The majority of the population and economic activity of the northern half of Vietnam is clustered in the Red River Delta and about half of the country’s rice production takes place here. There are significant problems associated with its geographical position and the intensive exploitation of resources by an overabundant population (population density of 962 inhabitants/km 2). Some thirty years after the economic liberalization and the opening of the country to international markets, agricultural land use patterns in the Red River Delta, particularly in the coastal area, have undergone many changes. Remote sensing is a particularly powerful tool in processing and providing spatial information for monitoring land use changes. The main methodological objective is to find a solution to process the many heterogeneous coastal land use parameters, so as to describe it in all its complexity, specifically by making use of the latest European satellite data (Sentinel-2). This complexity is due to local variations in ecological conditions, but also to anthropogenic factors that directly and indirectly influence land use dynamics. The methodological objective was to develop a new Geographic Object-based Image Analysis (GEOBIA) approach for mapping coastal areas using Sentinel-2 data and Landsat 8. By developing a new segmentation, accuracy measure, in this study was determined that segmentation accuracies decrease with increasing segmentation scales and that the negative impact of under-segmentation errors significantly increases at a large scale. An Estimation of Scale Parameter (ESP) tool was then used to determine the optimal segmentation parameter values. A popular machine learning algorithms (Random Forests-RFs) is used. For all classifications algorithm, an increase in overall accuracy was observed with the full synergistic combination of available data sets.
Coastal areas are subject to numerous pressures acting on a limited space. The current developments in the context, simultaneously climatic, economic and societal, make governance of changes in the coast areas land cover a fundamental issue. Land Use/Land Cover (LULC) is essential in planning and management activities especially for conserving the eco-environment as well as for urban planning in coastal areas. Nowadays, the use of remotely sensed data in order to produce a LULC map is rapidly increasing since it offers advantages in terms of cost effectiveness and accuracy [
The study site is a coastal sector of the Red River delta in Vietnam and the goal is to observe the changes in terms of the land cover of the coast zone. Some thirty years after the economic liberalization and the opening of the country to international markets, agricultural land use patterns in the Red River Delta, particularly in the coastal area, have undergone many changes. As a second phase, a change in land management took place within “Doi Moi” policy framework. A market economy has replaced the socialist economy. This policy, also known as “revival”, was proclaimed in 1986 by the Government and the Vietnamese Communist Party to reform and renovate the political and economic field of the country.
This article discusses the implementation of Geographic object-based image analysis (GEOBIA) for coastal mapping of changes of land use in River Red Delta in Vietnam. This is both an opportunity and challenge for GEOBIA in the mapping and monitoring of land use. GEOBIA is a newly evolving sub-discipline of geographic information science (GIScience) devoted to partitioning remote sensing (RS) imagery into meaningful image-objects, and assessing their characteristics through spatial, spectral and temporal scale. At its most fundamental level, GEOBIA requires image segmentation, attribution, classification and the ability to query and link individual objects (a.k.a. segments) in space and time. Geographic object-based image analysis (GEOBIA) has been gaining prominence in the fields of remote sensing and GIScience over the past decade, especially for the processing of high spatial resolution imagery [
This paper describes new developments in data sources, image segmentation, object-based feature extraction, and geo-object-based modeling frameworks, followed by a discussion about opportunities for future research and conclusions.
The majority of the population and economic activity of the northern half of Vietnam is clustered in the Red River Delta (
There are significant problems associated with its geographical position and the intensive exploitation of resources by an overabundant population (population density of 962 inhabitants/km2). Periodic floods threaten an agricultural activity with an accelerating yield, which weakens the environment. Food production in the delta is mainly based on the rice harvest during the second half of the
year. This is very often preceded by periodic flooding that threatens economic activities in general and agricultural activities in particular. The country’s recent transition to a market economy has resulted in increased yields following intensified agricultural activities which led to greater environmental vulnerability. After some forty years of economic liberalization and the opening of the country to international markets, agricultural land use patterns in the Red River Delta, and particularly in the coastal areas, have undergone many changes.
The production and marketing of aquaculture products in the coastal areas of Vietnam are very significant economic issues as they have allowed, and still allow, a rapid improvement in the income levels of the local population. This includes peasants and fishermen who rely heavily on the natural resources of the coastal zone to increase production. Mangroves and even rice paddies are being replaced by aquaculture fields at an unprecedented pace [
Current land use in the coastal area is farming, with rice as the main crop. Two rice crops and one dry crop is produced per year [
As a second phase, a change in land management took place within “Doi Moi” policy framework. A market economy has replaced the socialist economy. One year later (in April 1988), the so-called “khoan 10” decree no. 10 was adopted by the Political Bureau, marking a clear shift in economic orientation and a broad liberalization of private trade. Some fundamental points: Farmers obtained property rights to their land. Even though the State retains ownership of the land, private use thereof was permitted. Farmers were therefore granted land rights when the state planning system was abolished. Taxes paid by the farmers to the State were reduced to 10% and this rate remained unchanged for 5 years (this term was previously one year), which provided reassurance to farmers regarding the sustainability of their operations. The principle of egalitarian allocation of farmland was abandoned. Instead, these were allocated according to the farmers’ production capacity and the availability of cooperative land. The term of lease for rice paddies, previously set at between 1 and 3 years, was increased to 20 years. The “simplified package” was replaced by the “net package”, which allowed farmers to be responsible for 8 stages of rice growing, instead of the 5 stages traditionally granted. The majority of agricultural taxes were reduced (lower prices for chemical fertilizers, insecticides and fuel, the abolition of progressive taxes and other taxes levied by local authorities such as income tax or solidarity tax) which allowed farmers to reduce their operating costs. Farmers were then also allowed to sell their harvested produce on a free market, without working through a cooperative, as the household has since been considered an independent economic unit, and each farmer was therefore able to benefit from his own harvest. This decree has been enforced in the south of Vietnam since 1988. It only entered into force in the north of the country two years later, where conservative leaders, seeking to retain their power and privileges, opposed all forms of land management reforms.
In this article, we used Sentinel-2 optical satellite data and optical Landsat-8 data (
In this study the main data source originate from the Sentinel-1 satellites. The Sentinel series of satellites is part of Copernicus programme designed by the European Union (EU) and European Space Agency (ESA). The all Sentinel 2 images (Level 1C) covering the entire study area were downloaded from the ESA’s Sentinel Scientific Hub (ESA, 2018) [
Launched in June 2015 for Sentinel 2A and in March 2017 for Sentinel-2B, these satellites are equipped with the optical instrument MSI (Multi-Spectral instrument), which covers 13 spectral bands from the visible through infrared. The Sentinel-2 mission is the continuation of programs Landsat and SPOT. These two satellites are able to provide images with a width at the ground of 290 km, and a periodicity of 5 days. Acquisitions are carried out according to three resolutions in function of the spectral band: 10, 20 and 60 m (
Number | Date | Satellite |
---|---|---|
1 | 05/26/2018 | Sentinel 2 |
2 | 06/20/2018 | Sentinel 2 |
3 | 07/05/2018 | Sentinel 2 |
4 | 11/02/2018 | Sentinel 2 |
5 | 12/17/2017 | Sentinel 2 |
6 | 09/17/2017 | Landsat 8 |
7 | 07/30/2017 | Sentinel 2 |
8 | 04/21/2017 | Sentinel 2 |
9 | 06/15/2016 | Sentinel 2 |
10 | 04/23/2016 | Landsat 8 |
Sentinel-2 Bands | Central Wavelength (µm) | Resolution (m) | Bandwidth (nm) |
---|---|---|---|
Band 1―Coastal aerosol | 0.443 | 60 | 20 |
Band 2―Blue | 0.49 | 10 | 65 |
Band 3―Green | 0.56 | 10 | 35 |
Band 4―Red | 0.665 | 10 | 30 |
Band 5―Vegetation Red Edge | 0.705 | 20 | 15 |
Band 6―Vegetation Red Edge | 0.74 | 20 | 15 |
Band 7―Vegetation Red Edge | 0.783 | 20 | 20 |
Band 8―Near Infrared (NIR) | 0.842 | 10 | 115 |
Band 8A―Narrow NIR | 0.865 | 20 | 20 |
Band 9―Water vapor | 0.945 | 60 | 20 |
Band 10―SWIR―Cirrus | 1.375 | 60 | 20 |
Band 11―SWIR 1 | 1.61 | 20 | 90 |
Band 12―SWIR 2 | 2.19 | 20 | 180 |
Landsat 8, originally known as the Landsat Data Continuity Mission (LDCM), is a National Aeronautics and Space Administration (NASA)-US satellite. The conception of LDCM to the reality of Landsat 8 followed an arduous path extending over nearly 13 years, but the successful launch on February 11, 2013 ensures the continuity of the unparalleled Landsat record. Access to Landsat 8 data was opened to users worldwide. With a mission objective to enable the detection and characterization of global land changes at a scale where differentiation between natural and human-induced causes of change is possible, LDCM promised incremental technical improvements in capabilities needed for Landsat scientific and applications investigations [
We linked the shooting dates of the satellite images available for our study with the main cultivation stages of rice in the Red River Delta. Generally speaking there are two main harvests per year in the Red River Delta, thus two harvest seasons: the spring-summer season (from February to July) and the autumn-winter season (from August to January). The first stage concerns the preparations taking place in the rice fields: from mid-January onwards, with the harvest taking place 100 days later, at the end of April/May. The second stage is the soil preparations after the first harvest from the end of May onwards, thus the 2nd stage
Bands | Wavelength (micrometers) | Resolution (meters) | |
---|---|---|---|
Landsat 8 Operational Land Imager (OLI) And Thermal Infrared Sensor (TIRS) Launched February 11, 2013 | Band 1―Coastal aerosol | 0.43 - 0.45 | 30 |
Band 2―Blue | 0.45 - 0.51 | 30 | |
Band 3―Green | 0.53 - 0.59 | 30 | |
Band 4―Red | 0.64 - 0.67 | 30 | |
Band 5―Near infrared (NIR) | 0.85 - 0.88 | 30 | |
Band 6―SWIR 1 | 1.57 - 165 | 30 | |
Band 7―SWIR 2 | 2.11 - 2.29 | 30 | |
Band 8―Panachromatique | 0.50 - 0.68 | 15 | |
Band 9―Cirrus | 1.36 - 1.38 | 30 | |
Band 10―Thermal infrared (TIRS) 1 | 10.60 - 11.19 | 100 | |
Band 11―Thermal infrared (TIRS) 2 | 11.50 - 12.51 | 100 |
occurs until the end of August/beginning of September. From September to mid-January vegetables are grown (
The methodology deals with the processing and analysis of Sentinel-2 and Landsat-8 satellite imagery (
The first step of image processing was the NDVI calculation the vegetation index using Sentinel-2 and Landsat-8 images. A vegetation index can be an indicator to describe the greenness, density and health of vegetation. Vegetation indices were calculated for the Sentinel-2A reflectance image: NDVI (Normalized Difference Vegetation Index [
Vegetation index for Sentinel-2
NDVI = NIR − RED NIR + RED = B 8 − B 4 B 8 + B 4
Crop Season | Period | Planting Crops | Growth | Harvest |
---|---|---|---|---|
Spring-Summer | Early Harvest | 12/25-12/30 | 02/05-02/10 | 05/20-05/25 |
Annual Harvest | 01/05-01/20 | 02/20-02/25 | 06/01-06/15 | |
Late Harvest | 02/25-03/05 | 01/25-02/05 | 06/25-06/30 | |
Autumn-Winter | Early Harvest | 05/20-05/30 | 06/01-06/10 | 09/01-09/10 |
Annual Harvest | 06/01-06/10 | 06/10-06/20 | 10/25-11/10 | |
Late Harvest | 06/25-07/05 | 06/25-07/05 | 11/05-11/25 |
Vegetation index for Landsat 8
NDVI = NIR − RED NIR + RED = B 5 − B 4 B 5 + B 4
After layer normalization to same value range, layer stacking was performed by adding all the processed optical image layers. The following image layers were stacked into a 10-layers image for 2016: Sentinel-2A reflectance bands (4 × 2 = 8 bands); Sentinel-2A NDVI (2 bands); Total 2016 = 10 bands; Image 2017: Sentinel-2A reflectance bands (4 × 3 = 12 bands; Landsat-8 reflectance bands (4 bands); Sentinel-2A NDVI (3 bands); Landsat-8 NDVI (1 band); Total 2017 = 20 bands. 2018 image: Sentinel 2A reflectance bands (4 × 4 = 16 bands); Sentinel 2A NDVI (1 × 4 = 4 bands); Total 2018 = 20 bands.
An important step in image processing was the multi segmentation of images using the new Estimation of Scale Parameter (ESP2) tool. Multiscalar image segmentation is a fundamental step in GEOBIA, yet there is currently no tool available to objectively guide the selection of appropriate scales for segmentation, with the exception of the Estimation of Scale Parameter (ESP2) tool (
The ESP tool allows for a fast estimation of scale parameters for a multiresolution segmentation in the Definiens software environment. Segmentation is the process of dividing remotely sensed images into discrete regions or objects that are homogeneous with regard to spatial or spectral characteristics [
increasing scale parameter, 2) starting scale parameter for the analysis, 3) the use of an object hierarchy during segmentation, 4) number of loops, 5) shape weighting, and 6) compactness weighting. Parameters 2), 5), and 6) are used as implemented in the multiresolution segmentation [
The algorithm chosen for the supervised classification is the Random Forest. Random Forests (RF) of Breiman [
Classifying is then performed much in the same way as with regular decision trees, where the resulting class is the mean class of all the single decision trees in the forest. The trees in a random forest are generated in a way that attempts to split the data set at every node in half. For both halves, this algorithm is repeated until every leaf in the tree contains only one sample. Contrary to classical Decision Trees (DT), the RF trees are built without pruning and by randomly selecting at each node a subset of input variables. Currently, this number of variables used to split a RF node (denoted by m) corresponds to the square root of the number of input variables [
algorithm determines the classification accuracy. The more variables we use the more the correlations between decision trees and the more the classification accuracy (less misclassified pixels). Decreasing m decreases the correlation between classification trees but will be less accurate or its predictive power diminishes. The Random Forest is a very useful approach, it is not only nonparametric but it also provides a way of estimating the importance of the individual variables in classification. In ensemble classification, several classifiers are trained and their results combined through a voting process. Many ensemble methods have been proposed. Most widely used such methods are boosting and bagging [
In this work, the setting used for Random Forest was: Maximum depth of the tree (25), Minimum number of sample in each node (50) and Cluster possible values of a categorical variable into k (k = 8).
The methodological objective of this study was to evaluate the performance of Random Forest method when applied to land cover/land use mapping from Sentinel-2 and Landsat-8 time-series data. The methodology and the algorithm were used to improve the accuracy of recognition and mapping of land cover/land use in Red River Delta in Vietnam. A set of land cover/land use categories of interest were defined, namely: aquaculture, mangroves, rice, vegetables, villages, water and wetland (
In terms of the distribution of land use categories in the district, the agricultural fields (rice paddies) located inside the dikes cover more than 40% of the total surface area during the three years of observation. Water covers about 23% to 28% of the surface area and is mainly represented by the sea, the Red River and the rivers Lân and Tra Ly. Villages including gardens, cover approximately 14% to 15% and are located in the midst of the rice paddies. Wet soils are not really
found inside the dikes, except at the edge of the irrigation canals, but rather on the outside of the dikes, at the mouths of the Tra Ly River in the north and the Red River in the south of the district. Aquaculture ponds are often located outside the dikes but sometimes within the dikes and the percentage surface area occupied by it did not change in 2016 but increased by about 5% in 2017 and 8% in 2018. On the other hand, mangroves were found on the islands in the estuaries of these same rivers during 2016, but they have become very sparse by 2018. In 2018, new mangroves were planted around the aquaculture ponds. Submerged shrubs are scattered outside the dikes and sometimes within the dikes, covering about 1% to 3% of the area. Nuts edge and vegetables in general are often found in the lowlands near the banks of the Red River. Lastly, salt marshes cover very little territory, mostly in the center of the district and next to the dikes.
The results show very good classification performance for all combinations (kappa index 0.90 and 0.97). Following standard accuracy calculation procedures [
Looking at the changes by stellite images (
Land cover classes | Accuracy (%) optic + NDVI 2018 | Accuracy (%) optic + NDVI 2017 | Accuracy (%) optic + NDVI 2016 |
---|---|---|---|
Rice | 99 | 100 | 97 |
Villages | 100 | 100 | 100 |
Water | 100 | 100 | 94 |
Mangroves | 100 | 100 | 100 |
Aquaculture | 95 | 95 | 99 |
Vegetables | 93 | 91 | 47 |
Wetland | 76 | 81 | 71 |
Kappa index | 0.97 | 0.97 | 0.90 |
Overall accuracy | 0.98 | 0.97 | 0.93 |
comments can be made:
・ municipalities with the same biophysical conditions (such as Nam Phu and Dong Long) displayed several similar trends in changed land-use due to the fact that they have the same coastal ecosystems; what is more, their differences are often caused by human activities: the recent foundation of Nam Phu contributing to the expansion of its agricultural fields, the hasty commencement of Nam Phu aquaculture encouraging the development of ponds and the significant cutting down of mangroves; the second had such a pivotal impact on changes in the surface area of aquaculture ponds and mangroves that one could not see the differences caused by environmental protection regulations in the Nam Phu Nature Reserve;
・ municipalities with different biophysical conditions (such as Tây Tiên and the Dong Long municipalities) do not have the same types of land use; they have only two similarities: villages and agricultural fields that are inside the dikes and which do not vary much; the other land use types differ greatly in terms of surface area; there are only two types of land use in Dông Long that are not found in Tây Tiên, namely the nutsedge fields and aquaculture.
Physical and anthropogenic factors do not have the same impact on changes in land use. In addition, we note that, despite the intermingling of image processing that must be improved, the detection of land use changes with the help of satellite images is very useful. We were able to see how each type of land use has changed in the district during the years under observation to explain the relationship between economic liberalization and land use changes. These changes reflect the impact of physical factors (such as the characteristics of the coastal ecosystems that favor mangroves, submerged shrubs, salt marshes and aquaculture production) and anthropogenic factors (economic liberalization, the new policy of stimulating aquaculture and economic development plans by the government) on land use. Anthropogenic factors have a greater impact on changes than physical factors, especially in the rapid increase in aquaculture ponds, which in turn brought about changes in other types of land use. The increased demand for shrimp has also influenced other types of land use at Tiên Hai, namely the replacement of salt marshes and rice fields with aquaculture ponds. The reality is that, instead of earning only 10 million VND (Vietnamese currency) per hectare of rice, farmers can get 114 million VND for one hectare of shrimp ponds. This means that with shrimp production they can earn ten times more than with the production of salt or rice. As a result, 380 ha of salt marshes were changed to aquaculture ponds, increasing surface area occupied by aquaculture in the district to 1000 ha [
Despite some limitations, the image processing performed in this study is suitable for Vietnam’s highly fragmented and dynamic coastal area thanks to its effectiveness in detecting changes in land use and in identifying the categories themselves. The image processing shows the environmental degradation due to land-use changes, including mangrove cutting and the disappearance of rice paddies and salt marshes, as well as the pollution of aquaculture ponds.
Geographic Object-Oriented (GEOBIA) concepts and methods have been successfully applied to study of changes land cover in Red river delta, district of Tien Hai. The emerging trends were found in multiple sub fields of GEOBIA, including data sources, image segmentation, object-based feature extraction, and geo-object-based modeling frameworks. Segmentation is the partitioning of an array of measurements on the basis of homogeneity. It divides an image―or any raster or point data―into spatially continuous, disjointed and homogeneous regions referred to as “segments”. In feature extraction, it can be regarded as an end in itself. In GEOBIA, it is one step in a processing chain to ultimately derive “meaningful objects”. A very important concept to distinguish GEOBIA from per-pixel approaches is the ability to address a multiplicity of scales within one image and across several images. The ESP2 tool was used to select a parameter which is unique to the landscape structures. The image object is classified with machine learning methods (such as random forests) based on semantic network model and hybrid (top-down, bottom-up) control tactics.
In view of the rapid development of the global changes of coastal zone, with all its positive and negative effects on prevailing socio-ecological systems, the demonstrated low cost approach based on open source data is of high relevance to all aspects, i.e. planning, decision making, management, and monitoring, of an integrated and sustainable management of the coastal zone.
The authors declare no conflicts of interest regarding the publication of this paper.
Niculescu, S. and Lam, C.N. (2019) Geographic Object-Based Image Analysis of Changes in Land Cover in the Coastal Zones of the Red River Delta (Vietnam). Journal of Environmental Protection, 10, 413-430. https://doi.org/10.4236/jep.2019.103024