INTRODUCTION
The forest ecosystem plays a crucial role in regulating climate and supporting both human and animal life In Vietnam, approximately 44.5% of the land, or about 13,797,000 hectares, is covered by forests, with only 0.6% (80,000 hectares) classified as primary forest, which is the most biodiverse and carbon-dense type Historically, forest cover in Vietnam was around 43% in 1943, but by 2000, primary forest cover had decreased to 187,000 hectares, and further declined to 85,000 hectares by 2005, reflecting an annual change of -10.91% Research by scholars such as Rosyadi (1986) and Ringrose (1997) has identified key factors contributing to forest cover loss in developing countries, including inappropriate agricultural practices near forested areas and the centralization of forest management policies, which lead to the misuse of forest resources.
Conventional logging operations, particularly unplanned selective logging, significantly contribute to deforestation, with illegal logging and trade being the primary culprits (Atmopawiro, 2004) Changes in forest cover are a crucial factor in global environmental change and play a central role in sustainable development discussions These changes affect various environmental aspects, including water, land, and air quality, as well as ecosystem functions and climate systems through greenhouse gas emissions and surface albedo impacts Therefore, accurate information on forest cover changes is essential for updating forest maps and managing natural resources Between 1990 and 2010, Vietnam conducted four forest inventory surveys utilizing Landsat and SPOT imagery, and the current inventory program employs SPOT-5, SPOT-6, and VNREDSAT imagery.
Traditionally, forest resource investigations relied on time-consuming and costly manual survey and mapping methods, resulting in low accuracy and outdated information due to changing forest cover However, the advent of geographical information systems (GIS) and remote sensing technologies has revolutionized land and natural resource management, as highlighted by Bao et al (2013) Recent advancements in satellite imagery quality and accessibility now enable large-scale image analysis, significantly enhancing the efficiency and accuracy of forest resource assessments.
Forests in Thanh Hoa are invaluable national treasures, offering timber, grazing land, wildlife habitats, and water resources for local communities However, illegal logging and agricultural activities threaten their quality and quantity To monitor these changes, a change detection analysis was conducted to assess the nature, extent, and rate of forest cover alterations over time and space The findings highlight patterns of forest cover change and demonstrate the effectiveness of multi-temporal satellite data in analyzing these shifts within a spatial-temporal framework This research, focusing on Yen Nhan – Thuong Xuan from 2000 to 2015, provides essential insights for land management and policy decisions related to urbanization, water management, deforestation, and forest degradation.
GOALS AND OBJECTIVES
This goal of project is to classify forest and detect the changes from 2000 to 2015 in Yen Nhan commune, Thuong Xuan district, Thanh Hoa province
To segment SPOT 5 imagery based on object-based classification
To classify and map out the different land use / land cover and their spatial distribution in Yen Nhan in 2015
To identity, quantify and map out the forest cover changes in Yen Nhan from 2000 to
STUDY AREA AND DATA
Study Area
Yen Nhan is a mountainous commune in Thanh Hoa Province, characterized by diverse land covers and uses According to statistical reports, it spans a total area of 19,094.63 hectares and has a population of 4,850, resulting in a population density of 0.25 persons per hectare The region experiences four distinct seasons—spring, summer, autumn, and winter—with an average annual temperature ranging from 23 to 24°C and annual rainfall between 1,600 and 2,000 mm.
Figure 1: Location of the Yen Nhan Commune in Thanh Hoa Province.
Data Sources
This study utilizes multi-spectral SPOT5 remote sensing data, stored in GeoTIFF format and projected in UTM-48N with WGS-84 datum Configuration details of the SPOT5 imagery are summarized in Table 1 Additionally, provincial forest maps for the years 2000, 2005, 2010, and 2015 were obtained from the Institute for Forest Ecology and Environment at Vietnam Forestry University, along with an administrative map of the study area.
Table 1 Technical parameters and properties of the sensors used in this study
480–710 (pan) 500–590 (green) 610–680 (red) 790–890 (NIR) 1580–1750 (mid IR)
METHODOLOGY
Image segmentation (Step1)
The segmentation process was conducted using eCognition v8.9 image analysis software, which is essential for object-based image analysis Image segmentation involves partitioning an image into distinct regions based on specified parameters, focusing on the homogeneity and heterogeneity of these regions.
The segmentation algorithm applied in this study is the so-called “multi- resolution segmentation” The algorithm was applied to all four SPOT bands (green,
The segmentation algorithm utilizes equal weighting for the red, NIR, and SWIR bands, governed by a scale factor and a heterogeneity criterion The scale factor is linked to the average size of the objects to be identified, while the heterogeneity criterion influences the merging decisions based on spectral layers This criterion assesses two distinct properties: color, which pertains to spectral homogeneity, and shape, which considers the semantic characteristics of the objects Shape is further divided into three exclusive properties: scale, color, and form (Baatz et al., 2000).
The scale parameter set by the operator is influenced by pixel heterogeneity, while the colour parameter balances the homogeneity of a segment’s colour and shape Additionally, the form parameter balances the smoothness of a segment's border with its compactness The weighting of these parameters determines the homogeneity criterion for object primitives After testing various parameters, we found that a scale of 70, shape of 0.5, and compactness of 0.8 were optimal, resulting in the SPOT 5 image being segmented into 11,135 objects, with the smallest object measuring 0.11 ha and the largest 112.11 ha This detailed segmentation enhances the accuracy of the automated sorting process.
Figure 3: Objects with multi-resolution segmentation
(Scale parameter = 70, shape = 0.5, and compactness = 0.8)
Classification and accuracy (Step2)
The classification scheme was established using provincial map legends and field surveys, initially defining ten land cover types: rich evergreen forest, medium evergreen forest, poor evergreen forest, rehabilitation evergreen forest, bamboo forest, mixed wood and bamboo forest, plantation forest, bar land, shrub and grass, and water body An object-based classifier was created to categorize each object into these land cover classes, employing an object-based classification framework alongside interactive visual interpretation, expert knowledge, training data, and existing area maps.
Table 2: The object-based classification framework
SPOT 5 image Sample Plot Image Describes
Accuracy assessment is a crucial component of the image classification process, involving the evaluation of both positional and thematic accuracies Positional accuracy refers to the precision of a point's location in satellite imagery compared to its actual ground position, while thematic accuracy assesses the correctness of mapped land cover classes against the real conditions on the ground at a specific time In this study, 80 reference points were surveyed in the field to provide validation samples for classification, from which 50 sample points were randomly selected.
Figure 4: The distribution of sample points in the study area
To assess the accuracy of land cover classification, an error matrix and descriptive statistics were computed The study employed the Kappa coefficient, a discrete multivariate technique, to evaluate the classification accuracy.
[9] The Kappa coefficient represents the proportion of agreement obtained after removing the proportion of agreement that could be expected to occur by chance
The Kappa coefficient ranges from 0 to 1, indicating the level of error reduction, with values over 0.80 signifying strong agreement, those between 0.40 and 0.80 indicating moderate agreement, and values below 0.40 reflecting poor agreement According to Jensen and Cowen, Kappa is calculated using a specific equation.
In a matrix with N total sites, each row is represented by r, where the value at row i and column i indicates the specific entry Additionally, the total for each row i and the total for each corresponding column are calculated to provide a comprehensive overview of the matrix's structure.
Change Detection (Step 3)
The study analyzed forest cover changes over a 15-year period (2000-2015) by classifying land into 10 categories and calculating the area of each class A forest cover change map was created to visualize overall changes in the Yen Nhan commune, focusing on three short-term periods: 2000-2005, 2005-2010, and 2010-2015 Using ArcGIS 10.1, land cover changes were detected through overlay and post-classification comparison of maps from different years The resulting change maps included a cross-tabulation matrix to illustrate conversion pathways and quantify the changes, with dynamic change patterns displayed for enhanced clarity.
RESULTSDISCUSSION
Object – Based Classification
The classification results of SPOT 5 imagery, processed using eCoginition software in 2015, reveal ten distinct classes: rich evergreen, medium evergreen, poor evergreen, rehabilitation evergreen, bamboo, mixed wood and bamboo, plantation forest, bare land, shrub and grass, and water body.
Figure 5: Object-based classification result (2015)
The structural characteristics of forest ecosystems vary significantly, with natural forests exhibiting greater complexity compared to plantation forests Rich forests are more intricate than medium, poor, or rehabilitated forests Additionally, bare land and water bodies stand out as the most identifiable features within these landscapes.
Classification
Yen Nhan showcases a variety of forest types unique to the region, including rich evergreen, medium evergreen, poor evergreen, rehabilitation evergreen, bamboo, mixed wood and bamboo, and plantation forests The distribution of these forest classes has been quantified in hectares and percentages, providing a detailed overview of the area's ecological diversity.
The rich evergreen forest covers only 11.33 hectares, accounting for just 0.1% of the total land area, primarily situated in the Northern region, while medium evergreen forests are predominantly found in the same area with a smaller presence elsewhere.
The study area spans 1460.33 hectares in the southwest, accounting for 7.7% of the total land The poor evergreen forest covers 3759.15 hectares (24.1%), primarily located in the central and western regions The largest land cover is rehabilitation evergreen forest, encompassing 6386.6 hectares (33.7%) found in the southern, western, and eastern parts, with some patches in the north Plantation forest occupies 670.92 hectares (3.5%) in the central area, while bamboo forest covers 142.09 hectares (0.8%) in the southern region Additionally, mixed wood and bamboo forest spans 3647.46 hectares, concentrated in the north with smaller areas in the west and south Shrub and grasslands cover 1861.9 hectares (9.8%) in the western part, and water bodies, primarily in the south, account for 115.84 hectares (0.6%) Bare land near water bodies occupies 80.57 hectares (0.4%) Overall, forest cover totals 16077.9 hectares, representing 89.05% of the area, mainly in the northern and western parts, while non-forest areas cover 1977.7 hectares (10.95%), scattered around rivers like Khao, Chu, Dat, and Dan.
Figure 6: Land areas (ha) in 2015: Forest classification classes (a), their percentages (b) and their proportion forest and non-forest (c)
Figure 7: Land use/land cover in Yen Nhan 2015 5.2.2 Accuracy
The accuracy statistics are summarized in Table 3
Table 3: Summarized accuracy of objects in 2015 Map
RG TB RN PH HG RT TN DT DK MN Total
The article categorizes various forest types, including Rich Evergreen Forest (RG), Medium Evergreen Forest (TB), and Poor Evergreen Forest (RN), each representing different ecological health levels Additionally, it highlights Mixed Wood-Bamboo Forest (HG) and Rehabilitation Evergreen Forest (PH) as important ecosystems for biodiversity Bamboo Forest (TN) and Plantation Forest (RT) are also noted for their unique contributions to forestry, while Shrub and Grass (DT) areas serve as transitional habitats Lastly, Bare Land (DK) and Water Bodies (MN) are included to illustrate the broader landscape context, emphasizing the importance of diverse ecosystems for environmental balance.
The matrix above highlights discrepancies in the samples, as the spectral differences between certain objects within the interference range are not clearly defined This lack of clarity also reflects the trend of deviations observed in the image segmentation results compared to reality.
In a study involving 50 observations, 38 agreements were recorded, representing 76.00% of the total The expected agreements by chance were only 5.0, or 10.00% of the observations The analysis yielded a Kappa coefficient of 0.733, with a 95% confidence interval ranging from 0.603 to 0.864, indicating a good strength of agreement.
Classification errors arose from spectral mixing between various forest types, including rich, medium, poor, rehabilitation, mixed, and plantation forests, as well as shrub, grass, and bare land The low resolution of SPOT 5 imagery, coupled with cloud cover in certain areas, negatively impacted the results Additionally, accurately classifying spatial and temporal forest cover change patterns requires expertise in object-based classification methods To mitigate these challenges, careful selection of training sites and the application of majority filters post-supervised classification were implemented.
Land Cover Change
Land cover and land use classification maps of Yen Nhan Commune reveal significant changes over four time periods, as illustrated in Figure 8 From 2000 to 2010, rich evergreen forests predominantly occupied the northern region, but by 2015, their area had drastically reduced to only 11.33 hectares (0.1%), with medium evergreen forests emerging in the same northern areas and small patches in the west The decline in rich evergreen forests allowed for the expansion of poor evergreen forests, which became prominent in the central commune by 2015 Rehabilitation evergreen forests, initially found in small patches from the western to southeastern regions in 2000, expanded to the eastern and northern parts by 2005 Plantation forests were virtually non-existent until 2005 but began to develop in the central and southeastern areas, with further expansion noted in 2015 Mixed wood and bamboo forests were scattered throughout the commune until 2010, after which they became concentrated in the northern parts by 2015.
19 bamboo forest and shrub and grass could be find in many regions in commune, however, in
Between 2000 and 2005, bare land was limited to small areas in the western, central, and southeastern parts of the commune However, from 2010 to 2015, there was a significant and rapid increase in the conversion of bare land to residential areas, particularly in the central and southern regions Throughout the period from 2000 to 2015, the southern part of the commune was predominantly characterized by the presence of water bodies.
Table 4: Statistics of forest and non-forest area in Yen Nhan (2000-2015)
Looking at table 4, the largest forest area cover is in 2015, representing 89.5 % and the smallest area is 11988.9 ha (representing 63.6 %) in 2000
Figure 8: Land cover/land use classification maps of Yen Nhan commune in year 2000, 2005, 2010 and 2015
From 2000 to 2015, Yen Nhan commune experienced notable changes in land cover and land use, as illustrated in Figure 9 The data highlights the percentage distribution of various land cover types across different years, showcasing the evolution of land properties during this period.
Shrub and Grass Water body Bare Land Bamboo Forest Plantation Forest Mixed Forest Rehabiliation Evergreen Forest Poor Evergreen Forest Medium Evergreen Forest Rich Evergreen Forest (a)
Rich Evergreen Forest Medium Evergreen Forest Poor Evergreen Forest Rehabiliation Evergreen Forest
Water body Shrub and Grass (b)
22 the major land cover/land use types in hectare, (c) the magnitude of the land cover/land use changes in ha/year for each time interval
The analysis of land cover changes from 2000 to 2015 reveals significant trends in various classes, as illustrated in Figures 9a and 9b Shrub and grass, bare land, bamboo forest, and rich evergreen forest have gradually decreased, with bare land and rich evergreen forest becoming the least represented types by the end of the study period Notable changes occurred, particularly in rich evergreen forest, poor evergreen forest, bamboo, rehabilitation evergreen forest, bare land, and shrub and grass Initially, from 2000 to 2005, shrub, grass, and poor evergreen forest declined, while mixed and rehabilitation forests increased From 2005 to 2010, shrub, grass, and poor evergreen forest saw a gradual increase, but bamboo and rehabilitation forests grew slowly By the end of the study period, shrub, grass, and bamboo forests sharply declined, while bare land and poor evergreen forest experienced significant growth Rich evergreen forest and bare land remained relatively stable in the first two intervals but rapidly decreased from 2010 to 2015 Medium evergreen forest was stable from 2000 to 2010, followed by significant growth alongside plantation forest Water bodies and mixed forests maintained a consistent percentage throughout the 15 years.
Figure 9c demonstrates the annual changes in hectares, standardizing absolute changes by the duration of each analyzed year across nine primary land cover and land use categories Notably, only plantation areas exhibit a consistent upward trend, while other land cover types, including rich forest, poor forest, rehabilitation forest, bamboo forest, shrub and grass, and bare land, show fluctuations over various time periods.
23 magnitudes of change than the limited changes for medium forest, mixed forest and the negligible variation of water bodies
Figure 10 depicts the spatial distribution of changes across various time intervals Between 2000 and 2005, changes were observed in large patches scattered throughout the commune From 2005 to 2010, these changes became more fragmented By the end of the study period from 2010 to 2015, significant large-scale changes were noted.
Figure 10: Spatial distribution of land cover/land use changes in Yen Nhan commune from 2000-2015
The analysis of land cover and land use changes in Yen Nhan commune from 2000 to 2015 reveals significant transformations, as detailed in Table 5 The data indicates that the rich evergreen forest decreased dramatically from 700 ha in 2000 to only 11.3 ha in 2015, with 457.1 ha converted to medium evergreen forest and 133.8 ha to mixed forest Over the 15-year period, rich forests, poor forests, bamboo forests, and shrub and grass areas faced substantial reductions, totaling 694.1 ha, 1158.8 ha, 3029.8 ha, and 6076.3 ha, respectively In relative terms, rich forests and shrub areas experienced a staggering 99% reduction, while poor forests and bamboo forests saw declines exceeding 97% Conversely, plantation forests expanded by 670.5 ha, primarily from the conversion of shrub and grass (298.6 ha), mixed forest (151.1 ha), and bamboo forest (139.5 ha) Additionally, significant losses were observed in water bodies during this period.
56 percent (representing 50 ha) mostly to bare land (48.1 ha)
Table 5: Natural of land cover/land use changes in Yen Nhan commune from 2000 to 2015 in ha(figure are rounds up to entire numbers)
RG TB RN HG PH TN DT DK MN Total Expansion
The article discusses various forest types, including Rich Evergreen Forest (RG), Medium Evergreen Forest (TB), and Poor Evergreen Forest (RN), highlighting their ecological significance It also covers Mixed Wood-Bamboo Forest (HG) and Rehabilitation Evergreen Forest (PH), which play crucial roles in biodiversity conservation Additionally, the Bamboo Forest (TN) and Plantation Forest (RT) are mentioned for their economic benefits, while Shrub and Grass (DT) and Bare Land (DK) indicate areas of lesser vegetation Lastly, the presence of Water Bodies (MN) is emphasized for their importance in maintaining ecosystem balance.
RECOMENDATIONS
Between 2000 and 2015, Yen Nhan commune experienced a dramatic decline of over 97% in rich evergreen forests, bamboo forests, shrubs, and grasslands This loss can be attributed to the extraction of timber and non-timber forest products, along with the expansion of rehabilitation efforts for poor evergreen forests and bare land for roads and residential areas The increase in bare land during this period was driven by population growth, new road development, and socio-economic advancements Government policies, including programs 327, 661, and 147, facilitated the expansion of plantation forests since 2000 These land use and cover changes reflect the socio-economic dynamics of the area and highlight the government's efforts to balance rural development with ecological stability by removing compulsory grain crop quotas and promoting livestock.
CONCLUSIONS AND PERSPECTIVES
Conclusions
The object-based change detection method demonstrated high efficiency in identifying forest land cover changes in both deciduous and coniferous stands, achieving a moderate overall kappa of 0.73 This study represents the first comprehensive analysis of land cover and land use patterns in Yen Nhan commune, revealing its diverse landscape The object-oriented approach outperformed traditional pixel-based classification, which often misclassifies pixels in spectrally heterogeneous land covers.
Over the past 15 years, from 2000 to 2015, Yen Nhan commune has experienced significant changes in land cover and land use, characterized by categories such as rich evergreen, medium evergreen, poor evergreen, rehabilitation evergreen, bamboo, mixed wood and bamboo, plantation forest, bare land, shrub and grass, and water bodies Notably, plantation forests have seen substantial growth, while rich evergreen forests and bare land have notably decreased Variations in land cover types have been observed, with rich evergreen, poor evergreen, rehabilitation evergreen, bamboo, and shrub and grass experiencing more pronounced changes than others Additionally, the expansion of plantation forests has primarily resulted from the conversion of poor evergreen forests, mixed forests, rehabilitation evergreen forests, bamboo forests, shrub and grass, and bare land.
The plantation areas exhibit a consistent upward trend, while other land cover and land use types show fluctuations over time Notably, rich forest, poor forest, rehabilitation forest, bamboo forest, shrub and grass, and bare land experienced significant changes, contrasting with the minimal variations observed in medium forest, mixed forest, and water bodies.
Perspective
The findings from land cover and land use maps, along with change detection analysis, provide insights into the effects of historical policies and various factors, including socio-economic trends and environmental changes, on land use dynamics By comprehending the drivers behind land use changes, we can model future land use patterns more effectively and enhance land use planning policies at both district and provincial levels.
Future research will aim to identify the drivers and impacts of land cover and land use change in the study area Understanding the factors influencing individuals' decisions to alter land use and their spatial patterns is essential Additionally, investigating the causes of rapid land use dynamics will shed light on their environmental impacts, effects on local livelihoods and access to natural resources, and the increased vulnerability of communities to natural hazards and anticipated environmental changes.
APPENDICES
Period 2000-2005 RG TB RN HG PH TN DT DK MN Total Expansion
The article discusses various forest types, including Rich Evergreen Forest (RG), Medium Evergreen Forest (TB), and Poor Evergreen Forest (RN), highlighting their ecological significance It also covers Mixed Wood-Bamboo Forest (HG) and Rehabilitation Evergreen Forest (PH), emphasizing their roles in biodiversity and restoration Additionally, Bamboo Forest (TN) and Plantation Forest (RT) are mentioned for their economic value, while Shrub and Grass (DT) and Bare Land (DK) are noted for their environmental impact Lastly, the presence of Water Bodies (MN) is acknowledged as crucial for sustaining these ecosystems.
Appendix 2: Natural of land cover/land use changes in Yen Nhan commune from 2005 to 2010 in ha (figure are rounds up to entire numbers)
Period 2005-2010 RG TB RN HG PH TN DT DK MN Total Expansion
The article discusses various forest types, including Rich Evergreen Forest (RG), Medium Evergreen Forest (TB), and Poor Evergreen Forest (RN), each characterized by differing biodiversity and ecological significance Additionally, it highlights Mixed Wood-Bamboo Forest (HG) and Rehabilitation Evergreen Forest (PH) as important for ecosystem restoration Bamboo Forest (TN) and Plantation Forest (RT) are noted for their commercial value, while Shrub and Grass (DT) areas and Bare Land (DK) play crucial roles in land management Lastly, Water Bodies (MN) are emphasized for their importance in maintaining ecological balance.