Spatiotemporal Characteristics and Driving Factors of Land-Cover Change in the Heilongjiang (Amur) River Basin

Abstract

Monitoring land-use and land-cover change (LUCC) is extremely important in the sustainable development and management of terrestrial ecosystems. Taking the Heilongjiang (Amur) River Basin as the study area, this study aimed to identify the spatiotemporal characteristics of land cover at the entire basin and at the country levels of the three countries using the land-use change index, frequency statistics and land-cover change degree. The results showed that: (1) The LULC types were mainly forest land and grassland (accounting for nearly 83% in total) from 2001 to 2019. The main land-cover types in China, Russia and Mongolia were forest land, forest land and grassland, respectively. (2) The area of urban land, cropland and wetland increased significantly from 2001 to 2019, while the area of forest land and bare land decreased during this time. In general, the ecological environment has greatly improved over the last 19 years, although the relevant ecological restoration still needs to be further implemented. The findings can provide a scientific basis for ecological protection and sustainable development in the Heilongjiang (Amur) River Basin. The approaches here are also applicable to land-cover change research in other similar regions.

1. Introduction

Global environmental change and sustainability are being driven by land-cover change science [1,2]. By utilizing natural resources related to land, humans alter the surface cover of the Earth [3]. Changes in land cover can adversely affect natural elements such as the atmosphere, soil, vegetation, water resources and biological diversity, and then have a wide range of influences on the chemical cycles of the biological earth, energy exchange, water cycles and ecological processes. In the ecosystems of Earth, land-use and land-cover changes reflect human interaction with the environment and are increasingly regarded as factors contributing to global environmental change [4]. Ecosystem services and biodiversity are the results of the interaction between climate change and land-cover change [5]. Hence, understanding the driving factors of land-cover change and their impacts on ecological and anthropogenic processes is of great importance. At the end of the 20th century, under the strong promotion of the International Geosphere and Biosphere Program (IGBP) and the International Human Dimensions Programme on Global Environmental Change (IHDP), land use/land cover made significant progress in various aspects, such as spatiotemporal exploration, driving mechanism analysis, process characterization and simulation and macro-ecological effect evaluation, making effective contributions to understanding and improving global environmental change and achieving sustainable development [6].

Land-cover change has always been a central component in land-change science and plays a central role in current strategies for managing natural resources and monitoring environmental changes [7,8]. Earth observation services guarantee continuous land-cover mapping and are becoming of great interest worldwide [9]. Remote-sensing earth observation technology provides abundant, free, long-term, medium-resolution and high-resolution satellite imagery resources for land-use/land-cover research due to its rapid advancement [10]. With the development of remote sensing technology and classification algorithms, a series of land-use/land-cover products on regional, national and global scales have been developed at different spatial resolutions, such as GlobeLand30 [11], GLC_FCS30 [12], FROM-GLC [13], GLS_LC100 [14] and CCI_LC [15], among others.

The concepts of land use and land cover are fundamentally different but closely related. Land use refers to the purposes for which humans exploit the land cover and includes land management practices such as agricultural land, industrial land, transportation land, residential land, etc. [6,8]. According to the definitions of IGBP and IHDP, land cover refers to the layer of soils and biomass, including natural vegetation, crops and human structures that cover the land surface [8,16], such as various crops, forests, grasslands, houses, cement and asphalt pavements. On the one hand, land use is a direct reason for changing land cover, and on the other hand, its utilization is often constrained by land cover [6]. Land cover is thus directly observable, both in the field and in remote sensing images, while land use is not always easily observable [8]. However, from the perspective of spatial information product classification, land use and land cover can often be mutually transformed [6]. Since the late 20th century, global change issues have attracted a large amount of attention. As a key component of global changes, land-cover and land-use information have become increasingly important for an improved understanding of global environmental changes and feedbacks between social and environmental systems [2,17].

In the analysis of LULC-type changes, the application of the transfer matrix is extensive [18,19,20,21,22], but there are drawbacks such as less intuitive data expression and unfavorable trend analysis. For this reason, some scholars use a chord diagram to visualize the transition matrix in order to identify the flow, direction and diversity of LULC [23,24] more directly. However, chord diagrams cannot intuitively describe the comparative relationships between multiple time periods and have the drawbacks of complex graphics and difficulty in reading. The Sankey Diagram can effectively compensate for the shortcomings of chord diagrams and is very suitable for describing the transformation trajectories of LULC types over multiple research periods [25,26,27]. In terms of analyzing the intensity of LULC changes, comprehensive land-use dynamics [20,21,28], single land-use dynamics [20,28] and LULC change rate [24] are often used to describe the overall changes and severity of LULC. The land-use-type flow index [28] and the land-cover-type transfer rate [22] can further describe the conversion intensity between different LULC types. In addition, LULC has strong spatial differentiation, and analysis methods based on spatial features can better reveal the spatial pattern of LULC changes, such as the landscape pattern index [20], graph analysis [29,30], the grid unit method [30], etc. Among them, the grid element method is an effective means of expressing the spatial change pattern of LULC types [31], which can comprehensively reflect the spatial distribution of LULC change areas.

Affected by the natural environment and local economic policies, the characteristics of distribution and spatiotemporal changes of land-cover types in different countries and regions show obvious differences, especially in cross-border areas [32,33]. The Heilongjiang (Amur) River Basin (HARB) has a unique physical geography environment, rich and diverse species and natural resources, spanning China, Russia and Mongolia, with its southwestern side located within Mongolia, its southern side located within China and its northern side and eastern side located within Russia. In recent years, with various factors such as global population growth, climate change and frequent natural disasters, the ecosystem of the HARB has also been affected to a certain extent, and its function is gradually weakening.

Due to the function that land use and land cover (LULC) serve at a watershed scale, land-use and land-cover change (LULCC) are of paramount importance [34]. We used various methods to monitor land-use/land-cover change to analyze the status of land-use/land-cover change within the HARB. We also elucidated land-cover change and its driving factors in the past 19 years and achieved a comparative analysis of land-cover change characteristics in cross-border areas of China, Russia and Mongolia. The main reasons for land-cover change were also explored. The aim is to provide a reference for the study of human–land relations and land resource management in the HARB, lay the foundation for the sustainable use of land resources and ecological environment protection in China and Russia and provide basic data and a scientific basis for relevant research in the northeastern border area of China.

2. Materials and Methods

2.1. Study Area

The Heilongjiang (Amur) River Basin (HARB) is located in the middle- and high-latitude parts of eastern Eurasia [35] and is one of the ten biggest river basins in the world. It is situated between 107.051°E–141.128°E and 41.72°N–55.903°N. The total area of the HARB is 208.33 $\times$ 10⁴ km², and the flow path is 1.68 $\times$ 10⁴ km. The HARB is conceived as a whole in terms of its natural environment, but it is a cross-border region in the administrative division, which includes parts of China (41%), Russia (50%) and Mongolia (9%). The climate of the HARB varies widely because it is deeply influenced by the high- and low-pressure centers of the inland and seas and the alternation of the monsoon. In winter, the Siberian cold and dry monsoon from the west controls the HARB, while in summer, the warm and humid monsoon from the east controls the basin [36]. At the same time, the basin climate is sensitive to global climate change and is becoming more complicated due to the variable terrain. The terrain of the HARB is higher in the west and lower in the east. The western part of the basin is mainly composed of mountainous and plateau areas, such as the Greater Khingan Mountains and the Mongolian Plateau, and the southern and eastern parts are mainly flat plains, that is, the Songnen Plain and the Sanjiang Plain (see Figure 1). Because of the changeable topography and complex climate, the land-cover types in the basin are varied.

2.2. Data Sources

The datasets used in this paper mainly included land-cover data, the Digital Evaluation Model (DEM), meteorological data, field observation data and socioeconomic data.

(1) Land-cover datasets. The Collection 6 land-cover data (2001–2019) used in this study were the latest version of the Moderate Resolution Imaging Spectroradiometer (MODIS)-based Land-Cover-Type product (MCD12Q1) [37], available at https://earthexplorer.usgs.gov/ (Accessed on 10 May 2021). The MODIS Land-Cover-Type product contains multiple classification schemes that describe land-cover properties derived from observations spanning a year’s input of Terra and Aqua data. The MCD12Q1 product was derived through a supervised decision-tree classification method with five different land-cover classification schemes, including the International Geosphere Biosphere Programme (IGBP) legend [38,39], the University of Maryland (UMD) legend [40], the Leaf Area Index and Fraction of Photosynthetically Active Radiation (LAI/FPAR) biome legend [41], the Biome-Biogeochemical Cycles BGC model legend [42] and the Plant Functional Types (PFTs) legend [43]. The Collection 6 algorithm uses a hierarchical classification model, where the classes contained in each level of the hierarchy reflect structured distinctions between land-cover properties [37]. The Collection 6 algorithm also incorporates a state-space multitemporal modeling framework based on hidden Markov models, which can reduce the spurious land-cover changes caused by classification uncertainty in individual years [37]. The amount of spurious land-cover change has been significantly decreased in Collection 6, accounting for only 1.6%. The accuracy of the IGBP layer of the MCD12Q1 was estimated to be 73.6% globally. A Chinese scholar discussed the accuracy analysis and applicability evaluation of the MODIS land-cover product in the Yellow River source region from two aspects: quantity accuracy and shape consistency. The results showed that, in terms of shape, the overall shape consistency with weights is above 69%, with the consistency of the main grassland types reaching over 88%. In terms of quantity, the relative error of the overall area with weights is within 26%, and the error mainly occurs in unused land and other land types. The MODIS land-cover data product is still of importance and has application value in large-scale land-cover monitoring [44]. An evaluation of the consistency of five land-cover classification schemes of MCD12Q1 was performed based on the 30 m Chinese GlobeLand30-2010 GLC product in Anhui Province. The overall accuracy of IGBP, UMD and PFT in the three sub-regions of Anhui Province was more than 71%, while it was less than 57% for LAI/FPAR and NPP [45]. We also compared five classification schemes based on spatial consistency and area consistency and found that the consistency of the IGBP classification scheme was the best. Thence, the IGBP classification scheme was chosen to analyze the temporal and spatial changes of land cover in the study area.

The MODIS Land-Cover-Type product (MCD12Q1) supplies global maps of land cover at annual time steps and 500 m spatial resolution. MCD12Q1 data are provided as tiles that are approximately 10° × 10° at the Equator using a Sinusoidal grid in HDF4 file format [46]. The HARB contains a total of seven tiles, including H24V04, H25V03, H25V04, H26V03, H26V04, H27V03 and H27V04. We collected 133 tiles of MCD12Q1 products from 1 January 2001 to 31 December 2019. The MODIS Reprojection Tool (MRT) was used to finish the imaging mosaicking, format conversion, reprojection and resampling [47]. Here, the original HDF4 was converted into Geotiff, while the projection was converted to WGS_1984_Albers to ensure the smallest area deformation. In addition, we extracted the land-cover types of the study area by the boundary of the HARB and reclassified them to seven types, that is, Forest (F), Grassland (GL), Wetland (WL), Farmland (FL), Urban (UB), Barren (B) and Water (W) in ArcGIS 10.2.

(2) Boundary data. The boundary data for the Heilongjiang (Amur) River Basin are available at http://www.geodoi.ac.cn/WebCn/Default.aspx (Accessed on 12 May 2021). These data are based on ASTER-DEM data; preliminary results were extracted using watershed area extraction algorithms and then generated by matching them with Google Earth remote sensing images.

(3) DEM. The DEM data were obtained from the geospatial data cloud and had a resolution of 90 × 90 m (http://www.gscloud.cn/ (Accessed on 12 May 2021)). The DEM dataset was resampled from approximately 90 m to 500 m and reprojected to WGS_1984_Albers. In addition, we finished the imaging mosaicking and the image extraction by the boundary of the HARB and calculated the slope of the HARB in ArcGIS10.2.

(4) Meteorological data. The meteorological data used in this study were the monthly gridded Climatic Research Unit Time-Series data version 4.05 (CRU TS4.05), acquired from https://data.ceda.ac.uk/badc/cru/data/cru_ts/cru_ts_4.05/data (Accessed on 15 May 2021). CRU TS4.05 datasets with spatial resolution of 0.5° and long-time span from 1901 to 2020 are stored as NC files, produced by CRU at the University of East Anglia [35]. We chose factors ‘pre’ and ‘tmp’ from CRU TS4.05 as precipitation and temperature to analyze climate change and its relationship with land-cover change. The NC files were converted into Geotiff format by a Python script file. In ArcGIS10.2, the precipitation and temperature datasets were resampled to 500 m by using the cubic convolution method, and annual precipitation and annual average temperature were calculated from 2001 to 2019.

(5) Field observation data. In May 2019, we carried out a week-long field survey in Primorsky Krai, Russia, mainly on the west, east and south banks of Lake Khanka, Amur Bay, Ussuri Bay, the upper reaches of Ussuri and the Razdolnaya basin. We collected relevant data, including surface coverage and vegetation species.

(6) Socioeconomic data. The economic, population and agricultural data of the study area in China were acquired from the 2001–2019 statistical yearbooks of Heilongjiang Province, Jilin Province and Inner Mongolia Autonomous Region Liaoning Province; the relevant data in Russia and Mongolia were downloaded from the World Bank data platform (https://data.worldbank.org.cn/ (Accessed on 20 May 2021)) and the 2001–2019 Russian statistical yearbook, available at https://rosstat.gov.ru/ (Accessed on 15 May 2019).

2.3. Method

2.3.1. Spatiotemporal Changes of Land Cover and Ecological Value Judgment

In this study, the spatiotemporal changes of seven land-cover types were analyzed from four regions: the entire HARB, the Chinese region of the HARB, the Russian region and the Mongolian region. It was found that the ecosystem service value per unit area of each ecosystem type is different, and the order from high to low is wetland, water, forest, grassland, farmland, barren and urban [47,48,49,50]. In the analysis of land-cover dynamic change, we defined the conversion of land-cover types with increasing ecosystem service value as landscape improvement and the conversion with decreasing ecosystem service value as landscape degradation.

2.3.2. Dynamic Analysis of Land Cover

Land-use dynamic was used in the quantitative description of land-use change rate in different study periods, including the single land-use dynamic index and the comprehensive land-use dynamic index [51]. In this paper, these two indicators were used to represent a certain land-cover type and the rate of regional land-cover change during the study period [52] and then to analyze the rate of land-cover change in the HARB and explore the impact of different national policies on the land-cover change rate.

The single land-use dynamic index can be defined as:

$$D_s=\frac{A_{t2}-A_{t1}}{A_{t1}}\times \frac{1}{t2-t1}\times 100\%$$

(1)

where $A_{t1}$ and $A_{t2}$ are the areas of the land-cover type at times $t1$ and $t2$.

The comprehensive land-use dynamic index can be defined as:

$$D_c=\frac{{{\displaystyle \sum }}_{i=1}^n\left|\Delta A_{in-i}-\Delta A_{out-i}\right|}{2{{\displaystyle \sum }}_{i=1}^n{A}_i}\times \frac{1}{t2-t1}\times 100\%$$

(2)

where $A_i$ is the area of the land-cover type i in time $t1$; $\Delta A_{in-i}$ is the total area converted from other types to type i from $t1$ to $t2$; and $\Delta A_{out-i}$ is the total area converted from type i to other types from $t1$ to $t2$.

The dynamic degree of land use quantitatively describes the speed of land-use change, which plays a positive role in predicting the trend in land-use change in the future [51].

2.3.3. Land-Use Integrated Index

The change in extent of land use mainly reflects the breadth and depth of land use, not only in its natural attributes but also in the comprehensive effects of human factors and natural environmental factors. According to the land-use integrated index, which was put forward by Liu Ji-yuan [53], the change extent of land cover can be divided into four levels. The highest level is the use of land that completely destroys the natural environment, such as towns, residential areas and industrial development land; the land resources that have not been utilized by humans or cannot be utilized from the perspective of economy and technology are defined as the lowest level. Between these two levels, due to the different degrees of human intervention in the natural environment of the land, two intermediate levels are defined. Assign the value of its own category to the four land-use levels, and then calculate the grading index of the four land-use levels (

Table 1. Land-use degree classification assignment table.

Type	Unused Land	Forest, Grass, Water	Farmland	Urban
Land-cover type	Barren	Forest, Grassland, Water, Wetland	Farmland	Urban
Graded index	1	2	3	4

).

The degree of land-cover change can be used to better understand the trend and its driving forces [51,52,54]. The equation is shown as follows:

$$L_d={\displaystyle {\sum }_{i=1}^n}{A}_i{C}_i\times 100, L_j\in \left[100,400\right]$$

(3)

where $A_i$ is the given index of level i, $C_i$ is the percentage of the area of level i in the study area and n is the number of levels.

Furthermore, the change in quantity and the rate of land-use intensity can quantitatively reveal the comprehensive level and development trend in land use in this region.

$$\Delta L_{b-a}=L_b-L_a=\left({{\displaystyle \sum }}_{i=1}^n{A}_i{C}_{ib}-{{\displaystyle \sum }}_{i=1}^n{A}_i{C}_{ia}\right)\times 100$$

(4)

$$R=\frac{{{\displaystyle \sum }}_{i=1}^n{A}_i{C}_{ib}-{{\displaystyle \sum }}_{i=1}^n{A}_i{C}_{ia}}{{{\displaystyle \sum }}_{i=1}^n{A}_i{C}_{ia}}$$

(5)

where $L_b,L_a$ denote the change extent of land use at periods a and b and $C_{ia}$ and $C_{ib}$ are the area ratio of the ith level at periods a and b. When $\Delta L_{b-a}>0$ or $R>0$, land use is at the development stage or at the adjustment or recession stage [47,50].

3. Results

3.1. Spatiotemporal Changes of Land Cover

3.1.1. Spatiotemporal Changes of Land Cover in HARB

The spatial distribution maps of land cover in the HARB from 2001 to 2019 are given in Figure 2. The main land-cover type in the HARB during the study period was forest, accounting for about 50% of the total area and being mainly distributed in the north and east of the HARB. The second largest land-cover type was grassland, accounting for about 33% of the total area, distributed in the Mongolian plateau and the west of the Greater Khingan. Farmland accounted for about 15% of the total area, mainly distributed in the Sonnen Plain and the Sanjiang Plain, and a small part was distributed near Xingkai Lake. Other land-cover types, including wetland, urban, barren and water areas, accounted for less than 2% of the total area.

The area of different land-cover types from 2001 to 2019 is presented in Figure 3. The best trend-fitting equations of each land-cover type’s change with time were all in cubic form, and all passed the significance test (Figure 3,

Table 2. Land-cover types’ change from 2001–2019.

Change Types	Area (km2)	Ratio (%)	Change Types	Area (km2)	Ratio (%)
F–GL	66,891.25	3.21	B–GL	395.75	0.02
GL–F	57,156.75	2.74	WL–GL	381	0.02
GL–FL	35,741	1.72	W–GL	340.5	0.02
FL–GL	19,470	0.93	FL–W	275.5	0.01
FL–F	3693	0.18	GL–B	244	0.01
F–FL	2358	0.11	F–W	237.5	0.01
GL–WL	1642.5	0.08	FL–WL	218.25	0.01
GL–W	448	0.02	F–WL	209.5	0.01
FL–U	432	0.02	others	1069	0.05

). The data show that the area change in land-cover type in the study area was affected by many driving factors. We found that the speed of change in wetland areas was the highest, with an increase of 71.23% from 2001 to 2019. The change trend in the wetland area was decreasing first and then increasing, and the increased area of wetland was 1591 km². The change trend in the water area had the same trend as the wetland area; the area increased by 692.5 km², and the change rate was 3.86%. The area of urban land continued to increase, from 10,418.25 km² in 2001 to 10,973.75 km² in 2019, and the change rate was 5.33%. The change in farmland area also showed a trend of first decreasing and then increasing. On the whole, farmland increased by 14,031.5 km² during the study period. The change trend in forest and grassland areas was the opposite, but the change extent was the same. The area of forest was 1,070,843.75 km² in 2010, and it was the largest area in the study period. While the area of grassland was the smallest in 2010, it was 681,478.75 km². The barren area fluctuated around 1150 km² and decreased by 200.5 km² from 2001 to 2019.

During the study period, there were 36 types of land-cover-type transformation processes in the HARB, and the area where land-cover types changed accounted for 9.18% of the total area of the entire basin. The trend that the total area of land experienced a change in land-cover types every year was slow before 2016, but after that, the change intensified. The major land-cover-type transformation processes with the top 10 largest areas of change were selected to analyze the converted processes (Figure 4). The area of forest that converted to grassland was decreasing before 2010, but it increased afterwards, while there was an opposite trend of grassland converting to forest at the same time. The area of grassland converted to other land-cover types has been roughly increasing since 2001.

In order to analyze the macro-trends in landscape dynamic changes throughout the years, we further selected 17 land-cover-type transformation processes that were greater than 0.01% of the entire basin area for landscape quality change analysis (Table 2, Figure 5). The result showed that the area of landscape improvement accounted for 4.04% of the whole basin, and the main processes of landscape improvement included: (1) the transformation from grassland to forest, mainly distributed in the Greater Khingan and the north of the Russian region; and (2) the conversion of farmland to grassland or forest, mainly distributed in the low-altitude area of the Greater Khingan. However, landscape degradation occurred in 5.14% of the whole basin area, and the main processes were as follows: (1) forest degenerated to grassland or farmland, mainly distributed in Zabaykalsky Krai, both sides of the Heilongjiang River and the Greater Khingan; and (2) grassland was reclaimed into farmland, mainly distributed in the low-altitude areas of the Greater Khingan and around Xingkai Lake.

3.1.2. Spatiotemporal Changes of China’s Land Cover in HARB

The total area of the region that is located in the territory of China is 88.18 $\times$ 10⁴ km², accounting for about 41% of the HARB. The main land-cover type in this region in 2019 was forest, followed by farmland and grassland. Forests are distributed in the Greater Khingan, the Little Khingan and Changbai Mountain. The area of farmland in the region was 29.9 $\times$ 10⁴ km² in 2019, which accounted for about 95% of the farmland area in the whole basin. Grassland was mainly distributed in the north of Inner Mongolia, and its area was 22.88 $\times$ 10⁴ km² in 2019. Two-thirds of the urban area in the HARB was located in the Chinese region, and the area was 7421.5 km² in 2019.

The change trend in land-cover types in the Chinese region was consistent with the change trend in land-cover types in the entire basin, but the change rate of all land-cover types in the Chinese region was greater than that of the whole basin, which indicated that land-cover changes in the Chinese region were more intense. The change trend in forest, wetland and water was increasing first and then decreasing. In the study period, the areas of forest, wetland and water increased by 4708.25 km², 1178 km² and 655.25 km², respectively. The change rate of the wetland area, at 112.51%, is the largest among all the land-cover types. The farmland area and the urban area also showed an increasing trend and increased by 12,636 km² and 550.25 km², respectively, while the grassland area and the barren area decreased by 19,508 km² and 219.75 km², respectively (

Table 3. The area of land-cover changes in the Chinese region in 2001, 2010 and 2019.

Land-Cover Type	2001	2010	2019	2001–2010		2010–2019		2001–2019
	Area/km2	Area/km2	Area/km2	Change Area/km2	Change Rate/%	Change Area/km2	Change Rate/%	Change Area/km2	Change Rate/%
Forest	332,612.75	353,607	337,321	20,994.25	6.31	−16,286	−4.61	4708.25	1.42
Grassland	248,360	226,667.75	228,852	−21,692.3	−8.73	2184.25	0.96	−19,508	−7.85
Wetland	1047	1599	2225	552	52.72	626	39.15	1178	112.51
Farmland	286,430.5	286,120.25	299,066.5	−310.25	−0.11	12,946.25	4.52	12,636	4.41
Urban	6871.25	7072	7421.5	200.75	2.92	349.5	4.94	550.25	8.01
Barren	793.25	769	573.5	−24.25	−3.06	−195.5	−25.42	−219.75	−27.70
Water	5699.50	5979.25	6354.75	279.75	4.91	375.50	6.28	655.25	11.50

).

3.1.3. Spatiotemporal Changes of Russia’s Land Cover in HARB

Half of the total area of the HARB is located in Russia, covering an area of 101.02 $\times$ 10⁴ km². In the Russian region, forest was the most important land-cover type, accounting for about 70% of the total area of the Russian region and being evenly distributed in the region. Grassland was the second largest land-cover type, with a total area of 28.9209 $\times$ 10⁴ km² in 2019, accounting for about 27% of the total area of the Russian region. The total area of other land-cover types, that is, wetland, farmland, urban, barren and water, accounted for less than 3% (

Table 4. The area of land-cover changes in the Russian region in 2001, 2010 and 2019.

Land-Cover Type	2001	2010	2019	2001–2010		2010–2019		2001–2019
	Area/km2	Area/km2	Area/km2	Change Area/km2	Change Rate/%	Change Area/km2	Change Rate/%	Change Area/km2	Change Rate/%
Forest	702,521.75	714,935.75	689,516.25	12,414	1.77	−25,419.5	−3.56	−13,005.5	−1.85
Grassland	278,223.25	267,250.75	289,209.75	−10,972.5	−3.94	21,959	8.22	10,986.5	3.95
Wetland	1167.00	1159.50	1575.50	−7.5	−0.64	416	35.88	408.5	35.00
Farmland	13,369.50	11,822.25	14,800.25	−1547.25	−11.57	2978	25.19	1430.75	10.70
Urban	3396.50	3398.75	3401.75	2.25	0.07	3	0.09	5.25	0.15
Barren	173.00	236.25	196.00	63.25	36.56	−40.25	−17.04	23	13.29
Water	11,364.25	11,412.00	11,515.75	47.75	0.42	103.75	0.91	151.50	1.33

).

The change trend in land-cover types in the Russian region was consistent with the change trend in land-cover types in the Chinese region. However, except for the fact that the change rate of barren land cover was greater than that of the Chinese region, the change rate of other land-cover types was less than that of the Chinese region. The intensity of land-cover-type change in the Russian region was weaker than in the Chinese region. In the Russian region, only the forest area decreased by 13,005.5 km² from 2001 to 2019, while other land-cover types increased.

3.1.4. Spatiotemporal Changes of Mongolia’s Land Cover in HARB

The area of the Mongolian region in the HARB is 19.1 $\times$ 10⁴ km², accounting for 9% of the total area of the entire basin. In the Mongolian region, the grassland land-cover type, with an area of 18.7 $\times$ 10⁴ km², occupied a dominant position and accounted for more than 98% of the total area in 2019. The other land-cover types, that is, forest, farmland, wetland, urban, barren and water, only accounted for less than 2% and had a scattered distribution in the region.

In the study period, the areas of grassland and wetland in the region increased by 651 km² and 4 km², respectively. The urban area did not change, while the area of the other land-cover types decreased by 501.75 km², 35.25 km², 3 km² and 115 km², respectively (

Table 5. The area of land-cover changes in the Mongolian region in 2001, 2010 and 2019.

Land-Cover Type	2001	2010	2019	2001–2010		2010–2019		2001–2019
	Area/km2	Area/km2	Area/km2	Change Area/km2	Change Rate/%	Change Area/km2	Change Rate/%	Change Area/km2	Change Rate/%
Forest	2626.75	2251.50	2125	−375.25	−14.29	−126.5	−5.62	−501.75	−19.10
Grassland	187,139.25	187,546	187,790.25	406.75	0.22	244.25	0.13	651	0.35
Wetland	5.25	9.5	9.25	4.25	80.95	−0.25	−2.63	4	76.19
Farmland	236.5	288.75	201.25	52.25	22.09	−87.5	−30.30	−35.25	−14.90
Urban	150.5	150.5	150.5	0	0	0	0	0	0
Barren	192.25	197.75	189.25	5.5	2.86	−8.5	−4.30	−3	−1.56
Water	674.75	581.25	559.75	−93.5	−13.86	−21.50	−3.70	−115	−17.04

). The change intensity of all the land-cover types did not vary greatly.

3.2. Dynamic Degree Analysis of Land Cover

The results of the single land-use dynamic index, which was calculated by Equation (1), are shown in

Table 6. Dynamic index of single land cover during 2001–2019.

	F_Ds	GL_Ds	WL_Ds	FL_Ds	U_Ds	B_Ds	W_Ds
Chinese region	0.079	−0.436	6.251	0.245	0.445	−1.539	0.639
Russian region	−0.103	0.219	1.945	0.595	0.009	0.739	0.074
Mongolian region	−1.061	0.019	4.233	−0.828	0.000	−0.087	−0.947
HARB	−0.047	−0.061	3.957	0.260	0.296	−0. 961	0. 214

. It can be found that during the study period, the single land-use dynamic index of each land-cover type in the Chinese region was roughly consistent with that of the HARB. In the above two regions, wetland, water, urban and farmland increased at a high speed. Barren and grassland reduced at a high speed and forest increased at a slow speed in the Chinese region, whereas they decreased at a slow speed in the HARB. For forest in the Russian region, the reduced speed was higher than that in the HARB. Except for forest, the other land-cover types increased. In the Mongolian region, the reduced speed of forest was the highest in the four regions, and the area of reduced forest was converted to grassland.

Using Equation (2) combined with the land-cover transformation matrix, the comprehensive land-use dynamic index was calculated from 2001 to 2019 (see Figure 6). From Figure 6, we can see that the change rate of land-cover types in the Chinese region was the highest of the four regions. The comprehensive dynamics of land cover in the Russian and Mongolian regions were lower than the average level of the whole basin. The land-cover changes in the Chinese region play a major role in the change in land cover in the HARB. During the research period, the land-cover change within China’s watershed was more significant, and the intensity of human activities was greater.

3.3. Analysis of the Land-Cover Integrated Index

The integrated index of land use can reflect the overall land-use intensity in a region, which indicates the extent to which land is exploited by humans [55]. The land-use synthetic indices of the HARB, the Chinese region, the Russian region and the Mongolian region are shown in Figure 7. It can be found that, during the study period, the integrated index of land cover in the Chinese region was roughly consistent with that of the HARB. Before 2018, the integrated index of land use in the Russian region remained consistent with that of the HARB but suddenly declined thereafter. While the integrated index of land use in the Mongolian region remained consistent with that of the HARB before 2016, it suddenly declined thereafter.

Furthermore, we used Formulas (4) and (5) to calculate the change quantity and the rate of land-use intensity to reveal the comprehensive level and development trend in land use in the basin. The results (Figure 8 and Figure 9) showed that the development trend in land use in the other regions was a development stage from 2001 to 2019, except for Inner Mongolia. Throughout the entire research period, the state of land-use development went through five stages: (1) 2001–2002, development stage; (2) 2002–2007, recession stage; (3) 2007–2011, development stage; (4) 2011–2013, recession stage; (5) 2013–2019, development stage.

4. Discussion

4.1. The Validation of the Land Cover

Land-cover remote sensing products may have significant differences in product accuracy due to differences in data sources, classification techniques and producers [56,57]. Due to different application objectives, there are significant differences in spatial resolution, classification systems and other aspects of land-cover remote sensing products [57,58,59,60].

We validated the land-cover data using a total of 500 validation samples collected from Google Earth images and field investigations in 2019. The spatial distribution of the validation sample points is shown in Figure 10. In May 2019, we carried out a week-long field survey in Primorsky Krai, Russia, mainly on the west, east and south banks of Lake Khanka, Amur Bay, Ussuri Bay, the upper reaches of Ussuri, the Razdolnaya basin, the Ussuriysk Taijialin Forest Monitoring Station, etc. The obtained sampling points were used to verify the MCD12Q1 land-cover product using the Confusion matrix. The overall accuracy was 78.8%, and the Kappa coefficient was 0.78, which met the research requirements. The error in wetland was the largest.

From the verification results, it can be seen that the accuracy of the MODIS land-cover data within the HARB is relatively low, which will have a certain impact on the analysis of land-cover changes. In order to better study land-use/land-cover changes in the HARB, we need more accurate land-cover products. With the development of remote sensing and image recognition, we will obtain higher spatial resolution of remote sensing images and higher-precision land-cover products.

4.2. The Driving Forces of Land Cover

Land-use/land-cover change is a complex process, and the driving factors that affect land-use/land-cover change include natural and human factors. Understanding the correlation between land-cover change and its various driving factors is the basis for constructing a land-use/land-cover simulation model.

4.2.1. Natural Factors

In recent years, the gradual warming of the temperature and the fluctuation of precipitation in the HARB have been important physical geography factors of land-cover change in the sparsely populated region. The average annual temperature in the HARB from 2001 to 2019 showed a gradual upward trend (see Figure 11). The interannual precipitation fluctuates greatly. With the continuous development of agriculture, the demand for cropland will increase, resulting in the conversion of forest, grassland and wetland into cropland.

On the other hand, the precipitation in the basin grew steadily, and the interannual precipitation fluctuation was small (see Figure 12). By comparing the land-cover changes in the regions of China, Russia, Mongolia and the entire basin under nearly the same temperature and precipitation conditions, we found that the land-cover types in China changed significantly, while those in Russia and Mongolia changed slightly. It can be seen that, although physical geography factors have a certain impact on land-cover changes, they are not the main driving factor in the watershed [22].

Terrain factors are also one of the important factors affecting land-cover change [61]. With the increase in slopes, living costs continue to increase, water flow slows down and soil erosion intensifies, leading to difficulties in agricultural cultivation. The areas where land-cover types undergo transformation are mainly characterized by low altitude, low slopes, etc. Forest land, due to its unique characteristics, often appears in high-altitude and large slope areas. The transformation of farmland and wetlands mostly occurs in areas with slopes below 6° and elevations below 150 m, accounting for more than 90% of the total transformation area. From this, it can be seen that low-altitude and low-slope areas are the main areas where land-cover transformation occurs.

4.2.2. Anthropogenic Factor

Land-cover change is the result of a combination of various related factors. Natural factors mainly affect the spatial pattern of land cover, while human activities act on the spatiotemporal changes of land cover [62]. Population growth is the main driving force behind the intensification of human activities [63]. According to the statistical yearbook data of Heilongjiang Province from 2001 to 2019, the population change in Heilongjiang Province showed a slight downward trend from 38.11 million in 2001 to 37.51 million in 2019, indicating a decrease in the total population (Figure 13). However, the proportion of the agricultural population decreased from 47.62% in 2001 to 39.1% in 2019. While the proportion of the agricultural population decreased, urban migration intensified, resulting in an increase in artificial surfaces in China. This indicates that changes in population structure lead to the expansion of urban land use, and population factors have a certain impact on land-cover change but are not the main driving force of land-cover change in the HARB.

Describing the intensity of land-cover change from a spatial perspective is a challenging task. Widely used methods such as the landscape pattern index and graph analysis in existing research can describe the spatial distribution characteristics of land-cover change intensity. However, for areas with a staggered distribution of multiple land-cover types, the expression effect of these methods is not good, and they cannot effectively highlight the intensity and distribution patterns of their changes. The research method adopted in this article provides an effective means to analyze land-cover change in the Heilongjiang (Amur) River Basin during the past 20 years from multiple dimensions such as basic structure, temporal changes, spatial changes and driving factors. This can not only provide a useful reference for the ecological environment construction of the basin but also be extended to other regional land-cover change studies.

5. Conclusions

This article comprehensively analyzes the changes in land cover in the Heilongjiang (Amur) River Basin during the past 20 years from the perspectives of basic structure, temporal changes and spatial changes in order to provide useful references for the sustainable development of grassland ecological environments. The main conclusions are as follows:

From 2001 to 2019, the dominant type of LULC in the entire basin was forest land, covering the vast majority of the northern and eastern regions of the study area. The second was grassland, mainly distributed in the western region. The total area ratio between the two was about 83%. The main land-cover type within the Chinese region was forest, followed by farmland and grassland. The area of farmland accounted for about 95% of the farmland area in the whole basin. In the Russian region, forest was the most important land-cover type, accounting for about 70% of the total area of the Russian region and being evenly distributed in the region. In the Mongolian region, the grassland land-cover type, with an area of 18.7 × 10⁴ km², occupied a dominant position and accounted for more than 98% of the total area.

The area of the different land-cover types from 2001 to 2019 showed that the area change in land-cover type in the study area was affected by many driving factors. We found that the speed of change in wetland is the highest, with an increase of 71.23% from 2001 to 2019. The trend of changes in water areas and urban land use increased by 3.86% and 5.33%, respectively. The change in farmland area also showed a trend of first decreasing and then increasing. On the whole, farmland increased by 14,031.5 km² during the study period. The change trend in the forest and grassland areas was the opposite, but the change extent was the same.

It can be found that during the study period, the single land-use dynamic index of each land-cover type in the Chinese region was roughly consistent with that of the HARB. The comprehensive dynamics of land cover in the Russian and Mongolian regions were lower than the average level of the whole basin. The land-cover changes in the Chinese region played a major role in the change in land cover in the HARB.

However, while this study achieved an analysis of annual-scale land-cover changes and driving forces in the watershed, there are still some areas that need improvement. For example, the land-cover product used in this article has a spatial resolution of 500 m and low accuracy. In our next research work, we will consider introducing some of the latest time-series land-cover classification algorithms [37,63,64] to obtain higher-accuracy land-cover data.