Analysis of Spatial–Temporal Variations and Driving Factors of Typical Tail-Reach Wetlands in the Ili-Balkhash Basin, Central Asia

Abstract

The Ili River Delta (IRD) is the largest delta in the arid zone of Central Asia. Since the 1970s, the entire delta system has undergone a series of changes due to climate change and the impoundment of the Kapchagay Reservoir upstream of the delta, triggering an ecological crisis. Wetlands play a crucial ecological role in biodiversity conservation. Most studies have mainly focused on the response of vegetation and soil microbial to ecological changes in the delta, ignoring the dynamic processes of wetlands changes. Hence, such changes in the IRD and the underlying mechanisms need to be investigated in depth. In this study, wetlands in the IRD from 1975 to 2020 were extracted based on Landsat images using the object-oriented method; changes in the wetland area, wetland landscape pattern, NDVI, and NPP were analyzed; and the contributions of natural and human factors to wetland evolution were quantified. The results indicated the following: (1) From 1975 to 2020, the wetland area of the IRD showed an increasing trend, and changes in the wetland area were mainly found in the middle part of the delta near the Saryesik Peninsula. (2) The wetland landscape pattern in the IRD changed markedly from 1975 to 2020. The dominant patches of the wetland in the middle of the delta continued to expand; the patch aggregation index (AI) increased, and the landscape fragmentation index (LFI) decreased. (3) From 2000 to 2020, the average annual normalized difference vegetation index (NDVI) and net primary productivity (NPP) in the IRD increased, which is consistent with the change in wetland expansion. (4) Inflow to the delta from the Ili River and the water level of Balkhash Lake are significantly correlated with the wetland area, which are the dominant factors driving wetland evolution; and water evaporation from the Kapchagay Reservoir and irrigation water diversion on the left bank of the reservoir obviously intensified the process of lake water level decline and wetland degradation during 1970 to 1985. These results can provide scientific background for making informed ecological protection decisions in the IRD under the impacts of climate change and human activities.

1. Introduction

Wetlands play a vital role in maintaining ecological balance, reducing floods and droughts, conserving water sources, controlling soil and water loss, and bringing huge ecological, economic, and social benefits to human survival and development [1]. In addition, wetlands are also an important carbon reservoir and play a vital role in the global carbon cycle [2]. Particularly in the inland river basin in Central Asia, due to the impacts of global warming and unreasonable utilization of water resources, the wetland areas all over the world are decreasing, resulting in increasing challenges such as oasis degradation, salinization, and the decline of biological diversity [3]. Given the significance of ecological services provided by wetlands, it is imperative that they are sustainably managed.

The Ili-Balkhash Basin is a typical inland river basin in Central Asia [3]. The Ili River Delta (IRD), which is located in the tail-reach of the Ili River, is the largest natural delta in Central Asia. Previously, many studies focused on the analysis of the hydrological characteristics of Lakes Balkhash, such as the changes in water level, area, and water balance in the lake and delta [4,5,6]. From 1840 to the end of the 1960s, the delta ecosystem was relatively stable, and the water level of Lake Balkhash changed periodically. Since 1970, the water level of Lake Balkhash has continued to decrease [4,7], reaching a minimum water level in 1987 [5,8]. Deng [5] and Long [8] indicated that due to the construction of the Kapchagay Reservoir and irrigation water diversion in the middle and upper reaches of the main stream of the Ili River, the inflow of the Ili River into the delta in the 1970s was 24.28% lower than the average value from 1936 to 1969. Since the 1990s, with the slowdown of agricultural development, the amount of water entering the lake has increased, the water level of Balkhash Lake has gradually increased, reaching the historical highest water level of 343.01 m since its own observation data in 2005, and the lake area has been restored accordingly [6,9]. The delta is connected with Lake Balkhash, and the wetland has the ecological function of purifying water and degrading pollutants in inland rivers, playing the role of a natural manager and supporting the ecological environment of the lake area. It replenishes water for the lake during dry years and is an indicator of the ecological status of the whole lake system [5]. On the one hand, the construction of the Kapchagay Reservoir since 1970 has changed the annual distribution of runoff in the lower reaches of Ili River, and the floodplain in the delta cannot be supplied with water during flood, resulting in the degradation of vegetation and deterioration of ecological conditions in the delta. Kipshakbaev and Abdrasilov [10] indicated that the degradation of reed beds and meadow vegetation through the mid-1980s resulted in a reduction in pasture grounds by two thirds compared to 1970. On the other hand, a large amount of sediment is deposited in the delta to block the river channel, resulting in the reduction of water inflow into the lake. Since 1986, the impounding of the reservoir has finished, and with the arrival of the abundant water cycle in the Ili River Basin, the water resources in the delta have varied significantly and remain unstable. However, the monitoring of the IRD has been lacking for a long time, and most previous studies have focused on changes in vegetation species and structural changes in the communities in the delta region. Few studies have been conducted on wetland evolution processes, and there is a lack of quantitative assessments of the impacts of climate change and human activities on wetland ecosystems.

The continued degradation of wetlands and the decline or loss of ecosystem services have increased awareness of the importance of ecosystems [11]. With the advantages of macroscopic, rapid, and repeated observations, remote sensing has been widely applied to monitor the spatial and temporal evolution of large-scale wetlands [12]. Compared to MODIS data, radar images, and other data sources that have been used in wetland monitoring, Landsat series images have high temporal and spatial resolution, which is suitable for wetland information extraction and dynamic monitoring at small and medium regional scales [13,14]. The extraction of wetland information from remote sensing data has evolved to include manual visual interpretation, image-element-based extraction, and deep learning. The object-oriented classification method can not only make full use of the spectral characteristics of the features, but it can also fully combine the shape, color, texture, and spatial relationship of the target features, which can solve the feature confusion problem based on image element extraction to a greater extent and avoid problems such as model complexity [15]. For the analysis of wetland changes, in addition to the variation in wetland area, landscape indices based on the geometric characteristics of the landscape spatial structure (e.g., area, perimeter, and shape) have been widely used to describe the landscape patterns and dynamics of wetlands. The current landscape indices are selected from both the patch index type level and the landscape level [16]. Landscape indices, such as the number of patches (NP), patch density (PD), Shannon diversity index (SHDI), patch fractal dimension index (FRAC), and patch aggregation index (AI), can help reveal the characteristics of wetland landscape patterns and change rules, and have been widely used in wetland remote sensing research [17].

Therefore, the objectives of this study are extracting wetlands by using an object-oriented hierarchical method based on long time series Landsat images, analyzing the spatial–temporal variations in wetlands in the IRD from 1975 to 2020, and investigating the changes in ecological function based on normalized difference vegetation index (NDVI) and net primary productivity (NPP). Combining the hydrological data, LULC maps, and regional statistical data, the main driving factors and their impacts on wetland variation were quantified. The results provide a scientific implementation of wetland conservation and water resource management in the IRD.

2. Data and Methods

2.1. Study Area

The Ili River Delta (IRD) is located in southeastern Kazakhstan, southwest of Lake Balkhash, between 73°21′–79°30′E and 44°45′–46°44′N (Figure 1), covering an area of approximately 8000 km² [18]. It is the largest delta in the arid zone of Central Asia. The geomorphological unit is a sedimentary–alluvial plain in the lower reaches of the Ili River, and the overall topography is high in the southeast and low in the northwest [19,20]. Average annual precipitation in the study area is 192 mm [21], with an average annual temperature of 7 °C [22] and an annual evaporation of approximately 1000 mm [23]. The vegetation is mainly false reed fescue (Calamagrostis pseudophragmites) and reed (Phragmites communis) [24].

The Ili River flows into the delta and supplies the western part of Lake Balkhash. Lake Balkhash is the fourth longest lake in the world [4]. The lake body is narrow and long, with a peninsula within it: the Saryesik Peninsula. The middle part of the lake is separated by the Uzenalar waterway and is divided into the east and the west parts [6,8]. Lake Balkhash is mainly supplied by the Ili River, the Karatar River, the Lepse River, the Aksu River, and the Ayaguz River [25]. The Ili River accounts for 75% to 80% of the total water inflow, which is the main source of the water volume of Balkhash Lake [26].

2.2. Datasets

Table 1. Description of research data.

Dataset	Source	Source Scale	Spatial Resolution
MSS	USGS website https://earthexplorer.usgs.gov/ (accessed on 4 July 2022)	18-day	78 m
TM	USGS website https://earthexplorer.usgs.gov/ (accessed on 4 July 2022)	16-day	30 m
ETM+	USGS website https://earthexplorer.usgs.gov/ (accessed on 4 July 2022)	16-day	30 m
OLI	USGS website https://earthexplorer.usgs.gov/ (accessed on 4 July 2022)	16-day	30 m
MOD13A3	https://ladsweb.nascom.nasa.gov/search/ (accessed on 4 July 2022)	Monthly	1000 m
MOD17A2H	https://ladsweb.nascom.nasa.gov/search/ (accessed on 4 July 2022)	8-day	500 m
ERA5	https://www.ecmwf.int/en/research/climate-reanalysis (accessed on 4 July 2022)	Monthly	0.1° × 0.1°
Meteorological Data	CRU TS4.04 http://data.ceda.ac.uk/badc/cru/data/cru_ts/ (accessed on 4 July 2022)	Annually	0.5° × 0.5°
LULC	http://data.casearth.cn/ (accessed on 4 July 2022)	Annually	-
Hydrological data	http://www.ncdc.ac.cn (accessed on 4 July 2022)	Annually	-

lists the dataset we used in this study. Wetland information in the IRD were extracted based on Landsat MSS/TM/ETM+/OLI images. The MSS data from 1975 to 1980, the TM data from 1992 to 1998 and 2006 to 2011, the ETM+ data from 1999 to 2005 and 2012, and the OLI data from 2013 to 2020 were selected for interpreting wetland boundaries. Remote sensing image data were mainly downloaded from the United States Geological Survey (USGS) website. The study mainly selects Landsat MSS images (32 images), Landsat TM images (62 images), Landsat ETM+ images (39 images), and Landsat OLI images (40 images) as the main data sources. To improve the accuracy of the temporal and spatial change sequence of wetlands, the images with high quality and cloud coverage less than 10% were selected, mainly from August to October each year. As the downloaded images are raw L1 level products, all images need to be preprocessed one by one before they are interpreted. The preprocessing of remote sensing images mainly includes radiometric calibration, atmospheric correction, alignment, mosaic, cropping, and other operations. The Quick Atmospheric Correction (QUAC) and Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) atmospheric correction modules developed by ENVI software were used to radiometric calibration and atmospheric correction of the images; in addition, due to the failure of the Landsat ETM+ image sensor after 2003, the stripes were lost in the remote sensing images, so the stripes were repaired using the Landsat gap-fill extension tool in the ENVI App Store [27]. The images were then mosaicked and cropped to form wetland image time series data. The MOD13A3 (NDVI) product and MOD17A2H (NPP) product were selected for the analysis of wetland vegetation. As MODIS data are downloaded in HDF format, they need to be converted before they can be used on the data. The MODIS Reprojection Tool, which is a processing tool for MODIS data, can be used to convert data in HDF format to TIF format [28].

Annual precipitation and average temperature were collected from CRU TS4.04, which was published by the Climatic Research Unit (CRU) of East Anglia University. The download period was 1975–2020, and the spatial resolution was 0.5° × 0.5°.

Annual average evapotranspiration data were collected from the ERA5. The download period was 1975–2020, and the spatial resolution was 0.1° × 0.1°.

The land use/land cover change (LULC) maps in 1990, 2000, 2010, and 2015 were collected from the Chinese Academy of Sciences Earth (CASEarth), and were used to analyze land use and cover change in the Ili-Balkhash Basin.

The discharge observation data at hydrological stations along the Ili River were collected from the National Cryosphere Desert Data Center and the studies of Kezer and Matsuyama [7], Long et al. [8], Dostay et al. [29], Panyushkina et al. [30], and Duan et al. [31].

2.3. Methodology

2.3.1. Methods of Wetland Mapping and Accuracy Evaluation

Due to the separability of wetlands in remote sensing images, an object-oriented method of wetland information from remote sensing images was adopted [32]. The method first transforms image elements into primitives using an object-oriented segmentation technique; then clustering is based on the threshold-based normalized difference vegetation index (NDVI), and normalized difference water index (NDWI) dynamic threshold segmentations for wetland vegetation and wetland water bodies were automatically extracted separately. The threshold range set for extracting water bodies is NDWI ≥ 0.1 and NDVI ≤ −0.02, and the threshold range set for extracting vegetation is NDVI ≥ 0.3 and NDWI ≤ −0.02 [33,34,35].

$$\mathrm{NDVI}=\frac{{\rho }_{NIR}-{\rho }_R}{{\rho }_{NIR}+{\rho }_R}$$

(1)

$$\mathrm{NDWI}=\frac{{\rho }_G-{\rho }_{NIR}}{{\rho }_G+{\rho }_{NIR}}$$

(2)

where ${\rho }_{NIR}$ is the near-infrared band, ${\rho }_R$ is the red band, and ${\rho }_G$ is the green band.

Wetlands are shown in remote sensing images as dark tones, contrasting with the brighter tones of the surrounding desert, while cultivated land is shown as more regular rectangular plots that are easily interpreted. The extracted wetland information was exported and then manually visually interpreted to correct the automatically extracted wetland vegetation and wetland water body boundaries. The corrected results were checked for time series consistency with the help of the software of ArcGIS 10.2.2 (Environmental Systems Research Institute Inc., Redlands, CA, USA). Finally, the accuracy of the wetland extraction results was verified based on high-resolution images and land use data.

Sample points for the accurate evaluation of the wetland included 56 wetland samples and 144 non-wetland samples selected based on high-resolution images, and 413 wetland samples and 387 non-wetland samples generated by LULC maps. The evaluation indices include user accuracy (UA), producer accuracy (PA), overall accuracy (OA), and kappa coefficient [36]. This method was used to evaluate the accuracy of wetland extraction. The results showed that there were 407 correct points, 40 wrong points, 62 missing points, and 491 correct points in the non-wetland classification. The final calculation shows that the overall accuracy (OA) was 89.80%, the producer accuracy (PA) was 86.78%, the omission error (OE) was 13.22%, the user accuracy (UA) was 93.56%, and the kappa coefficient was 0.81, indicating that combined with visual correction, the wetland extraction results have high accuracy and can meet the research requirements (

Table 2. Accuracy assessment index.

Indicators	Accuracy
Overall accuracy	89.80%
Producer accuracy	86.78%
Omission error	13.22%
User accuracy	93.56%

).

2.3.2. Analysis of Spatiotemporal Variation in Wetlands

The linear trend method was used to describe the changes in wetland area, landscape index, precipitation, temperature, and other time series in the study area [37]. The Mann–Kendall test was applied to detect the mutation points of hydrological and meteorological time series, and the details of the functions can be found in Ma and Huang [38]. Moreover, landscape indices are commonly used to reflect changes in wetland landscapes. Using FRAGSTATS software, the class area (CA), number of patches (NP), patch density (PD), largest patch index (LPI), patch fractal dimension index (FRAC), and patch aggregation index (AI) were calculated. The FRAC can reflect the shape complexity of different spatial scales, and the ratio of NP and CA can reflect the landscape fragmentation index (LFI) [39].

The centroid analysis method was used to analyze the distribution pattern and changes in wetlands in different categories by describing the change characteristics of the centroid distribution and direction of geographical elements [40]; a detailed description of the method can be found in Liu et al. [41].

2.3.3. Analysis of Driving Factors of Wetland Changes

Changes in wetland area, NDVI, and NPP in the IRD are influenced by variable indicators. The Pearson correlation coefficient was used to test the correlation degree between the factors. The larger the Pearson correlation coefficient is, the higher the correlation degree [42].

$$\mathrm{r}=\frac{{{\displaystyle \sum }}_{i=1}^n\left(x_i-\overline{x}\right)\left(y_i-\overline{y}\right)}{\sqrt{{{\displaystyle \sum }}_{i=1}^n{\left(x_i-\overline{x}\right)}^2{{\displaystyle \sum }}_{i=1}^n{\left(y_i-\overline{y}\right)}^2}} \left(i=1\cdots N\right)$$

(3)

where $r$ is the correlation coefficient of the observation; $n$ is the number of samples; $x_i$ and $y_i$ are the sample values of the random variable $x$ and the random variable $y$, respectively.

3. Results

3.1. Temporal and Spatial Variation in Wetlands in the IRD

3.1.1. Wetland Area Change

Figure 2a shows the change in the wetland area from 1975 to 2020. The wetland area in the IRD showed an increasing trend, with a change rate of 12.56 km²/10a. Wetland vegetation accounts for a major part of the total wetland area, and the trend of wetland vegetation is consistent with the total wetland area, with a more pronounced decrease from 1979 to 1992 and an increase after 1992. Wetland water bodies remained consistent with the change in total wetland area until 1979, and after 1979 the area of wetland water bodies did not change much.

Long time series changes in the area of the western lake of Lake Balkhash are shown in Figure 2b. From 1975 to 2020, the area of the western lake of Lake Balkhash decreased from 12,724.07 km² to 12,502.04 km² with an average rate of change of −4.83 km²-a⁻¹. The maximum value of the area occurs in 1975 (12,724.07 km²), and the minimum occurs in 1992 (11,922.30 km²). Between 1975 and 1995, the area of the western lake of Lake Balkhash shrank significantly at a rate of −39.68 km²-a⁻¹. After the 21st century, the area of the lake began to recover gradually from 2000 to 2020, increasing from 11,999.66 km² to 12,502.04 km², a rate of increase of approximately 23.92 km²-a⁻¹.

3.1.2. Spatial Variation in Wetlands

The frequency map of wetland distribution in the IRD from 1975 to 2020 is shown in Figure 3a. The red areas with high coverage frequency represent the earliest wetlands in the study area, and the green areas with low coverage frequency represent the new wetlands. The maximum frequency of coverage was 33, and the color change from red to green shows the spatial expansion process of wetlands. The earliest wetlands in the study area are mainly distributed in the front and middle of the delta and along the bank of Lake Balkhash.

From 1975 to 2020, the wetland area of the IRD experienced a process of increase, decrease, and re-expansion, and the whole change process showed obvious fluctuations. As shown in Figure 3b, from 1975 to 1980, the wetland area increased, and the increase area was concentrated in the middle of the delta, with an increase of 1230.68 km² and an increase in range of 32.69%; from 1980 to 1992, the wetland area was reduced, and the shrunk areas were concentrated in the middle of the delta and Saryesik Peninsula, with an area of 2292.55 km², with a reduction in range of −45.90%; from 1992 to 2002, the wetland area again increased, and the expansion area was found in the middle of the delta and Saryesik Peninsula, with an increase of 3386.82 km² and an increase in range of 125.35%; from 2002 to 2011, the wetland area increased, the change was mainly found in the middle of the delta, near the periphery of the lake and Saryesik Peninsula, showing an outward expansion, with an increase of 938.74 km²; from 2011 to 2020, the wetland area maintained a slow increase, and the increased area was still centered in the central region of the delta, expanding outward, with an increase of 133.05 km², and an increase rate of 1.84%. The increase was the most significant from 1992 to 2002, with a rate of 125.35%; the decrease in wetland area was most significant from 1980 to 1992, with a rate of −45.90%.

3.1.3. Change in Wetland Centroid

According to the change in the centroid of the IRD wetland (Figure 4), the centroid of the IRD wetland moves from south to north, and the centroids of the wetlands in the six periods are located within the range of the IRD wetland.

The movement of wetland cores represents a spatial change in wetland patterns, reflecting the direction of core migration and the evolution of landscape patterns. According to the moving direction of the wetland centroid, the moving track of the wetland centroid can be roughly divided into two stages. The first stage was from 1975 to 1980, and the moving direction of the wetland centroid was moving towards the southeast; the second stage was from 1980 to 2020, and the wetland centroid gradually moved to the northeast. As the new wetlands migrate in the direction of development, the recovery of the delta vegetation tends to coincide with the direction of new wetland development and spreads along both sides of the river.

3.1.4. Change in Wetland Landscape Pattern

Landscape indices of the wetland in the IRD are presented in Figure 5. From 1975 to 2020, the NP in the study area decreased from 5002 in 1975 to 4507 in 2020, the PD decreased from 1.46 to 0.64, the LFI decreased from 0.015 to 0.006, the LPI increased from 50.94% to 81.36%, the FRAC increased from 1.044 to 1.06, and the AI increased from 95.5% to 97.50%, indicating that patch connectivity in the wetland landscape increased, the number of dominant patches of wetlands in the IRD increased, and the fragmentation of wetlands decreased.

3.2. Changes of Ecological Function in Wetland Vegetation

3.2.1. Variation in NDVI

From 2000 to 2020, the NDVI in the IRD showed an increasing trend. The trend of the maximum and mean NDVI variation of wetland was consistent. As shown in Figure 6a,b, trends of the mean annual NDVI and the wetland area are basically consistent. However, from 2011 to 2014, the maximum value of NDVI was opposite to the change trend of wetland area; the wetland area increased and the NDVI value decreased. The reason is that the wetland area in this study includes wetland water bodies and wetland vegetation. When the water body expands and the water level increases, the wetland vegetation is submerged, and the low value of water NDVI leads to a decrease in the average NDVI in the study area.

As shown in Figure 6c,d, the annual mean NDVI increased with the increase in elevation DEM in the study area. In the low-altitude area, that is, the area close to the lake, the water surface fluctuates greatly, the change in water area limits the growth of vegetation, and the wetland is staggered with the water body. With the increase in distance from the lake and elevation, vegetation growth was mainly affected by surface and subsurface hydrological conditions, increasing the NDVI.

The spatial variation in NDVI for different periods is shown in Figure 6e. From 2000 to 2020, the NDVI in the IRD wetlands showed an increasing trend. Spatially, the best coverage and most pronounced changes in vegetation were found in the middle of the delta, near the perimeter of the lake body and around the Saryesik Peninsula.

3.2.2. Variation in the NPP in the IRD

The value of NPP can directly reflect production capacity of plant communities under natural environmental conditions [43,44]. As shown in Figure 7a, from 2000 to 2020, the NPP of wetlands showed an increasing trend; the maximum value occurred in 2016, which was one of the years with a large wetland area, and the minimum value occurred in 2000, which was the year with the lowest wetland area.

The spatial–temporal variation in the annual NPP in the IRD from 2000 to 2020 is shown in Figure 7b. Similar to the change in the NDVI, the NPP in the study area has generally shown an increasing trend since 2000. The most obvious increments were found in the middle of the delta, near the periphery of the lake and around the Saryesik Peninsula. In 2016, the wetland area also increased, the vegetation coverage was enhanced, the NPP reached the maximum, and the spatial expansion of the wetland was the most obvious.

3.3. Analysis of the Impacts of the Main Driving Factors on Wetland Evolution within the IRD

3.3.1. Changes of Main Climate Factors

From 1975 to 2020, the annual average precipitation in the IRD was 215.01 mm, and the fluctuation in precipitation did not show an obvious trend (Figure 8a). Through the M–K test, it was found that the annual precipitation in the lake area changed suddenly in 1995. From 1975 to 1995, the annual precipitation in the IRD had a decrease range of 8.42 mm/10a. From 1995 to 2020, the annual precipitation in the IRD had an increase range of 1.58 mm/10a. The wetland area in the study area also increased, with a negative acceleration from 2005 to 2015, and continued to increase after 2015 until reaching the peak in 2020. The annual precipitation of the region was consistent with the change in wetland area. From 1975 to 2020, the average temperature in the IRD was 8.53 °C, and showed a significant increasing trend (Figure 8b). The change point of the annual average temperature was found in 1997 through the M–K test. From 1975 to 1997, the average annual temperature in the delta region decreased by −0.063 °C/10a, while from 1997 to 2020, the average annual temperature in the delta region increased by 0.054 °C/10a.

From 1975 to 2020, the average annual evapotranspiration of the IRD was 520.76 mm/a. The evapotranspiration in the lake area showed a clear increasing trend (Figure 8c). The average annual evapotranspiration from 1975 to 2020 was trended to obtain a curvature change in average annual evapotranspiration for 46 years (Figure 8d). It is evident that the areas where the change in the curvature of actual evaporation in the delta increases significantly are concentrated on the lakes, and the changes in the curvature of actual evaporation in the central part of the delta decrease.

3.3.2. River Runoff and Lake Hydrology

As shown in Figure 9a, both the annual discharges from the Uskerma station and wetland area in the IRD present an increasing trend from 1975 to 2020. The wetland area increased by 3581.94 km² and the runoff volume at the Uskerma station increased by 10.71 km³. Comparing the change curves of the delta wetland area with the water level of Lake Balkhash and the area of Lake Balkhash West Lake (Figure 9b), it can be seen that from 1975 to 1990, the decrease in the water level of Lake Balkhash caused the area of Lake Balkhash West Lake to decrease, affecting 1990 to 2020. The rise in the water level of Lake Balkhash increased the area of the western lake of Lake Balkhash and affected the increase in the area of wetlands in the delta.

In the past 46 years, the water inflow of the Ili River has increased each year, which not only provides necessary water conditions for the expansion of delta wetlands and the development of regional vegetation but also promotes the rise of the water level of Balkhash Lake and the further development of wetlands around the lake.

3.3.3. Correlation Coefficients among Multiple Factors

Pearson’s correlation coefficients were used in this study to quantify the potential drivers of wetland variation (Figure 10). Five environmental factors, i.e., the annual average temperature of the IRD, the annual precipitation in the IRD, the annual average evapotranspiration in the IRD, annual water level of the Lake Balkhash, Ili River runoff into the delta; and time series for wetland characteristic, i.e., total area of wetland, the area of wetland vegetation, area of wetland water bodies, and the average annual NPP and NDVI of wetland which were derived from Landsat/MODIS images, were used to calculate the correlation coefficient between each two time series and investigate the impact of environment driving factors on wetlands. Although the wetland was extracted based on the Landsat MSS, TM, ETM+, and OLI images, with a spatial resolution of 78 m, 30 m, 30 m, and 30 m, respectively, in the correlation analysis between wetland change and impact factors, time series of the total area of wetland, the area of wetland vegetation, and the area of wetland water bodies which were derived from the Landsat images, and annual total NPP and average NDVI of wetland derived from MODIS data, were used to represent overall change process of the wetland in delta region. The results showed that from 1975 to 2020, the wetland area illustrated a positive correlation of 99% statistical significance with the average temperature of the IRD and no obvious correlation (insignificant) with the total precipitation of the IRD, and the wetland area illustrated a strong positive correlation of 95% statistical significance with the water level of Lake Balkhash and river inflow from the Ili River. From 2000 to 2020, the NPP of the wetland illustrated a positive correlation with precipitation and illustrated a positive correlation of 99% statistical significance with the water level of Lake Balkhash. For the evapotranspiration factors, the wetland area and Ili River runoff showed a positive correlation of 99% statistical significance with the ET of the IRD.

4. Discussion

The driving factors of wetland change in inland river basins can be divided into two types: natural factors based on climate warming, precipitation, and snow/glacier changes, and human activity factors based on agricultural development, industrial and urbanization development, and water management policies [36]. The IRD is located in the lower reaches of the basin and is connected with Lake Balkhash, which is a natural and seminatural mixed area that is relatively weakly disturbed by human activities [18]. Wang et al. [45] indicated that the water level of Lake Balkhash has a significant positive correlation with the inflow (through the 95% confidence test) at the inlet of the delta. The Ili River originates from the north slope of Khan Tengri Peak in Kazakhstan west of Tianshan Mountain, which was the main water source of Balkhash Lake. The discharge of the Ili River is mainly affected by water produced upstream of the basin and water consumption in the middle and lower streams of the basin, which presented the joint impacts of natural climate changes and human activities. The water-producing areas of the Ili River are mainly distributed in China, while most of the water consumption area is located in Kazakhstan. The utilization degree of water resources in upstream China was low, and it was mainly affected by climate factors [21]. According to the research of Yuan and Wei [46], the temperature and the rainfall in the Tianshan Mountain area showed an upwards trend from 1957 to 2009. Duan et al. [47] indicated that the amount of water flowing into Kazakhstan from China was increasing significantly, and the average runoff outflow from 1998 to 2013 increased by 26.50% compared with 1931 to 2007; previous studies showed that the river runoff from the upper reaches of the Ili River was increasing, but the inflow from the Ili River to Balkhash Lake was decreasing in the recent 20 years, which indicated that the Ili River was excessively consumed in Kazakhstan [48,49]. Based on CRU data, time series of precipitation and temperature in the mountainous areas in the upper reaches of Ili River Basin were extracted. As shown in Figure 11, precipitation and temperature in the mountainous areas increased from 1961 to 2020; the runoff of hydrological stations in the upper reaches of the main stream of the Ili River in Kazakhstan also showed a slight increasing trend.

On the other hand, the construction of the Kapchagay Reservoir and agricultural development around the reservoir are the main factors affecting the runoff fluctuation into the delta. From 1970 to 1985, due to the construction of Kapchagay Reservoir, the flow into the delta decreased by nearly 3.5 km³ compared with that in the 1960s, which was the main reason for the rapid decline in the water level in Balkhash Lake [8]; it is also the main reason for wetland area shrinkage and vegetation degradation in the IRD. It is worth noting that this period is just in the dry season caused by long-term climate fluctuations. From 1986 to 1995, the discharges from the upper reaches of the basin were still at a low level, and the water consumption in the middle reaches of the main stream decreased 1.53 km³; from 1996 to 2020, with the significant increase in water from the upstream, although the water diversion from the middle reach also increased slightly, the overall flow into the delta was higher than that from 1970 to 1995 [45]. With the increase in water inflow in the delta, the wetland area gradually expanded, and the vegetation growth improved.

Before the construction of the Kapchagay Reservoir (before 1970), the impact of human activities was slight, and the flow into the delta was mainly dominated by climate change. From 1970 to 1985, the water inflow from the upper reaches entered a reduction period with climate fluctuations; the establishment of Kapchagay Reservoir has developed 4500 km² of cultivated land, resulting in an increase of 2.4 km³ of water consumption in the middle and upper reaches of the main stream of the Ili River, which is the main reason for the sharp decrease in the inflow into the delta, and the contribution rate was 47.47% (

Table 3. Contribution rate of climate change and human activities to inflow variation in the IRD.

		Factor	Average Temperature	Average Precipitation	Water Consumption in the Middle Reaches
	Contribution Rate (%)
Time
1961–1969			46.38	53.06	0.55
1970–1985			44.93	7.60	47.47
1986–2020			4.91	64.89	30.18
1961–2020			5.00	66.67	28.33

). In the same period, the water level of Lake Balkhash decreased from 343.0 m to 340.6 m, with a decline rate of 0.14 m/a, which is higher than the natural decline rate of 0.09 m/a recorded since the 19th century, indicating that human activities significantly exacerbated the process of lake water level decline during this period and impacted the wetland ecosystem in the IRD. Farmland in the Ili River Basin is mainly distributed between the hydrological stations of Kapchagay up 171 and Uskerma, accounting for 32.62% of the total farmland in the basin; from the early 1990s to 2015, the overall farmland area in the basin increased by 5774.13 km² (Figure 12). After the disintegration of the Soviet Union in the early 1990s, the cultivated land area in Kazakhstan first decreased by 2610 km² due to abandonment, with the reduction concentrated in the middle reaches of the mainstream of Ili River. Since 2000, the farmland area has significantly increased to 21,100 km², with the increase concentrated around the Almaty irrigation area which is located on the left side of the Kapchagay Reservoir [50,51]. The evaporation water consumption and irrigation water diversion of Kapchagay still maintain a large amount of water, and the impact of water consumption in the middle and upper reaches on the water inflow of the Ili River is weaker than that from 1970 to 1985, with a contribution rate of approximately 30%. Over a long period, climate change at the basin scale dominates the periodic runoff change of the Ili River into the delta, with a contribution rate of 71.67%, in which the contribution rate of precipitation in mountainous areas is higher than that of the warming effect.

5. Conclusions

This study analyzes the temporal and spatial changes in the wetland area of the delta from 1975 to 2020 based on 46 years of remote sensing interpretation data of the IRD. The main conclusions are as follows: (1) From 1975 to 2020, the wetland area of the IRD showed an increasing trend, with an average annual increase of approximately 111.94 km². The main areas of wetland area change in the IRD are found in the central part of the delta and near the Saryesik Peninsula. (2) For the landscape pattern of wetlands in the IRD, the dominant patches of wetlands in the middle of the delta continued to expand. (3) From 2000 to 2020, NDVI and NPP were consistent with changes in wetland area, and the natural vegetation growth in the delta continued to improve. (4) Before 1970, runoff into the delta and lake were mainly related to the amount of water coming from the upper Ili River, which was influenced by climate change. From 1970 to 1985, the construction of the Kapchagay Reservoir and irrigation water diversion on the left bank of the reservoir markedly intensified the process of water level decline of Lake Balkhash and wetland degradation; the effect of human activity was greater than the indirect effects of climate change in this period. From 1987, the Ili River entered the period of abundant water, and the streamflow from the upstream was significantly increased, the wetland area showed a continuous increase, and climate change at the basin scale should be the dominant factor. Since about 2015, with the periodic fluctuation of the climate in the basin, the streamflow of the Ili River may enter the process of downward reduction again. With the continuous reduction of the upstream water, the wetland in the delta may deteriorate again with the reduction of the river inflow.

It should be noted that due to the limitations of the spatial and temporal observation scale of the early remote sensing images, there were inevitably some biases in the wetland area statistics and the detailed process of change in the process of wetland remote sensing interpretation and analysis. However, this does not affect the overall trend analysis of the delta wetlands. These results can provide scientific background for making informed ecological protection decisions in the IRD under the impacts of climate change and human activities.