INTRODUCTION
B ACKGROUND
Mangroves are unique ecosystems characterized by their ability to thrive in both land and saltwater environments, existing between tidal boundaries These vital coastal resources link terrestrial and marine systems, offering valuable ecosystem goods and services Dominating low-energy tropical and subtropical coastlines, mangroves play a crucial role in stabilizing shorelines by dissipating wave energy, trapping sediment, and serving as barriers against wind Additionally, they are essential for local communities, providing economic resources and contributing to ecological and social well-being by serving as nursery habitats for fish, crabs, and shrimp.
Mangrove forests represent the highest biodiversity among coastal wetlands, thriving in nutrient-rich, saline environments According to Aubreville (1970), mangroves, or "mangals," are found in tropical coastal areas along seas, lagoons, and riverbanks, where they are submerged in brackish or saltwater during high tides (Puri et al., 1989) These ecosystems consist of various evergreen trees and shrubs that share similar physiological traits and structural adaptations to coastal habitats and tidal dynamics, predominantly in tropical and subtropical regions (Syed et al., 2001) Additionally, mangrove forests play a crucial role in stabilizing coastlines by trapping sediments from rivers and land through their dense root systems, effectively preventing erosion.
Likewise mangroves not only importance role in ecosystem but also define an economic resource for the local communities (Rửnnbọck, 1999) For instance, just the
Many people are drawn to coastal regions for their economic opportunities and aesthetic appeal, as these areas offer numerous job prospects and recreational activities However, the coastal zones face significant pressures from both natural processes and human activities It is crucial to recognize that there are limits to the resilience of coastal ecosystems against external threats, particularly those stemming from human-induced actions, which pose a serious risk to their integrity.
The rapid expansion of aquaculture development is a significant factor contributing to the loss of mangrove forests in Southeast Asia, particularly in Vietnam In areas like Thai Binh, coastal development has led to the degradation of these vital ecosystems (Alongi, 2002) Consequently, mapping the distribution and extent of mangroves in Vietnam and beyond is crucial for their effective conservation and management.
Cost-effective methods are essential for reducing the labor-intensive process of manually calculating biomass Remote Sensing (RS) is recognized for its effective classification of mangroves, making the combination of RS and Geographic Information System (GIS) a suitable choice for biomass assessment (Sellers et al., 1995) Research by Christensen (1993) indicates that biomass can be evaluated by deriving light interception from spectral reflectance ratios Utilizing remotely sensed satellite data enables the calculation of biomass over large areas, saving both time and resources (Tripathi et al., 2010) This study focuses on integrating RS and GIS to estimate the spatial extent and rate of change of mangroves along the coastline of Thai Binh province, as well as assessing the above-ground biomass in the region.
P RIOR STUDY
Numerous studies have been conducted on mangrove forests, including Dat's 2011 research that utilized multi-temporal satellite data to monitor these ecosystems along Vietnam's Northern Coast Additionally, Pham Tien Dat's 2012 analysis focused on assessing the current status of mangroves in Hai Phong, Vietnam, using various ALOS sensors in 2010, while also comparing the accuracy of post-processing techniques for ALOS imagery.
3 mapping mangroves (Pham & Yoshino, 2012) The research about implementation of mangrove management investigated by the authorities, community or local people has affected mangrove change in Vietnam (Pham & Yoshino, 2016)
Beland (2006) highlights a proposed change detection methodology for assessing mangrove forest alterations due to aquaculture development and the effectiveness of mitigation measures against deforestation in Giao Thuy, Thai Binh, Vietnam, during the years 1986, 1992, and 2001 Additionally, Mazda (1997) demonstrates the importance of mangrove reforestation for coastal protection in Thai Binh province Furthermore, Nguyen Hai Hoa (2016) employs Landsat imagery and vegetation indices differencing to identify changes in mangrove ecosystems.
R OLE OF REMOTE SENSING AND GIS IN MANGROVE MONITORING
Satellite remote sensing has revolutionized global data collection, enabling continuous monitoring of the Earth This technology captures both biological and physical data, which is essential for forest inventory and environmental monitoring Additionally, integrating Global Positioning System (GPS) technology enhances the accuracy of ground truth data collection on the Earth's surface (Parkinson, 2003).
To effectively address critical environmental challenges, it is essential to generate relevant and current spatial information that enhances our understanding of these issues and aids in developing sustainable strategies Remote Sensing and GIS technologies play a crucial role in mapping and monitoring mangrove ecosystems (Green, Clark, Mumby, Edwards, & Ellis, 1998).
Remote sensing serves as a crucial alternative to traditional field monitoring for effective management of extensive mangrove ecosystems Key sources of remote sensing data for mapping mangroves include aerial photographs and high-resolution satellite imagery, while medium or low-resolution satellite data and laser scanning technology also contribute valuable insights for assessing these vital ecosystems.
4 the scientific literature, there are a considerable number of studies related to mangrove forests, remote sensing data and various image-processing algorithms
Remote sensing studies predominantly utilize high-resolution spatial images with pixel sizes ranging from 5 to 100 meters, which significantly influence the accuracy of mangrove forest mapping To enhance precision, it is crucial to select appropriate data sources and effective processing methods for mangrove forests However, pixel-based classification algorithms face limitations, as factors such as the misalignment of mangrove forests, non-mangrove vegetation, urban areas, and mudflats can adversely affect classification accuracy (Gao, 1998).
Remote-sensing techniques have proven to be highly effective in detecting, identifying, mapping, and monitoring mangrove conditions and changes (Green, 1998) In recent years, there has been a significant increase in climate change-related remote-sensing studies focused on coastal zones (Green et al., 1998) These techniques provide timely and accurate information essential for the sustainable management of wetland vegetation, as well as for discriminating and mapping various types of wetland plants and estimating their biochemical and biophysical parameters (Adam, Mutanga, & Rugege, 2010).
P ROBLEM S TATEMENT
Ongoing activities significantly impact coastal areas and mangroves, leading to long-term cumulative effects These coastal regions serve as critical interchanges between land and sea, hosting unique geological, ecological, and biological sites essential for diverse terrestrial and marine life, including humans (Beatley, Brower, & Schwab, 2002) Due to the variability of tectonic and terrain processes, coastal ecosystems are particularly fragile and vulnerable.
Vietnam's coastal regions are undergoing significant changes due to natural factors and human activities Mangroves, a sensitive and vulnerable ecosystem, face threats from environmental changes such as rising sea levels and alterations in hydrology (Mitra, 2013) Additionally, rapid population growth and migration to coastal areas have heightened the demand for resources, putting further pressure on these vital ecosystems.
Weak governance, poor planning, and uncoordinated economic development have exacerbated the loss of vital coastal ecosystems Since 1980, over 3.6 million hectares of mangroves have been lost globally, with Vietnam experiencing a dramatic decline in its mangrove forests from approximately 400,000 hectares in the early 20th century over the past 50 years (T Q Vo, Kuenzer, & Oppelt, 2015).
Remote Sensing (RS) technology offers a cost-effective and efficient means of obtaining updated land cover information, making it a crucial tool for land use change detection This application is particularly valuable for monitoring changes in mangrove ecosystems in Vietnam (Muchoney & Haack, 1994).
The Vietnamese government prioritizes the preservation of mangrove ecosystem services and a healthy environment While numerous studies in Thai Binh province have explored the value of mangrove forests, significant knowledge gaps remain It is crucial to update baseline mangrove data and identify species that are vulnerable or facing mortality, as well as to assess changes in drainage caused by urban and rural development.
Therefore, it is necessary to monitor mangrove forest, and mapping of mangroves is important in order to support coastal zone management and planning programs.
R ESEARCH O BJECTIVES
This research aims to map mangrove forests from 1998 to 2018 and assess the above-ground biomass in these ecosystems using allometric equations and remote sensing techniques.
The primary objective can be subdivided into following tasks:
Mapping mangrove forest and using RS and GIS and assess of mangrove forest change using Remote Sensing
Estimate amount of aboveground biomass by different vegetation index within study area
Assessing the accuracy of each AGB estimation model
Estimate the changing of aboveground biomass from 1998 to 2018 within study area.
O RGANIZATION OF THE T HESIS
The content of the research is structured under the following chapters:
Chapter I: Chapter 1 introduces the research work It highlights on prior research work based on mangrove above ground biomass The objectives of the research is highlighted within this chapter This chapter also show the problem statement and research question
Chapter II: Chapter 2 gives a theoretical and conceptual of mangrove Literature review on mangroves and further talks about climate change, effect of climate change to mangrove This chapter further researches on the various RS methods that have been employed in similar study
Chapter III: Chapter 3 gives the method about establish survey pots, collecting data, analysis data, estimate above ground biomass and change detection
Chapter IV: Chapter 4 shows the results obtained from the research Analysis and discussions are carried out on the result
The conclusions and recommendations drawn from the research are presented in chapter five
LITERATURE REVIEW
M ANGROVES
Mangrove forests, characterized by salt-tolerant trees and shrubs, thrive in intertidal areas and estuary mouths, effectively bridging land and sea These unique ecosystems are predominantly located in tropical and subtropical regions, specifically between 30°N and 30°S latitude, and host between 54 to 75 species adapted to saline environments Mangroves play a crucial role in coastal ecosystems, found primarily in sheltered coastlines, shallow lagoons, estuaries, rivers, and deltas, with the highest species diversity observed in Southeast Asia.
In 2000, the global area of mangroves reached 137,760 km² across 118 tropical and subtropical countries and territories Asia hosts the largest proportion of mangroves at 42%, followed by Africa at 20%, North and Central America at 15%, Oceania at 12%, and South America at 11% Notably, around 75% of the world's mangrove coverage is concentrated in just 15 countries (Giri et al., 2011).
In recent years, the mangrove forests in Thai Binh province have significantly declined in both area and quality, particularly between 1995 and 2000, due to the conversion of mangrove land into aquaculture farms Currently, the coastal land of Thai Binh spans 9,167 hectares, with 3,709 hectares designated as forest and 5,908 hectares as non-forest In low tide areas, soil composition reveals a sand content ranging from 83.64% to 86.57%, with some locations reaching as high as 98.32% Conversely, in high tide areas, the sand content in the soil varies between 39.19% and 43.69% (Đỗ Quý & Bùi Thế, 2018).
P HYSICAL FACTORS AFFECTING THE GROWTH OF M ANGROVES
There are some important biological and abiotic factor influence to develop of mangroves That factor formed specific characteristic of mangroves forest They include:
Mangrove ecosystems face significant threats from climate change, particularly due to rising sea levels, which are considered the greatest risk to their survival Current knowledge indicates that most mangrove sediment surface elevations are failing to keep pace with this rise, highlighting the urgent need for long-term studies across various regions The impact of rising sea levels will be most severe in areas where mangroves are experiencing a net decrease in sediment elevation and have limited opportunities for landward migration Adaptation strategies must be explored to enhance the resilience of these vital ecosystems.
The mean annual temperature along the South coast of Vietnam averages around 27°C, decreasing to approximately 21°C in the North The northeast monsoon brings cold air to the northern regions, impacting the growth and composition of mangroves (Lugo & Patterson-Zucca, 1977) Mangrove species thrive best in equatorial and subtropical areas with higher annual temperatures and a narrow temperature range, ideally between 25°C and 30°C, as seen in southern Vietnam Consequently, the number and size of mangrove species in the north are generally lower due to colder winter temperatures and high summer temperatures, which can adversely affect mangrove health.
Tropical forests predominantly thrive in equatorial regions characterized by high rainfall, typically ranging from 1800 to 2500 mm annually Precipitation plays a crucial role in the distribution of mangrove forests across various tidal zones While mangroves are salt-tolerant, they require a specific amount of fresh water for optimal growth Rainfall not only regulates soil and plant salt concentrations but also supplements fresh water sources, enhancing the physiological processes of mangroves.
Vietnam, there are about 100 rainy days per year with average rainfall of 1.500 to 2.000 mm and air humidity of less than 80%
During the summer months, Vietnam experiences heavy rainfall from southwest monsoons, creating ideal conditions for dense mangrove forests, particularly at Ca Mau Cape, which receives 2000-2200 mm of rain annually and has 120-150 rainy days In contrast, mangrove growth is limited along the small estuaries of the Khanh Hoa coast, where rainfall is less than 1000 mm per year.
Mangroves thrive in sheltered coastal areas where waves and tidal activity are minimal, as their propagules and seedlings need low-energy habitats for successful growth The distribution and size of mangrove forests are influenced by surface slope and tidal range; larger tidal ranges typically support more extensive mangrove growth (De Vos, 2004).
Survival rates of vulnerable species decline with reduced irradiance, particularly in low salinity conditions that increase sensitivity to high light levels In understorey shade, survival is lower in high salinity environments compared to low salinity However, these salinity effects can be mitigated by decreasing below-ground interactions with adult trees (Ball, 2002) For instance, Excoecaria agallocha thrives in low salinity areas (below 5 psu), but salinity levels between 5-15 psu hinder root growth, and levels above 15 psu prevent rooting altogether Additionally, high salinity negatively impacts leaf growth and area, resulting in shorter mangrove heights and smaller leaf sizes This increased salinity also reduces leaf longevity and reproductive capacity, ultimately leading to long-term mangrove mortality (Chen & Ye, 2014).
Soil condition is also effect to the distribution of dominate mangrove species (McKee, 1993) The condition for develop mangrove in the area with substrate,
Mangrove forests thrive best in silty clay soils, as indicated by Hong & San (1993) These soils are formed from alluvial sediments brought by rivers and the sea, rich in essential nutrients like magnesium and sodium The physical and chemical characteristics of mangrove soil vary based on the sources of alluvial and sediment deposits, which in turn influences the distribution of mangrove forests (Tam & Wong, 1996).
T HE A PPLICATION OF R EMOTE S ENSING IN MONITORING M ANGROVES
Recent research highlights the significance of remote sensing as a cost-effective tool for large-scale mangrove forest studies (Giri et al., 2011; Giri et al., 2007; Winarso et al., 2017) This technology is primarily utilized for change detection and monitoring of mangrove ecosystems Remote sensing involves collecting data about objects or areas without direct contact, often using satellites or radar Its effectiveness in covering extensive geographic regions makes it invaluable for tracking vegetation changes, particularly in forest research (Lillesand, Kiefer, & Chipman, 2014).
Monitoring changes in vegetation cover is crucial for planners at both local and global levels, as it provides valuable data for resource management and future planning This information helps evaluate ongoing vegetation changes and anticipate future trends According to Macleod et al (1998), effective change detection in natural resource monitoring involves four key aspects: identifying whether a change has occurred, understanding the nature of the change, quantifying the extent of the change, and assessing its spatial distribution (Macleod & Congalton, 1998).
Remote sensing (RS) and geographic information systems (GIS) play a crucial role in the sustainable management of tropical coastal ecosystems, particularly mangroves These technologies facilitate long-term studies that integrate historical and current data to forecast future conditions and identify necessary actions to prevent natural resource degradation Their application has been notably significant in the research of mangrove ecosystems (Ramachandra & Ganapathy, 2007).
11 study, high-resolution satellite image can be show the forest structure characteristic These results can be used to predict future changes in forest structure (Dahdouh- Guebas, 2001)
Remote sensing data plays a crucial role in detecting changes in mangrove forest cover by analyzing variations in radiance values The rapid advancement of computer technology has significantly enhanced the use of remote sensing images for change detection According to Coppin et al (1996), there are ten key techniques for detecting changes, including mono-temporal change delineation, delta or post-classification comparisons, multi-dimensional temporal feature space analysis, and composite analysis Additional methods include image differencing, multi-temporal linear data transformation, change vector analysis, image regression, multi-temporal biomass index NDVI, background subtraction, and image rationing.
Aerial photography (AP) and high-resolution image system as Landsat and sentinel are the most common approaches to mangrove remote sensing (Newton et al.,
Aerial photography (AP) has proven to be an effective and cost-efficient method for mapping and assessing mangroves, especially in smaller areas, compared to satellite remote sensing According to Anderson (1997), aerial photographs remain valuable for wetland mapping, as they are relatively inexpensive to analyze and can facilitate quick assessments of environmental changes (M D Spalding, Blasco, & Field, 1997).
Aerial mapping faces several limitations that can impact the quality of the final product, primarily due to the restricted areal extent and high costs associated with acquiring data for extensive geographic regions Additionally, factors such as sensor capabilities, the airborne platform used, environmental conditions, and the expertise of the information interpreter also play a significant role in shaping the outcomes of aerial mapping projects.
The vast majority of mangrove remote sensing studies have employed high- resolution satellite imagery such as Landsat (MSS, TM, or ETM+), SPOT (HVR,
HRVIR, ASTER, and IRS (1C or 1D) are key satellite imagery sources for detecting and classifying mangrove forests Effective methods include unsupervised classification techniques like the ISODATA approach, and supervised techniques such as maximum likelihood classification (MLC) and Mahalanobis distance, as well as hybrid schemes Recent advancements in these techniques enhance the accuracy of mangrove classification, enable the identification of individual species, and yield reliable structural estimates like leaf area, canopy height, and biomass.
2.3.3 GIS, Remote Sensing and Change Detection
The creation of thematic maps using Remote Sensing and Geographic Information Systems (GIS) offers significant advantages in terms of effectiveness and efficiency These techniques are crucial for urban growth studies, land use change detection, and vegetation analysis, such as NDVI In this study, GIS plays a vital role in detecting changes in mangrove forests by utilizing advanced GIS software and remote sensing tools, which can integrate various datasets effectively.
Remote sensing, as defined by Lillesand and Kiefer (2014), is the science and art of acquiring information about objects, regions, or phenomena through data analysis without direct contact This technology enables the collection of variable and updated data sources, particularly valuable for land cover information.
2.3.4 Mangrove biomass estimation by Remote Sensing and GIS
Biomass measurements from field data are highly accurate but impractical for large-scale assessments In contrast, Remote Sensing offers a significant advantage by delivering data over vast areas at a lower cost compared to extensive sampling methods, while also allowing access to hard-to-reach locations.
13 from Remote Sensing satellites are available at various scales, from local to global, and from a number of different platforms (Kumar, Sinha, Taylor, & Alqurashi, 2015)
Estimates of forest biomass are crucial for understanding carbon storage and cycling in forests, particularly in mangrove ecosystems Traditional remote sensing methods have been effective in monitoring changes in mangrove areas, and recent advancements in satellite technology have enhanced the accuracy of mangrove classifications and biomass estimates Research over the past three decades has highlighted the growing focus on remote sensing techniques for estimating aboveground biomass (AGB) in forest ecosystems For instance, Proisy et al (2007) utilized Fourier-based textural ordination to assess mangrove forest biomass using very high-resolution IKONOS images, employing a standardized principal component analysis (PCA) on Fourier spectra to compute texture indices This approach, combined with multiple linear regression on key textural indices, has demonstrated reliable predictions of total aboveground biomass in mangrove forests.
Simard (2006) utilized elevation data from the Shuttle Radar Topography Mission (SRTM), calibrated with airborne LIDAR data and a high-resolution USGS digital elevation model (DEM), to create a landscape-scale map of mean tree height in mangrove forests He then employed field data to establish a relationship between mean forest stand height and biomass, enabling the mapping of the spatial distribution of standing biomass in mangroves through linear regression analysis (Simard et al., 2006).
Fatoyinbo (2008) analyzed the spatial distribution of mean tree height and biomass in mangrove forests by utilizing Landsat ETM+ and Shuttle Radar Topography Mission (SRTM) data The SRTM data were calibrated with a land-cover map derived from Landsat images and height calibration equations To estimate aboveground biomass, stand-specific canopy height-biomass allometric equations, developed from field measurements and existing published equations, were employed.
14 mangrove forests on a landscape scale (Fatoyinbo, Simard, Washington‐Allen, & Shugart, 2008)
In his 2016 survey, Lu examines current biomass estimation methods utilizing remote sensing data, highlighting four key issues: the collection of field-based biomass reference data, the extraction and selection of appropriate variables from remote sensing data, the identification of suitable algorithms for biomass estimation models, and uncertainty analysis to enhance estimation accuracy He also addresses the influence of scale on biomass estimation performance and outlines a general procedure for biomass estimation While optical sensors and radar data have been the primary sources for above-ground biomass (AGB) estimation, data saturation poses a significant challenge, leading to increased estimation uncertainty (Lu et al., 2016).
METHOD
S TUDY AREA
The study area includes the province of Thai Binh, located in northeastern coastal Viet Nam
Thai Binh is an eastern coastal province located in the Red River Delta, approximately 110 km from Hanoi, 70 km from Hai Phong, and 18 km from Nam Dinh It shares borders with Hai Duong, Hung Yen, and Hai Phong to the north, Nam Dinh to the south, Ha Nam to the west, and the Gulf of Tonkin to the east Characterized by its flat terrain with a slope of less than 1 percent, Thai Binh's elevation gradually decreases from the north to the south, ranging from 1 to 2 meters above sea level.
In administrative border, over natural land area of province, nowadays there is above
In Thai Thuy and Tien Hai districts, 16,000 hectares of coastal land have been surveyed and are currently being developed for aquaculture and afforestation This includes over 4,000 hectares dedicated to aquaculture and the planting of 7,000 hectares of salt-marsh forest.
This study examines Thai Binh province in Vietnam, characterized by its tropical monsoon climate and significant heat radiation, resulting in high average temperatures ranging from 23°C to 24°C These conditions are conducive to the growth of mangrove ecosystems The seasonal thermal amplitude reaches 13°C, with temperatures dropping below 20°C during three months, particularly in January and February when the lowest temperatures can fall even further.
5 o C This factor will be effect to the development of mangrove (Cúc, 2013)
The average annual rainfall ranges from 1,500 to 1,900 millimeters, with peak precipitation occurring in August and September, which is below the optimal levels for mangrove growth (Yinxia, 1995) During winter months, rainfall dips below 30 mm per month, while the average humidity remains high, between 85% and 90%.
The Tonkin Gulf influences the plain with a diurnal tide, exhibiting a tidal range of approximately 4 meters Each day features one high tide and one low tide, while each month includes one spring tide and one neap tide The tidal range gradually decreases from north to south and from the sea towards the inland rivers, although the variation is minimal due to the short distance between the estuary ends The highest recorded water level at Hon Dau (Hai Phong) was 2.66 meters above mean sea level in October 1955, while the lowest was -1.62 meters in January 1969 (Cat & Duong, 2006).
3.1.4 Mangroves forest in Thai Binh Province
3.1.4.1 Status mangroves in Thai Binh Province
The mangrove forests in Thai Binh province, although limited in area compared to the total province, play a crucial role in the food chain, coastal protection, and local economic value Most of the mangroves in this region are plantations, with a low percentage of natural mangrove areas that are unevenly distributed The majority of these mangroves have been established through funding from international organizations, while only a small portion has been supported by the Vietnamese government (Cúc, 2013).
Thai Binh mangrove forest distributed in the coastal area of 10 communities
17 belong to Thai Thuy and Tien Hai district The mangrove area in Thai Thuy district is 2000ha and in Tien Hai district are 1400 ha (Thụy et al., 2016)
The coastal area of Thai Binh is home to a diverse range of 12 plant species, which include Acrostichum aureum, Acathus ebracteatus, Acathus ilicifolus, Sensuvium portulacastrum, Avicennia marina, Lumnitzera racemosa, Derris trifoliata, Excoecaria agallocha, Aegiceras corniculatum, Bruguiera gymnorrhiza, Kandelia obovata, Rhizophora stylosa, and Sonneratia caseolaris (Cúc, 2013).
3.1.4.2 Effect of climate to mangroves in mangroves forest
There are some climate factor that effect to the development of mangrove forest are:
The winter season, spanning from December to February, brings frigid temperatures, particularly in January when temperatures can drop below 15°C and occasionally reach an absolute minimum of under 5°C During this period, mangroves experience a slow growth rate, and some may even perish due to the harsh cold.
Storms and tropical depressions significantly impact the mangroves in Thai Binh, where wind speeds can reach 40-50 m/s and waves can rise to 5-7 meters during landfall These extreme weather conditions, especially during high tides, lead to severe consequences such as the destruction of mangrove trees, alterations in salinity levels, and submersion of seedlings.
D ATA COLLECTION
In this study, we collected two type of data field survey data and satellite image data to detect mangrove change and estimate above ground biomass
The following list of instruments used for the fieldwork and the software used for this study (see Table 1)
Table 1: Instrument and Software are used
1 Instrument GPS: Garmin 7 channel Collecting ground truth coordinates
2 Instrument Diameter Tape Diameter Measurement
3 Instrument Measuring tape 50 meter Length of measurement
4 Field Datasheet Recording field data
5 Software Arc GIS 10.2 Image processing and data analysis, Spatial analysis Principal Component Analysis
6 Software MS Word For documental
7 Software MS Excel Data analysis
8 Software Envi 5.3 Image pre-processing and data analysis, classification data
This study utilized satellite images sourced from the United States Geological Survey (USGS) Global Visualization Viewer (GLOVIS), including Landsat and Sentinel images, available at no cost The images were captured in the years 1998, 2003, 2007, 2013, and 2018, as outlined in Table 2 The Landsat images, with a resolution of 30 meters, were obtained from the Landsat satellite constellation, focusing on the area of interest (AOI) within the World Reference System (WRS) path 126 and row 46, corrected to level 1-T The sensors on the Landsat satellites record the surface reflectance of electromagnetic radiation from the sun across seven distinct bands, detailed in Tables 3 and 4.
Sentinel 2 image was obtained from a constellation of two satellites, both orbiting Earth at an altitude of 786 km and they had a resolution of 10 meters The research was based on a decadal analysis of images but due to lack of clear images of cloud cover less than 10% SENTINEL-2 data are acquired on 13 spectral bands in the VNIR and SWIR The satellite image in this study was used in this research describe below:
Table 2: Satellite Images Used in Research
No Date of image acquisition Satellite Resolution Path/row
10x10 Sources: https://earthexplorer.usgs.gov/
3.2.2.1 Landsat 5 Thematic Mapper (TM) and Landsat 7 Enhanced Thematic Mapper
The Landsat Thematic Mapper (TM) sensor operated on Landsat 5 from July 1982 to May 2012, featuring a 16-day repeat cycle aligned with the Worldwide Reference System-2 Notably, there was a significant reduction in image acquisition from November 2011 to May 2012.
2012 The satellite began decommissioning activities in January 2013
Landsat 5 TM image data files consist of seven spectral bands (See Table 3) The resolution is 30 meters for bands 1 to 7 (Thermal infrared band 6 was collected at
120 meters, but was resampled to 30 meters.) The approximate scene size is 170 km north-south by 183 km east-west (106 mi by 114 mi) (Chander, Markham, & Barsi,
The majority of Landsat 5 TM scenes are processed using the Level 1 Product Generation System (LPGS) to achieve full Precision Terrain correction; however, some TM scenes lack the required ground-control or elevation data for these corrections.
Landsat 5 Thematic Mapper (TM) scenes held in the USGS archive can be searched using EarthExplorer, the USGS Global Visualization Viewer (GloVis), or the LandsatLook Viewer On EarthExplorer, Landsat 4-5 TM scenes can be found under the Landsat menu in the ―Landsat Collection 1 Level-1‖ section, in the ―Landsat 4-5
The Landsat Enhanced Thematic Mapper Plus (ETM+) sensor onboard the Landsat 7 satellite has acquired images of the Earth nearly continuously since July
Since its launch in 1999, Landsat 7 has operated on a 16-day repeat cycle However, all scenes collected after May 30, 2003, have experienced data gaps due to the failure of the Scan Line Corrector (SLC), leading to these images being classified as SLC-off The Landsat 7 Enhanced Thematic Mapper Plus (ETM+) captures eight spectral bands, with spatial resolutions of 30 meters for bands 1 to 7 and 15 meters for the panchromatic band 8 These bands can be collected in either high or low gain settings to enhance radiometric sensitivity and dynamic range, while Band 6 is unique in that it collects both gain settings for every scene The approximate size of each scene is significant for various applications.
The ETM+ sensor captures scenes measuring 170 km north-south and 183 km east-west (106 mi by 114 mi), generating approximately 3.8 gigabits of data per scene It features an Instantaneous Field Of View (IFOV) of 30 meters x 30 meters for bands 1-5 and 7, while band 6 has an IFOV of 60 meters x 60 meters, and band 8 offers a more detailed IFOV of 15 meters For comprehensive information on the spatial characteristics of ETM+, please refer to the L7 Science Data Users Handbook (Heckenlaible, Meyerink, Torbert, & Lacasse, 2007).
Table 3: The Band Designations for Landsat 5 Thematic Mapper (TM) and
Landsat 7 Enhanced Thematic Mapper Plus (ETM+)
Band 4 - Near Infrared (NIR) 0.77-0.90 30 Band 5 - Shortwave Infrared (SWIR) 1 1.55-1.75 30
Source: (Barsi, Lee, Kvaran, Markham, & Pedelty, 2014)
The Landsat 8 satellite, launched in February 2013, is equipped with the Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) This satellite captures Earth images every 16 days, following the Worldwide Reference System-2 The OLI sensor features spectral bands that enhance the capabilities of previous Landsat instruments, including the Landsat 7 ETM+ sensor.
Landsat 8 introduces two new spectral bands: a deep blue visible channel (band 1) for enhanced water resources and coastal zone analysis, and an infrared channel (band 9) for cirrus cloud detection The satellite features two thermal bands (TIRS) that provide data at a minimum resolution of 100 meters, which is registered and delivered alongside the 30-meter OLI data product Consequently, the file sizes of Landsat 8 are larger than those of Landsat 7, owing to the inclusion of additional bands and an improved 16-bit data product.
Table 4: The Band Designations for the Landsat 8 Satellites
Band 1 - Ultra Blue (coastal/aerosol) 0.435 - 0.451 30
Band 5 - Near Infrared (NIR) 0.851 - 0.879 30 Band 6 - Shortwave Infrared (SWIR) 1 1.566 - 1.651 30 Band 7 - Shortwave Infrared (SWIR) 2 2.107 - 2.294 30 Band 8 - Panchromatic 0.503 - 0.676 15
Band 10 - Thermal Infrared (TIRS) 1 10.60 - 11.19 100 * (30) Band 11 - Thermal Infrared (TIRS) 2 11.50 - 12.51 100 * (30) Source: (Barsi et al., 2014)
Launched on June 23, 2015, as part of the European Commission's Copernicus program, SENTINEL-2 is designed to provide extensive data and imagery It features an opto-electronic multispectral sensor that operates at a resolution of 10 to 60 meters across visible, near-infrared (VNIR), and short-wave infrared (SWIR) spectral zones, utilizing 13 spectral channels to effectively capture variations in vegetation state.
The mission utilizes two satellites operating at an average altitude of 785 km, enabling repeated surveys of the Earth's surface every 5 days at the equator and every 2-3 days at middle latitudes This approach effectively captures temporal changes while minimizing the impact on atmospheric photography quality.
Table 5: Wavelength Regions and Description of Each Sentinel Band
B12 20 2190 180 Snow / ice / cloud discrimination Source: (sentinel.esa.int, 2018)
Fieldwork is essential for research, as it enables the collection of crucial ground information for mapping purposes The primary goal of sampling is to acquire ground truth data and assess the Above Ground Biomass (AGB) of the study area Ground truth data were gathered using a Garmin handheld GPS for training and accuracy evaluation, while Google Earth images facilitated visual inspections Additionally, tree measurement data were collected through a random selection method within the study area (Cornelius, Sear, Carver, & Heywood, 1994).
Ground truth data refers to information collected on-site to validate satellite image pixels within a specific area This process involves comparing pixels from Landsat or other satellite images to actual objects on the ground, ensuring accurate verification of image data.
23 the case of classified images it will help to determine the accuracy of the image performed by remote sensing software
Transect lines were established throughout the study area using a Garmin 7-channel GPS to accurately record the locations of different habitat types However, the marshy terrain limited access to certain areas of the mangrove forest, preventing a uniform distance for site selection along the transect line.
3.2.3.2 Sample Size and Sampling Techniques
Sampling plots can vary in size and shape, including square, circular, or rectangular forms, with circular plots being the preferred choice in forest inventory due to their ease of establishment As noted by Wenger (1984), circular plots require only a single defined point—the center—allowing for straightforward measurement of the radius In this study, a circular plot measuring 1000 m² with a radius of 17.8 m was established from the center.
D ATA ANALYSIS
Figure 4: Diagram of Research Workflow
All Landsat data underwent pre-processing to enable inter-comparison, normalize values, correct atmospheric effects, and minimize noise This involved radiometric calibration, the generation of multispectral data, subsetting analysis, gap-filling analysis, and cloud masking The specific pre-processing techniques utilized in this study are detailed below.
Atmospheric correction is essential for eliminating scattering and absorption effects caused by the atmosphere, allowing for accurate surface reflectance characterization This process is crucial for generating image data used in the classification and monitoring of Earth's surface (Song et al., 2001).
The brightness value of an object is affected by various factors, including scene illumination, atmospheric conditions, instrument response characteristics, and viewing geometry Fortunately, these influences can be mitigated using specific tools or applications designed for this purpose.
26 though the below two step:
Landsat Calibration was utilized to transform digital numbers from Landsat TM and ETM+ into spectral radiance or exoatmospheric reflectance, employing established post-launch gains and offsets.
) The spectral radiance (L λ ) is calculated using the following equation:
QCAL is the calibrated and quantized scaled radiance in units of digital Numbers
LMIN λ is the spectral radiance that is scaled to QCALMIN in watts/(meter squared*ster*àm)
LMAX λ is the spectral radiance that is scaled to QCALMAXin watts/(meter squared*ster*àm)
QCALMIN is the minimum quantized calibrated pixel value in Digital Numbers
QCALMAX is the maximum quantized calibrated pixel value in Digital Numbers
The second step involved calculating the top of atmosphere (TOA) reflectance for each band, addressing illumination variations such as sun angle and Earth-sun distance This calibration is applied on a pixel-by-pixel basis for each sensor (Chavez Jr).
1989) The TOA reflectance of the Earth is computed according to the equation:
Where: ρλ= Planetary TOA reflectance [unitless] π= Mathematical constant equal to ~3.14159 [unitless]
Lλ= Spectral radiance at the sensor's aperture [W/(m2 sr μm)] d= Earth–Sun distance [astronomical units]
ESUNλ= Mean exoatmospheric solar irradiance [W/(m2 μm)] θs= Solar zenith angle [degrees]
3.3.2 Filling the Gaps of Landsat 7 ETM+ image
On May 31, 2003, the Scan Line Corrector (SLC) of Landsat 7 failed, leading to a permanent loss of its functionality This malfunction caused the Enhanced Thematic Mapper Plus (ETM+) to capture images in a zig-zag pattern along the satellite's ground track, resulting in duplicated areas and an increasing width towards the edges of the scenes.
The Landsat 7 ETM+ continues to provide valuable image data even with the SLC turned off, particularly in the central regions of scenes This "SLC-off" mode maintains high radiometric and geometric quality similar to pre-failure data, although it results in an estimated 22 percent data loss per scene The maximum data gaps along the image edges can reach up to one full scan line, approximately 390 to 450 meters wide.
On October 21, 2003, a Landsat 7 ETM+ image exhibited scan line errors due to a sensor malfunction To rectify this issue, the "Gaps Filling tool" in ENVI 5.3 software was employed This tool is designed to work with standard Level 1 terrain corrected (L1T) GeoTIFF format images from Landsat The gap filling process utilizes two sources of images: one working image and two reference images (usgs.gov, c2018).
To effectively utilize this method, multiple SLC-off images are necessary, and each image's individual bands must be gap-filled prior to creating a 3-band composite For example, to gap-fill Image 1 using Image 2, a mosaic of Band 1 from both images must be created Once the bands are combined, they can be stacked to form the RGB image The gap-filling process using ENVI 5.3 will be detailed below.
- Open the tif band files to be used
- Select Import -> Import Files and Edit Properties Click Open to choose the files you want to gap-fill; they will populate the left-hand frame
To adjust the Background See Through-Data Value, select a file and set it to 0 If necessary, perform colour balancing to correct any brightness discrepancies between the images.
- Select File -> Apply, and assign an output file name and select other applicable options
Figure 5: Landsat 7 Image (Band 4, 3, 2) Received On October 21 th 2003 before and After Gap Filling
Cloud and cloud shadow are common feature of visible and near infrared satellite image in the world especially in tropical and subtropical (Martinuzzi, Gould,
Masking clouds and cloud shadows is crucial for accurately mapping land surface attributes These elements pose significant challenges in land cover change analysis, as clouds can be misinterpreted as false changes This misrepresentation can lead to an overestimation of changes, ultimately diminishing the accuracy of land assessments.
In their 2010 study, Huang et al utilized a cover map but faced limitations in replacing masked areas with pixels from corresponding cloud-free images To address this, they employed the mosaicking tool in ENVI 5.3 for the automated placement of georeferenced images into a cohesive output mosaic The analysis aimed to identify and eliminate remaining cloud and cloud shadow cover This was achieved by interpreting true color images using a red, blue, and green (RGB) band combination, where cloud areas were marked in white These identified clouds were then compared against a clear image to accurately distinguish between cloud and non-cloud regions.
C LASSIFICATION
Classification is the process of recognizing patterns associated with each pixel in an image, reflecting the characteristics of objects or materials on the Earth's surface (Syed et al., 2001) This study utilized multispectral subset data from the multi-temporal Landsat series, including TM, ETM+ SLC-off, SLC-off gap-filled, and OLI_TIRS, for classification analysis.
This study utilized supervised classification techniques to analyze Landsat imagery for land use and land cover determination, employing composite bands that represented a false color combination.
Supervised classification methods in satellite image analysis rely heavily on input from an analyst, known as a training set, which is crucial for the accuracy of these methods The effectiveness of supervised classification is significantly influenced by the quality of the training samples, which come in two types: one for classification and another for assessing classification accuracy Additionally, supervised classification encompasses functionalities such as input data analysis, training sample creation, signature file generation, and evaluation of the quality of both training samples and signature files (Abburu & Golla, 2015).
Maximum Likelihood is a supervised classifier popularly used in remote
This study on image classification utilizes a maximum likelihood supervised classifier, focusing on the variance and covariance of class signatures to accurately assign each object or pixel to its respective class (Sisodia, Tiwari, & Kumar, 2014).
To begin, display the three-band overlay composite image, where the visible channels correspond to red, green, and blue This representation makes clouds appear white, vegetation green, water dark, and land without vegetation in various shades of brown Next, analyze the available features to identify the appropriate classes for image segmentation.
In Step 2, the 'box-cursor' tool is utilized to select representative training samples from a color composite image for each desired class, creating training data The selected pixels from the study area are represented as polygon shapefiles, which store the identity of each land cover type in an attribute table, as illustrated in Table 6.
- Step 3: Using the trained classifier to classify every pixel in the image into one of the desired classes
-Step 4: Color-encode and show the classified image Estimate the number of pixels and area for each class and show the statistics for each class
The study area is categorized into five distinct land cover types: sparse mangrove, dense mangrove, agricultural areas, water bodies, and other land uses, as detailed in Table 6.
Table 6: LULC ID and names
Some LULC photo from fieldwork:
Figure 8: Water body land use
A CCURACY ASSESSMENT
Accuracy assessment forms the most integral part of the classification process
No classification is truly valid without an accuracy assessment, which measures the alignment between the labels assigned by the classifier and the actual class data collected by the user This assessment is crucial, as it ensures that the sample selection is unbiased It is important to note that the reference data used for evaluating accuracy should be distinct from the data employed by the classifier itself.
This study involved the collection of 100 ground truth points using the random point tool in ArcGIS 10.2 for training data The accuracy of the classification was assessed using the Error Matrix method, which also facilitated the calculation of the kappa statistic.
An error matrix is a square grid that displays the number of sample units, such as pixels or polygons, categorized according to their actual ground truth In this matrix, the columns typically represent reference data, while the rows indicate classifications derived from remotely sensed data, as outlined by Congalton & Green (2008).
The 33 error matrix is an effective tool for representing classification accuracy, clearly outlining the accuracies of each category alongside errors of inclusion (commission errors) and exclusion (omission errors) (Congalton, 1991) This method allows for the evaluation of user’s accuracy, producer’s accuracy, and overall accuracy, with a brief description of these accuracy indices provided below.
Overall accuracy refers to the percentage of reference pixels that are correctly classified It is calculated by dividing the number of accurately classified pixels by the total number of reference pixels However, this metric is quite basic and does not offer insights into the accuracy of individual classes.
Where: D = total number of correct point
N = total number of cell in the error matrix
User 's accuracy reveals false positives, where pixels are misclassified as a known class instead of their correct classification For instance, an image may classify a pixel as impervious, while the reference data indicates it should be forest This misclassification results in the impervious class containing additional pixels that do not align with the reference data Known as errors of commission or type 1 error, user’s accuracy is calculated using data from the table's rows The Total row indicates the number of points that should have been identified as a specific class based on the reference data (Congalton & Green, 2008).
Producer's accuracy refers to the occurrence of false negatives, where pixels belonging to a known class are incorrectly classified as a different category For instance, in a classified image, a pixel may be identified as forest when it should actually be categorized as impervious.
In this case, the impervious class is lacking certain pixels when compared to the reference data Producer's accuracy, also known as errors of omission or type 2 error, is calculated using data from the table's columns The Total column indicates the number of points identified as a specific class based on the classified map.
The Kappa coefficient of agreement is a valuable discrete multivariate analysis technique used to assess the accuracy of change detection and classification maps derived from remotely sensed imagery It is derived from the error matrix and evaluates the classification performance against reference data The resulting Kappa (hat) statistic reflects the actual agreement between the reference data and an automated classifier, accounting for chance agreement that might occur with a random classifier This statistic encompasses all elements of the confusion matrix, providing a comprehensive measure of classification accuracy.
Kappa = (Observed agreement - Chance agreement)/(1 - Chance agreement)
The Kappa statistic measures the extent to which agreement in an error matrix is due to true agreement rather than chance As true agreement nears 1 and chance agreement nears 0, the Kappa value approaches 1, indicating strong reliability The primary benefit of using Kappa is its capacity to serve as a basis for assessing the statistical significance of a matrix or comparing differences between multiple matrices.
E STIMATING A BOVE G ROUND B IOMASS
Above-ground biomass can be measured using both destructive and non-destructive methods The destructive method, often referred to as the harvest method, involves cutting down trees to weigh them In some cases, a sample of trees is harvested, allowing for estimations of the entire population, particularly in uniform areas like plantations However, this method is limited to small areas due to its invasive nature.
Non-destructive methods for estimating tree biomass, such as allometric equations and remote imagery, offer significant advantages over traditional destructive techniques in terms of time, cost, and labor While allometric equations utilize tree dimensions like diameter at breast height (DBH) and height for estimation, their effectiveness is limited in heterogeneous forests, making them more suitable for uniform forests or plantations with similarly aged trees (Kumar & Mutanga, 2017).
In recent years, remote sensing has emerged as a valuable tool for field surveys, minimizing destructive sampling while saving time and costs Research has demonstrated a strong correlation between spectral reflectance values and biomass in remotely sensed data, supported by studies from Anaya, Chuvieco, & Palacios-Orueta (2009), Winarso et al (2017), and Muhd-Ekhzarizal et al (2018).
The total aboveground biomass (AGB) was estimated using species-specific allometric equations (Komiyama, Poungparn, & Kato, 2005) This estimation incorporated density values from the Global Wood Density Database for all mangrove forest species (Muhd-Ekhzarizal et al., 2018) Accurate AGB estimation was achieved by identifying all tree species, allowing for the application of species-specific wood density.
The estimation of AGB was based on D and wood density which were measured at the field The equation for AGB can be expressed as follows:
Where: AGB = above ground biomass (kg) p = wood density (g/cm 3 )
D = Diameter at 0.3m with Rhizophoraceae species and D Diameter at breath Height for other species
The AGB of Kandelia candel species are not include in list of species create by Komiyama (2005) Base on Khan (2005) AGB of K candel can estimate by bellow
AGB = 0.04117( H) Where: AGB = aboveground biomass
Table 7: Wood Density for Each Species in Mangrove Forest According To the
Species Vietnamese name Wood density (g cm -3 )
3.6.2 Vegetation indices and estimate above-ground biomass
Various vegetation indices (VIs) have been created to assess vegetation density from optical remote sensing images, with the normalized difference vegetation index (NDVI) being the most widely used for predicting tree biomass (Li et al., 2007) However, relying solely on NDVI can lead to significant underestimations of biomass in certain woody mangroves, as it primarily reflects canopy characteristics rather than the trunk properties essential for precise biomass estimation (Foody et al., 2001) Araujo (2000) similarly identified that the soil-adjusted vegetation index (SAVI) offers a more effective means of characterizing the biophysical profiles of forests.
Wicaksono (2016) discovered that the SAVI variable is more effective in predicting biomass compared to NDVI This is attributed to the fact that MSAVI minimizes the influence of background soil reflectance, enhancing the accuracy of vegetation reflectance measurements.
The plot sampling process was employed to extract Vegetation Index values from satellite images, specifically using the 2018 Sentinel-2 image Each ground plot, measuring 1000 m², corresponds to exactly 10 pixels of 10-meter resolution, ensuring accurate data collection at specified locations.
The sunlight spectrum consists of various wavelengths, which interact with objects by being absorbed or reflected The Normalized Difference Vegetation Index (NDVI) is utilized to assess the density of green vegetation on land by analyzing the distinct colors (wavelengths) of visible and near-infrared sunlight reflected from plants NDVI values are derived from composite images, employing band 3 (Red) and band 4 (Near Infrared) for Landsat 7, while Landsat 8 uses band 4 (Red) in conjunction with band 5 (Near Infrared) The NDVI formula is essential for evaluating plant health and vegetation cover.
NDVI = ((NIR – RED)/(NIR + RED))
The Normalized Difference Vegetation Index (NDVI) values for a pixel range from -1 to +1, with values near zero indicating no vegetation A value close to +1 (between 0.8 and 0.9) signifies the highest density of green leaves (Zaitunah, Ahmad, & Safitri, 2018).
In this study, NDVI was used for classification and estimate biomass of mangrove forest in Thai Binh province
In regions with low vegetative cover (less than 40%) and exposed soil surfaces, light reflectance in the red and near-infrared spectra can significantly affect vegetation index values This issue is particularly evident when comparing various soil types, as they reflect different amounts of light in these wavelengths To address this challenge, the soil-adjusted vegetation index was created as an enhancement of the Normalized Difference Vegetation Index (NDVI).
The Soil-Adjusted Vegetation Index (SAVI) is designed to mitigate the impact of soil brightness on vegetation assessments, particularly in areas with low vegetative cover Similar to the Normalized Difference Vegetation Index (NDVI), SAVI incorporates a soil brightness correction factor (L) to enhance accuracy in measuring vegetation health and density.
Where: is the reflectance value of the near infrared band is reflectance of the red band
L is the soil brightness correction factor
The value of the soil brightness correction factor (L) is influenced by the extent of green vegetation, ranging from L=0 in areas with dense vegetation to L=1 in regions devoid of greenery Typically, a default value of L=0.5 is effective for most scenarios In this study, we utilized L=0.5 as the correction factor for soil brightness.
3.6.2.3 Green Normalized Difference Vegetation Index
The Green Normalized Difference Vegetation Index (GNDVI) is an enhanced version of NDVI, specifically designed to better detect variations in chlorophyll content within forests Research indicates that GNDVI demonstrates the highest correlation with leaf nitrogen content and dry matter across various data acquisition periods and experimental phases It proves to be more effective than NDVI in identifying different chlorophyll concentration levels, which are closely linked to nitrogen in two plant species By utilizing the visible green band instead of the visible red band, GNDVI extends its sensitivity to higher chlorophyll concentrations, making it a valuable tool for measuring photosynthesis rates and monitoring plant stress.
GNDVI = (NIR – green)/(NIR + green) (Gitelson et al., 1996)
Where: GNDVI = Green Normalized Difference Vegetation Index
NIR: is the reflectance value of the near infrared band Green: is reflectance of the green band
GEMI (Global Environment Monitoring Index) outperforms NDVI in meeting environmental monitoring requirements across all vegetation values and atmospheric conditions While NDVI exhibits a broader transmission range under increasingly turbid atmospheric conditions, GEMI maintains comparable biological information content, as demonstrated in prior studies (Pinty & Verstraete, 1992).
NIR = pixel values from the near infrared band
Red = pixel values from the red band
R EGRESSION ANALYSIS
Regression models are key techniques for predicting Above Ground Biomass (AGB), alongside methods like K nearest neighbors and neural networks (Lu, 2006) These models establish a relationship between a dependent variable and one or more independent variables through regression analysis The most commonly utilized regression models include simple linear regression and multiple linear regression (Quinn & Keough, 2002).
Linear regression models are commonly utilized by researchers to illustrate the linear relationship between independent (x) and dependent (y) variables, allowing for the prediction of y values based on changes in x (Quinn & Keough, 2002) In this analysis, Above Ground Biomass (AGB) serves as the dependent variable, while NDVI, SAVI, GNDVI, and GEMI are the independent variables used to assess changes in AGB The coefficient of determination (R²) is also employed to evaluate the model's effectiveness.
40 was obtained to check the variability of vegetation indices can be caused or explained by its relationship to above ground biomass (Tang & Mayersohn, 2007)
A total of 37 plots observed in the field were used for model development and validation
Where: Y is the predicted biomass
3.7.2 Model validation and accuracy assessment
A validity check is crucial for assessing prediction accuracy, making the validation process essential prior to model application The correlation between the predicted Above Ground Biomass (AGB) from the model and the calculated AGB allows for the determination of the model's coefficient of determination (R²) Additionally, the Root Mean Square Error (RMSE) is computed using a specific formula to evaluate model performance.
Root Mean Square Error calculation:
RMSE is Root Mean Square Error
Y is biomass observed or calculated using allometric equation ̂ is biomass predicted or derived from the radar backscatter using the model n is the number of validating plots
RESULT AND DISCUSSION
M ANGROVE C LASSIFICATION
Table 8 presents five distinct classes, each designated with a specific name and color that accurately represent the real-world characteristics of the objects observed The assigned colors correspond to the identified regions of mangrove forests, water bodies, and agricultural areas These colors are consistently applied across all classified images to maintain clarity and prevent confusion during the accuracy verification process (see Table 8).
Table 8: Class Name and Assigned Class Colours
The area change map categorizes land into several color-coded classes: open mangrove, dense mangrove, agricultural land, water bodies, and other uses Open mangrove refers to low-density mangrove forests with fewer than 1,000 trees per hectare, while dense mangrove areas are characterized by a high density of vegetation along the coast Additionally, agricultural land includes areas used for temporary crops, pastures, and various types of cultivation Water bodies encompass streams, lakes, rivers, and seas, with this study identifying scattered aquatic regions Lastly, "other" land cover represents diverse uses such as residential areas, sediment zones, and construction sites.
In order to mapping land use changing in Thai Binh coastal zone, so the
Effective mangrove forest management requires 42 distinct classifications, primarily focusing on the differentiation between mangrove and non-mangrove areas Most mapping applications assess qualitative factors such as species diversity, growth status, and overall condition, categorizing forests into classes like "dense" or "sparse" (Q T Vo et al., 2013) These classifications are derived from map observations and literature reviews, providing valuable insights into land characteristics Consistent application of these values is essential for analyzing changes in land cover over time.
This study utilized supervised classification through the maximum likelihood method, employing training sites sourced from the field Five distinct land cover classes were identified: dense mangrove, open mangrove, water body, rice field, and other land use The results, illustrated in Figure 9, reveal the various land use and land cover (LULC) types A comparison of classified satellite images from 1998 and 2018 indicates significant changes in land use and land cover within the study area.
The accuracy of the land use and land cover maps was evaluated by comparing them to reference data, which was gathered through GPS sample points, field knowledge, and Google Earth During field visits, a handheld GPS was utilized to pinpoint the exact location of the areas being studied, recording their latitude and longitude, along with visual observations of their types This ground truth data served as a basis for verifying the classification accuracy of the maps.
We considered five images taken at approximately equidistant time points
Between 1998 and 2018, a study of a 55,201.86 ha area revealed changes in mangrove coverage by comparing classified images from adjacent years In 1998, open mangroves spanned approximately 4,508.73 ha, accounting for 8.17% of the total area, which increased to 4,563.45 ha by 2003.
In 2003, open mangrove areas covered 8.27% of the study area By 2007, this figure remained the same at 3230.28 hectares, still representing 8.27% In 2013, the mangrove area increased to 4511.07 hectares, which accounted for 7.88% of the total However, a slight decrease was observed in 2018, with the area reducing to 4148.83 hectares.
Between 1998 and 2018, Thai Binh province experienced a significant increase of 1,072.85 hectares of dense mangrove forest, averaging an annual growth of approximately 53.64 hectares Figure 9 illustrates the dense mangrove class in the study area during this period.
From 1998 to 2018, the area of dense mangroves experienced fluctuations, starting at 1366.2 hectares in 1998, increasing to 1372.32 hectares in 2003, and then decreasing to 1203.57 hectares by 2007 However, a significant recovery occurred, with the mangrove area rising to 1834.02 hectares in 2013 and reaching 2439.05 hectares in 2018 Statistically, this reflects an increase of 0.45% from 1998 to 2003, a decrease of 12.30% from 2003 to 2007, followed by a substantial increase of 52.38% from 2007 to 2013, and a further increase of 32.99% from 2013 to 2018.
A study revealed that the mangrove forest area increased from 1998 to 2018, with dense mangrove growth rising by 1,072.85 hectares, while open mangrove decreased by 359.9 hectares Previous research by Dat (2011) indicated a 987-hectare increase in mangrove forest in Thai Binh province from 1990 to 2007, averaging 58.06 hectares per year, which surpasses the 22.41 hectares per year increase noted in the current study Despite the overall increase in mangrove areas, a decline was observed along the Northern coast of Vietnam, particularly in Quang Ninh and Hai Phong provinces (Dat & Yoshino, 2011).
The increase in mangrove forest area in Thai Binh province can be attributed to two main factors Firstly, numerous projects have been initiated, including a 2006 initiative by the Asian Forest Cooperation Organization focused on afforestation, rehabilitation, and sustainable management of mangrove ecosystems This project aimed to raise awareness and enhance local community knowledge regarding the protection and sustainable development of mangrove forests, biodiversity conservation, climate change mitigation, and livelihood improvement strategies Additionally, the International Federation of Red Cross and Red Crescent Societies (IFRC) and the Japanese Red Cross (JRC) launched a project in the same year to plant additional mangrove trees and expand bamboo planting along river dykes and coastal areas, ultimately enhancing the protection of these vital ecosystems.
44 communities from hazards such as typhoons, storms and floods (VNRC, 2006) (2) The natural development of mangrove forest
Natural regeneration of seedlings in mangrove forests plays a crucial role in the secondary succession process, relying heavily on the presence of mother trees that disperse seeds The growth success of mangrove vegetation is influenced by environmental factors such as pH, COD, BOD, and TSS, which must remain within tolerable limits for both planted and naturally occurring mangroves However, soil fertility in the front zone of mangrove habitats is generally low due to high tidal frequencies and flooding, which wash away essential nutrients Research indicates that over time, particularly in rehabilitation areas, there is a trend of increasing soil fertility, highlighting the potential for improved mangrove growth (Wallacea, 2016; Salmo, Lovelock, & Duke, 2013).
Figure 9: Land Use Land Cover Map in 1998, 2003, 2007, 2013, 2018
Table 9: Area of LULC for Years 1998, 2003, 2007, 2013, 2018
Table 10: Percent (%) of Land Cover in Study Area
Figure 10: Land cover change from 1998 to 2018
Although the area of water body was increased from 1998 to 2003 but it decreased from 2003 to 2018 In 1998, the extent of water was 50.79% that is
28073.07ha In 2003, the extent is 56.16% that is 31008.6 ha In 2007, it decreased to 28746.09ha (52.07%) In 2013, the water body area was 26089.2 ha that is 45.55% In
In 2018, the water surface area experienced a significant reduction, dropping to 18,913.04 hectares, which is a 34.26% decrease This fluctuation in water bodies can be primarily attributed to the twice-daily inundation of water in mangrove regions, leading to river overflow and the formation of marshy ground Consequently, the extent of water bodies varies based on the specific day and time of observation (Yevugah, 2017).
Agriculture area reduced 208.7 ha from 12119.3ha in 1998 to 11910.6ha in
2003, however, increased by 2058.8ha from 2003 to 2007, while decreased by 2332.53ha from 2007 to 2013, following a decline of 3777.49 ha from 2013 to 2018
In generally, the agriculture area are not significant changing Their value around
12000 ha in 1998, 2003, 2013, and they was increased to 13969ha in 2007 and decreased to 11636ha in 2018 The agriculture area was effected by the urbanization of Thai Binh province (Van Suu, 2009)
Other land use, which includes sediment areas, construction zones, and bare land, is categorized collectively as "other land use." The sediment areas, influenced by tidal activities, undergo changes twice daily due to tidal movements In 1998, the total area designated as other land use was 7,400.52 hectares, accounting for 13.39% of the total land However, by 2003, this area had decreased to 6,355.17 hectares, representing 11.51%.
In 2007, this area increased 1705.68ha and they are keep increased to 13208.85ha in
2013 In 2018 other land used significantly increased to 21849.95 that is 39.58% in total study area
4.1.3 Land use land cover change Accuracy Assessment
The classified images underwent a quantitative evaluation through accuracy assessment of all land cover classes Producer's accuracy measures the correct classification of a class within an image, representing the percentage of pixels that should have been assigned to a specific class but were not Conversely, user’s accuracy reflects the confidence level of a class in the classified image Overall, producer's accuracy indicates the total accuracy of the classified image by showing the pixels incorrectly categorized into a specific class.
Table 13, Table 12, Table 14, Table 15) the values represent points The columns represent the actual values, and the rows represent the classified values
4.1.3.1 Accuracy Assessment of the Classified Images in 1998
M ANGROVE BIOMASS ESTIMATING
To estimate above-ground biomass (AGB) in Southeast Asia's mangrove forests, we utilized backscatter characteristics, supported by empirical functions identified by researchers such as G Anderson, Hanson, & Haas (1993), Zheng et al (2004), and Mutanga, Adam, & Cho (2012) These functions establish a relationship between vegetation indices and AGB measurements obtained from ground sample plots.
Vegetation indices, such as NDVI, SAVI, and GNDVI, serve as effective predictors of vegetation biomass due to their correlation with spectral data from optical remote sensing For each sampling location, a buffer zone with a radius of 17.8 meters was established, totaling an area of 1000 m², from which the mean values of the indices were calculated These values were then utilized to develop linear regression models, as illustrated in Figure 17 The scatterplots from the regression analysis, shown in Figures 11, 12, and 13, reveal a significant relationship between the vegetation indices and measured above-ground biomass (AGB) A simple linear regression model was created using 70% of the data (27 plots), while 30% was reserved for model accuracy assessment The correlation between estimated and observed AGB yielded strong coefficients of determination: R² values of 0.6762 for NDVI, 0.685 for SAVI, and 0.672 for GNDVI, as detailed in Table 18 The scatter graph comparing estimated and observed AGB is presented in Figure 11.
Table 18: Summary of simple linear regression models using single independent variable
No Vegetation index Model R R 2 Adjusted R 2
Figure 11: Scatterplots of correlations between aboveground biomass (AGB) and
Normalized Difference Vegetation Index (NDVI) y = 148.24x - 40.413 R² = 0.6618
Scatterplots of correlations between aboveground biomass (AGB) and NDVI
Figure 12: Scatterplots of Correlations between Aboveground Biomass (AGB) and Soil-Adjusted Vegetation Indices (SAVI)
Figure 13: Scatterplots of correlations between aboveground biomass (AGB) and green NDVI (GNDVI)
The coefficient of determination (R²) indicates the extent to which the variance in a dependent variable can be predicted from an independent variable An R² value of 0 signifies no predictive ability, while a value of 1 indicates perfect prediction According to Draper & Smith (1998), a coefficient of determination greater than 0.65 for each vegetation index suggests that over 65% of the variance is predictable In this context, the equation for aboveground biomass (AGB) is represented as y = 99.093x - 40.7, with an R² value of 0.6851, demonstrating a strong predictive relationship.
Scatterplots of Correlations between Aboveground Biomass (AGB) and SAVI y = 232.18x - 60.677 R² = 0.6726
Scatterplots of correlations between aboveground biomass (AGB) and green
54 estimate by the vegetation index NDVI, SAVI, GNDVI.
AGB A CCURACY A SSESSMENT
This study utilized an independent validation dataset to estimate model accuracy, addressing the challenge of limited sample sizes Validation was conducted through independent validation plots, demonstrating that the predicted Above Ground Biomass (AGB) aligned closely with the measured AGB.
Figure 10 illustrates the linear relationships between estimated above-ground biomass (AGB) derived from single measurements and field-based AGB For model validation, 30% of the dataset, consisting of 10 plots, was utilized to assess predictive accuracy, independent from the 70% used for model development, which included 27 plots The root mean square error (RMSE) calculated from the validation data was low, at 7.22861, and the simple linear model demonstrated a strong coefficient of determination (R²) of 0.92, as shown in Figure 14.
Figure 14: Relationship between NDVI linear regressions to estimated AGB and field‐based measured AGB
FIELD MESUAREED AGB (TON/HA)
The root mean square error (RMSE) for estimating above-ground biomass (AGB) using the Soil-Adjusted Vegetation Index (SAVI) was calculated from the validation data, yielding a low RMSE value of 7.22897 Additionally, the simple linear model demonstrated a robust coefficient of determination (R²) of 0.75, indicating a strong correlation in the estimation process.
Figure 15: Relationship between SAVI linear regressions to estimated AGB and field‐based measured AGB
The root mean square error (RMSE) for estimating above-ground biomass (AGB) using the Soil-Adjusted Vegetation Index (SAVI) was calculated from the validation data, yielding a low RMSE value of 7.975 Additionally, the simple linear model demonstrated a robust R² value of 0.68, indicating a strong correlation.
Figure 16: Relationship between GNDVI linear regressions to estimated AGB and field‐based measured AGB
Field mesuareed AGB (ton/ha)
Field mesuareed AGB (ton/ha)
In this study, the root mean square error (RMSE) for all estimated above-ground biomass (AGB) methods remained low, not exceeding 8 ton/ha, indicating a high accuracy in the estimates In contrast, Goh et al (2014) reported RMSE values ranging from 150 to 152 ton/ha for AGB estimation, highlighting the improved precision of the current study The overall RMSE obtained here is considered acceptable, aligning with the findings of Goh et al (2014) Additionally, using NDVI and SAVI, Hamdan et al (2014a) recorded RMSE values of 43.77 ton/ha (r² = 0.59) and 68.21 ton/ha, further demonstrating the variability in estimation accuracy across different methodologies.
In this study, the accuracy of each linear regression models are quite high Moreover, the highest accuracy is NDVI model with lowest RMSE (7.22861) and highest R 2 (0.9261)
The study was showed that the accuracy is quite high but for the study from Wicaksono (2016) map accuracy make by the model acquired from ALOS AVNIR-2
Research indicates that PC bands outperform vegetation indices in modeling mangrove carbon stock Specifically, indices reliant on visible bands, such as VARI, ARVI, and MSARVI, have proven to be less effective for this purpose (Wicaksono et al., 2016).
S PATIAL D ISTRIBUTION OF M ANGROVE V EGETATION B IOMASS IN 1998 AND 2018
This study utilized NDVI, SAVI, and GNDVI linear regression models to map aboveground biomass (AGB) The AGB was categorized into ten distinct levels, ranging from no biomass to over 80 tons per hectare, as illustrated in Figure 17.
Figure 17: Thai Binh AGB mapping base on vegetation indices in 2018
After build linear regression model for NDVI, SAVI, GNDVI in 2018, we were applied that models for 1998 to estimate the changing in aboveground biomass from
Figure 18: Thai Binh AGB mapping base on vegetation indices in 1998
The analysis of Above Ground Biomass (AGB) in mangroves from 1998 to 2018 reveals significant findings, as detailed in Table 20 The maximum estimated AGB using NDVI linear regression was 59.1 t/ha in 1998 and increased to 78.6 t/ha by 2018 The average AGB also saw an increase, rising from 22.569 t/ha in 1998 to 37.74 t/ha in 2018, reflecting the growth trends noted in the study by Darmawan (2014).
The average above-ground biomass (AGB) of mangroves varies across different regions in Vietnam, with Thai Thuy district in Thai Binh province measuring 13.87 tons/ha, Thanh An in Can Gio at 31.61 tons/ha, and Giao Thuy district in Nam Dinh province recording 13.12 tons/ha (Darmawan et al., 2014) Additionally, research by Hanh (2016) indicates that the average AGB in Dong Hung commune, Tien Lang district, Hai Phong city is significantly higher at 36.80 tons/ha.
The above-ground biomass (AGB) of mangroves in the study area is predominantly influenced by the environmental conditions of their habitat, similar to other natural forests Human activities have a minimal impact on the variation of mangrove AGB, as the forest is safeguarded by Xuan Thuy National Park and supported by replanting initiatives from NGOs and government programs.
Table 20: Table showing estimated AGB by NDVI in 1998 and 2018
Total mangrove AGB of the whole area (ton) 62880 187990 125110
Mean area of mangrove AGB (ton/ha) 22.569 37.745 15.180
Total area (detect by NDVI ) (ha) 2786 4980 2194