Introduction

Arid and semi-arid regions, which account for more than one-third of the global land surface, are among the most ecologically fragile areas on Earth1. Limited water resources, strong climatic variability, low vegetation cover, and human-induced land conversion have made these regions highly vulnerable to ecological degradation, desertification, and landscape instability2. In such environments, ecological risk—defined as the probability of ecosystem structural or functional deterioration under natural or anthropogenic disturbances—has become a key indicator for evaluating regional ecological security and sustainability3.

Landscape ecological risk assessment (ERA), which integrates landscape pattern characteristics with ecological processes, has been widely used to monitor ecosystem vulnerability and understand how land-use transitions reshape ecological resilience4. However, traditional ERA studies often face three methodological limitations5. First, many assessments rely on single-date satellite images, which are affected by cloud cover, atmospheric conditions, and phenological differences, limiting their ability to capture long-term ecological trajectories6. Second, the temporal discontinuity of remote sensing data prevents detection of regime shifts or nonlinear ecological responses7. Third, pixel-based classification approaches commonly applied in landscape analysis can produce high levels of noise, particularly in heterogeneous agricultural landscapes8.

Artificial oasis regions—intensively cultivated, human-constructed agricultural systems embedded within desert landscapes—represent an extreme case where ecological risk assessment is urgently needed9. These oases rely entirely on irrigation and are highly dependent on the balance between land expansion and water-resource availability10. In northwestern China, large-scale reclamation has transformed vast barren land into productive farmland, generating economic benefits while simultaneously imposing long-term ecological pressure11. Existing studies often emphasize the impact of irrigation or salinization but seldom incorporate three decades of continuous land-use dynamics to evaluate long-term ecological risks12.

Moreover, artificial oasis ecosystems exhibit complex nonlinear ecological responses13. Moderate reclamation can increase vegetation cover, reduce fragmentation, and temporarily stabilize the landscape. However, once reclamation exceeds ecological thresholds—particularly the maximum water-resource carrying capacity—ecosystem resilience declines rapidly, accelerating soil salinization, surface hardening, and ecological deterioration14. Despite its importance, this nonlinear mechanism has not been adequately explored in previous oasis-related ERA studies.

To address these gaps, this study integrates long time-series Landsat imagery, annual maximum NDVI composites, object-oriented land-use classification, landscape ecological risk indices, mutation detection, and CA–Markov scenario simulation. Specifically, the study aims to:

  1. (1)

    construct a consistent, long-term, high-quality landscape dataset using all available Landsat scenes from 1990 to 2019;

  2. (2)

    quantify spatiotemporal changes and mutation points in ecological risk across a 30-year period;

  3. (3)

    identify the driving mechanisms underlying ecological risk evolution in artificial oasis ecosystems; and.

  4. (4)

    simulate future ecological risk scenarios for 2050 and 2080 under continued land reclamation.

To reduce cloud contamination and phenological inconsistencies among multi-temporal images, we adopted the annual maximum NDVI composite, which preserves the highest-quality vegetation signal within each year. While this method effectively eliminates cloud cover and normalizes peak vegetation greenness, it does not correct for low NDVI values caused by crop rotation or long-term fallow conditions, where vegetation cover remains minimal throughout the year.

This research contributes to the understanding of ecological risk evolution in arid artificial oasis regions by revealing nonlinear ecological responses, threshold effects, and future trajectories of risk expansion. The findings provide a scientific foundation for sustainable land management and ecological protection in fragile arid environments.

Materials and methods

Study area

Alar Reclamation area is located in, south of Xinjiang Uygur Autonomous Region, northwestern China with a total area of 4105.92 km2, (80°30′0″~81°58′0″E, 40°22′0″~40°57′0″N) (Fig. 1a). It covers the region from the southern foot of the Tianshan Mountains and to the northern edge of the Taklimakan Desert, and forms a typical mountain-oasis-desert ecosystem. Alar reclamation area belongs to a warm temperate zone with extreme continental arid desert climate, arid and drier, with an average annual precipitation of 40.1–82.5 mm, and an average annual evaporation of 1 876.6–2 558.9 mm. The average annual sunshine time is 2 556.3–2 991.8 h. The main water sources in the reclamation area are the Tarim River and the three major reservoirs of Shengli, Duolang and the upper reaches. The soil type is mainly sandy loam. The terrain is high in the west and low in the east. It is rich in light and heat resources and is a typical oasis agricultural area.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

(a) Location of study area and (b) The ecological risk evaluation cells on simple map of Alar reclamation area.

Data acquisition and preprocessing

Data acquisition

The remote sensing data is downloaded from website of the United States Geological Survey (USGS) (https://earthexplorer.usgs.gov/). These images have orbit number of 146 − 32 and spatial resolution of 30 m, and with cloud coverage less than 40% (Multi-temporal image synthesis can eliminate the influence of cloud cover). Since the Landsat series of sensors have been continuously updated over the past 30 years, the sensor selections for remote sensing images in this study are: Landsat 5 from 1990 to 1998, Landsat 7 from 1999 to 2012, and Landsat 8 from 2013 to 2019. Through this way, we can avoid the differences caused by different sensors to a certain extent. 259 Landsat TM remote sensing images, which cover the study area, were collected in the past 30 years. The cloud coverage of each scene and the number of images in each year are shown in Fig. 2.

Although Landsat imagery after 2019 is available, scenes from 2020 to 2024 suffer from extensive cloud contamination and incomplete coverage in the study area, which prevents the construction of NDVI composites of comparable quality. Therefore, 2019 was selected as the final year to ensure temporal consistency.

Landsat imagery was obtained from the USGS EarthExplorer platform. The 30 m SRTM DEM was downloaded from the NASA SRTM database. Meteorological datasets were sourced from the China Meteorological Data Service Center, socioeconomic data were obtained from the Xinjiang Statistical Yearbook, and additional environmental datasets were acquired from the Resource and Environment Science and Data Cloud (RESDC).

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Remote sensing data: (a) percentage of unclear pixels in each image; (b) number of images per year.

NDVI synthesis of remote sensing images and landscape classification

Preprocess was employed for the Landsat images of from 1990 to 2019 in the ENVI software, including radiometric calibration, atmospheric correction, geometric correction, cropping, image enhancement, and band calculation. Then the Normalized Difference Vegetation index (NDVI) of each image was calculated. After that, we synthesized the maximum value of all NDVI in the same year to obtain the maximum NDVI composite map for each year. We used the eCognition software to employ object-oriented classification method for landscape classification. In this study, the landscape were divided into six types: cultivated land, forest and grassland, garden land, water body, construction land and unused land. The processing flow of the remote sensing images was presented in Fig. 3.

Accuracy assessment showed that the fused NDVI-based classifications achieved an overall accuracy of 89.7% and a Kappa coefficient of 0.87, indicating that the fused imagery is reliable for long-term ecological risk assessment.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Framework of this study.

Landscape ecological risk index

Dividing the area into ecological risk zones

Considering the spatial heterogeneity, patch size, area of the survey region and sample sampling density15, a 1 km×1 km square grid was used to conduct evenly spaced sampling fishing nets in the Alar reclamation area using ArcGIS software. The study area was divided into 4337 ecological risk communities, as shown in Fig. 1(b).

Construction of landscape ecological risk index

The landscape index can quantitatively reflect the relevant information of the landscape pattern, and the selection of appropriate landscape indicators can reveal the negative effects caused by human disturb16. In this study, the landscape fragmentation index (Ci), separation index (Si) and dominance index (Di) were selected and combined to construct the landscape disturbance degree (LDIi) index model17. Meanwhile, with the help of landscape, the landscape disturbance degree index (LDIi) and the vulnerability index (LFIi) were used to construct the landscape ecological risk index (ERI). Related specific calculation formulas are show in Table 1.

Fragility values were assigned based on ecological stability theory, vegetation stress-tolerance characteristics, and resistance to external disturbance. Cultivated land was assigned a higher fragility value due to its dependence on irrigation and lower natural resilience. Unused land, sensitive to salinization and desertification, received a moderate value. Construction land, with minimal ecological function, was assigned the lowest fragility score. This approach is consistent with previous ecological risk assessments18.

Table 1 The calculation formulas for ecological risk index.
Full size table

Identification of mutation points of landscape ecological risk

In this study, the cumulative anomaly method was used to explore the sudden change of landscape ecological risk in the Alar reclamation area in the past 30 years. First, we calculated the anomaly at each time point in the ecological risk time series, then we accumulated the anomaly at each time point. Finally, we plot the cumulative anomaly value at each time point. If significant decrease or increase occurs in certain point, then the inflection point is determined. Related formulas are as follow:

$$LR_{i} = \sum\limits_{i}^{v} {\left( {R_{i} - \bar{R}} \right)}$$
(1)

where \(L{R}_{i}\) is the cumulative value of anomalies in the i-th year, \({R}_{i}\) is the landscape index value in the i-th year, and \(\overline{R}\) is the multi-year average of the landscape index sequence.

Spatial-temporal analysis method of landscape ecological risk

Geostatistics are widely used for monitoring, simulating and estimating the spatial correlation and spatial pattern of target variables19. The experimental variogram is used to explore the spatial structure of the variables20. The calculation formula is:

$$\gamma \:\left( h \right) = \frac{1}{{2n\left( h \right)}}\sum\limits_{{i = 1}}^{{n\left( h \right)}} {\left[ {Z\left( {x_{i} + h} \right) - Z\left( {x_{i} } \right)} \right]^{2} }$$
(2)

Where: \(\gamma\:\left(h\right)\) is the variogram; \(n\left(h\right)\) is the number of the pairs of sample points with distance of \(h\); \(Z\) is the random variable of a certain system attribute; \(Z\left({x}_{i}\right)\) and \(Z\left({x}_{i}+h\right)\) are means the values of target variables of sample \({x}_{i}\) and \(\left({x}_{i}+h\right)\), respectively.

The Geostatistics software GS + is used to achieve the optimal fitting of the experimental semi-variogram model, based on which the ordinary kriging interpolation is employed21 in the ArcGIS software. The natural breakpoint method is used to divide the study area into five levels based on the maps of ecological risks in each year: extremely low-risk, low-risk, medium-risk, high-risk and extremely high-risk. Then we used the spatial overlay analysis to make quantitative analysis on the transform direction and area of regions with different ecological risk level.

Spatial autocorrelation analysis

Spatial autocorrelation analysis can test whether the values of the target variable in visited locations with are related to the adjacent points, which are divided into global spatial autocorrelation and local spatial autocorrelation22. In this study the Moran’s I index value was calculated to assess the degree of spatial autocorrelation of target variable using GeoDa software. In this study, using Moran’s I index we could detect whether there are statistically significant relationship of ecological risks in local areas23. The hot-spots means regions with high value of ecological risk cluster while the cold-spots represents regions with low value of ecological risk cluster.

Prediction of ecological risk based on cellular automaton–Markov (CA–Markov) model

In this study, in order to improve the prediction accuracy of ecological risk, we collected information of 15 potential influencing factor including elevation, slope, aspect, distance from highways, railways, highways and water systems, temperature, precipitation, population, GDP, social fixed asset investment, primary industry, gross agricultural production and cotton prices as well as landscape classification data. We used the CA–Markov model in the IDRISI software to simulate the future landscape pattern in the study area24. Firstly, we predicted the ecological risk in 2010 in the study area based on the data of 15 potential influencing factors and the 10 landscapes from 1990 to 1999. Then the prediction result in 2010 was compared with the real ecological risk of 2010, with which the performance of our CA–Markov model was validated. Finally, we predicted the ecological risk of the survey region in 2050 and 2080 using the model we developed above.

The CA–Markov model was validated by simulating the 2010 ecological risk map using 1990 and 1999 land-use data and comparing the result with the observed 2010 ERI. The model achieved an overall accuracy of 86.4% and a Kappa coefficient of 0.83, demonstrating strong predictive performance.

Results

Changes of ecological risk index of landscape

In this study we calculated the changes of indexes of various landscape types in Alar Reclamation area between 1990 and 2019 (Fig. 4) using the Fragstats and Excel software. Our results indicated that from 1990 to 2019, the area of arable land and garden land increased significantly, the area of unused land decreased significantly, and the area of forest, grassland, water bodies, and construction land kept stable, indicating that a large area of wasteland was reclaimed in the study area during the 30 years. The Fragmentation Index (\({C}_{i}\)) of cultivated land and construction land is relatively high with decreasing trend. Meanwhile, the Fragmentation Index (\({C}_{i}\)) value of water body and unused land is close to 0 which indicates that the spatial distribution of water body and unused land is very concentrated.

The separation index (\({S}_{i}\)) of construction land and cultivated land showed a decreasing trend, and the spatial distribution characteristics changed from random distribution to concentrated distribution. The dominance index (\({D}_{i}\)) of unused land shows a downward trend, and the degree of interference gradually increases, while the \({D}_{i}\) value of cultivated land shows an upward trend. Due to the increase of demand for cultivated land, large areas of wasteland are reclaimed which lead to increase of \({D}_{i}\) value of cultivated land.

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Landscape pattern metrics from 1990 to 2019.

The spatio-temporal variation of landscape ecological risk

Semi-variogram analysis

This study has obtained the results of ecological risk of all thirty years between 1990 and 2019 in the study area. Here we presented the prediction results of 1990, 2000, 2010 and 2019. The optimal fitting of the theoretical model of the variance function was performed on the sampling data of 4337 ecological risk communities in each year, and the relevant variance functions and parameters were shown in Table 2. In 1990 and 2019, the exponential model was fitted, while in 2000 and 2010, the spherical model was more suitable.

From 1990 to 2019, the sill value decreased from 0.0027 in 1990 to 0.0014 in 2000, and then increased to 0.0024 in 2019, indicating that the uneven spatial distribution of landscape ecological risk intensity decreased firstly and then increased between 1990 and 2019. The range value increased from 6,900 m in 1990 to 27,600 m in 2019, indicating that the relevance range of the ecological risk index is gradually expanding. The Nugget/Sill value of the ecological risk index showed a decreasing trend firstly and then increased, ranging from 0.641 to 0.846. This indicates that the spatial correlation of the ecological risk value decreased firstly and then increased.

Table 2 Parameters of theoretical model of variogram.
Full size table

The Spatial characteristics of landscape ecological risk

The spatial distribution of landscape ecological risk (ERI) and the proportions of each risk level in the Alar reclamation area in 1990, 2000, 2010 and 2019 are shown in Fig. 5. As showed in Fig. 5, in 1990, most of areas in the survey region had high ecological risk and few areas had extremely low and low risk, which mainly distributed along the coast of the Tarim River. The main landscape types in 1990 were unused land, and the ecological environment is very fragile, therefore the ecological risk level was at high level. When it comes to 2000, the area occupied by low-risk areas had expanded, while the areas of regions with high ecological risk decreased. Concerning 2010, the ecological risk in the northwest of the study area changed from high level to medium level. This mainly attributed to the fact that this region was reclaimed and expanded, which lead to increase of planting density as well as ecosystem stability, and reduced the ecological risks. By 2019, the proportion of regions with extremely low ecological risk and low ecological risk kept decreasing trend. Most of regions have low or medium ecological risk. Regions with high ecological risk mainly distributed in Southeast and Northeast of the study area.

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Spatial patterns and area proportion of each level of the ERI in 1990, 2000, 2010 and 2019(ERI: ecological risk index).

Spatial characteristics of dynamic changes of landscape ecological risk

Here we presented the changes of ecological risk levels in different areas between 1990 and 2019 (Fig. 6). Between 1990 and 2000, the areas with ecological risk levels reduced were concentrated in the northwest of the study area and coastal areas of Tarim River. Regions with increased ecological risk level were mainly distributed in near the three major reservoirs. From 2000 to 2010, the change of ecological risk is more complex and irregular in the study area. It is worth noting that large areas in the Northwest part of the survey region showed decreased ecological risk level. It is rooted from extensive reclamation of unused land in this region. Increase of cultivated land reduced the ecological risk. From 2010 to 2019, the areas with reduced ecological risks mainly located along the coast of the Tarim River, and the areas with increased ecological risks are mainly located at the north and south part of the study area. The expansion of cultivated land showed a trend of spreading outward from the Tarim River.

Overall, the ecological risk level showed a decreasing trend in the survey region between 1990 and 2019. Areas with changed ecological risk level were mainly distributed along the coast of the Tarim River and the northwest part of survey region. This is because these areas close to the water source and it are conducive for agricultural farming. With the development of agriculture, the planting structure in this area is diversified and the ecosystem is more stable which increase the ability to resist external interference. Therefore the dominance is increased, which reduces the ecological risk level.

Fig. 6
Fig. 6The alternative text for this image may have been generated using AI.
Full size image

Changes in the spatial distribution and area proportion of each level of the ecological risk change class in the Alar reclamation area for (a) 1990–2000, (b) 2000–2010, and (c) 2010–2019. Note: I: Extremely ecologically deteriorated area, II: ecologically deteriorated area, III: ecologically stable area, IV: ecologically improved area, and V: extremely ecologically improved area.

Spatial autocorrelation analysis

Global Spatial autocorrelation

The global Moran’s I index value of the ecological risk index of 4337 plots in Alar reclamation area in 1990, 2000, 2010 and 2019 were calculated using GeoDa. As shown in Fig. 7, the global Moran’s I value of ecological risk index in 1990, 2000, 2010 and 2019 are 0.317, 0.625, 0.716, and 0.692, respectively. All these four values are positive and showed increasing trend, indicating that the ecological risk index of the Alar reclamation area has a significant positive spatial correlation. The increased value of global Moran’s I index means the spatial correlation of ecological risk in the study area was strengthened.

Fig. 7
Fig. 7The alternative text for this image may have been generated using AI.
Full size image

Global Moran’s I scatter figures of the landscape ecological risk index in the Alar reclamation area for (a) 1990, (b) 2000, (c) 2010, (d) 2019.

Local spatial autocorrelation

Since the global spatial autocorrelation can only reveal the spatial autocorrelation in the whole study areas, the local Moran’s I index was also calculated to analysis the local spatial cluster of areas with high or low ecological risks. The maps of local autocorrelation of ecological risks in the study area in 1990, 2000, 2010, 2019 were showed in Fig. 8. As presented in Fig. 8, the area of regions with high-high cluster of ecological risk significantly decreased between 1990 and 2019. The regions have a cluster of high-high ecological risk show a shrinking spatial trend of from the outside to the inside of the survey region. The main landscape type in the regions with cluster of high-high ecological risk is unused land with low vegetation coverage, single vegetation structure. Thus, the ecological environment is fragile in these areas. In contrast, areas with cluster of low-low ecological risk expanded from the outward to the central part of the survey region. These regions featured by low ecological risk. The main landscape types are cultivated land, garden land and construction land with relatively stable ecosystem.

Fig. 8
Fig. 8The alternative text for this image may have been generated using AI.
Full size image

Local spatial autocorrelation of the landscape ecological risk in the Alar reclamation area for (a) 1990, (b) 2000, (c) 2010, (d) 2019.

Forecasting future changes in ecological risks

As revealed by Fig. 9a and b, the predicted map of ecological risk is highly consisted with the real map of ecological risk in the study area in 2010. This confirms that our developed CA–Markov model showed good ability to predict the ecological risk, which make it, is a good choice to forecast ecological risk in 2050 and 2080. The forecasted results showed that in 2050, the proportion of areas with high ecological risk will reach 20% and the high ecological risk regions would patchily distributed in whole the survey region. The main landscape is cultivated land.

In 2080, areas of regions with low ecological risk will continue decrease while the areas of regions with low ecological risk will continue increase. The regions with high ecological risk would cover the entire study area. The main landscape type is cultivated land.

Fig. 9
Fig. 9The alternative text for this image may have been generated using AI.
Full size image

Spatial patterns and area proportion of each level of the ERI in 2010 Prediction, 2010 Real, 2050 and 2080.

Discussion

Long-term NDVI composites improve the reliability of ecological risk assessment

Conventional ERA studies commonly employ remote sensing images collected on a single date, which are highly susceptible to cloud contamination, phenological differences, and seasonal vegetation variability25. Such inconsistencies often produce misclassification or incomplete classification, which in turn biases the calculation of landscape fragmentation, dominance, and ecological risk26. In this study, using the annual maximum NDVI composite allowed us to incorporate all available satellite observations each year, ensuring that the highest-quality vegetation signal was preserved while eliminating the influence of clouds and phenological shifts. The object-oriented classification method further avoided pixel-level “salt-and-pepper” noise, yielding coherent landscape patches and more robust landscape indices. These improvements provide a reliable foundation for assessing ecological risk dynamics in long-term time series.

Nonlinear response and mutation characteristics of ecological risk in artificial Oasis systems

The 30-year ecological risk trajectory in the Alar Reclamation Area revealed a clear mutation point around 1995. Before 1995, ecological risk increased slightly due to dispersed and fragmented land reclamation, which caused local disturbances without forming stable agricultural systems. After 1995, large-scale, contiguous reclamation replaced scattered unused land with irrigated cropland, substantially increasing vegetation cover and improving landscape connectivity. This transition resulted in enhanced ecosystem stability and a continuous decline in ecological risk.

This finding reflects the nonlinear and threshold-dependent ecological behavior of artificial oasis ecosystems. Moderate reclamation can initially stabilize the landscape by reducing fragmentation and promoting vegetation cover. However, once reclamation intensity approaches or exceeds water-resource carrying capacity, ecosystem resilience may decline sharply. These threshold dynamics are critical for understanding why past reclamation reduced ecological risks, whereas future reclamation—under constrained water resources—is projected to increase them27.

Mechanisms driving observed changes in ecological risk

The analysis of landscape-specific ecological risks demonstrated that anthropogenic landscapes (cultivated land, garden land, construction land) experienced a continuous decline in ecological risk from 1990 to 2019. Population growth, expansion of irrigation infrastructure, and application of modern agricultural technologies contributed to the establishment of a more stable artificial oasis. These processes enhanced vegetation cover, reduced fragmentation, and increased ecological buffering capacity28.

Conversely, natural landscapes (unused land, forest and grassland, water bodies) showed an overall increase in ecological risk. Unused land—characterized by low vegetation cover and high sensitivity to erosion and salinization—was progressively reclaimed, exposing these areas to disturbance and reducing their ecological resilience. Limited natural water availability further intensified ecological vulnerability in non-irrigated areas29.

Taken together, the contrasting trends between natural and anthropogenic landscapes highlight the dual role of human activities: while structured reclamation can enhance ecological stability in reclaimed zones, the broader desert ecosystem may face higher risk due to declining natural buffering capacity and intensified pressure on water resources.

Future ecological risks under sustained reclamation pressure

The CA–Markov model projections reveal diverging ecological pathways30. The 2050 scenario suggests moderate expansion of high-risk zones, while the 2080 scenario indicates a clear intensification of ecological risk across the entire oasis. This shift reflects long-term resource depletion, particularly the growing imbalance between agricultural water demand and surface water supply. As reclaimed land expands, the ecological cost of maintaining agricultural production increases, diminishing ecosystem resilience. The projected expansion of high-risk zones in 2080 underscores the necessity of regulating land reclamation intensity, optimizing water allocation, and protecting natural vegetation31.

Implications for sustainable land management in artificial Oasis regions

This study demonstrates that artificial oasis ecosystems respond strongly to both land-use structure and water-resource constraints32. Policies should therefore prioritize: limiting reclamation beyond ecological thresholds; enhancing water-use efficiency in agriculture; conserving ecologically sensitive unused land; strengthening ecological restoration and salinization control; integrating long-term remote sensing into monitoring programs33.

The methodological framework developed here—long-term NDVI composites, landscape metrics, mutation detection, and scenario simulation—provides a replicable approach for assessing ecological risks in similar arid oasis environments worldwide.

Driving factor detection

To quantify the influence of natural and human factors on ecological risk, the Geodetector q-statistic model was applied. The results indicated that cropland expansion (q = 0.41), NDVI (q = 0.38), distance to rivers (q = 0.33), elevation (q = 0.27), and population density (q = 0.25) were the dominant contributors to the spatial differentiation of ecological risk.

Conclusion

In this study, we collected data from 259 Landsat image in Alar reclamation area and then used method based on landscape ecology theory and spatial statistical analysis methods to construct a landscape ecological risk index to assess ecological risk in Alar reclamation area. Based on that, we revealed the sudden change points and potential driving factors of the ecological risk in the study area, and forecasted the ecological risk in 2050 and 2080 in the survey region. The main conclusions of this study are as follows:

  1. (1)

    From 1990 to 2019, the area of arable land increased in the Alar reclamation area, the degree of fragmentation decreased, and the degree of dominance increased. The spatial pattern of the arable land changed from small random scattered distribution to large patches of concentrated distribution. The area of unused land decreased, and the degree of dominance of the unused land decreased. The ecological risk of the study area showed decreasing trend, with a sudden change occurred in 1995, which accelerated the reduction of ecological risk.

  2. (2)

    The ecological risk level decreased from 1990 to 2019 in the survey region. Clear change of the ecological risk was detected in the area along the Tarim River and the northwestern part of the reclamation area. The spatial cluster of ecological risk intensity increased, and clear cluster of high-high value of ecological risk index and low-low value of ecological risk index was appeared in the study area.

  3. (3)

    The ecological risk of Alar reclamation area will increase in the next 60 years. It is mainly due to excessive land reclamation and used land expansion, which make the ecosystem more susceptible to external interference, and as results, the ecological risks would increase.