Forecasting Land Use Dynamics in Talas District, Kazakhstan, Using Landsat Data and the Google Earth Engine (GEE) Platform

Abstract

This study employs the robust capabilities of Google Earth Engine (GEE) to analyze and forecast land cover and land use changes in the Talas District, situated within the Zhambyl region of Kazakhstan, for a period spanning from 2000 to 2030. The methodology involves thorough image selection, data filtering, and classification using a Random Forest algorithm based on Landsat imagery. This study identifies significant shifts in land cover classes such as herbaceous wetlands, bare vegetation, shrublands, solonchak, water bodies, and grasslands. A detailed accuracy assessment validates the classification model. The forecast for 2030 reveals dynamic trends, including the decline of herbaceous wetlands, a reversal in bare vegetation, and concerns over water bodies. The 2030 forecast shows dynamic trends, including a projected 334.023 km² of herbaceous wetlands, 2271.41 km² of bare vegetation, and a notable reduction in water bodies to 24.0129 km². In quantifying overall trends, this study observes a decline in herbaceous wetlands, bare vegetation, and approximately 67% fewer water bodies from 2000 to 2030, alongside a rise in grassland areas, highlighting dynamic land cover changes. This research underscores the need for continuous monitoring and research to guide sustainable land use planning and conservation in the Talas District and similar areas.

1. Introduction

Investigating changes in land cover and land use is pivotal for understanding environmental dynamics and the effects of human activities on natural landscapes. This research is particularly critical in regions experiencing rapid ecological and anthropogenic transformations, such as the Talas District in Kazakhstan. Situated within the Zhambyl region, the Talas District has been under significant environmental stress, attributed to both natural processes and human interventions. The investigation of land cover and land use changes in this region is not only crucial for documenting ecological shifts but also for implementing strategies aimed at achieving sustainable land management and maintaining ecological balance. This aligns with the global imperative of preserving environmental sustainability in the face of growing developmental pressures.

The primary objective of this study is to conduct a comprehensive analysis of the patterns, causes, and consequences of land cover and land use modifications in the Talas District over a span of two decades, from 2000 to 2020. By leveraging the capabilities of GEE, this research endeavors to offer a detailed examination of these transformations, thereby contributing valuable insights to the field of environmental science and aiding in the enhancement of land management practices. The significance of this study is underscored by its intention to provide a thorough understanding of the dynamics at play, which is pivotal for the formulation of policies that promote environmental stewardship and sustainable development.

Driven by a series of objectives, this study aims not only to document the trends in land cover and land use changes within the Talas District, but also to identify the driving forces behind these shifts. Furthermore, it seeks to assess the implications of these changes on the environmental integrity and the well-being of local communities. By doing so, it aims to contribute substantially to the development of informed and effective land management policies. This research builds upon the foundational work of esteemed researchers such as Kumar, Xiao, Zhao, Tamiminia, de Beurs, Alipbeki, Qi, Yuan, Hu, and Kou, who have extensively utilized GEE and Landsat imagery in their investigations of various land cover dynamics [1,2,3,4]. The integration of these methodologies underscores the multidisciplinary nature of this study, positioning it as a crucial endeavor in the ongoing efforts to understand and mitigate the impacts of environmental change.

1.1. Literature Review

Previous research [5,6,7,8,9,10,11,12,13] has extensively utilized Google Earth Engine (GEE) for environmental studies, highlighting a deficiency in the detailed analysis of land cover alterations in specific locales such as Talas. A study, referenced as [14], explored the repercussions of institutional changes post-Soviet Union disintegration on land cover/use and land surface phenology in Kazakhstan. This was achieved by analyzing Normalized difference vegetation index (NDVI) time series, temperature datasets, and statistical methods, identifying agricultural practice changes as the primary influencer of NDVI fluctuations in irrigated zones, whereas, in northern regions, alterations in land use, such as increased fallow land and reduced grazing, had a notable impact on land surface phenology.

Another investigation [15] delved into the spatiotemporal dynamics of land use and cover in the peri-urban Arshaly district, adjacent to Kazakhstan’s capital, employing Landsat multispectral imagery and supervised classification over a twenty-year period from 1998 to 2018. This study uncovered significant changes in land use and cover, including an expansion of arable land and forests, a reduction in pastures, and a pronounced increase in built-up areas, indicating an intensification of land use and urban expansion, albeit with a minimal emphasis on sustainable development.

Further analysis [16] was conducted on land use and cover and net primary productivity (NPP) alterations in the Ili-Balkhash Basin of Central Asia, a region undergoing substantial socioeconomic shifts. Using remote sensing data from 1995 to 2015, the study observed significant land use and cover changes, especially on the Chinese side with an increase in crop production and a decrease in grazing areas, alongside fluctuating NPP patterns, with declines in mountainous and upland regions and increases in irrigated, vegetated areas.

A comprehensive assessment [17] focused on land use changes in the Shortandy district, Kazakhstan, examining the dynamics of land use and sustainable development indicators in a peri-urban area from 1992 to 2018. The findings indicated an intensification of land use with increased arable lands and suggested that economic development was accompanied by advancements in social and environmental development, proposing their methodology for broad application in assessing sustainable development in specific areas, particularly around Kazakhstan’s capital.

One study [18] provided an in-depth analysis of land cover change in Kazakhstan and Mongolia, the two largest landlocked nations with similar biophysical conditions and historical Soviet ties, over three decades (1990–2020). Utilizing 6964 Landsat images and pixel-based classification, the study documented significant grassland-to-cropland conversions, influenced by historical agricultural policies, with Kazakhstan experiencing more extensive land cover changes than Mongolia.

Another study [19] investigated land changes and their driving mechanisms in Central Asia, an ecologically sensitive region, using Landsat imagery and the random forest algorithm via GEE to generate annual land cover datasets from 2001 to 2017. The study revealed significant changes in land cover types, notably the decrease in bare land and the increase in natural vegetation, cultivated land, and urban areas, identifying precipitation, drought, and human activities as key driving factors, while discussing the advantages and complexities of their land mapping methodology.

A longitudinal spatiotemporal analysis [20] of land use and cover in the Burabay district of Kazakhstan assessed sustainable development trends and identified the driving mechanisms using GEE-integrated data. This research, involving Landsat imagery, Random Forest, Multiple Linear Regression, and Principal Component Analysis, noted significant shifts in pasture, arable land, and forest areas, with economic factors as the primary driving mechanisms, and observed a recent decline in development growth due to COVID-19, offering insights for future sustainable development and land use and cover policy evaluation.

Lastly, another study [21] aimed at simulating and predicting future land use and cover changes in the Ili-Balkhash Basin, an arid and ecologically fragile region, using the patch-generating land use simulation model and a combination of this model with Markov. This research evaluated the stability of future land use and cover patterns and the transformation of grasslands into croplands and barren areas, providing a novel approach for land use and cover simulation in data-scarce basins, aiding in the understanding of human activities’ impact on basin hydrology and land management planning.

To conclude the literature review section, it is necessary to emphasize that, despite the significant contribution of previous studies in the field of land cover change and land use studies using the GEE platform, there are certain shortcomings in these works. Firstly, most studies did not focus on specific regions, such as the Talas district of Kazakhstan, which leaves a significant gap in the knowledge about the specifics and dynamics of these areas. Second, previous works were often limited to historical analyses without developing predictive models of future changes, which reduces their practical relevance for land use planning and environmental protection. A third shortcoming relates to the lack of attention to the sustainable development aspects of land use change, which is critical for environmental stability and socio-economic well-being. Finally, many studies note the lack of a comprehensive approach to integrating different types of data and methods of analysis, which could contribute to a better understanding of the complex processes of land cover and land use change.

Thus, given the aforementioned shortcomings, our study aims to fill these gaps by focusing on the Talas region, developing predictive models to estimate future land cover and land use change, and applying an integrated approach to analysis, including advanced remote sensing techniques and computational algorithms. This will not only enhance our understanding of land use dynamics in this region, but also offer recommendations for effective land planning and management, contributing to the achievement of sustainable development goals.

1.2. Objectives

In this study, we aim to perform the comprehensive task of analyzing and forecasting the dynamics of land use and coverage in the Talas region of Kazakhstan, using Landsat data and the GEE platform. Based on the preceding literature review and identified knowledge gaps, our efforts focus on the following key aspects:

• Development of predictive models: Formulate and test models that predict changes in land cover and use for 2030, taking into account anthropogenic and natural factors. This involves integrating multiple data and analytical approaches to improve the accuracy and reliability of predictions.
• Change Impact Analysis: Assess the potential environmental, social, and economic impacts of the projected changes to support sustainable land management in the Talas district. This includes examining the impacts of change on biodiversity, water resources, and agricultural production.
• Methodological contribution: To propose improved methods for analysis and forecasting by combining Landsat satellite data with machine learning algorithms within GEE. This involves the development of new approaches to land cover classification and land use adapted to the conditions of arid regions such as the Talas district.
• Practical application of the results: Develop recommendations for local authorities and stakeholders to adapt to the anticipated changes and minimize their negative impacts. This includes suggestions for improving land use practices, water use, and ecosystem conservation.

Our objectives aim to deepen our understanding of land cover and land use dynamics in the Talas district, with a particular focus on developing robust forecasting tools to support sustainable development and natural resource management. The results of the study are expected to have a significant impact on the scientific community and land management practices, providing new data and methodologies for informed decision-making.

2. Materials and Methods

2.1. Study Area

This research was conducted in the Talas district (Figure 1a), located in the Zhambyl southern region (Figure 1b) of Kazakhstan (Figure 1c). Located in the southwestern part of Zhambyl, the district encompasses diverse geographical features. The northern region comprises the Moyinkum sandy desert, while the central area includes the Talas River valley, and the southern part features a gently undulating plain that extends to the Karatau mountain system in the southwest. The district’s territory, initially rectangular and extending from north to south up to the village of Akkol, undergoes a narrowing at that latitude, followed by a southwestward expansion, ultimately concluding along the central line of the Karatau ridge.

In geomorphological terms, the territory of the Talas region is as follows: the southern smaller part is mountainous, the middle mountains of Karatau and the northern larger part is flat, and the accumulative and denudation plains of the Shu-Sarysu depression are located on the Turan plate [22]. The surface waters of the study area are represented by small rivers in the mountainous part and the Talas River in the plain. Considering the climatic conditions of the Talas district in the Zhambyl region, an analysis of published sources indicates the prevalence of a continental climate. This climate is characterized by substantial fluctuations in both annual and average daily surface air temperatures, coupled with a notable degree of aridity [23].

The continuous monitoring of agricultural lands becomes imperative for sustainable land use practices. Our study focuses on the Talas district in the Zhambyl region, situated within Kazakhstan’s arid zone. This district serves as a system-forming agricultural region and its local landscapes have evolved under the intricate interplay of climatic factors, topography, water dynamics, soils, vegetation, and ongoing anthropogenic influences.

While our research specifically delves into the nuances of the Talas district, the findings hold a broader relevance for understanding land cover dynamics and challenges in arid regions of Kazakhstan and, by extension, Central Asia.

2.2. Materials and Methods

In this study, conducted in in the Talas district, a systematic approach was adopted to choose and process satellite imagery. Data for this study were primarily acquired from GEE, utilizing its extensive repository of remotely sensed images. A reasonable consideration of data availability, temporal relevance, and optimal spatial resolution underpinned the choice of Landsat imagery for this study. Landsat’s extensive archive and consistent data collection made it a pragmatic choice, particularly for the targeted timeframe between April and July (2000, 2020). This period from April to July was strategically selected to align with the peak of the vegetation season, corresponding to the peak of the growing season in the study area. Focusing on this period captures the maximum extent of vegetation activity, allowing for more accurate assessments of vegetation dynamics. Additionally, it ensures that satellite imagery reflects the landscape during a crucial phase of vegetation development, providing valuable data for monitoring. To narrow down the selection of images, a bounds filter was applied to constrain them to the region of interest. Specifically, images from the Landsat 7 Tier 1 TOA image collection, taken between 1 April 2000 and 30 July 2000, were selected and filtered to remove cloud cover, ensuring data quality and accuracy [24].

Figure 2 presents a model for predicting changes in land use and soil vegetation cover for the study region. This methodology integrates both spatial as well as knowledge-based and field studies. For the analysis, 30 m resolution images from Landsat-7 Thematic Mapper and Landsat-8 Operational Land Imager satellites acquired in 2000 and 2020, respectively, were used. These data were downloaded from the Earth Resources Observation and Science (EROS) Center website at the USGS. The details of the satellite observations for the change analysis are presented in

Table 1. “LANDSAT/LC08/C01/T1_TOA” collection.

Image ID	Sensor	Acquisition Date	Path	Row	Bands	Cloud Cover (%)	UTM Zone
LANDSAT/LE07/C01/T1_TOA/LE07_153029_20000721	Landsat 7	21 July 2000	153	29	10	0	42
LANDSAT/LE07/C01/T1_TOA/LE07_154030_20000728	Landsat 7	28 July 2000	154	30	10	3	42
LANDSAT/LE07/C01/T1_TOA/LE07_153030_20000721	Landsat 7	21 July 2000	153	30	10	0	42
LANDSAT/LE07/C01/T1_TOA/LE07_154029_20000728	Landsat 7	28 July 2000	154	29	10	3	42
LANDSAT/LC08/C01/T1_TOA/LC08_153029_20200618	Landsat 8	18 June 2020	153	29	12	0.009	42
LANDSAT/LC08/C01/T1_TOA/LC08_154029_20200524	Landsat 8	24 May 2020	154	29	12	0.009	42
LANDSAT/LC08/C01/T1_TOA/LC08_153030_20200618	Landsat 8	18 June 2020	153	30	12	0.72	42
LANDSAT/LC08/C01/T1_TOA/LC08_154030_20200524	Landsat 8	24 May 2020	154	30	12	0.11	42

. The analyses were focused on the post-monsoon period due to the predominantly clear weather and minimal clouds during this time.

Figure 2 details the software implementation of the 2030 land use and vegetation change forecast using the Random Forest algorithm, involving a number of key steps. As input, Landsat-7 and Landsat-8 satellite data are used with atmospheric and geometric distortion correction and image quality enhancement. The data are then subjected to a classification process using the maximum likelihood method (MLC), based on training samples.

Next, using the resulting classification maps for the years 2000 and 2020, change detection is performed to determine land use dynamics. In the software implementation, Random Forest is used to generate the transition matrix and transition probabilities to predict changes in the landscape. The Random Forest algorithm is chosen for its ability to efficiently process large amounts of data and for its high prediction accuracy, due to its multiple decision trees working as an ensemble. This allows the algorithm to improve its predictive power and robustness to overfitting. In addition, Random Forest performs well on classification tasks in situations where classes may overlap and where there are a large number of predictors.

To fulfil the 2030 prediction, a software implementation may include the following steps:

• Data preprocessing: Cleaning and preparation of satellite images.
• Model training: Using training samples to train Random Forest algorithm.
• Model validation: Checking the accuracy of the model on test data.
• Prediction: Using the Random Forest model to estimate the probability of transitions between different land use categories.
• Forecast generation: Generating a projected land use map for 2030 based on the derived probabilities and current landscape trends.
• Accuracy assessment using Kappa statistics and the confusion matrix: The final step involves a quantitative evaluation of the model’s accuracy in predicting land use changes. This is achieved through the use of Kappa statistics and the analysis of a confusion matrix. The confusion matrix compares the predicted classifications of land use against the actual observed classifications (or ground truth data) and includes specific values such as True Positives (TPs) = 120, True Negatives (TNs) = 105, False Positives (FPs) = 15, and False Negatives (FNs) = 10. These values detail the number of correct predictions (true positives and true negatives) and incorrect predictions (false positives and false negatives) across different categories, serving as the foundation for calculating various accuracy metrics, including overall accuracy, precision, recall, and the Kappa statistic. The Kappa statistic measures the agreement between the predicted and actual classifications, adjusted for the agreement that might occur by chance alone. A higher Kappa value signifies a higher level of accuracy and agreement beyond what could be expected by random chance, offering a robust measure of the model’s predictive performance.

Incorporating the detailed values from the confusion matrix into the accuracy assessment provides a more comprehensive and nuanced understanding of the model’s performance. This approach ensures the predictions of land use and land cover change are both accurate and reliable, crucial for effective sustainable land use planning and management strategies aimed at achieving the objectives set for 2030.

Several filters were applied to narrow down the selection of images. A bounds filter was employed to constrain the images to the region of interest and a date filter was used to select images captured between 1 April 2000 and 30 July 2000. Subsequently, the filtered images were sorted in ascending order based on cloud cover, prioritizing images with lower cloud cover. This method effectively retrieved the image with the least cloud cover within the specified time and region, ensuring optimal data quality for analysis.

Between 1 April 2000 and 30 July 2000, a total of 32 images were available within the study area. These images were segmented into three distinct scenes, including one on 21 July 2000, with a path of 153 and a row of 29; another on 28 July 2000, with a path of 154 and a row of 30; the third one on 21 July 2000, with a path of 153 and a row of 30, and another on 28 July 2000, with a path of 154 and a row of 29. These three scenes were subsequently chosen for analysis and their coverage was merged to form a comprehensive dataset. Notably, one of the scenes exhibited a mere three percent cloud cover, ensuring minimal interference with data quality. Additionally, across all selected scenes, no cloud cover was observed within the study area.

A similar procedure was executed for the time period of 1 April 2020 to 30 July 2020, utilizing images from the “LANDSAT/LC08/C01/T1_TOA” collection (Table 1). In this timeframe, a total of 67 images were found to be available within the study area. The images selected through this rigorous process were then subjected to further processing and analysis.

Then, using the pixel values from the collection of imagery, a mosaic was built from the chosen scenes, choosing the pixel with the least amount of cloud cover or the highest priority for each pixel position. To visualize the study area, the image was displayed using the specified visualization parameters (Figure 3). ‘B3’, ‘B2’, ‘B1’ and ‘B4’, ‘B3’, ‘B2’ were selected to specify the bands for display, corresponding to the red, green, and blue bands of the Landsat 7 and Landsat 8 image in natural color [25]. For ‘B1’ (Landsat 7) and ‘B2’ (Landsat 8), blue is used for bathymetric mapping and distinguishing soil from vegetation and deciduous from coniferous vegetation. For ‘B2’ (Landsat 7) and ‘B3’ (Landsat 8), green emphasizes peak vegetation for assessing plant vigor. For ‘B3’ (Landsat 7) and ‘B4’ (Landsat 8), red discriminates variations in vegetation slopes [26].

The points sampled across different landscapes were amalgamated into a singular feature collection. This consolidated collection encompassed a variety of land cover categories, namely herbaceous bogs, bare vegetation, shrublands, salt marshes, water, and grasslands. In the study area, six land cover types were predominantly observed, namely herbaceous bogs, bare vegetation, shrublands, salt marshes, and grasslands.

The collection of all training and validation samples was meticulously executed based on the manual visual interpretation of high-resolution images sourced from Google Earth and corroborated topographic maps [27]. This method of sample collection and validation, grounded in visual interpretation, is well-documented and extensively utilized, as substantiated by a wealth of existing literature [28,29].

In culmination, a total of 196 and 242 training/test points were diligently gathered and subsequently employed for the study. The integration of these diverse points facilitated a comprehensive analysis, thereby contributing to the depth and breadth of the study’s findings on land cover variations within the specified region.

The gathered sample points were amalgamated into a unified feature collection, integrating diverse land cover categories including herbaceous wetland, bare vegetation, shrubland, solonchak, water, and grassland (Figure 4). A specified list of spectral bands, namely B2, B3, B4, B5, and B7 were employed for training the classifier. The choice of these specific bands for the Land Use and Land Cover (LULC) classification was based on their unique spectral characteristics and applications [30].

The points sampled were systematically associated with their corresponding land cover classes (Figure 5), facilitated through the comprehensive land cover feature collection. Incorporating field research data, land cover class pictures captured during July 2021, April 2022, and July 2022 have been included in Figure 5, providing a visual representation of the land cover. Consequently, four field campaigns were carried out between 2021 and 2023, covering various landscapes such as the Moyinkum sand massif, the Talas River terraces, foothill plains, part of the Karatau ridge and agricultural landscapes. The morphological description and the collection of soil samples were carried out through excavated soil profiles at 38 different locations. In addition, descriptions of soil sections, collection of samples from soil horizons, and descriptions of facies and vegetation were carried out at the reference points.

A Random Forest classifier, composed of 100 trees, was trained, employing the curated training data and the defined spectral bands. The class labels for this training data were derived from the ‘landcover’ property and the input bands were selected from the predetermined spectral bands.

Subsequent to the training phase, the adeptly trained classifier was applied to classify the input Landsat imagery. This application yielded a detailed land cover classification, effectively representing the diverse landscape characteristics inherent to the region under study. The classifier’s results, manifested in this classification, offer invaluable insights into the multifaceted nature of land cover variations and transitions within the study area (Figure 6).

To strengthen the classification and analysis of land cover in the Talas district, a thorough accuracy assessment was carried out using a split training and testing approach. The dataset of known land cover points was divided into training (70%) and testing (30%) subsets, which were randomly distributed to ensure model validity. The Random Forest Equation (1) classifier was thoroughly trained on the larger subset and subsequently used to predict classes in the test subset to evaluate its predictive ability.

$$Gini=1-{\sum }_{i=1}^c{\left(p_i\right)}^2,$$

(1)

This formula represents the Gini index, which is often used in the context of machine learning algorithms, in particular in the Random Forest algorithm to evaluate the quality of data partitioning in a tree node. The Gini index is a measure of the cleanliness of a node, where p_i denotes the fraction of objects of class i in the node. If all objects in a node belong to the same class, the Gini index is 0, indicating perfect purity. The Gini index value increases when a node contains a mixture of objects of different classes and reaches a maximum when all classes are equally represented in the node.

The performance of the classifier was evaluated using a mixture matrix for each evaluation year, in which predicted land cover types were contrasted with actual observations from the test set. The resulting matrices are an indispensable measure of accuracy, highlighting the strengths of the classifier and indicating areas for improvement in future classification iterations. Building on this foundation, the change analysis was extended to a land cover forecast for the year 2030. To build this forecast, temporal trends were analyzed from the 2000 and 2020 classified maps using an algorithm that extrapolated existing trends into the future. The predictive model was trained based on land cover trends observed over a 20 year period, taking into account variables such as climate data, anthropogenic activities, and land management policies that may affect land cover status in the future. A land cover map for 2030 was then produced according to an established classification system and using the predictive capabilities of the Random Forest model, now enriched with time trend data. The accuracy of this predictive model was assessed based on the consistency of trends and historical accuracy rates, assuming similar underlying processes are ongoing.

For the years 2000 and 2020, the accuracy of the classification models based on confusion matrices was 74.63% and 92.98%, respectively. These figures show an improvement in the model over time, reflecting improvements in classifier training and improvements in the resolution and quality of satellite imagery. The accuracy of the 2023 forecast has been cautiously assessed in light of the observed improvements in accuracy, expected technological advances, and improved algorithms. The 2000 Talas LULC map provides a historical baseline and the 2020 map reflects the most recent land cover condition (Figure 7).

3. Results

Table 2. The analysis of land cover areas between 2000, 2020, and 2030.

Class	Area 2000 (km2)	Area 2020 (km2)	Area 2030 (km2)
Herbaceous Wetland	351.08	340.35	334.02
Bare Vegetation	3639.99	2056.23	2271.41
Shrubland	3518.83	2003.07	1833.95
Solonchak	25.41	73.83	102.52
Water	73.25	53.71	24.012
Grassland	4430.09	7511.48	8828.77

presents a longitudinal analysis of land cover changes within the Talas District over a 30 year span, culminating in a forecast for the year 2030. The data delineate the areas covered by different land classes in square kilometers for the years 2000, 2020, and the projected areas for 2030 [31]. Here is a detailed description of the land cover changes and their potential implications.

• Herbaceous Wetland: There has been a gradual decrease from 351.08 km² in 2000 to 340.35 km² in 2020, with a further slight decline anticipated by 2030 to 334.02 km². This trend may reflect changes in wetland conservation and water management practices or possibly shifts in climate conditions affecting wetland ecosystems.
• Bare Vegetation: Initially, this class showed a substantial decrease from 3639.99 km² in 2000 to 2056.23 km² in 2020. Interestingly, a reversal of this trend is forecasted by 2030, with an increase to 2271.41 km². This could suggest a recovery of previously bare areas, perhaps due to natural regrowth or effective land rehabilitation efforts.
• Shrubland: There has been a consistent decline in shrubland from 3518.83 km² in 2000 to 2003.07 km² in 2020, with the decline projected to continue to 1833.95 km² by 2030. This ongoing reduction might be caused by land use changes, such as agricultural expansion or urban development, that replace shrubland with other land cover types.
• Solonchak: Contrary to most other classes, solonchak areas have been increasing, from 25.41 km² in 2000 to 73.83 km² in 2020, with a forecasted continuation of this trend to 102.523 km² in 2030. The growth of these saline soils can have negative impacts on soil quality and agricultural productivity, potentially indicating over-irrigation or poor drainage in the region.
• Water: This class has exhibited a concerning decrease from 73.25 km² in 2000 to 53.71 km² in 2020, with the forecast predicting a more dramatic reduction to 24.01 km² by 2030. This significant drop could have serious implications for regional water security, aquatic biodiversity, and could be indicative of the effects of climate change, increased water extraction for human use, or changes in land use that affect water retention.
• Grassland: In contrast to the general trend of decrease in other classes, grassland areas have expanded significantly from 4430.09 km² in 2000 to 7511.48 km² in 2020, with a projection of further increase to 8828.77 km² by 2030. This might reflect changes in land management, such as the abandonment of agricultural lands reverting to grassland, or could be due to conservation efforts promoting grassland restoration.

As well as the results, our extensive fieldwork spanning from 2021 to 2023 has provided valuable insights into the dynamics of natural processes in the Talas region. The lack of surface water inflow was identified as a key factor contributing to the increased aridization of natural landscapes. This phenomenon, observed during the autumn–spring periods and partially in the summer, has profound implications for agricultural landscapes. Our meticulous soil studies across diverse ecosystems, including the Moyinkum sand massif, Talas River float terraces, foothill plains, Karatau ridge, and agricultural landscapes, revealed significant variations in salt and dust content. Agrochemical analysis of soil samples from the Moyinkum sandy massif indicated elevated dust fractions on the surface and within the soil profile, potentially leading to the formation of a dust source in the Moyinkum territory. The negative impacts on agricultural areas were evident, as salt and dust removal affected the dried salt lake, Ashchykol, and surrounding areas. Findings from our fieldwork spanning 2021 to 2023 reveal a noticeable increase in salinity and individual ion levels in distinct regions. Additionally, the excessive salinization of crop fields and pastures, vital for the local population’s main source of income, results in soil erosion and decreased productivity. The analysis of soil salt in the study area highlights a significant surface accumulation in the topsoil of the irrigation massif.

The observed and projected changes in land cover within the Talas District signal significant ecological transformations. The expansion of grasslands and solonchak, coupled with the decline in wetlands, bare vegetation, shrublands, and especially water bodies, point to a complex interplay of environmental, climatic, and anthropogenic factors. These trends warrant a comprehensive strategy for sustainable land and water resource management to mitigate potential adverse impacts on the district’s ecosystems and the services they provide.

To complement the results depicted in Table 2, a precision assessment was carried out using a confusion matrix and the Kappa statistic, commonly utilized in remote sensing studies to evaluate the accuracy of land cover classifications. The Kappa statistic, a measure of agreement corrected for chance, yielded a value of 0.912, indicating an overall accuracy of 91.12%. This high level of precision in the classification process strengthens the reliability of the observed and forecasted land cover changes within the Talas District, confirming the effectiveness of the applied methodologies in capturing the complex dynamics of the region’s ecosystems.

4. Discussion

The extended analysis into the Talas District’s land cover changes now includes a forecasted perspective for the year 2030, revealing trends that could potentially reshape the region’s ecological framework. The ongoing decline in the “Herbaceous Wetland” class from 2000 to 2020 is anticipated to continue into the next decade. This persistent shrinkage necessitates a closer examination of environmental policies and practices, as wetlands are critical for biodiversity, water purification, and flood regulation. The forecast also indicates a reversal of the decreasing trend in the “Bare Vegetation” class seen from 2000 to 2020, with an expected increase by 2030. This unexpected change could suggest a recovery or adaptation of these ecosystems to prevailing climatic conditions or might reflect the success of land restoration initiatives.

Several scholars have observed notable fluctuations in precipitation patterns in Central Asia. Yao et al. [32] documented that the era from 1936 to 1957 exhibited the lowest precipitation levels, whereas the period from 1997 to 2004 experienced the highest levels of rainfall. The years 1969 and 1944 exhibited the highest and lowest levels of precipitation on record, with roughly 332.1 mm and 152.0 mm, respectively, representing a twofold disparity. In contrast, the study conducted by Hu et al. [33] investigated the temporal fluctuations in precipitation across three different time intervals, namely 1901–2013, 1951–2013, and 1979–2013. The researchers utilized the Global Precipitation Climate Center (GPCC) dataset and reported three significant discoveries. In the context of Central Asia, it can be observed that the period spanning from 1901 to 2013 exhibited a general tendency to increase, with a rate of 0.66 mm per decade. However, it is worth noting that this rate was comparatively lower than that observed during the period from 1951 to 2013. Furthermore, the regional annual precipitation exhibited significant changes at high frequencies, characterized by quasi-periods of 3 and 6 years, as well as at low frequencies, with quasi-periods of 28 years. Furthermore, it should be noted that variations in the quantity of precipitation, which can sometimes be as high as a twofold increase or decrease, within a brief timeframe do not necessarily signify the initiation of a drought.

Zhang et al. [34] presented a contrasting scenario, highlighting the susceptibility of southern Kazakhstan to drought and desertification. They emphasized that persistent drought conditions played a pivotal role in the deterioration of pastures and the process of desertification in the broader Central Asian region. A remote sensing analysis conducted over Central Asia revealed a decrease of 12.5% in pasture area from 2000 to 2014. Henebry et al. [35] conducted an extensive nonparametric examination of land surface dynamics in the drylands of Central Asia, spanning the years 2003 to 2017. The study revealed noteworthy upward trends in MODIS 0.05° nighttime land surface temperature (LST) in approximately 11.0% of the region, while significant downward trends in the diel temperature range (DTR) were observed in approximately 7.5% of the area. Furthermore, when considering the Central Asian grasslands, researchers observed noteworthy positive trends in daytime land surface temperature (LST) in 62.2% of the region. Similarly, substantial positive trends in nighttime LST were identified in 56.1% of the area. Conversely, significant negative trends in diurnal temperature range (DTR) were observed in 52.4% of the studied region. The daytime land surface temperature (LST) exhibited a notable increase of 44.5% in barren lands and 29.6% in open shrublands. Similarly, the nighttime LST experienced a substantial increase of 17.8% in barren lands and 19.2% in open shrublands. Conversely, there were large decreases in the diurnal temperature range (DTR) of 18.2% in barren lands and 33.2% in open shrublands. The aforementioned patterns collectively indicate the widespread occurrence of substantial aridification throughout Central Asia.

The Talas District classified specific land cover classes such as “Herbaceous Wet-land”, “Bare Vegetation”, “Shrubland”, “Solonchak”, and “Water”. Comparatively, Hu, Y. et al. [19] provided a more extensive list of land cover types in Central Asia, including grassland, bare land, cultivated land, shrubland, water bodies, wetlands, forests, permanent snow and ice, and artificial surfaces.

In comparing the land cover changes in the Talas District with the broader Central Asia study, it is evident that both highlight certain key classes such as grassland, water bodies, wetlands, and shrublands. While the Talas District’s analysis specifically delves into the implications of “Herbaceous Wetland”, and “Solonchak”, the broader Central Asia study expands the scope by addressing additional land cover types like bare land, cultivated land, forests, permanent snow and ice, and artificial surfaces.

Natural vegetation and grassland expansion appear as positive trends, while concerns about water-related areas are evident in both cases. However, the divergence in certain categories underlines the region-specific factors influencing land cover dynamics.

The shared emphasis on grassland, water bodies, wetlands, and shrublands in both discussions underscores their significance in understanding ecological dynamics. The Talas District’s focused analysis provides insights into local nuances, whereas the broader Central Asia study offers a comprehensive view of land cover changes across a larger and more diverse geographical area. This comparative approach ensures a nuanced understanding of both specific regional intricacies and overarching trends within Central Asia.

However, “Shrubland” areas are projected to continue their decline, reinforcing concerns over habitat loss and urging a reassessment of land use planning and conservation efforts. Remarkably, the “Solonchak” soils are projected to expand further, compounding the challenges of managing soil salinization. This increasing trend underscores the need for strategies to manage salinity and protect agricultural productivity. It also raises questions about the long-term sustainability of current water management and agricultural practices. Conversely, the “Water” class’ forecasted substantial reduction by 2030 is alarming, as it suggests escalating stress on aquatic ecosystems. The drivers of this decline, likely a combination of climatic variability and anthropogenic demand, call for urgent action in water conservation and sustainable usage to safeguard these critical resources. For example, the authors of [36] reviewed a compelling case where the widespread use of irrigation practices led to a marked decline in natural water bodies in an arid region, demonstrating how anthropogenic activities can directly affect aquatic ecosystems. The “Grassland” class, however, is expected to continue its expansion, which may be linked to ecological succession and changes in land management post-1990s. This continued increase underscores the dynamism of grassland ecosystems, but also highlights the need to understand the balance between land recovery and the demands of increasing livestock numbers. It should be noted that, in the 1990s, the Central Asian countries experienced a catastrophic economic crisis; in particular, pastoral animal husbandry decreased several times, as a result of which the quality and quantity of pastures improved, as a result of the absence of transhumance pastoral animal husbandry [14]. This statement is indirectly confirmed by the above studies, as well as by a significantly rapid increase in the number of livestock during the period of relative economic recovery (conditionally, the period 2002–2015).

The evolving land cover trends in the Talas District reflect complex ecological transformations that necessitate proactive and adaptive management strategies. It is imperative that we continue to conduct robust monitoring and research to unravel the complex interplay of factors driving these changes. Understanding these dynamics is crucial for crafting informed policies aimed at conserving biodiversity, sustaining human livelihoods, and ensuring ecosystem resilience in the face of ongoing environmental change. This analysis acknowledges the limitations of our prediction model, particularly the challenges of accounting for future uncertainties such as climate change, policy shifts, and other factors. It emphasizes the need for a comprehensive approach to discussing potential uncertainties in more depth, thereby enhancing our discussion with a broader perspective on the implications of these projections.

5. Conclusions

In light of the projections for the year 2030, our thorough analysis has elucidated the evolving dynamics of land cover in the Talas District, revealing intricate changes over the past three decades. The data forecast persistent transformations, indicative of significant ecological and anthropogenic influences on the landscape. The observed increase in grasslands and solonchak soils, alongside the decrease in shrubland and wetland areas, and the concerning reduction in water bodies, highlight the complex interplay between environmental factors and human activities. These trends reveal both the ecological resilience and vulnerability of the district, emphasizing the critical need for informed and strategic interventions.

This research underscores the imperative for the continuous, detailed monitoring of land cover changes to proactively respond to and mitigate their impacts. Effective management, guided by scientific understanding, is crucial for maintaining the ecological integrity of the district. Addressing these changes necessitates not just a reactionary stance to current trends, but a forward-thinking approach that integrates conservation efforts with sustainable land use and water resource management.

The conservation of Talas District’s natural resources is a complex task that goes beyond conventional environmental stewardship. It is an ethical commitment, a pledge to future generations, and a recognition of the interconnected nature of life in this region. Our findings advocate for a holistic strategy that harmonizes economic development with ecological conservation.