Spatial–Temporal Changes in the Distribution of Populus euphratica Oliv. Forests in the Tarim Basin and Analysis of Influencing Factors from 1990 to 2020

Abstract

Understanding the spatiotemporal evolution patterns of Populus euphratica Oliv. (P. euphratica) forests in the Tarim Basin (TB) and their influencing factors is crucial for regional ecological security and high-quality development. However, there is currently a lack of large-area, long-term systematic monitoring. This study utilized multi-source medium and high-resolution remote sensing images from the Landsat series and Sentinel-2, applying a Random Forest classification model to obtain distribution data of P. euphratica forests and shrublands in 14 areas of the TB from 1990 to 2020. We analyzed the effects of river distance, water transfer, and farmland on their distribution. Results indicated that both P. euphratica forests and shrublands decreased during the first 20 years and increased during the last 10 years. Within 1.5 km of river water transfer zones, P. euphratica forests more frequently converted to shrublands, while both forests and shrublands showed recovery in low-frequency water transfer areas. Farmland encroachment was most significant beyond 3 km from rivers. To effectively protect P. euphratica forests, we recommend intermittent low-frequency water transfers within 3 km of rivers and stricter management of agricultural expansion beyond 3 km. These measures will help maintain a balanced ecosystem and promote the long-term sustainability of P. euphratica forests.

1. Introduction

Populus euphratica Oliv. (P. euphratica) is a unique deciduous broad-leaved tree species found in desert riparian zones throughout North Africa, the Middle East, Central Asia, and Western China [1]. P. euphratica can only form riparian forests in desert areas, known as desert riparian forests, because it has strong drought and salt tolerance [2]. It serves as a natural barrier between deserts and oases, playing a crucial role in windbreak, sand fixation, biodiversity conservation, and maintaining ecosystem balance [3]. However, the habitat of P. euphratica forests is fragile and susceptible to external disturbances, with significant variations in health across the entire landscape [4]. Approximately 61% of the global distribution of P. euphratica is in China, with 89% of that found in the Tarim River Basin in Xinjiang [5]. Numerous studies have shown that climate warming and increased humidity lead to increased frequencies of drought stress on P. euphratica forests [6]. Excessive agricultural irrigation and inter-basin water transfers have significantly reduced or damaged P. euphratica forests far from rivers [7]. Field surveys indicate that the tree mortality rate in P. euphratica forests is increasing [8]. Therefore, long-term monitoring of the health of P. euphratica forests over time and space is essential.

Previous studies have primarily focused on the physiological ecology of P. euphratica [9,10], including aspects such as individual trees [11], reproduction [12], populations [13], and communities, and the limiting factors affecting its occurrence [14,15]. Numerous studies have investigated the effect of water resources on the distribution of P. euphratica, suggesting that hydrological conditions are the primary constraint on its distribution, and ecological water conveyance can alleviate water stress on P. euphratica. Numerous studies have investigated the impact of water resources on the distribution of P. euphratica, suggesting that hydrological conditions are the primary constraint on its distribution, and ecological water conveyance can alleviate water stress on P. euphratica [16,17]. However, few studies have discussed the effects of human activities and climate change on P. euphratica [18,19]. Aishan et al. (2024) [20] found that drought leads to hollowing in P. euphratica trees, thereby accelerating their aging. Peng et al. (2022) [21] studied the impact of climate change on water resource allocation in the Ejina P. euphratica forests and found that groundwater depth is the main factor limiting the growth of P. euphratica. Zhou et al. (2020) [4] analyzed tree ring data from P. euphratica forests and discovered that when groundwater depth exceeds 6 m, climate warming accelerates the decline of these forests. Zhang and Chen (2022) [22] revealed that human activities, such as population growth, agricultural expansion, irrigation, and grazing, lead to habitat loss and a decline in biodiversity in P. euphratica forests. Zhang et al. (2023) [23] found that higher salinity correlates with fewer P. euphratica trees. These studies widely used traditional field sampling and on-site surveys to investigate P. euphratica [24,25]. However, obtaining these indicators through field surveys is time-consuming and costly due to the sparse distribution of P. euphratica forests and their widespread presence along riparian zones in Xinjiang [26].

Remote sensing provides an effective tool for large-area monitoring of forest health, allowing the extraction of most health indicators from long-term remote sensing data. However, remote sensing technology has not been fully applied to the investigation of P. euphratica. Han et al. (2024) [27] used Sentinel-2 data to extract the Normalized Difference Vegetation Index (NDVI) of P. euphratica forests and combined it with field-measured radial growth indicators to analyze interannual changes. Li et al. (2021) [24] used GLASS LAI time series data from the upper reaches of the Tarim River to study the phenology of P. euphratica. Yan et al. (2022) [26] used multisource remote sensing data to study the relationship between the distribution of P. euphratica over nearly 60 years in Tarim National Nature Reserve and factors such as water systems, farmland, and soil salinity. The results showed that farmland within 1 km of the Tarim River significantly affects P. euphratica, whereas the impact diminishes beyond 1 km. The occurrence of P. euphratica decreases with increasing soil salinity. Li et al. (2023) [28] utilized NOAA-AVHRR, EOS-MODIS, and Landsat-TM remote sensing data to construct the Enhanced Vegetation Index (EVI) for the canopy of P. euphratica forests in the lower reaches of the Tarim River. Based on this, they explored the response of P. euphratica to groundwater. Wang et al. (2024) [29] used high-resolution satellite imagery to extract vegetation coverage in the lower reaches of the Tarim River with higher accuracy. Su et al. (2017) [30] employed airborne hyperspectral data to classify land cover in the Ejina Oasis. These studies mainly focused on utilizing remote sensing technology to study the spatial patterns of P. euphratica forests, lacking in-depth research on the driving factors of changes in these forests. Some studies have shown that shrubland encroachment is one of the reasons for the reduction in the area of arboreal forests [31]. Therefore, this paper proposes using multisource remote sensing technology to study the spatiotemporal changes in the distribution of P. euphratica forests and further explore the impact of human activities and natural factors on their distribution.

This study utilized Sentinel-2 data and Landsat time series data to analyze the spatiotemporal changes in the distribution and mutual transformation of P. euphratica forests and shrublands in the TB from 1990 to 2020. The study focuses on the data from 1990 to 2020 because this period encompasses significant ecological and anthropogenic changes in the TB. The EWTP began in 2000, making it a critical node for observing changes before and after its implementation. Additionally, studies have shown that 1995–2020 marked a period of rapid arable land expansion, further impacting the region’s vegetation dynamics. Therefore, this timeframe allows us to comprehensively analyze the effects of human activities such as water transfer and agricultural expansion on the distribution of P. euphratica forests by employing the intensity analysis, transition matrix, and buffer zone methods. The study aimed to address the following four key research questions: (1) What was the spatiotemporal distribution of P. euphratica forest and shrublands in the TB over the period of 1990–2020? (2) What were the spatiotemporal changes in the interconversion of P. euphratica forest and shrublands in the TB over the period of 1990–2020? (3) How did farmland affect the interconversion of P. euphratica forest and shrublands in the TB over the period of 1990–2020? (4) How did water affect the interconversion of P. euphratica forest and shrublands in the TB over the period of 1990–2020? This study could provide decision support for the conservation of P. euphratica forests in the TB.

2. Materials

2.1. Study Area

The study area is located in the southern part of Xinjiang, within the Tarim Basin (TB), encompassing the desert riparian zones of the Tarim River, Kongque River, Yarkand River, Kashgar River, Sangzhu River, Hotan River, Cele River, Keriya River, Cherchen River, and the rivers on the northern slopes of the Kunlun Mountains, covering an area of approximately 10.06 km² (Figure 1). Geographically, it spans latitudes 36°04′ N to 42°16′ N and longitudes 76°16′ E to 88°40′ E. The region features a simple topography, consisting entirely of plains, with elevations ranging from 823 m to 1599 m, gradually decreasing from south to north (Figure 1). The TB is situated in the inland arid region of northwestern China, characterized by an annual average precipitation of less than 120 mm. However, the evaporation rate exceeds the annual precipitation by more than 20 times [32]. The temperature variation in the TB is substantial, with daily maximum temperatures exceeding 40 °C and daily minimum temperatures potentially dropping below −40 °C [33]. The desert riparian forests are one of China’s most important arid forest vegetation communities, primarily composed of P. euphratica forests [7]. The understory herbaceous vegetation includes species such as Alhagi sparsifolia and reeds, while the shrub vegetation is mainly composed of Tamarix species [34]. Referring to the data of the second national land survey in 2009, the TB was divided into 14 P. euphratica forest areas based on the characteristics of P. euphratica forests as desert riparian forests (Figure 1). The administrative divisions where the 14 P. euphratica forest areas are located are shown in

Table 1. P. euphratica forest distribution rivers in the TB included in this study.

Number	Code	P. euphratica Forest Distribution Rivers	Administrative Subdivision
1	A	Up-Tarim River	Shojak-Imbazar section of Shaya County
2	B	Mid-Tarim River	Yingbazha-Qala section bugur and Kuqa counties
3	C	Down-Tarim River	Yuli County from below Daxihaizi to above Lake Daitama
4	D	Up-Yarkand River	Shamal Forest and Bachu County
5	E	Down-Yarkand River	The three missions in Keping County and the three harvest fields in Awati County
6	F	Mid-Hotan River	The Krakash River in Lopu County and the Moyu River in Moyu County
7	G	Down-Hotan River	P. euphratica forests along the Bostan Ranch and Hotan River in Awati County
8	H	Kongque River	Acha Junction to Gili Changkou Gate Section
9	I	Kashgar River	Payzawat County
10	J	Cherchen River	Qiemo County
11	K	Rivers in the northern Kunlun Mountains	P. euphratica forests along the Niya River and scattered along small rivers
12	L	Keriya River	Primitive P. euphratica forests along the rivers of the Keriya River and Dariyabui
13	M	Cele River	Qira County
14	N	Sangzhu River	Pishan County

.

2.2. Data Sources

2.2.1. Remote Sensing Dataset

In this study, high-quality, cloud-free, multi-band Landsat series remote sensing images from the growth season (July–September) of P. euphratica in the years 1990, 1995, 2000, 2005, 2010, 2015, and 2020 were selected. Landsat 8 OLI (Operational Land Imager) data were used to extract the spectral characteristics of P. euphratica forests, shrublands, water bodies, and croplands for the years 2015 and 2020. For the years 1990, 1995, 2000, 2005, and 2010, Landsat 5 TM (Thematic mapper) and Landsat 7 ETM+ (Enhanced Thematic mapper Plus) were supplemented. The full time series of Landsat SR Collection 2 Level 2 dataset from 1990 to 2020 was used for surface water extraction. The remote sensing dataset used in this study spans from 1990 to 2020, chosen to capture the ecological impact of the water transfer project initiated in 2000 and the rapid expansion of arable land from 1995 onwards. By analyzing data over these three decades, we aim to understand the long-term effects of these significant environmental changes on the P. euphratica forests in the TB. The Landsat image sequence numbers involved are P141R032, P142R032, P142R033, P143R031, P143R032, P143R033, P143R034, P144R031, P144R032, P144R034, P145R031, P145R032, P145R033, P145R034, P146R031, P146R032, P146R033, P146R034, P147R032, P147R033, P147R034, P148R032, P148R033. The data cover the 1990, 1995, 2000, 2005, 2010, 2015, and 2020 P. euphratica growing seasons (July~September) with 1114 Landsat images (Figure 2).

Sentinel satellite images were selected for their multispectral surface reflectance data (L2A), which underwent orthorectification, radiometric calibration, and atmospheric correction. High-quality, cloud-free Sentinel-2 multispectral remote sensing images from the years 2015 and 2020 were chosen to extract the spectral characteristics of P. euphratica forests, shrublands, water, and farmlands, and these characteristics were fused with those extracted from Landsat 8 OLI. Due to the difference in spatial resolution between Landsat series images (30 m) and Sentinel-2 images (10 m), the cloud-free Sentinel images were resampled to 30 m, and all images were clipped in accordance with the study area boundaries. All remote sensing image data in this study were processed on the GEE platform (https://Developers.Google.Com/Earth-Engine/Datasets/Catalog) (accessed on 20 November 2023). The Sentinel-2 image sequence numbers involved are T43SFD, T43SFE, T43SGH, T44TLK, T44SLK, T44SLJ, T44SLH, T44SLG, T44SKG, T44TMK, T44SMK, T44SMJ, T44SMH, T44SMG, T44SMF, T44SNF, T44SNG, T44SNH, T44SNJ, T44SNK, T44TNL, T44TNK, T44TPM, T44TPL, T44TPK, T44SPJ, T44SPH, T44SPG, T44SPF, T44SQG, T44SQH, T44TQK, T44TQL, T45TTG, T45TTF, T45STC, T45STB, T45SUB, T45SUC, T45TUE, T45TUF, T45TUG, T45TVF, T45TVE, T45SVD, T45SVC, T45SWC, T45SWD, T45TWE, T45TWF, T45TXF, T45TXE, T45SXE, T45SXP. The data cover the 2015 and 2020 P. euphratica growing seasons (July~September), with 137 Sentinel-2 images (Figure 2).

2.2.2. Data for Validation

The validation data for this study originate from two sources: actual observation plots and interpreted image plots. The actual observation plots were collected during various field surveys: the core germplasm survey of P. euphratica was conducted by the research team in Northwest China during July and August 2021, the biodiversity survey of P. euphratica in the TB was performed during July and August 2022, the P. euphratica census in Hotan was conducted in February 2023, and the community structure survey in Northwest China was performed during July and August 2023. These surveys covered the entire northwest region, resulting in over 2000 sampling plots, including categories such as P. euphratica forests and shrublands. After the raw validation data were filtered, a total of 1255 plots were selected, comprising 956 plots of P. euphratica forests and 299 plots of shrublands.

The actual observation points also include 162 instances of multispectral UAV data. In the TB, 54 sampling plots were established, each containing three subplots, including P. euphratica forests and shrublands. A DJI P4 multispectral UAV was used, which integrates one visible-light camera and five multispectral cameras (blue, green, red, red edge, and near-infrared), totaling six image sensors to capture data above the canopy, thus enabling the acquisition and fusion of various spectral data. The flight altitude was set at 189 m (with a resolution of 10 cm) and the speed was 15 m/s. The aerial survey covered areas of 500 m × 500 m and 1000 m × 1000 m, with autonomous flight paths over the target areas. The resulting aerial images were processed using DJI Terra v3.4.4 software for geographic coordinate correction, radiometric calibration, and image fusion. The fused aerial images were then subjected to precise extraction of spectral band data by using the remote sensing image processing software ArcGIS 10.8.2 and ENVI 5.6.

The interpreted image plot data were derived from high-resolution remote sensing images with a spatial resolution exceeding 2 m (Google Earth Online Images, https://code.earthengine.google.com/) (accessed on 20 November 2023). Due to the distinct visual characteristics of farmlands, the obvious spatial distribution patterns of P. euphratica, and the clear texture information in high-resolution remote sensing images, this study determined the types of interpreted image plots on the basis of the tone, texture, and landscape features of the actual observation plots in the corresponding years. This process resulted in the creation of a historical ground sample dataset. A total of 4169 validation points were selected, including farmlands (750), water (1045), P. euphratica forests (1516), and shrublands (858). Farmlands were defined as a combination of standard farmlands, basic farmlands, and fallow lands for each year [35]. In accordance with the land cover classification system [36] and the current land use classification system in China [37], P. euphratica forests were classified as such when the vegetation cover of P. euphratica exceeded 10%, encompassing dense and sparse P. euphratica forest plots.

3. Methods

In order to study the spatial and temporal changes of P. euphratica forests in the TB from 1990 to 2020, the information on P. euphratica forests, shrublands, water, and farmlands was extracted based on the Random Forest model using Landsat and Sentinel-2 images. On this basis, the spatial and temporal changes in the distribution of P. euphratica forests and shrublands in the TB from 1990 to 2020 were analyzed, and the impacts of farmland, river systems, and ecological water transport on the interconversion of P. euphratica forests and shrublands were explored. Figure 3 shows the technical flow chart.

3.1. Classification Scheme

3.1.1. Random Forest Model

Following the methodology described in Peng et al. (2022) [26], we have improved the methodology. The study indicated that the Random Forest model alone provided high accuracy and reliability for poplar trees [38]. The Random Forest model, a graph-theoretic forest, classifies by randomly selecting training samples and features, then performing classification votes, and integrating the votes of each tree to obtain the final classification [39,40]. The key to the Random Forest model is determining the number of samples, input features, and trees. We carefully considered different types of target objects and manually selected training samples by combining field survey data, aerial photography data, and high-resolution remote sensing images. In this study, we selected dense, sparse, healthy, and unhealthy P. euphratica and shrubland samples. For farmland samples, we also considered different types of farmland. If the number of decision trees is set too low, underfitting occurs, while setting the number too high does not significantly improve the model. By comparing experimental results with different numbers from 1 to 151 at intervals of 5, we ultimately set the number of trees to 100. The detailed table of the classes are shown in

Table 2. Detailed table of the classes.

Class ID	Class Name	Description
1	Farmland	Areas with a combination of standard farmland, basic farmland, and fallow farmland for each year. The texture characteristics of farmland are block shaped distribution and bright red color.
2	P. euphratica forest	Areas with dense, sparse, healthy, and unhealthy P. euphratica. Vegetation coverage greater than 10%. The texture features of P. euphratica forest are dark red dots with rough edges.
3	Shrubland	Areas with dense, sparse shrubland. The texture characteristics of shrubbery are circular clumps of light red, with delicate edges, and generally larger individual coverage areas than P. euphratica.

Notes: Texture features are based on Landsat band4/3/2 and Sentinel-2 band8/4/3.

.

3.1.2. Input Features

This study selected 10 Landsat image features and 13 Sentinel-2 image features from the two dimensions of spectrum and spectral index and constructed a multidimensional classification feature set. The detailed input characteristics are shown in

Table 3. Input features.

Type	Feature	Resolution (m)	Source
Spectrum	Red, Green, Blue, NIR, SWIR1, SWIR2	30	Landsat
	Red, Green, Blue, RE1, RE2, NIRn1, NIRn2, NIR, SWIR1, SWIR2	10–60	Sentinel-2
Spectral index	NDVI = (NIR − Red)/(NIR + Red)	30	Landsat
	MNDWI = (Green − SWIR)/(Green + SWIR)
	NDBI = (SWIR − NIR)/(SWIR + NIR)
	EVI = 2.5 × ((NIR − Red)/(NIR + 6 × Red − 7.5 × Blue + 1))
	NDVI704 = (RE2 − RE1/RE2 + RE1)	20	Sentinel-2

.

The modified normalized difference water index (MNDWI) calculates the difference ratio between green and near infrared [41] and is used for identifying vegetation with vegetation indexes [42,43]. The normalized difference vegetation index (NDVI) can quantitively reflect vegetation growth by calculating the difference ratio between the NIR and red band [44]. The enhanced vegetation index (EVI) is an ‘optimized’ vegetation index with improved sensitivity in high biomass regions and improved vegetation monitoring through a de-coupling of the canopy background signal and a reduction in atmosphere influences [45]. The normalized difference built-up index (NDBI) can accurately reflect the information of building land use [46]. NDVI704 is a modified vegetation index of the red-edge band based on NDVI. It is susceptible to small changes in leaf canopy, gap fragments, and senescence, and can be used in precision agriculture and forest monitoring.

3.1.3. Accuracy Verification

In this study, the Kappa coefficient was calculated using the confusion matrix method to validate the results of the classification of P. euphratica forests, shrubland, and farmlands. The Kappa coefficient is a metric used to evaluate the classification ability of a classifier, which takes into account the balance between the proportion of correct predictions and the proportion of incorrect predictions made by the classifier [47]. The formulas are shown in (1).

$$Kappa={\displaystyle \frac{P_0-P_e}{1-P_e}}$$

(1)

where P₀ is the ratio of the sum of the diagonal elements of the matrix to the sum of the elements of the matrix as a whole; P_e is the ratio of the sum of the “product of actual and predicted quantities” corresponding to each of the categories to the square of the total number of samples of all the elements. The kappa coefficient calculation results are located between −1 and 1. The larger the value, the higher the consistency between the actual value and the predicted value, and the more accurate the classification effect.

3.2. Spatiotemporal Variation of P. euphratica Forest, Shrublands, and Farmland

3.2.1. Strength Analysis

With reference to Aldwaik and Pontius (2012) [48] and Ke et al. (2019) [49], the intensity of conversion between P. euphratica forests and shrubland in the TB from 1990 to 2020 was analysed using intensity analysis. Intensity analysis based on multi-temporal land use results analyzes the characteristics of land cover changes over each time period by calculating a cross-matrix. In this study, to analyze the spatiotemporal change characteristics of shrublands and P. euphratica forests from 1990 to 2020, the change intensity of these two land cover types was calculated at two levels as follows:

Category Layer

This analysis primarily indicates the total loss and gain of each land use category within each time period, as well as the characteristics of their intensity changes. The formulas are shown in (2) and (3).

$$G_{tj}={\displaystyle \frac{\frac{\left[\left({\sum }_{j=1}^J{C}_{tij}\right)-C_{tij}\right]}{Y_{t+1}-Y_t}}{{\sum }_{j=1}^J{C}_{tij}}}\times 100\%$$

(2)

$$L_{ti}={\displaystyle \frac{\left[\left({\sum }_{j=1}^J{C}_{tii}\right)/\left(Y_{t+1}-Y_t\right)\right]}{{\sum }_{j=1}^J{C}_{tii}}}\times 100\%$$

(3)

where $G_{tj}$ represents the annual total gain intensity of category j during the time period [$Y_t,Y_{t+1}$]; $L_{ti}$ represents the annual total loss intensity of category i during the time period [$Y_t,Y_{t+1}$].

Conversion Layer

We analyzed the transformation relationship and characteristics between a certain land use category and other categories at different time periods for each land use category, as shown in Formulas (4) and (5).

$$R_{tin}={\displaystyle \frac{C_{tin}/\left(Y_{t+1}-Y_t\right)}{{\sum }_{j=1}^J{C}_{tij}}}\times 100\%$$

(4)

$R_{tin}$ represents the annual intensity of the transformation from category i to category n during the time period [$Y_t,Y_{t+1}$].

$$W_{tn}={\displaystyle \frac{\left[\left({\sum }_{i=1}^J{C}_{tin}\right)-C_{tnn}\right]/\left(Y_{t+1}-Y_t\right)}{{\sum }_{j=1}^J\left[\left({\sum }_{j=1}^J{C}_{tij}\right)-C_{tii}\right]}}\times 100\%$$

(5)

$W_{tn}$ represents the uniform intensity value for all non-category n transitions to category n during the time period [$Y_t,Y_{t+1}$].

3.2.2. Transition Matrix

With reference to Kondum et al. (2024) [50] and Guo et al. (2023) [31], this study employed the Markov transition matrix method to analyze the conversion characteristics between P. euphratica forests and farmlands from 1990 to 2020. The calculations were performed according to Equations (6) and (7):

$$P_{ij}=\left[\begin{array}{ccc}{P}_{11}& \cdots & P_{1n}\\ ⋮& \ddots & ⋮\\ P_{n1}& \cdots & P_{nn}\end{array}\right]$$

(6)

$$A_i={\sum }_{j=1}^n{P}_{ij};B_j={\sum }_{i=1}^n{P}_{ij}$$

(7)

where represents the magnitude of the transition from category i to category j, represents the area of category i in the period t₁ (km²), and represents the area of ecosystem type j in the period t₂ (km²).

3.3. Occurrence Frequency of Surface Water

Surface water extraction employed a multispectral index algorithm based on decision tree classification [51]. This algorithm, which analyzes the spectral response mechanisms of inland surface water, uses multiple comprehensive indices and incorporates the JRC Global Surface Water Dataset [52] as prior knowledge, randomly selecting samples. The modified normalized difference water index (MNDWI), normalized difference vegetation index (NDVI), and land surface water index (LSWI) were extracted using the full time series Landsat SR dataset from 1990 to 2020. LSWI, the land surface water index, refers to the ratio of reflectance in the near-infrared band to that in the shortwave infrared band, which can identify the characteristics and location of water [53]. Surface water is determined when MNDWI is greater than 0, and both MNDWI > NDVI and LSWI > NDVI occur simultaneously. Additionally, the frequency of occurrence of surface water is the percentage of the 1990–2020 time period in which a pixel is covered by water. When the occurrence frequency approaches 1, the frequency of surface water presence is higher, whereas a frequency of 0 indicates no surface water. At the same time, the average frequency of occurrence of surface water and the area of surface water overflowing the water surface of different rivers from 1990 to 2020 were counted. Thus, by utilizing the occurrence frequency and overflow water surface area of surface water in the study area, the impact of water resources on the distribution of P. euphratica is explored.

3.4. Buffer Analysis

This study established symmetrical buffer zones (0.5, 1, 1.5, 2, 2.5, 3, 4, 5, 6, 8, 10 km) around the major rivers (Tarim River, Yarkand River, Kashgar River, Hotan River, Keriya River, Niya River, Cele River, Sangzhu River, Cherchen River, and Kongque River) in the TB where P. euphratica forests are predominantly distributed. By analyzing the spatiotemporal changes in the distribution of P. euphratica forests and shrublands from 1990 to 2020, we investigated the impact of farmland, river systems, and EWTP on the mutual transformation between P. euphratica forests and shrublands.

4. Results and Analysis

4.1. Accuracy Verification of Extracting P. euphratica Forest, Shrublands, and Farmland

The ground sample data from field surveys were matched with contemporaneous Landsat series and Sentinel-2 images, and the pixels corresponding to the sample points were annotated. The high-resolution image sample data, along with the ground sample data, constitute the sample set for different land cover types in this study. Of this sample set, 70% of the data were used as the classification sample set and 30% were used as the validation sample set. P. euphratica forests, which ‘thrive with water’, are highly dependent on groundwater and primarily distributed in areas close to water bodies such as alluvial and flood plains. The forests in the TB are predominantly composed of P. euphratica, exhibiting a marked uniformity [7]. Therefore, considering the growth characteristics of P. euphratica, medium- and high-resolution images can be used to accurately determine the distribution range of P. euphratica forests over different periods.

Considering the characteristics of land cover in arid regions and the research requirements, the distribution ranges of P. euphratica forests, shrublands, and farmlands were extracted. The sample points for P. euphratica forests included dense and sparse forests (excluding artificial forest sample points). This study used random validation points, field survey validation points, and high-resolution image validation points to assess the accuracy of the seven-period data by using the confusion matrix method. The overall interpretation accuracy exceeded 85% (Figure 4 and Tables S1 and S2). A cross-validation was conducted using the confusion matrix method and the 30 m LUCC land use/cover change data (1980–2020) [37] from the Chinese Academy of Sciences for the TB, specifically focusing on P. euphratica forests (dense and sparse forests), shrublands, and farmlands. The resulting kappa coefficient was 85.37%. The main areas of improvement were the P. euphratica forests along the Hotan River, the rivers on the northern slopes of the Kunlun Mountains, the Keriya River, and the Sangzhu River.

4.2. The Temporal and Spatial Changes of P. euphratica and Shrubland

As shown in Figure 5h, for the period 1990–2020, the P. euphratica forests and shrublands exhibited a trend of initially decreasing and then increasing. The area changes in shrublands were more pronounced than those in P. euphratica forests. Specifically, during the periods 1995–2000, 2000–2005, 2010–2015, and 2015–2020, the area changes in P. euphratica forests were significant, with changes exceeding 1000 km² in each period. The areas of P. euphratica forests decreased in the Cele River, Cherchen River, lower Tarim River, Kashgar River, Konque River, middle Tarim River, Sangzhu River, upper Tarim River, and upper Yarkand River, with the most substantial reduction observed in the middle Tarim River. Conversely, the areas of P. euphratica forests increased in the lower Hotan River, lower Yarkand River, Keriya River, the rivers on the northern slopes of the Kunlun Mountains, and middle Hotan River, with the most significant increase observed in the lower Yarkand River. The spatial changes in P. euphratica forests and shrublands in the TB from 1990 to 2020 during various time periods are shown in Figure 5a–g. The distribution of P. euphratica forests exhibited a high degree of fragmentation. By using the Taklamakan Desert as the central axis, the southern region showed higher fragmentation than the northern region. The P. euphratica forests distributed along the rivers on the northern slopes of the Kunlun Mountains were scattered amongst shrublands, exhibiting a fragmented distribution pattern. From 1990 to 2010, the reduction in P. euphratica forest area was mainly characterized by the fragmentation of large land parcels, particularly along the lower Tarim River and the rivers on the northern slopes of the Kunlun Mountains. From 2010 to 2020, a significant increase in the area of P. euphratica forests could be observed along the main course of the Tarim River, extending into shrubland areas.

The area of P. euphratica forests distributed along different rivers in the TB was statistically analyzed for seven periods, considering various distances from the rivers. As shown in Figure 6, from 1990 to 2020, the area of P. euphratica forests first decreased and then increased with increasing distance from the river channels. Specifically, 27.53% of the P. euphratica forests were within 1.5 km of the rivers, 18.87% were between 1.5 and 3 km, and 53.60% were beyond 3 km. This finding indicates that most of the P. euphratica forests were sparse woodlands growing more than 3 km away from the rivers in desert areas. Spatially, the P. euphratica forests along the Tarim River, Yarkand River, the rivers on the northern slopes of the Kunlun Mountains, and the Cherchen River were mainly distributed within 1 km and beyond 3 km from the rivers. By contrast, the P. euphratica forests along the Hotan River, Kongque River, and Keriya River were primarily within 1 km of the rivers. The P. euphratica forests along the Kashgar River were located further away, beyond 3 km from the river. From 1990 to 2020, the area of shrublands showed an increasing trend with increasing distance from the river channels. The majority of shrublands (34.84%) were distributed 8–10 km from the rivers and each buffer zone between 0.5 and 8 km from the rivers had an average of 7.24% shrubland distribution.

4.3. Mutual Transformation between P. euphratica Forest and Shrublands

As shown in Figure 7a,b, the mutual transformation includes four components: P. euphratica to shrublands, shrublands to P. euphratica, P. euphratica to others, and shrublands to others. Before 2000, the intensity of the area change from P. euphratica to shrublands increased, whereas after 2000, it decreased, and the intensity of the area change from shrublands to P. euphratica began to rise. This finding indicates that before 2000, shrublands encroached on P. euphratica forests, but after 2000, shrublands gradually converted back to P. euphratica forests. As shown in Figure 6c,d, the encroachment of shrublands on P. euphratica forests mainly occurred in the middle and lower Tarim River, the Kongque River, the Cherchen River, and the Kashgar River. Within 2 km of the rivers, the encroachment of shrublands on P. euphratica forests was greater than the conversion of shrublands to P. euphratica forests; between 2 and 3 km, the conversion areas were equivalent; beyond 4 km, the area of shrublands converting to P. euphratica forests was greater than the encroachment. Additionally, the further from the rivers, the greater the area of conversion between P. euphratica forests and shrublands.

4.4. Farmland Effects on P. euphratica and Shrubland Distribution

Overall, the distribution of farmland was relatively concentrated, with a small portion scattered within P. euphratica forests, mainly along the main course of the Tarim River (Figure S1). Figure 8a shows that during the periods 1995–2000, 2005–2010, and 2010–2015, a significant conversion of P. euphratica forests to farmland occurred, resulting in a substantial decrease in P. euphratica forest area and an increase in fragmentation. From 1990 to 2020, the area of farmland in the TB expanded nearly threefold. During this period, 19.79% of P. euphratica forests and 10.61% of shrublands were converted to farmland. Figure 8b–d indicates that from 1990 to 2020, the area of P. euphratica forests and shrublands converted to farmland far exceeded the area of farmland converted back to P. euphratica forests and shrublands. The conversion of P. euphratica forests to farmland mainly occurred in the middle and upper reaches of the Tarim River, the upper and lower Yarkand River, the upper Kongque River, the Kashgar River, and the upper Cherchen River. Except for the rivers on the northern slopes of the Kunlun Mountains, where the area of P. euphratica forests converted to farmland was smaller than that of shrublands converted to farmland, all other rivers showed a higher conversion of P. euphratica forests to farmland. The conversion of shrublands to farmland was mainly distributed in the upper Tarim River, the Cherchen River, and the rivers on the northern slopes of the Kunlun Mountains. Figure 8e,f shows that within 0.5–3 km of the rivers, the area of P. euphratica forests and shrublands converted to farmland was relatively small, gradually increasing from 3 km to 10 km. Meanwhile, the trend of farmland converting back to P. euphratica forests and shrublands was similar. However, the area of farmland converting to P. euphratica forests was far greater than the area converting to shrublands.

4.5. Water Effects on P. euphratica and Shrublands Distribution

The ecological water transfer process in the TB was analyzed on the basis of the frequency of surface water occurrence. Figure 9a,b summarizes the water transfer frequency, inundation area, and the distribution of the conversion between P. euphratica forests and shrublands for different rivers from 1990 to 2020. The results show that the upper reaches of the Tarim River had the largest inundation area in the ecological water transfer zone, followed by the lower Hotan River, with the middle Tarim River, the upper Yarkand River, and the Cherchen River having similar inundation areas. The Sangzhu River had the smallest inundation area. The water transfer frequencies in the upper and middle Tarim River, the Yarkand River, the Hotan River, the Kashgar River, and the Cherchen River were mostly medium-to-low frequency. This finding indicates that the area of P. euphratica forest receiving water transfer in these regions is large and widespread. According to investigations, the lower Tarim River began receiving water transfer in 2000. However, high-frequency regions are still rare, possibly because the water transfer area in the lower Tarim River is too singular.

Figure 9c shows that from 1990 to 2020, in the water transfer zones of the upper Tarim River, the upper Yarkand River, the Cherchen River, the lower Hotan River, and the rivers on the northern slopes of the Kunlun Mountains, the area of shrublands converting to P. euphratica forests was greater than the area of P. euphratica forests converting to shrublands. However, in the water transfer zones of the lower Tarim River, the Keriya River, and the Kongque River, the area of P. euphratica forests converting to shrublands was greater than that of shrublands converting to P. euphratica forests, with the largest conversion areas in the low-frequency water transfer zones. This finding indicates that the ecological water transfer in certain regions of the TB from 1990 to 2020 has led to the restoration of shrublands. Figure 9d shows that within 1.5 km of the rivers in the ecological water transfer zones, the area of P. euphratica forests converting to shrublands was greater than that of shrublands converting to P. euphratica forests. Between 1.5 and 3 km, the area of P. euphratica forests converting to shrublands decreased with increasing distance from the river channels, whereas the area of shrublands converting to P. euphratica forests increased. Beyond 4 km from the rivers, the areas of mutual conversion between P. euphratica forests and shrublands increased, with the area of shrublands converting to P. euphratica forests significantly exceeding that of P. euphratica forests converting to shrublands.

5. Discussion

Long-term remote sensing monitoring often faces challenges due to a lack of plot data and the presence of pseudo-changes at boundaries, making it difficult to verify the accuracy of the results. This limitation introduces uncertainties in subsequent analyses. In this study, historical period results were extracted on the basis of the 2020 data, and the results for change areas were updated, thereby reducing pseudo-changes to some extent. Additionally, the 2020 results were validated using basic ground truth data, ensuring reliability for subsequent analyses. Consequently, the results for other periods can also be considered reliable.

5.1. Shrublands Encroachment Effects on P. euphratica Distribution

From 1990 to 2020, the area of P. euphratica forests in the TB generally showed a trend of initial decrease followed by an increase, with 2010 as the turning point. This trend is consistent with the findings of Peng et al. (2014) [54] and Li et al. (2020) [55]. This pattern is closely related to the implementation of the Grain for Green policy in 2012 and the rapid population growth in Xinjiang between 2005 and 2010. The trends in changes of P. euphratica forests and shrublands in the TB showed a good level of consistency, though differences in the magnitude of change were observed. The rate of change in the shrubland area was higher than that in P. euphratica forests, but the degree of fragmentation in P. euphratica forest distribution was greater than that in shrublands, which may be due to ecological water transfers favoring the recovery of species associated with P. euphratica forests [56]. From 1990 to 2010, the reduction in the area of P. euphratica forests was mainly characterized by the fragmentation of large land parcels, a period during which agricultural development was rapid [57]. This reduction was primarily observed along the lower Tarim River and the rivers on the northern slopes of the Kunlun Mountains. From 2010 to 2020, a significant increase was found in the area of P. euphratica forests along both sides of the main stream of the Tarim River, extending into shrubland areas.

The results of this study indicate that before 2000, shrublands encroached upon P. euphratica forests, but after 2000, shrublands gradually transitioned back to P. euphratica forests. However, the reduction in P. euphratica forests was not solely due to their conversion to shrublands. The majority were converted to other types such as farmlands and deserts. This finding is consistent with the research of I. Säumel et al. (2011) [58]. The encroachment of other types into P. euphratica forests led to desertification, thereby reducing the ecological service value of these forests [59]. However, the encroachment of shrublands into P. euphratica forests is part of a natural succession process, which increases fragmentation and negatively impacts soil moisture, altering soil nutrient cycling or availability [60]. Similar findings were observed in the studies by Ma [61] and Pompeu [62]. Therefore, the expansion of shrublands may lead to widespread forest loss and landscape fragmentation, disrupting the natural habitats of specific species or populations. Even small-scale conversions should be given due attention.

5.2. Farmland Expansion Effects on P. euphratica Distribution

From 1990 to 2020, the farmland area in the TB expanded nearly threefold. During this period, 19.79% of P. euphratica forests were converted to farmland, whereas only 10.61% of shrublands were converted to farmland. This finding indicates that the expansion of farmland primarily affected the distribution of P. euphratica forests in the TB. As the farmland area rapidly expanded, agricultural water consumption significantly increased, leading to a reduction in the available water for P. euphratica forests and severe degradation. Meanwhile, unreasonable agricultural irrigation practices exacerbated soil salinization and groundwater depletion [63]. In the lower Kongque River, the Yarkand River, the rivers on the northern slopes of the Kunlun Mountains, the middle Tarim River, and the upper Yarkand River, the farmland area showed a trend of first increasing and then decreasing, with the decrease starting after 2010. This change is related to agricultural development policies. The establishment of the socialist market economy in 1992, the abolition of agricultural taxes in 2006, and the strengthening of basic farmland protection all actively promoted rapid agricultural development, causing a rapid expansion of farmland before 2010. After 2015, efforts to modernize agricultural and rural development prioritized resource conservation and environmental protection [64], slowing the rate of farmland expansion.

Furthermore, the study results indicate that farmland primarily encroached on P. euphratica forests located more than 3 km away from rivers. Therefore, the P. euphratica forests beyond 3 km should be restored in time by creating ecological protection zones and returning farmland to forests. At the same time, remote sensing technology can be used to monitor the cultivation of farmland and the growth of P. euphratica forests beyond 3 km for a long time, so that early detection can allow it to be stopped as soon as possible. For P. euphratica forests with good growth conditions, it is sufficient to let them grow naturally. Figure 10 quantifies the indirect effects of farmland on the mutual conversion between P. euphratica forests and shrublands, analyzing the correlation between farmland and the conversion of P. euphratica forests and shrublands within a 1 km radius of farmland. It can be observed that areas near farmland predominantly exhibit the conversion of shrublands to P. euphratica forests. This indicates that farmland expansion reduces shrubland areas, and the subsequent reforestation efforts mainly increase P. euphratica forests. Studies have shown that an increase in shrublands leads to greater fragmentation of P. euphratica forests [31]. Therefore, converting shrublands to farmland followed by reforestation with artificial planting can enhance the connectivity of P. euphratica forests. The expansion of farmland can affect the microclimate and soil structure, thereby impacting the root distribution and water uptake of P. euphratica, increasing the fragmentation of natural habitats and reducing the suitable growth area for these forests [65,66]. Unreasonable land reclamation increased the proportion of over-mature forests far from water sources [15,65], thus highlighting the necessity of rational land reclamation to protect P. euphratica forests. Overall, sustainable agricultural practices and effective water resource management are crucial for the protection and restoration of P. euphratica forests in arid regions.

5.3. River Distance and Water Transfer Effects on P. euphratica Distribution

The most significant distribution characteristic of P. euphratica is its robust growth in areas with water resources [67]. Analysis of the impact of river systems on the distribution of P. euphratica forests revealed that from 1990 to 2020, the area of P. euphratica forests first decreased and then increased with increasing distance from river channels. Specifically, 27.53% of P. euphratica forests were within 1.5 km of rivers, 18.87% were between 1.5 and 3 km, and 53.60% were beyond 3 km. By contrast, shrublands were primarily distributed 8–10 km from rivers, accounting for 34.84%. Wang et al. (2023) [68] used precise species distribution models to show that the suitable habitat for P. euphratica is mainly along riverbanks. P. euphratica seedlings primarily use surface water, whereas mature trees rely more on deep soil water and groundwater, implying that younger trees within 1.5 km are more susceptible to excessive watering, which can lead to their death [69]. However, most P. euphratica forests are located beyond 3 km, and 5.20–5.31 m is the range of groundwater thresholds most suitable for P. euphratica growth [28]. Additionally, river system interruptions affect the distribution of P. euphratica forests [70]. As shown in

Table 4. Changes in river systems in the P. euphratica forest area of the TB, 1990–2020.

River	P. euphratica Forest Area	Changes in River Systems
Cele River	Decrease	Disconnection by 2010 [73]
Cherchen River	Decrease	Disconnection by 2000 [74]
Up-Tarim River	Decrease	The change is not obvious
Mid-Tarim River	Decrease	The change is not obvious
Down-Tarim River	Decrease	Disconnection by 2000 [67]
Kashgar River	Decrease	The change is not obvious
Keriya River	Increase	Decrease
Kongque River	Decrease	Decreased until 2000, increased thereafter [75]
the rivers on the northern slopes of the Kunlun Mountains	Increase	Increase [76]
Mid-Hotan River	Increase	The change is not obvious
Down-Hotan River	Increase	The change is not obvious
Up-Yarkand River	Decrease	The change is not obvious
Down-Yarkand River	Increase	The change is not obvious
Sangzhu River	Decrease	The change is not obvious

, rivers that experienced prolonged interruptions demonstrated a gradual reduction in the distribution range of P. euphratica forests. After river restoration, the runoff primarily recharged groundwater, leading to the gradual recovery of P. euphratica forests [71,72].

As shown in

Table 5. Overview of EWTP in P. euphratica forest in the TB, 1990–2020.

River	Year of Water Transfer	Total Water Transfer (billion m3/a)	Number of Water Transfer	Mode of Water Transfer	P. euphratica Forest Area
Kongque River	2016	20.88	6	River-Lake-Reservoir Water System Connections for Drainage [78]	decrease
Up-Tarim River	2016 [79]	90.15	24	Ecological channels and streams	decrease
Mid-Tarim River	2016 [79]			Ecological channels and streams	decrease
Down-Tarim River	2000 [80]			Artificial water channels	decrease
Up-Yarkand River	2016 [81]	13.46	7	Ecological and artificial water channels	decrease
Down-Yarkand River	2016 [81]			Ecological and artificial water channels	increase
Kashgar River	2022	8	1	Natural rivers and streams	decrease
Sangzhu River	/			Natural rivers and streams	decrease
Mid-Hotan River	2017 [82]			Artificial water channels	increase
Down-Hotan River	2017 [82]			Natural rivers and streams	increase
Cele River	2019			Natural rivers and streams	decrease
Keriya River	2016 [83]			Natural rivers and streams	increase
the rivers on the northern slopes of the Kunlun Mountains	2019			Natural rivers and streams	increase
Cherchen River	2019 [84]			Natural rivers and streams	decrease

, EWTP was primarily concentrated in the Tarim River, Konque River, Yarkand River, and Kashgar River, where artificial water channels have been constructed. Although other rivers have implemented ecological water transfer, they rely on river flood overflow. The results indicate that the upper and middle Tarim River, Yarkand River, Hotan River, Kashgar River, and Cherchen River had large overflow water surface areas, with water transfer frequencies mostly being medium to low. The remaining rivers had smaller overflow water surface areas and lower water transfer frequencies, with the initiation of water transfer occurring relatively later. Despite the implementation of EWTP, some areas still had a decreased P. euphratica forest area. One reason is the lack of human intervention during the ecological water transfer process, often resulting in a large water area at the terminal lake and a narrow impact range of floods on the desert riparian forests along the riverbanks [55]. Another reason is the overly singular high-frequency water transfer areas. Scientific ecological water supply should rely on a combination of natural and artificial interventions to support the completion of ecosystem protection in arid basins [77].

This study shows that within 1.5 km of rivers in water transfer zones, the area of P. euphratica forests converting to shrublands is greater than that of shrublands converting to P. euphratica forests. Beyond 4 km from rivers, the mutual conversion area between P. euphratica forests and shrublands gradually increases, with shrublands converting to P. euphratica forests far exceeding the reverse. This indicates that from 1990 to 2020, long-term water transfer within 1.5 km of rivers has promoted the restoration of shrublands. Therefore, it is recommended to switch to intermittent water transfer within 1.5 km of rivers. Studies have shown that P. euphratica forests and shrublands have similar ecological niches and cannot coexist in the same habitat. P. euphratica forests reduce their density to ensure individual growth [15]. Therefore, it is advisable to establish ecological protection zones for P. euphratica forests within 1.5 km of rivers. In these critical areas, water transfer measures should be complemented by seedling re-generation and artificial planting. The restoration of degraded forest areas can also be prioritized based on proximity to water sources. Areas close to water sources should first adopt artificial planting to ensure sufficient water supply for P. euphratica propagation. Using distance from rivers as a benchmark allows for better monitoring and evaluation of conservation efforts. This approach helps assess the effectiveness of water resource management and other conservation measures over time.

From 1990 to 2020, the restoration of P. euphratica forests and shrublands in the TB was mainly distributed in low-frequency water transfer zones, with very little mutual conversion between P. euphratica forests and shrublands in medium- to high-frequency water transfer zones. This finding suggests that while water transfer has a certain role in the restoration of P. euphratica forests, high-frequency water transfer in the same area does not increase the area of P. euphratica forests. This is consistent with the results of Xue et al. (2021) [3]. Therefore, the protection and restoration of P. euphratica forests within 1.5 km of rivers should adopt intermittent low-frequency water transfer; high-frequency water transportation requires the assistance of artificial planting. It is also recommended to prioritize the implementation of this measure in the distribution areas of P. euphratica forests in the lower reaches of the Tarim River, the Keriya River, and the Kongque River. Additionally, vegetation’s response to water has a lag effect [85], often requiring a prolonged natural succession process for the effective restoration of P. euphratica forests. Therefore, long-term continuous monitoring of the effects of water transfer is crucial for the implementation of P. euphratica forest restoration measures.

5.4. Suggestions for the Protection of P. euphratica Forests

Table 6. Statistical table of changes and driving factors of P. euphratica forests in the TB.

River	P. euphratica Forest Area	River System	Overflow Water Surface Area (km2)	Frequency of Water Transfer	Farmland Area
Cele River	Decrease	Disconnection by 2010	5.34	53.69% extremely low	Increase
Cherchen River	Decrease	Disconnection by 2000	166.8	20.91% high	Increase
Up-Tarim River	Decrease	Not obvious	325.68	40.21% extremely low, 26.31% low, 13.09% medium.	Increase
Mid-Tarim River	Decrease	Not obvious	238.86	45.76% extremely low, 39.93% medium, 14.31% high	Decrease
Down-Tarim River	Decrease	Disconnection by 2000	67.75	70% high1% low	Increase
Kashgar River	Decrease	Not obvious	94.32	74.88% low	Increase
Keriya River	Increase	Decrease	44.92	49.58% extremely low	No change
Kongque River	Decrease	Disconnection by 2000	21.61	92.67% low	Decrease
The rivers on the northern slopes of the Kunlun Mountains	Increase	Increase	18.91	79.93% low	Increase
Mid-Hotan River	Increase	Not obvious	114.98	90.39% low	Increase
Down-Hotan River	Increase	Not obvious	353.53	87.57% low	Increase
Up-Yarkand River	Decrease	Not obvious	198.68	42.95% extremely low, 31.94% medium	Decrease
Down-Yarkand River	Increase	Not obvious	150.83	89.65% low	Decrease
Sangzhu River	Decrease	Not obvious	0.73	No water transfer	Increase

summarizes the distribution of P. euphratica forests along 14 rivers in the TB and their influencing factors in this study. In 1990, the areas of P. euphratica forests in the Cele River, Cherchen River, the main course of the Tarim River, Kashgar River, Kongque River, and the upper reaches of the Yarkand River and Sangzhu River decreased, whereas the areas in other rivers increased. The following suggestions are proposed to address the changes in the distribution of P. euphratica forests:

In regions where the area of P. euphratica forests decreased due to the expansion of farmland, limited inundation areas from ecological water transfer, very low water transfer frequencies, and river system interruptions, measures, such as reforesting reclaimed farmland and increasing water transfer frequencies, should be taken. However, in some areas, the overly singular water transfer range prevented the effective restoration of P. euphratica forests. For instance, in the main course of the Tarim River, high-frequency water transfer often occurs within the same area within 3 km of the river. If this situation continues, it could lead to a decline in P. euphratica forests within 3 km. The water transfer range should be expanded, increasing the frequency of water transfer to P. euphratica forests beyond 3 km and reducing the frequency within 3 km. Additionally, the encroachment of farmland on P. euphratica forests in these areas should be controlled. In the lower reaches of the Tarim River, a significant reduction in the area of P. euphratica forests occurred due to river system interruptions even though this region has the highest water transfer frequency. During the periods of river system interruptions, natural succession may have led to the encroachment of shrublands on P. euphratica forests. During the recovery of the river system, shrublands were restored first, and P. euphratica forests continued to decrease. Therefore, measures, such as seedling regeneration or artificial planting, should be adopted to restore P. euphratica forests in this region.

Experimental studies on soil seed banks have shown that river flooding in these areas can trigger seed germination. The timing should ensure the use of river overflow for irrigation once or twice a year [86]. Additionally, maintaining a groundwater depth of 2–6 m within 1 km perpendicular to the river is necessary to ensure the normal growth of trees, shrubs, and herbaceous plants because this groundwater level supports the healthy growth of P. euphratica [87].

6. Conclusions

From 1990 to 2020, P. euphratica forests and shrublands decreased in the first 20 years and increased in the last 10 years. P. euphratica forests were primarily distributed within 1.5 km (27.53%) and beyond 3 km (53.60%) of rivers, while shrublands were mainly distributed 8–10 km (34.84%) from rivers. P. euphratica forests were converted to shrublands more frequently in water transfer areas within 1.5 km of rivers, whereas shrublands were converted to P. euphratica forests more frequently beyond 4 km. Additionally, from 1990 to 2020, the farmland area in the P. euphratica distribution areas of the TB nearly tripled, with 19.79% of P. euphratica forests and 10.61% of shrublands converted to farmland, mainly beyond 3 km from rivers. Therefore, to ensure the sustainability of P. euphratica forests, it is crucial to regulate water transfer and control agricultural land expansion. We recommend implementing intermittent low-frequency water transfer within 1.5 km of rivers. Additionally, agricultural land expansion should be strictly managed, particularly in areas beyond 3 km from rivers, to minimize the impact on P. euphratica forests. These measures will help maintain a balanced ecosystem and promote the long-term health and sustainability of P. euphratica forests.