remote sensing

Article
Analysis of Land Use Change and Driving Mechanisms in
Vietnam during the Period 2000–2020
Xuan Guo 1,2 , Junzhi Ye 1 and Yunfeng Hu 1,2, *

1 State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences
and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China;
[email protected] (X.G.); [email protected] (J.Y.)
2 College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 100049, China
* Correspondence: [email protected]

Abstract: High-accuracy, long-time-series and large-scale land classification mapping are essential
for assessing the evolutionary patterns of land systems and developing sustainability studies. In this
paper, using Google Earth Engine (GEE) and Landsat satellite remote sensing images, based on the
Random Forest (RF) algorithm, we carried out remote sensing classification to obtain a year-by-year
land use/cover data set in Vietnam over the past 21 years (2000–2020). Further applying principal
component analysis and multiple linear regression methods, we examined the spatio-temporal
characteristics, dynamic changes and driving mechanisms of land use change. The results show
the following: (1) The RF classification algorithm supported by the GEE can quickly and accurately
obtain a land use/cover data set. The overall classification accuracy is 0.91 ± 0.01. (2) The land cover
types in Vietnam are dominated by woodland and cropland, with an area share of 54.62% and 37.90%,
respectively. In the past 20 years, the area of built-up land has increased the most (+93.49%), followed



by the area of water bodies (+54.19%), while the area of woodland has remained almost unchanged.


(3) The expansion of built-up land is driven by regional economic development; the area changes in
Citation: Guo, X.; Ye, J.; Hu, Y.
cropland, water bodies and woodland are influenced by both national economic development and
Analysis of Land Use Change and
climate change. The results of the study provide a basis for assessing land use policies in Vietnam
Driving Mechanisms in Vietnam
and a reference methodological framework for rapid land mapping and analysis in other countries in
during the Period 2000–2020. Remote
the China–Indochina Peninsula.
Sens. 2022, 14, 1600. https://
doi.org/10.3390/rs14071600
Keywords: spatial pattern; dynamic change; driving mechanisms; random forest; regional development;
Academic Editors: Parth Sarathi Roy, time-series analysis; remote sensing
Mi Wang, Hanwen Yu, Jianlai Chen
and Ying Zhu

Received: 8 March 2022

Accepted: 24 March 2022 1. Introduction
Published: 26 March 2022 Land Use/Land Cover Change (LULC) is one of the most direct results of climate
Publisher’s Note: MDPI stays neutral change and human activities on the Earth’s surface systems [1]. LULC not only profoundly
with regard to jurisdictional claims in affects land surface energy balance, carbon and water cycles and species diversity [2–5],
published maps and institutional affil- but also has strong links to local ecosystem services, food security and socioeconomic
iations. sustainability. Therefore, it is of great scientific importance to comprehensively grasp and
monitor the spatial distribution and dynamic changes in surface cover, which is basic
research into an important element of global environmental change. Since the 1990s,
the International Geosphere-Biosphere Programme (IGBP) and the International Human
Copyright: © 2022 by the authors. Dimensions Programme on Global Environmental Change (IHDP) have jointly initiated
Licensee MDPI, Basel, Switzerland. the Land Use/Land Cover Change (LULC) core research project [6], and land use/cover
This article is an open access article change monitoring, driving mechanisms and effects on regional economies and climate
distributed under the terms and
have become key research directions in geography, ecology and geoeconomics [7–9].
conditions of the Creative Commons
Before 2015, most global land cover data products had a resolution of 300–1000 m,
Attribution (CC BY) license (https://
with low regional accuracy and spatial consistency of 53–57%, generally. Spatial detail was
creativecommons.org/licenses/by/
insufficient to meet the needs of applications, such as global and regional land use analysis
4.0/).

Remote Sens. 2022, 14, 1600. https://doi.org/10.3390/rs14071600 https://www.mdpi.com/journal/remotesensing

Remote Sens. 2022, 14, 1600 2 of 17

and climate change simulation [10]. In 2014, Chinese scholars developed the first global
30-metre resolution land cover data product—GlobeLand30 [11–13]. The product is valued
by researchers for its wide spatial extent, high resolution and free sharing [14]. However,
there are still many manual interventions in GlobeLand30 products, which make it difficult
to extract and compare land cover information on an interannual scale [15]. At the regional
scale, the study of high precision and faster land use/land cover mapping methods and
the development of a spatio-temporal continuous LULC database remain the motivation
for scholars to research regional LULC evolution patterns, driving mechanisms and climate
and ecological environmental effects.
Regional-scale land use/cover classification and mapping are mainly carried out
with the help of satellite remote sensing image data, and can be broadly classified into
several method types, such as image segmentation, unsupervised classification, supervised
classification and deep learning. Currently, the more commonly used LULC classification
algorithms include minimum distance classification (MDC) [16], maximum likelihood
classification (MLC) [17], random forest algorithm (RF) [15], classification and regression
tree (CART) [18] and support vector machine algorithm (SVM) [19,20]. Of these, RF is
an integrated classifier based on the combination of multiple CART, as a result of its
nonparametric nature, ability to handle dimensionality and overfitting [21]. RF algorithms
have a higher classification accuracy compared with other algorithms and are widely used
in the field of satellite remote sensing land mapping [22–24].
However, limited by the computing power and storage space of remote sensing
platforms, RF-based land cover classification could only rely on limited band data from
multispectral data for a long time, with few applications of the full range of satellite band
data. Few studies have considered other environmental background data and human activ-
ity characteristics, such as the normalized difference vegetation index (NDVI), normalized
difference water index (NDWI), digital elevation model (DEM) and the nighttime light
index. In recent years, the rapid development of satellite remote sensing big data and cloud
computing technology represented by Google Earth Engine (GEE) has provided a new
platform for global and regional land cover/use research. GEE combines data collection
and download with powerful spatial analysis and processing capabilities, not only pro-
viding global time-series satellite maps and other ancillary data (such as digital elevation
models and weather and climate data), but also integrating algorithms for processing large
amounts of satellite data to easily meet the major needs of processing remote sensing data,
which are widely used in studies of vegetation cover, settlement and population and urban
development [25]. In this context, based on the GEE platform, exploiting the potential of
the RF algorithm to fuse long time series and full elements of satellite surface parameters
features have become an important direction for regional LULC exploration research.
In Vietnam, due to the limitations of basic data and large-scale processing capabilities,
past LULC studies have mainly explored land use/cover change processes at smaller spatial
scales and shorter time scales using the manual visual interpretation method, and few
studies have explored land use/cover change and its driving mechanisms over long time
periods on a national scale. Disperati and Virdis analyzed the land use change in Tam
Giang-Cau Hai Lagoon, central Vietnam, from 1965 to 2014, using Landsat MSS/TM/ETM+
satellite imagery with the help of computer-aided visual interpretation. They found that
urban areas have encroached on peri-urban agricultural areas, while aquaculture has
spread into agricultural land and lagoons; scattered vegetation in hilly areas was replaced
by woodlands [26]. Van et al. analyzed the mangrove vegetation area change in Mui Ca
Mau, Vietnam, from 1953 to 2011, with the help of aerial photographs and satellite images,
and found that the mangrove areas showed a trend of a sharp decline in the first period
and a slow increase in the later period, using the year 1992 as the boundary [27]. These
trends were mainly influenced by war, overexploitation and land management policies. In
summary, existing studies only focus on regional-scale land use/cover dynamics, and lack
national-scale, high-precision, automated spatio-temporal continuous land use/cover data
Remote Sens. 2022, 14, x FOR PEER REVIEW 3 of 18

Remote Sens. 2022, 14, 1600 dynamics, and lack national-scale, high-precision, automated spatio-temporal continuous 3 of 17
land use/cover data sets, in addition to comprehensive studies on land use change dy-
namics and analysis of driving mechanisms.
sets, To address to
in addition thecomprehensive
above issues, this study
studies on proposes the usedynamics
land use change of GEE toandachieve high-
analysis of
precision and long-term national-scale land use/cover mapping of Vietnam based on sum-
driving mechanisms.
mer Landsat 5/7/8
To address theimage
abovedata and other
issues, parameters,
this study proposes such
theasuse
EVI,ofNDVI,
GEE toNDWI,
achieveSRTM3
high-
data and the
precision andglobal nighttime
long-term light index,
national-scale landduring the period
use/cover 2000–2020;
mapping furthermore,
of Vietnam based on it
applies
summerprincipal
Landsat component
5/7/8 image analysis andother
data and multiple linear regression
parameters, methods
such as EVI, NDVI,to explore
NDWI,
the
SRTM3influence of the
data and economic development
global nighttime light and regional
index, during climate factors
the period on land
2000–2020; use/cover
furthermore,
change processes
it applies principalincomponent
the study area. We and
analysis attempt to answer
multiple linearthe following
regression three questions:
methods to explore
the influence of economic development and regional climate factors on land
(1) How can we achieve high-accuracy, high-precision land use mapping over long time use/cover
change processes
series across in the study area. We attempt to answer the following three questions:
Vietnam?
(1) How are
(2) What can the
we spatial
achievepatterns
high-accuracy, high-precision
and evolution land use mapping
of land use/cover change inover long time
Vietnam dur-
series
ing theacross
periodVietnam?
2000–2020?
(2) Whatare
(3) What arethe
thekey
spatial patterns
factors drivingand evolution
land use/coverofchange
land use/cover
in Vietnamchange
duringinthe
Vietnam
period
during the
2000–2020? period 2000–2020?
(3) What are the key factors driving land use/cover change in Vietnam during the period
2000–2020?
2. Materials and Methods
2.1. Study Area
2. Materials and Methods
Vietnam
2.1. Study Areais located in the eastern part of the China–Indochina Peninsula (8°10′–
23°24′N, 102°09′–109°30′E), bordering Guangxi and Yunnan, China, in the north, and Laos
Vietnam is located in the eastern part of the China–Indochina Peninsula (8◦ 100 –23◦ 240 N,4
and◦ Cambodia
0 ◦ 0 in the west, with a long and narrow land area of approximately 33 × 10
102 09 –109 30 E), bordering Guangxi and Yunnan, China, in the north, and Laos and
km2 (Figure 1). The region has a tropical monsoon climate with high temperatures and
Cambodia in the west, with a long and narrow land area of approximately 33 × 104 km2
rainfall, with an average annual temperature of approximately 24 °C and an average
(Figure 1). The region has a tropical monsoon climate with high temperatures and rainfall, an-
nual an
with rainfall of annual
average 1500–2000 mm. In the
temperature north, there are
of approximately 24 ◦four seasons:
C and spring,
an average summer,
annual rainfallau-
of
tumn and winter; in the south, there are two distinct rainy and dry seasons,
1500–2000 mm. In the north, there are four seasons: spring, summer, autumn and winter; with mostin
areas having
the south, therea are
rainy
twoseason
distinctfrom
rainyMay
and to
dryOctober
seasons,and
witha most
dry season from aNovember
areas having rainy seasonto
April.
from May to October and a dry season from November to April.

Figure 1. Location,
Figure 1. Location, topography
topography and
and main provinces of
main provinces of Vietnam.
Vietnam.

The study area has a long and narrow terrain. Three-quarters of the area consists of
mountains and plateaus. The northern and northwestern parts are high mountains and
Remote Sens. 2022, 14, 1600 4 of 17

plateaus. The eastern coast is plain with low and flat terrain, a dense network of rivers, and
an elevation of approximately 3 m.
According to geographical divisions, Vietnam can be divided into the northern area,
middle area and southern regions. In detail, it can be divided into 8 parts: northeast-
ern provinces (NE), northern and central provinces (NC), northwestern provinces (NW),
Mekong Delta (MD), Red River Delta (RRD), southeastern provinces (SE), southern and
central provinces (SC) and western provinces (WS).

2.2. Data Sets

The satellite remote sensing information sources used for land cover mapping in this
study are as follows: Landsat 5/7/8 image data, EVI, NDVI, NDWI index data, SRTM3
data and global night light products of the study area, year-by-year during the period
2000–2020.
Before 2013, we used Landsat 5/7 image data, including 6 bands: B1 (0.45–0.53 µm), B2
(0.52–0.60 µm), B3 (0.63–0.69 µm), B4 (0.76–0.90 µm), B5 (1.55–1.75 µm) and B7 (2.08–2.35 µm).
After 2013, we used Landsat 8 data, including 9 bands: B1 (0.43–0.45 µm), B2 (0.45–0.51 µm),
B3 (0.53–0.59 µm), B4 (0.64–0.67 µm), B5 (0.85–0.88 µm), B6 (1.57–1.65 µm), B7 (2.11–2.29 µm),
B10 (10.6–11.19 µm) and B11 (11.50–12.51 µm). Specific data we used are USGS Landsat 5
Level 2, Collection 2, Tier 1; USGS Landsat 7 Level 2, Collection 2, Tier 1; and USGS Landsat
8 Level 2, Collection 2, Tier 1. The data level of processing is level 2 and the Landsat
Collection Tier is Tier 1. All of them have a spatio-temporal resolution of 30 m and 16 days.
In order to improve the classification accuracy of woodland and water body cover areas,
the study also used EVI, NDVI and NDWI data, which were derived from Landsat 5/7/8
Collection 8-day composites [28].
To improve the recognition accuracy from the perspective of elevation, this study also
used SRTM3 (Shuttle Radar Topography Mission Version 3), with a spatial resolution of
30 m, which was jointly measured and produced by the National Aeronautics and Space
Administration (NASA) and the National Imagery and Mapping Agency (NIMA).
To improve the accuracy of the classification of built-up land, the study also used
two nighttime light products: DMSP-OLS Nighttime Lights Time Series during the period
2000–2011 from the Defense Meteorological Satellite Program (DMSP), and NPP-VIIRS
Day/Night Band (DNB) during the period 2012–2020 from the Suomi National Polar-
Orbiting Partnership (Suomi NPP), with spatial resolutions of 900 and 450 m, respectively.
To establish the high-confidence training sample and validation sample data set
required by the RF model, this study referred to the principles of “multi-source consistency”
and “time-series stability” [15,29] and used multi-phase and multi-source land cover
products. Specifically, we used GlobCover in 2000, 2010 and 2020 [30] with a spatial
resolution of 300 m derived from http://due.esrin.esa.int/page_globcover.php (accessed
on 5 March 2022), and the GlobeLand30 in 2000, 2010 and 2020 [31] with a spatial resolution
of 30 m derived from http://globeland30.org/home_en.html (accessed on 5 March 2022).
In order to analyze the impact of climate change and human activities on regional land
cover, this research also collected meteorological data and regional social development data.
The meteorological data (2000–2020) were mainly day-by-day rainfall data from
PERSIANN-CDR [32] with a spatial resolution of 0.25 radians. From the period 2000–2020,
we summed the rainfall amount for the whole year and summer (July–September) to obtain
the annual and summer rainfall data, respectively. Finally, the 2 values we obtained refer to
annual and summer rainfall amounts for each year. These procedures were conducted in
the GEE platform.
Regional economic and social development data, including population, gross domestic
product and other indicators, were obtained from https://data.worldbank.org.cn/country/
vietnam?view=chart (accessed on 5 March 2022); the specific data items obtained are as
follows: total population, rural population, urbanization rate, gross domestic product
(GDP), manufacturing value added, industrial value added and agricultural value added.
 (2) Based on the principle of “multi-source consistency” and “time-series stability” of
multi-source land cover products, we selected sample points with no change in land
use/cover type, deployed RF models to train and validated sample points and ex-
tracted land cover information of sample points.
(3) Using 70% of the sample points as training points, combined with the synthetic im-
Remote Sens. 2022, 14, 1600 5 of 17
ages year-by-year, RF model training was carried out to interpret the land use/cover
result data set. The remaining 30% of the sample points were used as validation sam-
ple points to evaluate thebase
All of the accuracy of classification
data sets results.
were integrated into the GEE platform and can be used directly,
except change
(4) Based on climate for GlobCover, GlobeLand30
and regional andsocial
economic regional socioeconomic
development statistical
factors, we de- yearbook data
termined the(Supplementary
categories of Table
land S1).
use/cover change drivers via principal component
analysis and applied multiple linear regression methods to quantitatively determine
2.3. Mapping Methods
the contribution of each driver.
2.3.1. Overall Technical Process
The overall technical process can be divided into 4 steps (Figure 2).

Figure 2. TechnicalFigure
flowchart of landflowchart
2. Technical use/land of
cover
landmapping
use/land and driving
cover mechanism
mapping analysis.
and driving mechanism analysis.

(1)of Long
2.3.2. Preparation BasedTime
on Landsat satellite Images
Series Satellite images and related auxiliary data, such as EVI, NDVI,
With the supportNDWI and
of the Nighttime
GEE Light
platform, weIndex,
used using
cloud the method
mask of image
and image synthesis and cloud
synthesis
masking under the support of GEE, we obtained year-by-year cloud-free and shadow-
methods to produce cloud-free composite images based on SR (surface radiation) images
free synthetic image data of Vietnam from 2000 to 2020.
during the period 2000–2020. Specifically, referring to the built-in cloud mask method in
(2) Based on the principle of “multi-source consistency” and “time-series stability” of
GEE, we constructed multi-source
the function land
to mask clouds based on the pixel_qa band of Landsat
cover products, we selected sample points with no change in
land use/cover type, deployed RF models to train and validated sample points and
extracted land cover information of sample points.
(3) Using 70% of the sample points as training points, combined with the synthetic images
year-by-year, RF model training was carried out to interpret the land use/cover result
data set. The remaining 30% of the sample points were used as validation sample
points to evaluate the accuracy of classification results.
(4) Based on climate change and regional economic social development factors, we
determined the categories of land use/cover change drivers via principal component
analysis and applied multiple linear regression methods to quantitatively determine
the contribution of each driver.

2.3.2. Preparation of Long Time Series Satellite Images

With the support of the GEE platform, we used cloud mask and image synthesis
methods to produce cloud-free composite images based on SR (surface radiation) images
during the period 2000–2020. Specifically, referring to the built-in cloud mask method in
GEE, we constructed the function to mask clouds based on the pixel_qa band of Landsat
5/7/8 SR (surface radiation) data. Bits 3 and 5 in the pixel_qa band are cloud shadow
and cloud, respectively. Both flags should be set to zero, indicating clear conditions. We
replaced and supplemented each year’s low-quality pixels with data from the previous
Remote Sens. 2022, 14, 1600 6 of 17

and following year using the median synthesis method to produce the best available pixel
image synthesis, resulting in an annual cloud-free image data set for the summers during
the period 2000–2020.
For the nighttime imagery, we used the built-in algorithms “ee.reduceResolution” and
“ee.reproject” for nighttime imagery to aggregate smaller pixels into a larger pixel and force
the image to be computed in a given projection (EPSG:4326) and resolution (30 m, same
with Landsat data).

2.3.3. Construction of Sample Data Set

Automated satellite remote sensing supervised classification usually requires a certain
number of model training samples and validation samples. Traditional studies generally
use manual visual interpretation to obtain training and validation sample points. However,
for a large geographical unit such as Vietnam, it is very difficult to obtain enough samples
by manual visual interpretation, which will cost a large amount of labor and time, and
cannot achieve the purpose of automated and fast mapping. In order to solve the above
problems, this study implemented a highly reliable and automated method for training
and validation sample point selection and attribute extraction based on the principles of
“multi-source consistency” and “time-series stability” [15,29].
The specific steps are as follows:
(1) The main land use types in the study area were cropland, woodland, water bodies and
built-up land; however, there are very few and extremely scattered wetlands, shrub
and grassland types. Therefore, we only focused on the four main types. We unified
the land cover classification system of GlobeLand30 and GlobCover products and
reclassified the land cover types in Vietnam into four categories (Table 1), specifically
cropland, woodland, water bodies and built-up land.
(2) The GlobeLand30 (2000, 2010 and 2020) and GlobCover (2000, 2010 and 2020) data
sets were overlaid to select pixels with identical land cover type definitions in the
multi-source and multi-period data products.
(3) To avoid the risk of sample bias (over-representation of correct or incorrect points), a
stratified random sampling method (“stratifiedSample” algorithm in GEE) was used
to randomly deploy sample points in the above target image elements. A total of
2000 sample points were deployed in this paper, with 1400 (70%) as training points
and 600 (30%) as validation points (Figure 3). This same sample data set was applied
for each year.

Table 1. The category of land use/land cover in Vietnam.

Code 1st Classes Description

Lands used for agriculture, horticulture and gardens, including
1 Cropland paddy fields and irrigated and dry farmland and vegetation.
Lands covered with wetland plants and water.
Lands covered with trees, with vegetation cover over 30%,
2 Woodland including deciduous and coniferous woodlands, and sparse
woodland with cover 10–30%, etc.
Water bodies in the land area, including rivers, lakes, reservoirs
3 Water Body
and fish ponds.
Lands modified by human activities, including all kinds of
4 Built-up Land habitation, industrial and mining areas, transportation facilities
and interior urban green zones and water bodies.
 Lands modified by human activities, including all kinds of
Remote Sens. 2022, 14, 1600 4 Built-up Land habitation, industrial and mining areas, transportation facili-
7 of 17
ties and interior urban green zones and water bodies.

area.
Figure 3. Distribution of sample points in the study area.

2.3.4.
2.3.4. Random
Random ForestForest Method
Method
Random forest
Random forest (RF) (RF) is aisdata mining
a data miningmethod and aand
method potential machinemachine
a potential learninglearning
method
based on the idea of integrated learning by building a large number
method based on the idea of integrated learning by building a large number of decision of decision trees
and applying a voting mechanism to the decision results. Specifically,
trees and applying a voting mechanism to the decision results. Specifically, first, the boot-first, the bootstrap
sampling technique
strap sampling is used
technique to extract
is used N training
to extract setssets
N training from thethe
from original
originaldata set;set;
data then,
then,a
decision tree (“forest”) is built for each training set, and each decision tree
a decision tree (“forest”) is built for each training set, and each decision tree is independ- is independent
and unrelated to each other; when new samples are input, each decision tree is judged
ent and unrelated to each other; when new samples are input, each decision tree is judged
separately; finally, voting is used to decide the category of the new samples. When a new
separately; finally, voting is used to decide the category of the new samples. When a new
sample is an input, each decision tree is judged separately, and the class of the new sample
sample is an input, each decision tree is judged separately, and the class of the new sample
is decided by voting.
is decided by voting.
As the RF algorithm optimizes the decision accuracy by combining the results of
As the RF algorithm optimizes the decision accuracy by combining the results of mul-
multiple decision trees, the RF algorithm usually has better accuracy than traditional
tiple decision trees, the RF algorithm usually has better accuracy than traditional auto-
automated classification algorithms; at the same time, the algorithm can make the training
mated classification algorithms; at the same time, the algorithm can make the training and
and classification processes highly parallelized. Especially when the sample features
classification processes highly parallelized. Especially when the sample features are of
are of high dimensionality, the model can still run efficiently and reduce the overfitting
high dimensionality, the model can still run efficiently and reduce the overfitting phe-
phenomenon [22], which further improves the overall prediction performance of the model.
nomenon [22], which further improves the overall prediction performance of the model.
The parameters of random forest methods in GEE include the number of decision trees, the
The parameters
number of variablesof random
per split,forest methods
the fraction in GEE
of input to include
bag per the
tree,number of decision
the maximum number trees,
of
the number of variables per split, the fraction of input to bag per tree,
leaf nodes and the randomization seed. In this article, we trialed many parameter settings the maximum num-
ber of
and leaf nodes
recorded the and the randomization
accuracy, seed. In between
and found a balance this article, we trialedefficiency
operational many parameter
and the
highest operational accuracy for the following parameter settings: number ofefficiency
settings and recorded the accuracy, and found a balance between operational decision
and the
trees: 120;highest
the numberoperational accuracy
of variables for the
per split: the following
square rootparameter
of the numbersettings: numberthe
of variables; of
fraction of input to bag per tree: 0.5; maximum number of leaf nodes in each tree: no limit;
randomization seed: 0.5.
In total, there are 4 land use classes used in this study, namely cropland, woodland, water
bodies and built-up lands. We obtained a land classification map for each year, containing
Remote Sens. 2022, 14, 1600 8 of 17

four land use types using random forest classification under the GEE platform. Then, we used
the Tabulate Area tool of ArcGIS10.2 software to calculate the area of each land use type. After
that, we obtained 4 area values for each land use type every year from 2000 to 2020.

2.3.5. Evaluation of LULC Data Accuracy

In order to evaluate the sample point construction method under the principle of
“multi-source consistency” and “time-series stability”, and to confirm whether these meth-
ods are capable of automating long-time series and high-accuracy land use/cover mapping
tasks, we evaluated the results of the land use/cover data set from 2000 to 2020 year-by-
year based on the accuracy assessment ideas proposed by Olofsson [33]. The classification
accuracy was assessed using the confusion matrix (CM) method, a widely used approach
based on the comparison of the RF classification mapping outputs with the known reference
data (ground truth values) [34,35]. The confusion matrix was obtained using the built-in
algorithm “errorMatrix” in the GEE platform. Some specific accuracy metrics were derived
from the following matrix: Overall Accuracy (OA) through the built-in algorithm “accu-
racy”, Producer’s Accuracy (PA) (also known as “recall”) through the built-in algorithm
“producersAccuracy”, and User’s Accuracy (UA) (also known as “precision”) through the
built-in algorithm “.consumersAccuracy”.
The confusion matrix represents the relationship between the known reference data
(ground truth values) on each land use/cover type and the corresponding results of the
classification process; user accuracy represents the frequency that the classifications on the
map will appear on the ground; and producer accuracy represents the proportion of the
results of each classification judged correctly on the map to the total observed values.

2.3.6. Principal Components Analysis

We introduced several socioeconomic and natural climate indicators to analyze the
drivers of land use change, but there may be strong correlations (expressing similar infor-
mation) between different indicators. Therefore, principal components analysis was first
applied to filter the variables in this study.
Principal components analysis (PCA) is a multivariate statistical method to examine
the correlation between multiple variables, which replaces the original variables with a
new set of mutually uncorrelated composite indicators by regrouping several indicators
with a certain correlation. The core idea is to reduce the dimensionality of the variable data
set, while preserving the information of the variable data set variation as much as possible.
This is performed via the orthogonal transformation of a series of potentially correlated
variables into a new set of linearly ordered uncorrelated variables, with the top few variables
retaining most of the variation in the original variables [36]; the transformed set of new
variables is called principal components and is applied in the subsequent analysis.
In total, 9 variables (total summer rainfall, total annual rainfall, total population,
agricultural population, urbanization rate, gross domestic product, manufacturing value
added, industrial value added and agriculture value added) (Supplementary Table S2)
were input into the PCA to obtain the principal component. We collected each data set
year-by-year during the period 2000–2020 in the pre-processing. The principal component
analysis was performed with the support of SPSS.

2.3.7. Multiple Regression Model

Multiple linear regression refers to regression analysis with two or more independent
variables. In fact, in the study of practical problems, the change in the dependent variable
is often affected by several important factors. At this time, it is necessary to use two or
more influencing factors as independent variables to explain the change in the dependent
variable, which is more effective and realistic than using only one independent variable for
the estimation. The parameter estimation of the multiple regression model is the same as
the linear regression equation, and the least square method is used to solve the parameters
under the premise that the sum of squares of the error is required to be the smallest.
Remote Sens. 2022, 14, 1600 9 of 17

The multiple regression model is as follows:

Y = β + α 1 X1 + α 2 X2 + . . . + α n X n

where α1 , α2 , . . . , αn denote the regression coefficients and β is a constant term.

In the multiple regression analysis, we used various land use/cover type areas as
the dependent variables and the principal components as the independent variables to
obtain the regression model, the significance level and the R2 of typical land use types.
Various land use/cover area values were obtained from RF classification mapping and
area calculation process based on ArcGIS software mentioned in Section 2.3.4. Principal
components are obtained from PCA based on 9 natural and social indicators, which have
been expressed in Section 2.3.6.
The multiple linear regression analysis was performed with the support of R software
with the MASS package in the study.

3. Results
3.1. Accuracy Assessment of Mapping Results
With the support of GEE, this study obtained the land use/cover data set of Vietnam
for 2000–2020 (spatial resolution of 30 m) using the above method. The overall accuracy of
this data set was 0.91 ± 0.02 (Table 2). In terms of different land use types (Table 3) for the
period 2000–2020, the user accuracy and producer accuracy of water bodies and built-up
land types were relatively high (0.95 ± 0.01 and 0.90 ± 0.01 for water bodies; 0.96 ± 0.01
and 0.95 ± 0.02 for built-up land), but those of cropland and woodland were relatively low
(0.82 ± 0.03 and 0.82 ± 0.03 for cropland; 0.89 ± 0.03 and 0.95 ± 0.04 for woodland).

Table 2. Accuracy assessment of different land use/land cover data sets from 2000 to 2020.

Year Overall Accuracy PA UA

2000 0.93 0.93 0.93
2001 0.92 0.92 0.92
2002 0.97 0.87 0.88
2003 0.86 0.84 0.85
2004 0.90 0.90 0.91
2005 0.91 0.91 0.90
2006 0.91 0.91 0.91
2007 0.91 0.91 0.91
2008 0.91 0.92 0.91
2009 0.91 0.91 0.91
2010 0.90 0.90 0.90
2011 0.89 0.89 0.89
2012 0.92 0.92 0.92
2013 0.91 0.91 0.91
2014 0.92 0.92 0.92
2015 0.89 0.89 0.89
2016 0.92 0.92 0.92
2017 0.91 0.91 0.91
2018 0.9 0.90 0.90
2019 0.91 0.91 0.91
2020 0.91 0.91 0.91
Remote Sens. 2022, 14, 1600 10 of 17

Table 3. Accuracy assessment of different land use/land cover types.

Classification Accuracy
Land Use/Land Coverage
UA PA
Cropland 0.82 0.82
Woodland 0.89 0.95
Water 0.95 0.90
Built-up land 0.96 0.95

The results (Tables 2 and 3) show that the classification accuracy of water bodies and
built-up land was high, which may be related to the introduction of additional environ-
mental background data and human activity element data, such as the nighttime light
products DEM, and NDWI in this study. Compared with the high accuracy of water bodies
and construction land, the classification accuracy of cropland and woodland is relatively
poor. The reason is that some croplands are easily misclassified as woodlands (e.g., bush)
and wetlands (e.g., paddy land), and some woodlands are easily misclassified as cropland
(e.g., bush). (Table 4).

Table 4. Confusion matrix of validations (round to integers) for land use/land cover data sets over
the past 21 years.

Land Cover Type Cropland Woodland Water Body Built-Up Land

Cropland 121 15 6 3
Woodland 7 144 0 0
Water Body 12 2 140 1
Built-up Land 5 0 1 153

3.2. Spatial Distribution and Dynamic Changes

In 2020, woodland and cropland were the main types in the Vietnam region in general
(Figure 4). Woodland has the largest distribution area, followed by cropland, with an area
Remote Sens. 2022, 14, x FOR PEER REVIEW 11 of 18
share of 54.62% and 37.90%, respectively; water bodies, with 5.20% of the national land
area; and the smallest area was built-up land, with only 2.28%.

Figure
Figure4.4.Land
Landuse/land
use/landcover
covermap
mapininVietnam
Vietnaminin2020.
2020.

Spatially, cropland is mainly concentrated in the low elevation plains, such as the
Mekong River Delta in the south and the Red River Delta in the northeast of the study
area; woodland is mainly distributed in the middle-altitude and high-altitude mountains
and hilly areas, such as the Giai Song Chao Re mountain in the north of the study area
Remote Sens. 2022, 14, 1600 11 of 17

Spatially, cropland is mainly concentrated in the low elevation plains, such as the Mekong
River Delta in the south and the Red River Delta in the northeast of the study area; woodland
is mainly distributed in the middle-altitude and high-altitude mountains and hilly areas,
such as the Giai Song Chao Re mountain in the north of the study area and the Truong Son
mountain in the middle and south of the study area; water bodies are mainly distributed
in the Red River Delta in the northeast and Mekong River Delta in the south of the study
area, including Ca Mau, Bac Lieu and Kien Giang provinces; built-up land is sporadically
distributed in the eastern coastal areas, especially in the two major plains in the northeast and
south, where there are some large cities distributed, such as Ha Noi and Ho Chi Minh.
During the period 2000–2020, the study area generally showed a decrease in the area
of cropland, a significant increase in the area of water bodies and built-up land and an
almost constant area of woodland (Figure 5A). The changes were also consistent with the
area percentage change in LULC composition (Figure 5B), where the 12area
Remote Sens. 2022, 14, x FOR PEER REVIEW of 18percentages of

water bodies and built-up lands increased; however, that of cropland decreased.

Figure 5. Change tendencies (A) and area percentages (B) of various land use/land cover types from
Figure 5. Change tendencies (A) and area percentages (B) of various land use/land cover types from
2000 to 2020.
2000 to 2020.
Specifically, the largest increase in the study area was built-up land (0.35 × 104 km2,
Specifically,
93.49%). the largest
The rapid increase increase in
in construction landthe studyoccurred
mainly area was built-up
after 2011. Theland
area(0.35
of × 104 km2 ,
93.49%).
water bodiesThe
alsorapid
showedincrease in construction
a continuously increasingland
trendmainly occurred
(0.59 × 10 after 2011.
4 km2, 54.19%). The The area of
area
waterof woodland
bodies also onlyshowed
showed aa slight change. However,
continuously the area
increasing of cropland
trend (0.59 × 10 4 km2a, 54.19%). The
showed
general
area ofdecrease
woodland × 104 km
(1.17only 2, 8.64%). Cropland showed a yearly decrease during the
showed a slight change. However, the area of cropland showed a
period 2000–2005, a slight increase between 2005 and 2009 and another decrease during
general decrease (1.17 × 104 km 2 , 8.64%). Cropland showed a yearly decrease during the
the period 2009–2020.
period 2000–2005, a slight increase between 2005 and 2009 and another decrease during the
From the year 2000 to 2020, the area of cropland transferred out reached 3.28 × 104
period 2009–2020.
km , mainly from woodland and water bodies (Table 5), mostly in the southern (Ca Mau,
2

From
Bac Lieu, Socthe year
Trang, 2000
Long Antoand
2020, theNai)
Dong areaandof central
cropland(Dak transferred
Lak and GiaoutLai)reached
provinces3.15 × 104 km2 ,
mainly
of the studyfrom
area,woodland and waterinbodies
with little distribution (Table
the northern part5),(Lang
mostly
Son inandthe southern
Quang Ninh); (Ca Mau, Bac
during the same
Lieu, Soc Trang,period,
Long theAn
areaandof cropland
Dong Nai) transferred in reached
and central 2.11
(Dak × 10and
Lak 4 km2, mainly
Gia Lai) provinces of
from woodland
the study area,and water
with bodies,
little with the most
distribution in thedistribution
northern in the(Lang
part northern
Son(Tuyen
and Quang Ninh);
Quang, Yen Bai and Thai Nguyen) and south-central areas (Dak Nong, Lam Dong and 4
during the same period, the area of cropland transferred in reached 2.11 × 10 km2 , mainly
Ninh Thuan) of the study area; in general, cropland showed a shrinking trend (Figure 6).
from woodland and water bodies, with the most distribution in the northern (Tuyen Quang,
Remote Sens. 2022, 14, 1600 12 of 17

Yen Bai and Thai Nguyen) and south-central areas (Dak Nong, Lam Dong and Ninh Thuan)
of the study area; in general, cropland showed a shrinking trend (Figure 6).

Table 5. Land-use/land cover transfer matrix in Vietnam during the period 2000–2020.

2020

2000 Croplands Woodlands Water Bodies Built-Up Lands

Area (104 km2 ) Area (104 km2 ), Area (104 km2 ), Area (104 km2 ),
Percentage (%) Percentage (%) Percentage (%) Percentage (%)
Croplands 10.14, 75.6% 1.86, 13.8% 0.99, 7.4% 0.3, 2.3%
Woodlands 1.70, 9.6% 15.93, 89.9% 0.08, 0.5% 0.00, 0%
Water Bodies
Remote Sens. 2022, 14, x FOR PEER REVIEW 0.41, 37.3% 0.03, 2.7% 0.61, 55.5% 13 of0.05,
18 4.5%
Built-up Lands 0.00, 0% 0.00, 0% 0.00, 0% 0.38, 100%

Figure Spatialdistributions
6. Spatial
Figure 6. distributions of gains/losses
of gains/losses of cropland
of cropland (A), woodland
(A), woodland (B), water(B),
bodywater body (C) and
(C) and
built-up land
built-up land(D) during
(D) the period
during 2000–2020.
the period Loss indicates
2000–2020. Loss that the landthat
indicates cover category
the was con-
land cover category was
verted to other
converted land land
to other covercover
categories, while gain
categories, shows
while gainthat otherthat
shows landother
use types
land were transferred
use types were transferred
to the given land cover type from 2000 to 2020.
to the given land cover type from 2000 to 2020.
Table 5. Land-use/land cover transfer matrix in Vietnam during the period 2000–2020.

2020
Croplands Woodlands Water Bodies Built-up Lands
2000
Area (10 km )
4 2 Area (10 km ), Area (10 km ),
4 2 4 2 Area (104 km2),
Remote Sens. 2022, 14, 1600 13 of 17

The conversion area of woodland to cropland and water bodies (Table 5) was
1.78 × 104 km2 and was distributed in the northwest (Lao Cai, Yen Bai and Son La) and
most of the central part (Dak Nong, Lam Dong and Binh Phuoc) of the study area; the
transferred area of woodland land reached 1.89 × 104 km2 , mainly from cropland and
water bodies, and was mainly distributed in a small part of the south-central parts of
the study area (Dak Lak and Gia Lai) (Figure 6); in general, the area of woodland in the
study area barely changed.
The transferred out area of water bodies was 0.49 × 104 km2 , and the main transferred
out types were cropland and built-up land (Table 5), which are mainly distributed in
the Mekong River basin in the south of the study area, specifically, An Giang, Dong
Thap and Can Tho; the transferred in area of water bodies was 1.07 × 104 km2 , and the
main transferred in types were cropland and woodland, which are also distributed in the
southern parts of Vietnam, including Bac Lieu, Ca Mau and Kien Giang (Figure 6). In
general, the area of water bodies in the study area showed an increasing trend.
The area of built-up land transferred in was 0.35 × 104 km2 , and the main transfer
types were cropland and water bodies (Table 5), which are mainly distributed in the
northern parts of the study area (Bac Ninh, Ha Noi and Hung Yen) and the southern core
urban areas (Long An, Tien Giang and Binh Duong) (Figure 6).

3.3. Analysis of Driving Mechanism Based on Statistical Relationship

Through principal component analysis, information on the changes in all nine indi-
cators could be concentrated on two principal components (Table 6). The first principal
component (F1) is a characterization of economic development, mainly reflecting the
changes in the gross domestic product (X6), industrial value added (X8), manufacturing
value added (X7), total population (X3) and urbanization rate (X5). The second principal
component (F2) is a characterization of regional climate change, mainly reflecting total
summer rainfall (X1) and total annual rainfall (X2).

Table 6. Rotated component matrix of the principal component analysis.

Component
Variables Description
F1 F2
X1 Total summer rainfall −0.402 0.761
X2 Total annual rainfall −0.182 0.876
X3 Total population 0.997 0.053
X4 Agricultural population 0.897 0.008
X5 Urbanization rate 0.997 0.057
X6 Gross Domestic Product 0.989 0.084
X7 Manufacturing Value Added 0.967 0.088
X8 Industrial Value Added 0.990 0.089
X9 Agriculture Value added 0.983 0.093

Through multiple linear regression analysis, the statistical relationship model of land
area change and the principal components of the driving factors could be obtained (Table 7).
Among them, the driving relationship models of cropland, and built-up land passed the
significance test, while the driving relationship models of woodland area change did not
pass the significance test. The specific models showed the following:

Table 7. Relationships between land cover change and principal components in different types.

Cropland Woodland Water Body Built-Up Land

Y1 = 14.08 − 0.23 × F1 + 1.29 × F2 Y2 = 18.23 − 0.38 × F1 + 2.20 × F2 Y3 = 1.15 − 0.009 × F1 + 0.069 × F2 Y4 = 0.29 + 0.0026 × F1 ***
(R2 = 0.47, p < 0.05) (R2 = 0.04, p > 0.05) (R2 = 0.28, p > 0.05) (R2 = 0.67, p < 0.05)
Note: the significance test symbol *** is p < 0.001.
Remote Sens. 2022, 14, 1600 14 of 17

Cropland area showed a negative relationship with economic development (F1) and
a positive relationship with climate change (F2). This implies that urban construction
accompanying national economic development results in the shrinkage of the cropland
area; however, the cropland area in the study area, especially paddy fields, may increase
along with increased rainfall due to global climate change, etc.
The area of water bodies was negatively correlated with economic factors (F1) and
positively correlated with regional climate components (F2), but was mainly influenced by
regional climate change. This suggests that national economic development in Vietnam may
result in the encroachment of some traditional water bodies by urban and rural construction
sites, but it is clear that this encroachment process will not be a major factor. However,
climate change, especially increased rainfall, will result in a significant expansion in the area
of densely packed waters in the deltas of the two rivers in the north and south of Vietnam.
There was a very significant positive relationship between urban and rural built-up
land area and macroeconomics (F1), and no relationship with climate change. Obviously,
this reflects the historical process of the rapid expansion of urban areas with the rapid
development of the national economy and the gradual acceleration of the urbanization
process in Vietnam.
The area of woodland was negatively correlated with the economy (F1) and positively
correlated with climate change, but mainly showed the influence of the remaining climate.
This reflects the encroachment of urban development/construction land expansion on
traditional woodland, but this encroachment process is not a major factor (R2 = 0.04,
p > 0.05). However, climate change (mainly the increase in precipitation) will increase the
area of woodland cover within Vietnam.

4. Discussion
Using the GEE platform and the principle of “multi-source consistency” and “time-
series stability” to build a sample data set, combined with the random forest machine
learning algorithm, this study completed the entire land use/cover mapping task quickly
with high accuracy, at a national regional scale in Vietnam. However, the “classification
followed by comparison” approach we used probably ignores the continuity of the temporal
changes in land use/cover types, which may have led to irregular and jumpy changes
in meta-land cover types at different time points [8,15,37–39]. This possibly affected the
accuracy and credibility of the analysis of land cover change patterns in the study area.
On the other hand, constrained by the accuracy of RF classification, this study did not
distinguish between natural woodland, scrub woodland and economic woodland among
woodland types; open water and wetland in water body types; or dryland and paddy fields
in cropland types. This may have led to the land mapping results obtained from the study
being too coarse and not precise enough in terms of the land classification system. It should
be noted that Landsat 8 replaced Landsat 5/7 after 2013. There are minor differences in
sensor features, which may affect the land mapping process supported by the same RF
model and parameters. In addition, the authors also carried out accuracy assessment for
the land data sets in 2000, 2010 and 2020 based on random verification samples. The results
of those validation processes are very close to those presented in Table 4. However, in any
case the practice based on 600 “time-series stable” samples shown in this paper is different
from the classic method normally supported by random samples. It may be necessary to
carry out a comparative study based on two different methods in the future.
The spatial pattern of land use/cover and change trends obtained in this study are
generally consistent with the results of previous studies. For example, Minderhoud et al.
found that cropland in the Mekong Delta was significantly transformed into built-up land
from 1988 to 2009 [40], and Leonardo et al. found a similar trend in the Tam Giang-Cau Hai
Lagoon area in the central part of the study area [26]. Muller et al. found that, since the
government proposed the intensive agricultural production and forest protection policy
in 1992, the forest coverage in the Dak Lak area has shown an increasing trend [41]. Tran
et al. found that the area of water bodies in Ca Mau province in the southern part of the
Remote Sens. 2022, 14, 1600 15 of 17

study area increased significantly from 1973 to 2011, which was mainly reflected in the
development of the aquaculture industry [42]. Our results revealed the temporal dynamics
of major land use/cover types over long time periods on national and provincial scales.
In this paper, we applied a combination of principal component analysis and multi-
ple linear regression analysis to determine the combined effects of natural climatic and
socioeconomic factors on land use/cover type changes. The area change in built-up land
was significantly correlated with macroeconomic indicators, such as population and urban-
ization growth, which means that some other land types were transformed into built-up
land as the urbanization process advanced. This is also consistent with previous findings
in the local areas of Vietnam (e.g., Mekong Delta region, Tam Giang-Cau Hai Lagoon
region) [26,40]. However, for cropland, woodland, and water bodies, although the multiple
regression models of area change with each key factor reached a statistical level of 0.05 over-
all, specific significant influencing factors were not clarified. This modeling approach based
on statistical analysis has limitations. The multiple linear regression method is mainly
based on the correlation between factors to measure, which does not imply a strict causal
relationship between the independent and dependent variables.

5. Conclusions
To address the key issues of the lack of national-scale, high-precision, automated
spatio-temporal continuous land use/cover data sets, and the scarcity of studies on the
driving mechanisms of land use change in Vietnam, this study used the GEE platform,
based on Landsat satellite imagery and several other important auxiliary data. We applied
“multi-source consistency” and “time-series stability” and RF methods to generate a high-
precision land use/cover data product in Vietnam during the period 2000–2020. The
technical route and modeling approach adopted in this paper are important references for
similar studies to be conducted in other regions in the future.
This study found that the land cover types in the Vietnam region are dominated by
cropland and woodland, and the area of built-up land and water bodies has increased
significantly over the past 20 years; the area of woodland did not show significant changes.
The expansion of built-up land is driven by regional economic development, and the area
changes in cropland, water bodies and woodland are influenced by both national economic
development and climate change. The results of the study have important referential value
for analyzing the relationship between land use/cover change and regional economic
development and industrial structure in Vietnam. This research helps others gain an
understanding of the response processes and mechanisms of regional land systems in the
context of global climate change and regional economic and social development, and with
promoting sustainable development in the study area.

Supplementary Materials: The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/rs14071600/s1, the detailed data introduction uploaded as a
DOC file, Table S1: Datasets used in the study; Table S2: Natural and Social indicators used in
the PCA.
Author Contributions: Conceptualization, Y.H.; data curation, X.G. and J.Y.; investigation, X.G.;
methodology, Y.H.; supervision, Y.H.; writing—original draft, X.G.; writing—review and editing,
X.G., J.Y. and Y.H. All authors have read and agreed to the published version of the manuscript.
Funding: This research was supported by the National Natural Science Foundation of China
(42130508, 41977421), National Key Research and Development Program of China (2021YFD1300501,
2016YFB0501502) and Strategic Priority Research Program of the Chinese Academy of Sciences
(XDA19040301, XDA20010202, XDA23100200).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Remote Sens. 2022, 14, 1600 16 of 17

Acknowledgments: We would like to express our sincere thanks to the anonymous reviewers, as the
comments and suggestions were of great help to improve the quality of this paper.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Lawler, J.J.; Lewis, D.J.; Nelson, E.; Plantinga, A.J.; Polasky, S.; Withey, J.C.; Helmers, D.P.; Martinuzzi, S.; Pennington, D.; Radeloff,
V.C. Projected land-use change impacts on ecosystem services in the United States. Proc. Natl. Acad. Sci. USA 2014, 111, 7492.
[CrossRef] [PubMed]
2. Foley, J.A.; DeFries, R.; Asner, G.P.; Barford, C.; Bonan, G.; Carpenter, S.R.; Chapin, F.S.; Coe, M.T.; Daily, G.C.; Gibbs, H.K.; et al.
Global consequences of land use. Science 2005, 309, 570. [CrossRef] [PubMed]
3. Turner, B.L., II; Lambin, E.F.; Reenberg, A. The emergence of land change science for global environmental change and sustain-
ability. Proc. Natl. Acad. Sci. USA 2007, 104, 20666. [CrossRef] [PubMed]
4. Alkama, R.; Cescatti, A. Biophysical climate impacts of recent changes in global forest cover. Science 2016, 351, 600. [CrossRef]
[PubMed]
5. Fan, Z.; Li, S.; Fang, H. Explicating the mechanisms of land cover change in the New Eurasian Continental Bridge Economic
Corridor region in the 21st century. J. Geogr. Sci. 2021, 31, 1403. [CrossRef]
6. Zhang, Y.; Chen, J.; Chen, L.; Li, R.; Zhang, W.; Lu, N.; Liu, J. Characteristics of land cover change in Siberia based on GlobeLand30,
2000–2010. Prog. Geogr. 2015, 34, 1324.
7. Dewan, A.M.; Yamaguchi, Y. Land use and land cover change in Greater Dhaka, Bangladesh: Using remote sensing to promote
sustainable urbanization. Appl. Geogr. 2009, 29, 390. [CrossRef]
8. Ye, J.; Hu, Y.; Zhen, L.; Wang, H.; Zhang, Y. Analysis on Land-Use Change and Its Driving Mechanism in Xilingol, China, during
2000–2020 Using the Google Earth Engine. Remote Sens. 2021, 13, 5134. [CrossRef]
9. Piao, Y.; Jeong, S.; Park, S.; Lee, D. Analysis of Land Use and Land Cover Change Using Time-Series Data and Random Forest in
North Korea. Remote Sens. 2021, 13, 3501. [CrossRef]
10. You-hua, R.A.N.; Xin, L.I.; Ling, L.U. Accuracy Evaluation of the Four Remote Sensing Based Land Cover Products over China. J.
Glaciol. Geocryol. 2009, 31, 490.
11. Bai, W.; Guo, D.; Tian, Q.; Liu, N.; Cheng, W.; Li, L.; Zhang, W.-H. Differential responses of grasses and forbs led to marked
reduction in below-ground productivity in temperate steppe following chronic N deposition. J. Ecol. 2015, 103, 1570. [CrossRef]
12. Jun, C.; Ban, Y.; Li, S. Open access to Earth land-cover map. Nature 2014, 514, 434. [CrossRef] [PubMed]
13. Chen, J.; Chen, J.; Liao, A.; Cao, X.; Chen, L.; Chen, X.; Peng, S.; Han, G.; Zhang, H.; He, C.; et al. Concepts and Key Techniques
for 30 m Global Land Cover Mapping. Acta Geod. Cartogr. Sin. 2014, 43, 551.
14. Ran, Y.; Li, X. The world’s first comprehensive high-resolution land cover map—Review of China’s 30m resolution global land
cover map. Sci. Sin. 2015, 45, 1243.
15. Hu, Y.; Hu, Y. Land Cover Changes and Their Driving Mechanisms in Central Asia from 2001 to 2017 Supported by Google Earth
Engine. Remote Sens. 2019, 11, 554. [CrossRef]
16. Su, B.; Noguchi, N. Discrimination of Land Use Patterns in Remote Sensing Image Data using Minimum Distance Algorithm and
Watershed Algorithm. Eng. Agric. Environ. Food 2013, 6, 48–53. [CrossRef]
17. Li, G.; Lu, D.; Moran, E.; Hetrick, S. Land-cover classification in a moist tropical region of Brazil with Landsat Thematic Mapper
imagery. Int. J. Remote Sens. 2011, 32, 8207. [CrossRef]
18. Bittencourt, H.R.; Clarke, R.T. Use of classification, and regression trees (CART) to classify remotely-sensed digital images. In
Proceedings of the 2003 IEEE International Geoscience and Remote Sensing Symposium, Toulouse, France, 21–25 July 2003; pp.
3751–3753.
19. Huang, C.; Davis, L.S.; Townshend, J.R.G. An assessment of support vector machines for land cover classification. Int. J. Remote
Sens. 2002, 23, 725. [CrossRef]
20. Shao, Y.; Lunetta, R.S. Comparison of support vector machine, neural network, and CART algorithms for the land-cover
classification using limited training data points. Isprs J. Photogramm. Remote Sens. 2012, 70, 78. [CrossRef]
21. Tassi, A.; Gigante, D.; Modica, G.; Di Martino, L.; Vizzari, M. Pixel- vs. Object-Based Landsat 8 Data Classification in Google
Earth Engine Using Random Forest: The Case Study of Maiella National Park. Remote Sens. 2021, 13, 2299. [CrossRef]
22. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5. [CrossRef]
23. Dong, J.; Xiao, X.; Menarguez, M.A.; Zhang, G.; Qin, Y.; Thau, D.; Biradar, C.; Moore, B., III. Mapping paddy rice planting area in
northeastern Asia with Landsat 8 images, phenology-based algorithm and Google Earth Engine. Remote Sens. Environ. 2016, 185, 142.
[CrossRef] [PubMed]
24. Akar, Ö.; Güngör, O. Classification of multispectral images using Random Forest algorithm. J. Geod. Geoinf. 2012, 1, 105.
[CrossRef]
25. Zhang, D.D.; Zhang, L. Land Cover Change in the Central Region of the Lower Yangtze River Based on Landsat Imagery and the
Google Earth Engine: A Case Study in Nanjing, China. Sensors 2020, 20, 2091. [CrossRef] [PubMed]
26. Disperati, L.; Virdis, S.G.P. Assessment of land-use and land-cover changes from 1965 to 2014 in Tam Giang-Cau Hai Lagoon,
central Vietnam. Appl. Geogr. 2015, 58, 48. [CrossRef]
Remote Sens. 2022, 14, 1600 17 of 17

27. Van, T.T.; Wilson, N.; Thanh-Tung, H.; Quisthoudt, K.; Quang-Minh, V.; Xuan-Tuan, L.; Dahdouh-Guebas, F.; Koedam, N. Changes
in mangrove vegetation area and character in a war and land use change affected region of Vietnam (Mui Ca Mau) over six
decades. Acta Oecol.-Int. J. Ecol. 2015, 63, 71. [CrossRef]
28. Chander, G.; Markham, B.L.; Helder, D.L. Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+,
and EO-1 ALI sensors. Remote Sens. Environ. 2009, 113, 893. [CrossRef]
29. Hu, Y.; Zhang, Q.; Dai, Z.; Huang, M.; Yan, H. Agreement analysis of multi-sensor satellite remote sensing derived land cover
products in the Europe Continent. Geogr. Res. 2015, 34, 1839.
30. Defourny, P.; Schouten, L.; Bartalev, S.; Bontemps, S.; Cacetta, P.; De Wit, A.; Di Bella, C.; Gérard, B.; Giri, C.; Gond, V. Accuracy
Assessment of a 300 m Global Land Cover Map: The GlobCover Experience. 2009. Available online: https://library.wur.nl/
WebQuery/wurpubs/fulltext/14725 (accessed on 5 March 2022).
31. Chen, J.; Chen, J.; Liao, A.; Cao, X.; Chen, L.; Chen, X.; He, C.; Han, G.; Peng, S.; Lu, M.; et al. Global land cover mapping at 30 m
resolution: A POK-based operational approach. ISPRS J. Photogramm. Remote Sens. 2015, 103, 7. [CrossRef]
32. Ashouri, H.; Hsu, K.-L.; Sorooshian, S.; Braithwaite, D.K.; Knapp, K.R.; Cecil, L.D.; Nelson, B.R.; Prat, O.P. PERSIANN-CDR Daily
Precipitation Climate Data Record from Multisatellite Observations for Hydrological and Climate Studies. Bull. Am. Meteorol. Soc.
2015, 96, 69. [CrossRef]
33. Olofsson, P.; Foody, G.M.; Herold, M.; Stehman, S.V.; Woodcock, C.E.; Wulder, M.A. Good practices for estimating area and
assessing accuracy of land change. Remote Sens. Environ. 2014, 148, 42. [CrossRef]
34. Congalton, R.G. A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote Sens. Environ. 1991, 37, 35.
[CrossRef]
35. Liu, C.R.; Frazier, P.; Kumar, L. Comparative assessment of the measures of thematic classification accuracy. Remote Sens. Environ.
2007, 107, 606. [CrossRef]
36. Jolliffe, I.T.; Cadima, J. Principal component analysis: A review and recent developments. Philos. Trans. R. Soc. A-Math. Phys. Eng.
Sci. 2016, 374. [CrossRef] [PubMed]
37. Zhu, Z.; Woodcock, C.E. Continuous change detection and classification of land cover using all available Landsat data. Remote
Sens. Environ. 2014, 144, 152. [CrossRef]
38. Kennedy, R.E.; Yang, Z.; Cohen, W.B. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1.
LandTrendr—Temporal segmentation algorithms. Remote Sens. Environ. 2010, 114, 2897. [CrossRef]
39. Hu, Y.F.; Dong, Y.; Batunacun. An automatic approach for land-change detection and land updates based on integrated NDVI
timing analysis and the CVAPS method with GEE support. ISPRS J. Photogramm. Remote Sens. 2018, 146, 347. [CrossRef]
40. Minderhoud, P.S.J.; Coumou, L.; Erban, L.E.; Middelkoop, H.; Stouthamer, E.; Addink, E.A. The relation between land use and
subsidence in the Vietnamese Mekong delta. Sci. Total Environ. 2018, 634, 715. [CrossRef]
41. Muller, D.; Zeller, M. Land use dynamics in the central highlands of Vietnam: A spatial model combining village survey data
with satellite imagery interpretation. Agric. Econ. 2002, 27, 333. [CrossRef]
42. Hanh, T.; Thuc, T.; Kervyn, M. Dynamics of Land Cover/Land Use Changes in the Mekong Delta, 1973-2011: A Remote Sensing
Analysis of the Tran Van Thoi District, Ca Mau Province, Vietnam. Remote Sens. 2015, 7, 2899. [CrossRef]