Assessing Spatial–Temporal Characteristics of Land Desertification from 1990 to 2020 in the Heihe River Basin Using Landsat Series Imagery

Abstract

Monitoring the status and dynamics of desertification is one of the most important parts of combating it. In this study, 30 m high-resolution information on land desertification and restoration in the Heihe River basin (HRB) was extracted from the land cover database. The results indicate that land desertification coexists with land restoration in the HRB. In different periods, the area of land restoration was much larger than the area of land desertification in the HRB, and the HRB has undergone land restoration. Upstream of the HRB, there is a continuing trend of increasing land desertification associated with overgrazing in a context where climate change favors desertification reversal. In the middle and lower reaches, although climate variability and human activities favor land desertification, land desertification is still being reversed, and land restoration dominates. Implementing the eco-environmental protection project and desertification control measures, especially the Ecological Water Distribution Project (EWDP), contributes to the reversal of desertification in the middle and lower reaches of the HRB. However, the EWDP has indirectly led to the lowering of the water table in the middle reaches, resulting in local vegetation degradation. Therefore, there is an urgent need to transform the economic structure of the middle reaches to cope with water scarcity and land desertification.

1. Introduction

Land desertification, known as the “cancer of the Earth”, is one of the top ten major environmental problems in the world [1] and a challenging issue in global ecological governance [2]. The United Nations Convention to Combat Desertification (UNCCD) [3] defines land desertification as a form of land degradation in dryland, which includes arid, semi-arid, and dry subhumid areas, driven by several interrelated factors and triggered by the combination of climate change with the irrationality of human activities. The combination of global warming, rapid economic development, urban expansion, and population growth [4,5,6] has increased the extent and intensity of desertification in some dryland areas over recent decades [7,8]. Desertification causes serious environmental issues and significant negative impacts on socioeconomic development, such as food shortages, poverty, and health problems [9,10,11,12], which have become important factors hindering sustainable development in drylands [13,14]. Considering the potential negative impacts of desertification on the environment and socioeconomic development, desertification prevention and control were included as part of the SDGs in the 2030 Agenda for Sustainable Development, which aims to “protect, restore and promote sustainable use of terrestrial ecosystems, sustainably manage forests, combat desertification, and halt and reverse land degradation and halt biodiversity loss” [15]. To achieve the goal of land degradation neutrality by 2030, combating desertification plays a critical role in conserving biodiversity, mitigating human-induced global warming, and eradicating global poverty.

China is one of the countries most affected by desertification. According to the Sixth Survey of Desertification in China, as of 2019, the area of desertified land nationwide reached 2,537,000 km², accounting for nearly 27% of the total land area, of which 65.6% is aeolian desertified land (ADL), and more than 16% of the ADL still tends to be desertified [16]. In China, the most serious areas of desertification are mainly concentrated in the northwestern and northeastern regions, which not only seriously jeopardize the local ecological environment but are also the main source area of spring sandstorms in China [17]. The Heihe River basin (HRB) in northwest China’s inland semi-arid and arid zones is a typical ecologically fragile area because of desertification. In the HRB, water scarcity is the most limiting factor for ecological restoration and socioeconomic development. Against the backdrop of climate change, although the climate tends to be warm and humid in northwestern China, increased precipitation is insufficient to offset increased evaporation from a higher temperature due to a small precipitation base [18]. Coupled with the surge in water demand from oasis agriculture and urbanization in the middle reaches of the HRB [19], the sensitivity and vulnerability of ecosystems are further exacerbated, which has made the downstream area of the HRB one of the major sources of sandstorms [20]. To resolve water resource conflicts in different regions of the HRB, the Chinese government has implemented the Ecological Water Diversion Project (EWDP) to ensure that the ecological water demand downstream of the HRB is met, thereby mitigating and restoring ecological desertification in the downstream oases since 2000. However, it is still unclear whether the EWDP contributes to reversing desertification and how climate change and human activities influence the desertification process. Therefore, studying the spatio-temporal characteristics of desertification and its driving mechanisms is of great importance for the adaptation and formulation of subsequent desertification control and ecological restoration measures in the HRB.

The accurate monitoring of the status and dynamics of desertification is one of the most important parts of combating desertification. However, the traditional methods of desertification monitoring have the disadvantages of requiring long consumption; being labor intensive, costly, and inefficient; having small spatial scales; and so on [21]. With the rapid development of remote sensing and geographic information system technology, satellite remote sensing provides an economical, efficient, effective, and objective approach to assessing land desertification over long periods from regional to global scales [6,22,23,24]. Methods of applying remote sensing and a geographic information system (GIS) to monitor land desertification can be broadly categorized into three groups: (1) the use of vegetation indices to estimate desertification processes (normalized difference vegetation index (NDVI), modified soil adjusted vegetation index (MSAVI), enhanced vegetation index (EVI), net primary productivity (NPP), etc.) [25,26,27] and, in some cases, land surface and climate parameters (albedo, temperature vegetation dryness index (TVDI), topsoil grain size index (TGSI), land surface temperature (LST), precipitation, temperature, evapotranspiration, etc.) for more accurate results [28,29,30]; (2) the extraction of desertification information by visual interpretation or automatic classification using Landsat imagery [22,31,32]; and (3) the use of feature space models for classification, such as the albedo–NDVI [33] and the point–point model [34]. However, the use of method one to monitor desertification relies on data with a coarse spatial resolution (e.g., 8 km or 1 km) and can provide macro distribution and general trends with a high temporal resolution, but details of desertification characteristics are not available. Methods two and three use data sources with a high spatial resolution (e.g., Landsat series data with a 30 m resolution). However, the monitoring results are overly dependent on the time of image acquisition and the classification method chosen, resulting in large differences in the interpretation results of images at different times. The monitoring results do not truly reflect the characteristics and trends of desertification.

Considering the above issues, this manuscript aimed to use object-oriented classification methods to assess desertification dynamics from 1990 to 2020 in the HRB based on Landsat imagery acquired from the Google Earth Engine (GEE). A 30 m resolution land cover dataset for 1990, 2000, 2010, and 2020 was established, and spatial analysis methods provided by GIS were used to analyze the changes in land cover types in 1990, 2000, 2010, and 2020 to obtain the characteristics of desertification in the HRB over the past 30 years. Finally, the dominant driving factors of the desertification and restoration processes were examined and discussed.

2. Materials and Methods

2.1. Study Area

The Heihe River is the second-largest inland river in China, originating from Lenglongling in the Qilian Mountains; flowing through Qinghai Province, Gansu Province, and the Inner Mongolia Autonomous Region; and finally flowing into the terminal lake Juyan Sea, with a total length of 821 km and a drainage area of 271,000 km², of which the area within the territory of China is about 87.6%, and the rest is located in the territory of Mongolia (Figure 1). The Heihe River basin is characterized by significant zonal differentiation, with a globally unique glacier–frozen soil–river–oasis–desert multidimensional natural landscape zone sequentially distributed with the altitude. Due to the surrounding mountains and the distance from the sea, the HRB has a typical temperate continental climate with characteristics of low precipitation and higher evaporation. Depending on the altitude, the temperature, precipitation, and evaporation in the HRB vary greatly on spatial and temporal scales. From south to north, as the altitude decreases, the annual mean temperature increases from 2 to 10 °C, precipitation decreases from 500 to less than 50 mm, and evaporation increases from 700 to more than 3500 mm. As of the end of 2014, the total population within China in the HBR was 2.52 million, with a population density of 10.5 people per km², of which 56.9% was agricultural. The population of the Heihe River basin is concentrated in the middle reaches of the basin, accounting for 95.3% of the total population. The HRB is backward regarding economic development, with a gross domestic product (GDP) of only CNY 115.1 billion in 2014 [24].

2.2. Data Sources

In this study, we mapped fine-resolution land cover distributions in 1990, 2000, 2010, and 2020 in the HRB by using an object-oriented classification method. The Landsat Collection 2 Surface Reflectance Tier 1 images (including TM, ETM+, and OLI) derived from the Google Earth Engine (GEE) in the study area were used to establish the land cover dataset for each period [35,36]. All of these images were subjected to geometric and atmospheric correction and cross-calibration between the different sensors [37,38,39]. For each image, the CF mask, which is a cloud-masking method, was used to remove clouds, cloud shadows, and snow pixels by using the data quality layer [31,40,41]. All the remaining pixels were considered good-quality Landsat observations that could be used for land cover mapping. To reduce the influence of atmospheric noise and inter-annual phenological differences, the images from three different years were used to acquire high-quality observation composites for each period [42]. For example, the 2020 data synthesized the good-quality growing-season pixels from the 2019, 2020, and 2021 images. Due to cloudy weather and snow-covered areas above the snow line of the Qilian Mountain in the southern HRB, there were no good-quality observations in these areas. Considering that the unobserved area is small and the land cover type is glacier and alpine desert, we believe it has little impact on this study. Finally, four phases of high-quality growing-season Landsat image collections of the HRB with a 30 m spatial resolution, including bands of blue, green, red, near-infrared (NIR), short-wave infrared 1 (SWIR1), and short-wave infrared 2 (SWIR2), were generated based on the GEE.

In addition, the auxiliary data used to assist in the interpretation of the land cover data and the analysis of the driving factors behind desertification include the following: (1) vegetation type map of China, with a scale of 1:1,000,000 (source: National Cryosphere Desert Data Center, Lanzhou, China, http://www.ncdc.ac.cn (accessed on 12 October 2022)); (2) DEM and slope data, with a spatial resolution of 30 m (source: ASTER Global Digital Elevation Model V003, NASA Earth science data, Washington, DC, USA, https://www.earthdata.nasa.gov (accessed on 21 May 2014)); (3) meteorological data (Ejina Banner in Alax League of Inner Mongolia, Zhangye in the Hexi Corridor of Gansu Province and Qilian in the Qilian of Qinghai Province), including precipitation and the annual mean temperature (source: the China Central Meteorological Bureau, Beijing, China, http://cdc.cma.gov.cn (accessed on 26 March 2021)); (4) hydrological data (Yingluoxia (YLX), Zhengyixia (ZYX), and Langxinshan (LXS), source: the Heihe River Bureau); and (5) field survey data of land use/cover types in the HRB acquired in 2015.

2.3. Classification of Land Cover

According to the China Land Cover Classification System, the National Land Use Remote Sensing Mapping Classification System [43], and the characteristics of land cover in the HRB, we identified seven primary types of land cover in the HRB, including cropland, woodland, grassland, water bodies, wetland, artificial surface, and desert, which were divided into 16 secondary types (

Table 1. The land cover type classification system used in this study.

Primary Types	Secondary Types
Cropland	Paddy field and non-paddy field
Woodland	Forest and shrubs
Grassland	Meadow steppe, typical steppe, and desert steppe
Water bodies	Lake, reservoirs, and ponds
Wetland	Marsh
Artificial land	Residential area, industrial area, traffic land, and mining land
Desert	Sandy and barren

). Considering the small river area in the water area type, the low runoff, and the long breakup period in the middle and lower reaches of the HRB, we categorized the river type as bare soil to improve the accuracy of the classification.

Based on the availability of reliable Landsat imagery, the study period was divided into three intervals (1990–2000, 2000–2010, and 2010–2020). The 2020 land cover dataset of the HRB was generated using Landsat OIL imagery, and the land cover datasets in the other periods were generated using Landsat TM imagery. In this study, the automatic object-oriented classification based on decision trees was used to derive fine-resolution data for land cover. First, we utilized the spectral difference segmentation and multiresolution segmentation algorithms provided by eCognition Developer version 8.64 (Definiens Imaging, Germany) to segment the images from 2020 into homogeneous polygons of different sizes according to the spectral differences and segmentation scale, which were named objects. Then, the segmented objects were classified according to texture, spatial, spectral, and shape features by using the decision tree method (

Table 2. Interpretation rules and reference index thresholds for land cover classification.

Land Cover Types	Rule and Reference Index Threshold
Cropland	Compactness > 2.5, visual interpretation
Woodland	NDVI > 0.4, red < 0.06
Meadow steppe	NDVI > 0.4, 0.1 > red >0.06
Real steppe	0.4 > NDVI > 0.25, 0.15 > red > 0.1
Desert steppe	0.25 > NDVI > 0.1, red > 0.15
Water bodies	NDWI > 0
Artificial land	Visual interpretation
Wetland	NDVI > 0.4, Nir < 0.24, slope < 10
Sand	NDVI < 0.1, 2000 > brightness > 1000
Barren	NDVI < 0.1, NDSI > 0.09

). In addition, to assist in the classification and improve classification accuracy, the NDVI, fraction of vegetation cover (FVC), TGSI, normalized difference water index–blue (NDWI_B), and normalized difference soil index (NDSI) were calculated using the following equations:

$$\begin{array}{cc}NDVI& =(Nir-Red)/(Nir+Red)\\ FVC& =(NDVI-NDV{I}_{soil})/(NDV{I}_{veg}-NDV{I}_{soil})\\ TGSI& =(Red-Blue)/(Red+Blue+Green)\\ NDW{I}_B& =(Blue-Nir)/(Blue+Nir)\\ NDSI& =(Swir1-Nir)/(Swir1+Nir)\end{array}$$

where Blue (TM = B1, OLI = B2), Green (TM = B2, OLI = B3), Red (TM = B3, OLI = B4), Nir (TM = B4, OLI = B5), and Swir1 (TM = B5, OLI = B6) denote the reflectivity of the blue, green, red, near-infrared, and short-wave infrared 1 bands, respectively.

Finally, 5% of the generated 10 × 10 km grid points, 468 in total, were randomly selected, and the land cover type of each point was manually interpreted using Google Earth ultra-high-resolution imagery, which was used to assess the accuracy of land cover classification. If the accuracy of the water area, grassland, woodland, cropland, and wetland types was less than 90%, these types met the accuracy requirement by increasing the decision tree classification index or adjusting the index threshold. Artificial surfaces, sand, and desert types are not easily confused with the above types. However, they are confused with each other, so we used the visual interpretation method combined with a human–computer interaction to improve classification accuracy.

After completing the 2020 land cover dataset, we utilized a knowledge-based object-oriented spectral vector change detection method to generate land cover datasets for 1990, 2000, and 2010. Taking 2010 as an example, the images of 2010 and 2020 land cover classification results were segmented by using eCognition Developer version 8.64 to generate objects. The reflectance of each band in different years of the object formed a spectral vector, and the similarity between the two vectors was calculated. The object whose similarity was less than the threshold value was the object that had been changed. According to the land cover type, 5% of the unchanged objects were randomly selected to calculate the average spectrum vector. Then, the similarity between the vector of the changed object and the vector of each land cover type was calculated. The changed objects were classified into the land cover type with the largest vector similarity pair. The details of classification and the change detection method were described by Wu et al. [44] and Song and Yan [45].

2.4. Extraction of Land Desertification Information and Processing

Land desertification is the reduction in or loss of biological or economic productivity and diversity of cropland, woodland, grassland, or pasture in a dryland region as a result of the utilization of the land or the impact of one or multiple factors [46]. Therefore, the process of land desertification can be characterized by changes in land cover. Based on the land cover change dataset, we classified woodland, meadow grassland, and typical grassland as non-desertified land (NDL) and the areas that used to be NDL but were transformed into desert grassland, sandy land, and barren land as desertified land (DL). Using the spatial analysis module provided by Arcmap version 10.8.2, the land cover datasets of the four periods were spatially overlapped to generate the land cover change datasets for 1990–2000, 2000–2010, and 2010–2020, respectively. Then, the distribution maps of desertification development and reversal were obtained by extracting the changed areas of DL and NDL types as defined above in the HRB from 1990 to 2000, from 2000 to 2010, and from 2010 to 2020. By analyzing the status of land desertification in the HRB over time, we identified the trends of land desertification over a 30-year period and analyzed the driving factors of desertification development and reversal.

3. Results

3.1. Pattern of Land Cover Changes from 1990 to 2020

In this study, to better understand the process, extent, and trend of land desertification or restoration, NDL, DL, and other land cover types (including cropland, water bodies, artificial land, and wetlands) were counted.

Table 3. Areas of different land cover types in the HRB from 1990 to 2020 (km²).

		Cropland	Water Bodies	Artificial Land	Wetland	NDL	Desert Steppe	Sand	Barren
1990	Upstream	115.69	575.36	10.45	842.65	7231.48	8650.89		10,226.86
	Midstream	5513.55	493.93	478.75	123.36	1466.92	3963.32	3474.57	39,167.62
	Downstream	99.26	198.45	40.92	2.18	120.26	1046.25	45,832.48	141,342.70
	Total	5728.50	1267.75	530.12	968.19	8818.66	13,660.47	49,307.05	190,737.18
2000	Upstream	114.19	464.73	12.79	845.41	11,789.02	8128.68		6298.55
	Midstream	6160.47	496.85	541.09	144.04	1644.88	4564.01	3413.22	37,717.46
	Downstream	98.65	134.02	46.00	0.78	130.79	1447.49	45,804.67	141,020.12
	Total	6373.31	1095.60	599.87	990.23	13,564.69	14,140.18	49,217.89	185,036.13
2010	Upstream	103.24	450.00	18.77	1169.58	11,809.60	8730.88		5371.30
	Midstream	6808.06	459.50	756.18	190.50	2252.93	6124.60	3261.41	34,828.83
	Downstream	157.51	206.67	81.92	7.80	220.70	1318.31	45,736.07	140,953.53
	Total	7068.81	1116.17	856.87	1367.88	14,283.23	16,173.79	48,997.48	181,153.67
2020	Upstream	92.84	430.89	31.02	1215.02	12,323.23	9504.85		4055.54
	Midstream	7540.87	426.20	1218.57	244.81	3131.72	9116.96	2974.12	30,028.77
	Downstream	160.35	202.99	111.40	35.44	542.10	9015.42	43,244.05	135,370.76
	Total	7794.05	1060.08	1360.98	1495.27	15,997.05	27,637.23	46,218.17	169,455.08

presents the statistical data on the area of different land cover types in different parts of the HRB. Figure 2 illustrates the spatial distribution patterns of different land cover types.

In 1990, barren land was the most prevalent land cover type throughout the study period, accounting for 70.38% of the HRB. This land cover type is predominantly found in the west–central region of the HRB and is characterized by the Gobi, bare rock, bare land, and salinity due to extremely dry and scarce rainfall. Because the Badain Jaran Desert, the third-largest desert in China, is located entirely downstream, sand is the second-largest land cover type, accounting for 18.19% of the HRB. Additionally, patches of quicksand are distributed adjacent to the edge of the midstream oasis. The vegetation type of the HRB is dominated by desert grassland, which accounts for 5.04% of the HRB and 63.32% of the area covered by vegetation. It is mainly distributed around oases in the middle and lower reaches and in the pre-mountain floodplains in the upper reaches. Attributed to natural conditions, the proportion of NDL is relatively small, accounting for only 3.25% of the HRB. It is mainly found in the mountainous regions upstream, including coniferous forests, shrubs, alpine meadows, and alpine steppes, and there are also sporadic shrub forests and deciduous forests in the middle and lower reaches of the HRB along both sides of the river. Since the population of the HRB is concentrated in the midstream region, the vast majority of cropland and artificial land is located here, which accounts for 2.11% and 0.20%, respectively. Water bodies and wetlands are predominantly located upstream of the HRB, accounting for 0.47% and 0.36%, respectively. Water bodies and wetlands are dominated by rivers and marshy meadows in the upstream and reservoirs, lakes, and riparian wetlands in the midstream and downstream.

With climate variability and rapid economic development, between 1990 and 2020, the coverage of cropland, artificial land, wetland, NDL, and desert steppe increased by 36.06%, 156.73%, 54.44%, 81.40%, and 102.32%, respectively, while the coverage of water bodies, sand, and barren decreased by 16.38%, 6.26%, and 11.16%, respectively. From the perspective of the different subregions of the HRB, the observed trends in the land cover types are also different. The NDL, sand, barren, wetland, and artificial land demonstrate a similar trend across the HRB, with the NDL, wetland, and artificial land exhibiting an increasing trend and the sand and barren land exhibiting a decreasing trend. Although the cropland area continues to increase, the upstream cropland continuously decreases due to the implementation of ecological protection and restoration programs, such as the “Grain for Green” project and the rapid development of urban construction. In contrast, the downstream cropland shows a decreasing and then increasing trend due to the changes in water resources. In this study, river floodplains are also classified as water bodies. It can be observed that the area of water bodies in the middle and upper reaches of the HRB continues to decrease due to vegetation restoration, leading to the conversion of areas with good moisture conditions to wetlands and of those with poor moisture conditions to desert steppe. In the lower reaches of the HRB, the area of water bodies decreased in the 2000s due to the disappearance of the terminal lakes because of the disintegration of the Heihe River. However, the area of water bodies increased due to the guaranteed downstream water supply since the implementation of the Heihe River Basin Plan in 2000. Although desert steppes are more vulnerable to precipitation variability, the desert steppe area in the basin as a whole has continued to expand, particularly since 2010, with only minor decreases in the upper reaches in 2000 and in the lower reaches in 2010.

3.2. Pattern of Land Desertification Development and Reversal from 1990 to 2020

Usually, the increase or decrease in DL and NDL can be employed to describe the process of land desertification and restoration. However, it should be noted that degraded land accounts for a significant proportion of the HRB. Consequently, a more comprehensive understanding of the process of land desertification and restoration can be achieved by considering not only the changes between DL and NDL but also the transformation of DL, NDL, and other land cover types. The transfer matrices of DL, NDL, and other land cover types in 1990–2000, 2000–2010, and 2010–2020 are shown in

Table 4. Area of land desertification or land restoration in the HRB from 1990 to 2000 (km²).

	Cropland	Desert Steppe	Water Bodies	Artificial Land	Sand	Wetland	Barren	NDL
Upstream
Cropland		1.29					0.08	1.23
Desert steppe	0.81		1.37	0.17		93.65	320.62	4144.96
Water bodies		1.26					0.47	2.68
Artificial land
Wetland		8.34					17.25	193.82
Barren	0.74	3962.79	0.50	0.43		0.61		437.63
NDL	0.03	65.70	0.80	0.76		19.54	135.95
Midstream
Cropland		53.05			0.03		19.51	12.77
Desert steppe	285.41		20.85	7.38	7.12	26.60	448.92	268.05
Water bodies		16.90			0.22		7.77	2.67
Artificial land
Sand	7.32	29.88	3.33	0.06			31.37	0.23
Wetland		8.84			0.17		0.52	7.89
Barren	369.89	1488.42	22.94	36.32	3.23	0.89		39.28
NDL	76.75	67.53	1.23	0.73	0.04	4.04	2.67
Downstream
Cropland		4.17					5.19	0.26
Desert steppe	4.61		1.41	0.02	14.59	0.02	235.58	47.96
Water bodies		1.43			0.10		66.31	0.02
Artificial land
Sand		28.36	0.23				14.77	0.17
Wetland		0.58					0.06	0.63
Barren	2.92	633.94	1.64	4.96	0.97	0.04		0.85
NDL	1.47	36.94	0.04	0.04	0.05	0.01	0.83
Whole basin
Cropland		58.50			0.03		24.78	14.26
Desert steppe	290.83		23.62	7.58	21.72	120.27	1005.12	4460.98
Water bodies		19.59			0.33		74.55	5.37
Artificial land
Sand	7.32	58.24	3.56	0.06			46.14	0.40
Wetland		17.76			0.17		17.83	202.35
Barren	373.56	6085.15	25.07	41.72	4.21	1.54		477.77
NDL	78.25	170.16	2.07	1.52	0.09	23.59	139.45

,

Table 5. Area of land desertification or land restoration in the HRB from 2000 to 2010 (km²).

	Cropland	Desert Steppe	Water Bodies	Artificial Land	Sand	Wetland	Barren	NDL
Upstream
Cropland		0.24						16.72
Desert steppe	2.85		2.03	2.19		10.80	381.44	1003.10
Water bodies		1.79					0.27	2.82
Artificial land
Wetland		10.26					5.15	11.88
Barren	3.84	1573.00	0.31	0.45		19.48		130.95
NDL	0.98	419.30	1.70	1.43		307.54	413.93
Midstream
Cropland		100.09					14.65	226.04
Desert steppe	344.20		6.45	9.31	1.48	21.21	229.86	460.13
Water bodies		30.09			0.42		7.29	4.09
Artificial land
Sand	18.23	102.16	3.17	0.66			26.31	4.11
Wetland		2.69					0.06	3.93
Barren	615.16	2369.10	18.99	163.70	0.93	1.23		30.20
NDL	43.65	29.21	0.56	2.66		11.90	32.56
Downstream
Cropland		2.92					0.51	3.79
Desert steppe	29.83		1.10	1.64	14.47	0.50	529.14	113.85
Water bodies		3.77			1.31		0.54	0.28
Artificial land
Sand		56.70	0.94	0.14			29.38	0.26
Wetland		0.04						0.17
Barren	24.40	475.72	77.04	32.90	3.01	6.10		7.49
NDL	12.99	22.19	0.06	0.15	0.02		0.52
Whole basin
Cropland		103.25					15.16	246.54
Desert steppe	376.87		9.57	13.15	15.96	32.51	1140.44	1577.08
Water bodies		35.65			1.73		8.09	7.18
Artificial land
Sand	18.23	158.86	4.11	0.80			55.69	4.37
Wetland		12.99		0.15			5.21	15.98
Barren	643.41	4417.81	96.34	197.06	3.94	26.81		168.64
NDL	57.62	470.70	2.32	4.23	0.02	319.44	447.01

and

Table 6. Area of land desertification or land restoration in the HRB from 2010 to 2020 (km²).

	Cropland	Desert Steppe	Water Bodies	Artificial Land	Sand	Wetland	Barren	NDL
Upstream
Cropland		0.49						9.34
Desert steppe	1.02		1.09	4.89		12.21	397.80	902.75
Water bodies		9.10					0.43	9.12
Artificial land
Wetland		6.32					3.40	7.60
Barren		1374.68	0.39	0.39		5.70		373.50
NDL	1.46	703.13	1.75	3.82		41.26	37.27
Midstream
Cropland		37.88					2.53	89.56
Desert steppe	328.79		12.91	42.90	4.27	17.47	118.81	862.46
Water bodies		39.54			0.09		2.31	11.81
Artificial land
Sand	44.14	233.25	0.75	7.53			12.04	2.47
Wetland		2.11					0.30	2.79
Barren	505.55	3990.09	23.78	344.22	8.52	2.59		70.47
NDL	50.27	77.14	2.30	1.69	0.01	20.36	9.17
Downstream
Cropland		1.61					0.38	8.52
Desert steppe	4.79		5.21	2.22	6.07	4.80	50.28	280.51
Water bodies		7.30			0.07		1.12	2.36
Artificial land
Sand		2493.67	4.75				1.98	0.48
Wetland		0.10					0.05	0.13
Barren	4.79	5536.33	15.05	25.49	2.72	4.34		48.02
NDL	5.20	11.97	0.30	0.23		0.81	0.11
Whole basin
Cropland		39.98					2.91	107.42
Desert steppe	334.60		19.21	50.00	10.35	34.49	566.90	2045.72
Water bodies		55.95			0.16		3.87	23.29
Artificial land
Sand	44.14	2726.92	5.50	7.53			14.03	2.95
Wetland		8.53					3.75	10.52
Barren	510.35	10,901.10	39.22	370.10	11.23	12.63		491.99
NDL	56.93	792.24	4.35	5.74	0.01	62.43	46.55

, respectively.

In this study, we define the conversion of DL to NDL, DL to other, other to NDL, and sand or barren to desert steppe as land-restoration processes and the conversion of NDL to DL, other to DL, NDL to other, and desert steppe to sand or barren as land-desertification processes. Figure 3 shows the spatial distribution characteristics of land desertification and restoration between 1990 and 2000, 2000 and 2010, and 2010 and 2020 in the HRB. Figure 4 shows the statistical characteristics of the different land-desertification or -restoration processes in the different subregions of the HRB for each period.

3.2.1. Spatial Distribution Patterns of Land Desertification and Restoration in the HRB from 1990 to 2000

From 1990 to 2000, the area of the HRB undergoing desertification was 1655.54 km², accounting for 0.61% of the total area. The main process of land desertification was the conversion of desert steppe to sand or barren, which accounted for 62.02% of the degraded area, followed by the conversion of NDL to DL, other to DL, and NDL to other, which accounted for 18.71%, 12.90%, and 6.37% of the degraded area, respectively. The newly increased DL area was 523.26 km², accounting for 31.61% of the degraded area, of which 59.19% was degraded NDL. The reduced NDL area was 415.15 km², accounting for 25.08% of the degraded area, of which 74.60% was converted to DL. Regarding the various divisions of the HRB, land desertification was primarily concentrated in the midstream region, which accounted for 43.25% of the total degraded area. This was followed by the upstream and downstream regions, which accounted for 34.56% and 22.19% of the degraded area, respectively. Although the process of land desertification in different subregions of the HRB was mainly desert steppe to sand or barren, the upstream region had a much larger proportion of NDL to DL than other regions, accounting for 35.25% of the degraded area, while the downstream region had a minimal area of less than 1% of NDL to other. Furthermore, the midstream region had a relatively large area of NDL to other. The newly increased DL in the upstream region was primarily derived from NDL, while that in the middle and downstream regions was mainly derived from other. Conversely, the reduced NDL was primarily converted to other in the middle reaches and to DL in the upstream and downstream regions. Regarding the spatial distribution of land desertification, it can be observed that in the upstream region, the phenomenon was concentrated in the flat areas of the broad valleys of the river. In the midstream region, it was primarily located in the desert steppes in the southeast of Shandan County. Finally, in the downstream area, it was mainly observed on the banks of the river, within the oases, and in the terminal lakes and their surroundings.

From 1990 to 2000, the area of the HRB where land restoration was taking place was 11,757.74 km², accounting for 4.34% of the total area. The main process of land restoration was the conversion of sand or barren to desert steppe, which accounted for 52.25% of the restored area, followed by the conversion of DL to NDL, DL to other, and other to NDL, which accounted for 42.01%, 3.85%, and 1.89% of the restored area, respectively. The newly increased NDL area was 5161.11 km², accounting for 43.90% of the restored area, of which 95.70% was converted from DL. The reduced area of DL was 5392.38 km², representing 45.86% of the restored area, of which 91.59% was converted to NDL. Regarding the different divisions of the HRB, land restoration was mainly concentrated in the upstream region, which accounted for 75.18% of the total restored area. This was followed by the midstream and downstream regions, which accounted for 18.71% and 6.11% of the restored area, respectively. The main land-restoration process in mid- and low catchments, consistent with the whole catchment, was sand or barren to NDL. Especially in the downstream region, more than 90% of the restoration was performed in this way. For the upstream region, the main land-restoration process was the conversion of DL to NDL and the conversion of sand or barren to NDL, which accounted for 96.68% of the restored area, the former being slightly larger than the latter. The midstream region differed from the other regions in that the conversion of DL to other was relatively large and was the second major land-restoration process, accounting for 15.95% of the restored area, in contrast to the upstream and downstream regions, which both accounted for about 1% of the restored area. A common feature of the different regions of the HRB was that the conversion of other to NDL was small in all of them. Regarding the spatial distribution of land restoration, upstream land restoration occurred mainly in the floodplains and more arid western regions. Midstream land restoration was mainly concentrated around oases and desertified grasslands in the eastern parts. Land restoration in the downstream area occurred mainly along the western and northern margins of the Badain Jaran Desert, in addition to small patches in the northern parts of the region.

3.2.2. Spatial Distribution Patterns of Land Desertification and Restoration in the HRB from 2000 to 2010

From 2000 to 2010, the area of the HRB undergoing desertification was 2285.66 km², accounting for 0.84% of the total area. The rate of land desertification was slightly higher in this period than in the previous period. Although the main process of land desertification remained the conversion of desert steppe to sand or barren, its proportion decreased. Unlike in the previous period, the proportion of conversion of NDL to DL increased considerably, accounting for 36.09% of the degraded area, while the proportion of the other two desertification processes decreased. The newly increased DL area was 1009.02 km². The reduced NDL area was 945.14 km², almost twice as much as in the previous period. The main sources of DL and destinations of NDL were the same as in the previous period but with a greater weight. In terms of the different subregions of the HRB, land desertification occurred mainly in the upstream region, accounting for 52.06% of the degraded area, followed by the downstream and midstream regions, which accounted for 25.75% and 22.19% of the degraded area, respectively. The main process of land desertification in the midstream and downstream regions was the conversion of desert steppe to sand or barren, consistent with the whole basin, especially in the downstream region, where it dominated with more than 90%. The main process of land desertification in the upstream region was the conversion of NDL to DL, which accounted for 62.23% of the degraded area, followed by the conversion of desert steppe to sand or barren, with a proportion of 32.06%. The newly increased DL in the midstream region was primarily derived from other, while the upstream and downstream regions were mainly derived from NDL. Conversely, the reduced NDL was primarily converted to DL in different subregions, but the proportion of NDL converted to other in the midstream region was much larger than that in the other regions. In terms of spatial distribution, land desertification in the upstream region was predominantly seen in the more arid regions of the northwest, while in the midstream and downstream regions, it was predominantly sporadic in areas along rivers and adjacent to oases.

From 2000 to 2010, the area of land restored in the HRB was 7189.88 km², representing 2.65% of the total area. The rate of land restoration in the HRB slowed down, with only 61.14% of the area restored in the previous period. The conversion of sand or barren to desert steppe remained the primary process of land restoration, although its proportion increased. Furthermore, the proportion of the conversion of DL to NDL was only half of what it was before, while the proportion of conversion of DL to other increased significantly. The newly increased NDL area was 2156.65 km², accounting for 30.00% of the restored area, of which 87.74% was converted from DL. The reduced area of DL was 2348.85 km², representing 32.67% of the restored area, of which 80.56% was converted to NDL. The decrease in the area of newly increased NDL and in the area of reduced DL indicated a slowdown in the rate of land restoration in the HRB during this period. In terms of the different subregions of the HRB, land restoration during this period took place mainly in the midstream region, which accounted for 50.11% of the entire restored area, followed by the upstream and downstream regions, which accounted for 40.26% and 9.63%, respectively. The main process of land restoration in all watershed subregions was the conversion of sand or barren to desert steppe, with the conversion of DL to NDL as the second most important process of land restoration, especially in the upstream region. In addition, the other two land-restoration processes in the midstream region accounted for a much larger proportion of land restoration than in the upstream and downstream regions. The main source of newly increased NDL in the different watershed subregions was from DL, but the proportion of other converted to NDL was much larger in the midstream than in the other regions. Reduced DL converted primarily to NDL, especially in the upstream region, where this process of land restoration was absolutely dominant. The spatial distribution of land restoration in the midstream and downstream areas was roughly the same as in the previous period, except that, in the downstream region, there was a significant increase in land restoration along the banks of the river, in the oases, and in the area of the terminated lakes.

3.2.3. Spatial Distribution Patterns of Land Desertification and Restoration in the HRB from 2010 to 2020

From 2010 to 2020, the area of the HRB undergoing desertification was 2014.78 km², accounting for 0.74% of the total area. The rate of desertification was slower than in the period between 2000 and 2010 and was comparable to the period between 1990 and 2000. In contrast to the previous two periods, the main process of land desertification was the conversion of NDL to DL between 2010 and 2020, which accounted for 46.24% of the total degraded area, followed by the conversion of desert steppe to sand or barren, NDL to other, and other to DL, which accounted for 28.65%,19.50%, and 5.61%, respectively. The newly increased DL area was 1044.74 km², accounting for 51.85% of the degraded area, of which 89.17% was degraded NDL. The reduced NDL area was 1324.43 km², accounting for 65.74% of the degraded area, of which 89.17% was converted to DL. Regarding the various divisions of the HRB, land desertification was primarily concentrated in the upstream region, which accounted for 77.45% of the total degraded area, followed by the midstream and downstream regions, which accounted for 20.23% and 4.25% of the total degraded area, respectively. The main process of land desertification in the upstream region was the conversion of NDL to DL, and in the midstream and downstream regions, it was the conversion of desert steppe to sand or barren. The rate of desertification in the midstream and downstream regions was not only significantly smaller than that in the upstream region but also significantly smaller than that in the other periods. The upstream region accounted for the majority of both newly increased DL and the reduction in NDL, suggesting that land desertification was more severe in the upstream region. In terms of spatial distribution, land desertification was mainly concentrated in the pre-mountain plains and broad-valley steppe areas in the central and western regions of the upstream area. Land desertification in the midstream region was found in small patches in the southeastern regions and sporadically within the oases on a merit basis. There were very few areas of land desertification in the downtown area, and they were mainly concentrated on the western edge of the desert and along the rivers.

From 2010 to 2020, the area of land restored in the HRB was 16,667.17 km², accounting for 6.15% of the total area. The rate of land restoration accelerated significantly during this period. The conversion of sand or barren to desert steppe remained the primary land-restoration process, accounting for 81.77% of the total restored area. The newly increased NDL area was 2545.03 km², an increase in area over the previous period but a significant decrease in proportion to only half of the previous period. The reduction in the area of DL was 2892.59 km², accounting for 17.35% of the total restored area, of which 82.92% was converted into NDL. In terms of the different subregions of the HRB, land restoration occurred mainly in the downstream region, which accounted for 50.35% of the total restored area, followed by the midstream and upstream regions, which accounted for 34.30% and 15.35% of the total restored area, respectively. The conversion of sand or barren to desert steppe remained the main process of land restoration in all watershed subregions. In addition, the conversion of DL to NDL was the second major land-restoration process, especially in the upstream region, accounting for 44.43% of the total restored area. The main source of new NDL was from DL, and the reduced DL area was mainly converted to NDL. It was only in the midstream region that the proportion of the conversion of DL to other and of other to NDL was much larger than in the other regions. In terms of spatial distribution, the areas where land restoration was occurring were roughly the same as in the previous period, with an increase in the area restored. Furthermore, a large restoration area occurred in the northern mountainous areas of the corridor in the northeastern of the midstream region, and in the downstream region, large restoration areas occurred in the eastern desert region.

In summary, land desertification coexisted with land restoration in the HRB. In different periods, the area of land restoration was much larger than the area of land desertification in the HRB (Figure 4), and the HRB has been in the process of land restoration. From Figure 4, it can be seen that the area of land restoration in the whole watershed decreased and then increased, while the area of land desertification increased and then decreased, indicating that land desertification was more serious from 2000 to 2010. It can also be seen that the area of land restoration in the upstream region continued to decrease, and the area of desertification continued to increase; the area of land restoration in the middle region continued to increase, and the area of desertification continued to decrease; and the area of land desertification in the downstream region continued to decrease, except for the area of desert steppe that was converted to sand or barren, which was more sensitive to the effects of precipitation, which increased from 2000 to 2010, suggesting that land desertification in the upstream region continued to worsen.

4. Discussion

Climate change and human activities are the two critical factors affecting land-desertification and -restoration processes and could influence land desertification on a regional scale [47]. In this study, the driving factors of land desertification and restoration were revealed by analyzing changes in precipitation, temperature, population, cropland, livestock, and political measures.

4.1. Impact of Climate Change on Land Desertification

The HRB is a typical inland basin in the northwest of China, with a unique “glacier-permafrost–river–oasis–desert–terminal lake” multifaceted natural landscape linked by water. The upper, middle, and lower reaches of the HRB are three distinct geographical units. Therefore, we chose the annual mean temperature and precipitation data recorded at the Qilian, Zhangye, and Ejina meteorological stations from 1975 to 2015 to characterize climate change in different subregions, respectively (Figure 5).

In the upstream region of the HRB, both precipitation and mean annual temperatures show a significant upward trend, with a rate of change in the mean annual temperature of 0.46 °C/10a, which is 0.5 to 1 times the average warming of the Qing–Tibetan Plateau over the past four decades [48]. Due to the high altitude, warmer temperatures help to alleviate the thermal constraints on vegetation growth, which promotes vegetation growth. Meanwhile, increased precipitation also contributes to vegetation growth. However, some scholars believe that an increased temperature is a major climatic factor responsible for vegetation desertification in alpine regions [49], as it causes the upper permafrost to thaw or disappear, and the disappearance of the impermeable layer formed by the presence of permafrost decreases the water content of the soil in the vegetation root zone, thus leading to changes in soil structure and composition, which is also the main reason for the desertification of alpine meadows and swampy meadows [50]. In summary, taken together, the coexistence of land-desertification and -restoration processes is consistent with climate variability in the upstream region of the HRB.

In the mid- and downstream regions, the mean annual temperature also shows a similarly significant upward trend, with warming rates greater than 0.5 °C/10a. As the HRB is located in the northwestern arid zone of China, the increase in temperature has not been accompanied by a significant increase in precipitation, especially in the midstream region, where precipitation has actually tended to decrease, which has led to an increase in the degree of aridity, further aggravating the water constraints on vegetation growth, both of which are unfavorable to vegetation growth, leading to the development of land desertification. In addition, agricultural droughts caused by increased dryness will inevitably lead to a reduction in ecological water use as a result of enhanced irrigation to reduce water stress. As the cropland in the HRB is concentrated in the midstream region, the reduction in the amount of water released into the downstream region due to heavy irrigation has been a major factor contributing to downstream land desertification, especially between 1990 and 2000 [24].

4.2. Impact of Human Activity on Land Desertification

The impact of human activities on desertification is twofold: on the one hand, irrational human activities, such as overgrazing, over-cultivation, and over-exploitation of water resources, can exacerbate or accelerate the process of land desertification; on the other hand, rational human activities can slow down or even reverse desertification under adverse climatic conditions. In this study, we illustrate the impact of human activity on land desertification in five dimensions: population, the area of cropland, the number of livestock, political measures, and water resources.

The over-utilization of natural resources as a result of rapid population expansion is the root cause of all irrational human activities. Take Zhangye City as an example (Figure 6a); the population has increased from 68.25 × 10⁴ in 1964 to 113.0 × 10⁴ in 2020, with a particular peak in 2010 (130.83 × 10⁴), when the total number of people nearly doubled. A larger increase in population will inevitably lead to a significant increase in cropland, residential land, industrial land, and land for transport. This is the main reason for the significant decrease in DL since 2000. However, the lack of adequate protection of the land surface and the destruction of natural vegetation and soil structure due to the over-exploitation of natural resources by population growth and irrational economic activities, coupled with strong wind erosion and the accumulation of surface soil salts caused by the over-utilization of water resources, are primary factors contributing to land desertification.

Because the basin is landlocked, natural precipitation is not sufficient to support agricultural production, and, therefore, a large number of water resources is needed for irrigation to meet the demands of agricultural production. With the increasing area of cropland and the adverse effects of climate warming, the irrigation demand in the midstream oasis has been increasing, and the downstream ecological water use has been drastically reduced (Figure 7), leading to serious ecological desertification downstream. In addition, when surface water resources are unable to meet irrigation, the over-exploitation of groundwater to meet irrigation demand has caused the water table to fall, leading to the withering of natural vegetation around the oases. To cope with water shortages, the implementation of water-saving measures, such as drip pipes and furrow lining, has reduced irrigation use to some extent, but this has also reduced the amount of water that infiltrates, which has resulted in a lack of water recharge for the natural vegetation in oases and the drying out of the vegetation, leading to land desertification [23].

Overgrazing is another major cause of land desertification in the HRB. The population of livestock has been increasing linearly since the 1980s. Taking sheep farming in Zhangye City as an example (Figure 6b), the sheep population in 2020 was 446.6% of that in 1984. Rapidly increasing livestock numbers is making overgrazing widespread in the HRB. Furthermore, the over-exploitation of valuable medicinal plants, such as Ephedra, Cistanche, Cordyceps, and Licorice, is also a major factor in the desertification of natural grassland.

To curb land desertification and protect the ecological environment, the Chinese government has implemented a series of ecological protection and restoration measures, such as the “Three-North Shelterbelt Project”, “Grain for Green”, “the Natural Forest Protection Project”, the “Grazing Withdrawal Program”, and the “Construction of Nation Ecological Security Barriers”. In the HRB, the exclusion of grazing in degraded grassland, closed breeding, rehabilitation, and the use of saline land and the EWDP has been adopted to combat land desertification and rehabilitate the degraded environment. In addition, as one of the main bases for developing and constructing new energy sources in China, the construction of ecological PV power plants using the “PV + sand control” model has also significantly reduced desertified land in the HRB. These measures and policies play an important role in controlling desertification and restoring vegetation.

5. Conclusions

In this study, 30 m high-resolution information on land desertification and restoration in the HRB was extracted using remote sensing interpretation and GIS spatial analysis methods. Our results show that land desertification coexists with land restoration in the HRB. At different periods, the area of land restoration was much larger than the area of land desertification in the HRB, and the HRB has been in the process of land restoration. Upstream of the HRB, there is a continuing trend of increasing land desertification and a slowdown in land restoration. In the middle and downstream regions, desertification continues to decrease while land restoration continues to increase. The increasing desertification upstream is associated with overgrazing in a context where climate change favors desertification reversal. Although both climate variability and human activities favor land desertification, the process of land desertification in the middle and lower reaches is still being reversed, and land restoration dominates. Implementing the eco-environmental protection project and desertification control measures, particularly the EWDP, contributes to the reversal of desertification in the middle and lower reaches of the HRB. However, the EWDP indirectly contributes to lowering the water table in the middle reaches through the expansion of cropland and the implementation of water-saving irrigation systems. There is, therefore, an urgent need to transform the economic structure of the middle reaches, which is dominated by water-intensive agriculture, to cope with water scarcity and land desertification.