of what was once an original mangrove growing area, we were able to determine the following history of mangrove distributions and changes. With the population increasing, from 1961 the local government allowed private occupation and development of mangroves. Since the 1960s, mangroves in the Bay of Bangkok have experienced private and governmental deforestation and occupation [35], including aquaculture, shrimp farming, salt pans, and expansion of urban areas. The destruction was particularly serious on the coast from Samut Sakhon province to Chachoengsao province, where mangrove coverage was reduced to a minimum, with information showing that there was only 1,047,390 rai (A rai is a unit of area equal to 1600 square meters.) left in the Bay of Bangkok area around 1996. Since the 1990s, aquaculture has become less profitable than in the past due to damage to the coastal environment. Moreover, fluctuations in the price of aquaculture products in the market have led to the abandonment of many fish ponds. The local government began restoring mangroves around 2000 so that mangroves began to gradually enter the recovery phase.

Field Measurements and Investigations in the Study Area
Field investigations were carried out in the Bay of Bangkok during September 2018 over the mangrove protection areas and the fish pond areas adjacent to the mangroves. Data collected included locations of verification points, and interviews were conducted to determine the history of mangrove coverage and changes in the area including human disturbances. In the study area, the spectral difference between water body (including pond, salt field, and sea water) and mangrove is significant, while the spectral difference between mangrove and other vegetation is small, thus requiring field verification of the vegetation classifications. The main aim of the field investigations was determining the border between mangrove and other vegetation. The Global Positioning System (GPS) and Open Street Map (OSM) Tracker were used to record the position and land cover properties at the 38 field verification points, as shown in Figure 1.

Satellite Data Processing and Mangrove Mapping
For time-series analysis of the dynamic changes in mangroves, we used Landsat due to the availability of a long-term sequence of continuous data [36]. We used the 30-year Landsat collection [37] including sensors of TM\ETM+\OLI. The U.S. Geological Survey provided a Landsat archive, reorganized as a tiered collection structure, to ensure that the Landsat Level-1 products provide a consistently accessible stack of known data quality suitable for time-series analyses and data stacking. Data from 1987 to 2017 (except for 2012 due to the lack of data) of Landsat yearly data were processed via Google Earth Engine (GEE) [38].
Taking 2017 as an example, we generated 394 sample points in the study area, including 38 field collection points (as shown in Figure 1). Because of extensive changes in the mangroves over the 30 years, we used the sample points of 2017 as a standard, and changed the sample points according to the image year-by-year, before finally generating the 1987-2017 sample points. The sample points were divided into four classes (mangroves, water, bare land, and other vegetation), and the classification and counts in each class are shown in Table 1 below. We randomly selected 75% of these points in the sample point dataset as training points and 25% as verification points. In addition to the spectral bands of Landsat data (blue to Swir2), to distinguish between mangrove and other vegetation, three types of vegetation indexes were used: NDVI, enhanced vegetation index (EVI), and soiladjusted vegetation index (SAVI). For differentiating between mangroves and water bodies, the normalized difference water index (NDWI) was used in the classification. Finally, considering that mangroves grow in low-lying areas near the sea, elevation collected from SRTM Digital Elevation Data [39] was used.
We mentioned in the introduction that effective mangrove change detection over long timeseries requires removal of the tide influence, or by aggregating over periods much longer than the tidal period. To resolve the problem, we extracted the mangrove information from 1987 to 2017 using GEE according to the following process: (i) Generate yearly fusion images, which were used to classify mangroves of each year, by taking the median value of the annual Landsat images with less than 30% cloud in the GEE dataset; (ii) Use random forest to classify the four classes using the following bands: Blue, Green, Red, Nir, Swir1, Swir2, NDVI, EVI, SAVI, NDWI, together with Elevation; (iii) Verify the mangrove classification results with the field observations at the sample points. Furthermore, the proportion of common area of mangrove classification results of two published land cover or mangrove distribution datasets was calculated to verify the reliability (as shown in Table 2). Result of 2000 87.39%

Relationship Resolution
The relationship between mangrove stability and response to marine erosion is confounded by a number of factors, such as the direction of the coast relative to wave direction, coastline geomorphology, and sediment substrate [42]. At a smaller scale of mangrove patches, there are also spatial differences in marine erosion. Thus, it is difficult to establish and deterministically model the relationship between variables governing responses to marine erosion. For this study, we chose to investigate statistical correlations between parameters and yearly erosion distance, in order to provide a rapid and objective method from which we can derive abstract quantitative methods and clear association rules [43].
The flow chart, shown in Figure 2, graphically summarizes our method to quantify the relationship between mangrove responses to marine erosion, spatial distribution, and vegetation growth.

Kendall Correlation
Equally spaced with 30m interval; Perpendicular to the shoreline; Based on the same baseline;

Width-Change
Correlation coefficient

Cross-shore vertical lines
Indices extraction Output result Sampling rule

NDVI-Change
Correlation coefficient

Two time scales:
Yearly; 10-year Figure 2. A flow chart describing the data and methodology.

Sampling Rule and Indices Extraction
In order to account for changes in the intensity of marine erosion as a function of the bay's direction, the study shoreline was selectively divided into nine partitions based on shoreline direction and shape, the natural division caused by estuaries, and the observed mangrove distribution. These divisions of the study area are shown in Figure 1.
The methodology we employed was based on the following ideas: (i) Simplify the quantification method: To account for the directionality of mangrove erosion relative to the coast, we used equally spaced lines perpendicular to the shoreline to provide transect statistics for the sampled mangrove [44]. To ensure transects sampled mangrove for 30 years, the baseline was established offshore at an average distance of 1.27 km. Based on the same baseline, transects with a length of 5 km were established with samples at 30 m intervals along the transect. The schematic diagram in Figure 3 shows how the transects were established.
(ii) Control the variables spatially: Regional variations in the propensity of marine erosion were controlled by dividing the study area into nine units/sub-regions (Zone 1 to Zone 9, shown in Figure  1) in order to control marine erosion factors such as current, sea breeze, soil quality, and sea level rise in each unit. We used these units as the spatial basis to explore the different sub-regional responses of mangroves, and performed a 30-year time-series analysis of the average width, NDVI, and seaside values from the nine partitions.
Special considerations were required to define the following key parameters of this study: The mangrove response to mangrove erosion; the spatial distribution of mangrove ecosystem; and vegetation growth. First of all, the change of mangrove on the seaside, including retreat and growth (positive values represent growth seaward, and negative values represent retreat landward), were measured to represent the response to marine erosion. Considering the narrow strip distribution of mangroves in the study area, cross-shore mangrove extent was used to define mangrove spatial distribution characteristics, and NDVI was used for mangrove growth characteristics. Thus, the values sampled by the sampling lines included: The long-term sequence of mangrove change on the seaside; cross-shore mangrove extent; mean NDVI of each line, which represent mangrove response to marine erosion; the spatial distribution; and vegetation growth, respectively. The detailed descriptions of each index are provided in Table 3.  Table 3. Description and formulas for the parameters and indices used in the calculations in the paper.

Spatial distribution
Cross-shore mangrove extent The length of the mangrove width along the sample line.
Vegetation growth NDVI 1 ∑ The mean NDVI of the mangrove pixels ( ) along the sample line.
Response to marine erosion (RME) Change on the seaside −( 2 − 1 ) The difference between the length of mangrove in the first year ( 1 ) and the length at the end of the relevant year ( 2 ); where i = cell number, k = transect number, and y = year; y2 > y1.

Time-series regression of yearly width and NDVI changes from 1987 to 2017:
Time-series based on Landsat images provide an opportunity to observe and characterize relative trends in disturbance and resilience at a regional scale, by disturbance type and ecozone, and at a spatial level that is relevant to both forest management and science [45]. The interpretation of geographical phenomena at different time scales can represent procedural and phased characteristics [46]. In order to reveal the trend of mangrove width, NDVI, and changes on the seaside over 30 years, a yearlybased Loess regression analysis was used to examine temporal variations from 1987-2017 of mangrove width, NDVI, and responses to marine erosion in the nine mangrove ecosystems. Loess regression is a nonparametric technique that uses local weighted regression to fit a smooth curve, which can reveal trends and cycles in data [47].

Kendall correlation analysis at yearly and 10-year scale:
The relationship of mangrove Width-Change and NDVI-Change was computed using the Kendall rank correlation coefficient (often called Kendall's τ or tau)-a non-parametric test which measures the strength of the relationship between two variables [48]. In calculating the correlations, we used rank correlations based on Kendall's tau, which is often said to be robust in the sense of capturing patterns and being resistant to outlying observations [49]. The tau correlation coefficient (tau-CC) of Width-Change and NDVI-Change were calculated at yearly and 10-year scales in order to compare the relationship between parameters and marine erosion responses in different regions and growth stages.

Dynamics of Coastal Mangrove Change Across 30 Years
We began the analysis by comparing the overall changes of mangroves between 1987 and 2017 across the nine sub-regions. Our field observations revealed that the distribution of mangrove patches were small and strip-type. This led us to use spatial and temporal aggregation approaches in order to study local and regional changes. Local changes were estimated by the width and NDVI parameters along transects (Figure 4), and regional changes were examined from the trends of mangrove change from 1987 to 2017 for each partition ( Figure 5). These analyses provided both regional and fine-scale understanding of the mangrove changes across the bay and across time.    Table 3) is summarized in Table 4, which shows that the order of change is Zone 3 > Zone 1 > Zone 2 > Zone 8 > Zone 9, with the largest value in Zone 3 which grows 317.1 m seaward. The order of retreating was Zone 7 > Zone 5 > Zone 4 > Zone 6, with the largest retreating value in Zone 7, which retreats 142.8 m landward. The maps of coastal mangrove distribution in Figure 5 display an obvious spatial differentiation of mangrove response to marine erosion in the nine partitions: Mangroves retreated in three zones in the northern coast of the Bay of Bangkok (Zone 4, Zone 5, and Zone 7), while growth occurred in three zones of the west coast (Zone 1, Zone 2, Zone 3) and in two zones of the east coast (Zone 8 and Zone 9). Mangrove retreats were, in general, landward, apart from mangroves near the estuaries, which retreated to a lesser extent. Distribution and trends of NDVI show five zones where growth occurred (Zone 1-3 and Zone 8-9), and NDVI growth was relatively stable internally. Zones where retreat occurred were low in NDVI status, regardless of whether it was 1987 or 2017.
Comparative analysis of trends within and between the regions were analyzed from annual changes across 1987 to 2017 of three Loess-regression trends: (1) Change of mangrove boundary at the seaside; (2) width; and (3) NDVI. The trends of the width and NDVI are displayed in Figure 6, and the trend of coastal mangrove change on the seaside are shown in Figure 7. Regressions were computed across the 30 years between the annual change of mangroves on the seaside (in meters), the annual mangrove width (in meters), and the annual mangrove NDVI.  In summary, the different mangrove responses to marine erosion and its impact on width and NDVI across 30 years showed that: (i) Mangroves in Zones 1, 2, 3, 8, and 9 grew at a constant stable rate out to the sea; the mangroves in Zones 4, 5, and 7 retreated landward, with narrow width and poor growth state; (ii) The broad trends of width and NDVI change across the nine regions showed a consistent but fluctuating rise. For growing regions, mangroves with wider extent and higher NDVI show more significant growth out to the sea. For retreating regions, mangroves with wider extent and higher NDVI retreated shorter distances; (iii) The retreat rates of regions 4 and 5 gradually stabilized after 2011, with width and NDVI increasing, while region 4 stopped retreating in 2015-2017 and showed seaward growth.

Relationship between Width-Change and NDVI-Change
The 30-year time-series mangrove change in the Bay of Bangkok showed different responses to marine erosion, represented by periods of growth and retreat. In this section, we analyze those changes in relation to changes in mangrove width and NDVI by computing the Kendall-tau correlation coefficients (tau-CCs) between the Width-Change and NDVI-Change. The characteristics of the relationship above were analyzed at short and long-time scales corresponding to yearly and 10year scales, respectively. The tau-CCs were computed between mangrove width at those two time scales, and similarly for NDVI.

Correlations at the Yearly Scale
The annual tau-CCs, displayed in the heat map in Figure 8, shows that annual retreats or growth were small-as indicated by low values for tau-CCs. (i) tau-CCs: Overall, the tau-CCs for width are greater than those for NDVI, implying that the pattern of yearly changes in width are more consistent than those for NDVI; possibly also implying that changes in NDVI are more dynamic at yearly time-scales and less consistent; (

Correlations at the 10-Year Scale
Three periods were used to compute the correlations at the 10-year time scale : 1987-1997, 1997-2007, and 2007-2017. For these periods, tau-CCs were calculated for width and NDVI. Correlations between Width-Change and NDVI-change for the three periods are listed in Table 5.

Quantification Method for Long-Term Analyses
Our study focused on using satellite remote sensing in order to provide rapid and relevant information on different recovery and protection measures during different future stages of mangrove growth. Reviews of mangrove classification suggest that mangroves grow in tropical and subtropical intertidal zones where it is difficult to avoid errors caused by cloudy images and tidal flooding [19]. In addition, NDVI of mangrove, as a correlation factor of mangrove change calculated in our study, also has seasonal differences. Thus, long-term mangrove classification and impact comparisons requires removal of the uncertainty from tidal influence and NDVI seasonal changes.
In this study, the tidal influences were eliminated by aggregating and median-filtering Landsat data to annual images. We calculated the median value of all bands for the whole year of the Landsat collection Level-1 images by using only images with under 30% cloud cover. Mangroves were classified using the yearly images. At the same time, the NDVI from the yearly images was used, thus avoiding problems from seasonal changes.

The Impacts of Spatial Distribution and Vegetation Growth in the Various Growing Stages
In this study, we chose the yearly scale to display trends and relationships at short time scales and a 10-year scale for longer-term changes. Yearly changes showed low correlations, and we used the 10-year analyses to highlight the relationships between spatial distribution and vegetation growth in different long-term stages. Both the yearly and 10-year tau-CC results revealed that the positive effects of width were greater than those for NDVI.
The long-term sequence quantification showed that there were differences in growth and relationships between the mangrove parameters. The positive impacts of mangrove spatial distribution and vegetation growth status on stability varied in the different growing stages and zones: Zone 2 and Zone 3 were regions where mangrove width increased the most, and Zone 8 and Zone 9 had a large number of young mangroves with low NDVI. The relationship between width and response was more obvious during the high-speed recovery and development phase, 1987-2000, but that relationship declined with time. Hence, as mangrove width increased, in other words as mangroves grew in time, the positive effect of width decreased. Zone 1 had the widest mangrove forest of the nine regions, with the highest NDVI and second highest mangrove growth rate, but the positive effect of width was weak at the sixth rank. These results imply that, during the early growth stage, the width of the original mangrove ecosystem had a strong effect on stability, as opposed to NDVI.
The tau-CCs of NDVI-change (see Figure 8 and Table 5) and the 30-year NDVI trends (see Figure  6), show that strong growth of NDVI is consistent with high initial NDVI values. This implies that vegetation biomass in the original mangrove forest ecosystem displayed a strong positive impact on its later development when NDVI increases. For example, in Zone 1 mangroves grew significantly and had the highest NDVI. At the same time, the positive impact of NDVI was the highest among the nine regions. Additionally, Zones 2, 3, 4, 5, and 6 all had high tau-CCs during years with higher NDVI (1996-2003 and 2013-2017). In Zones 8 and 9, mangrove recovery was lagging behind Zones 2 and 3, reflected in the very low positive, and negative, impact of NDVI in the early stage of mangrove development. As the width of mangrove in these two regions gradually increased, the positive effects of NDVI appeared after 2010.
In summary, mangrove width rather than NDVI appeared to offer greater resistance to marine erosion. At the early growth stage of mangroves, width impacts were particularly prominent. As mangrove width increased, at a certain width with a high NDVI value, positive impacts of vegetation growth status, like health and density, appeared.
Mangrove ecosystems are a complex sea-land link system, and future research effort is needed to verify and extend the findings of this study by using multi-source data, multiple indicators, and models. For example, the stability of the mangrove ecosystem is not only related to the spatial distribution and vegetation growth parameters but also to its environmental factors such as climate, tidal fluctuation, sediment, and wave energy [50].

Conclusions
Our investigation and analyses of mangrove dynamics in the Bay of Bangkok show that two parameters most commonly used in forest ecosystem evaluation-spatial distribution and vegetation growth-affect mangrove dynamics differently for different growing stages. The main conclusions of the study are as follows: Broadly, both spatial distribution and vegetation growth of mangroves display positive impacts on their defensive ability to marine erosion.
Mangroves with small width and low NDVI appear to continuously retreat landward as part of a coastal squeeze phenomenon, while mangroves of wide width and high NDVI perform stable or seaward extensions.
The positive effect of the spatial distribution was greater than vegetation growth, especially for mangroves at an early growth stage. However, as mangroves mature and grow, the vegetation growth status becomes more relevant than spatial distribution.
Thus, overall, we find that the impact of spatial distribution is higher at the early growing stage, while the impact of vegetation growth is higher at mature growing stages. The implication is that at the initial stage of mangrove restoration or cultivation, planting should focus on width and area, while for mature growth stages, attention should focus on increasing the vegetation health and density to maintain ecosystem stability.
Our research, which uses readily available remotely sensed images, provides objective guidance on planting structure and coverage relevant to the management, restoration, and development of mangroves, especially for fragile mangrove ecosystems that are in need of urgent restoration in the Bay of Bangkok.
In future, we aim to extend the research to explore the relationship between environmental factors and parameters computed from remotely sensed images. Such studies are needed in order to develop predictive models of how mangrove extent and health will be affected by changes in environmental factors that affect marine erosion. A further aim of future research is to use high-resolution data to separate mangrove tree species in order to understand species-related responses to changes in the environment.
Author Contributions: All authors contributed to this manuscript: F.S. and C.H. conceived the research and collected all the data; F.S., H.X, and C.H. designed the experiment and drafted the manuscript; C.H., D.F., and Q.W. provided help with the language and reviewed the manuscript. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.