Positive impact of urbanization on vegetation growth has been continuously strengthening in arid regions of China

The ecological environment is fragile in arid regions, and the direct and indirect impacts of continuing urbanization on vegetation growth in cities still need to be studied in depth. In this study, we focused on four provincial capital cities (Urumqi, Lanzhou, Yinchuan, and Hohhot) in arid regions of China. We used continuous 30 m land cover and vegetation greenness (VG) data from 1990 to 2021 to extract the impact of urbanization on vegetation growth by separating the impact of natural conditions. Our results showed that the study area’s urban area (UA) had expanded rapidly at a rate of 39.2 km2/a, increasing by 3.39 times between 1990 and 2021. While urban expansion occurred rapidly, the overall VG of the study area also increased (slope = 2.16 × 10−3), with the enhancement of VG increasing gradually from west to east. The VG and its trend in the UA were significantly higher than those in the natural vegetation area (NA). In addition, the duration of the urban vegetation growth season was longer than that of the NA, which also confirmed the positive impact of urbanization on the vegetation growth period. Furthermore, we found that the positive impact of urbanization on vegetation had continuously strengthened over time. In 1990 and 2021, the VG areas had 83% and 87% above the ‘zero-impact line’ in the UA, respectively, indicating a very significant impact of urbanization on vegetation growth in arid regions with fragile natural conditions. Our study identified the long-term dynamic trends in urbanization and VG in arid regions and clarified the non-linear relationship between the two concurrent growth factors. This has significant implications for correctly understanding the impact of urbanization on vegetation in arid regions and can provide a scientific reference for the ecological construction of urban environments in the region.


Introduction
As global climate change intensifies, arid events will become more severe and frequent [1].However, this does not necessarily imply that the growth status of urban vegetation in arid regions is deteriorating [2].Continuous urbanization in arid regions has significantly altered the landscape ecology and local climate conditions within urban areas (UAs) [3][4][5].The vegetation greenness (VG) (usually characterized by normalized difference vegetation index, NDVI), has a strong indicative function for the surrounding environment and intuitively reflects the spatiotemporal characteristics of vegetation growth [6,7].To clarify the intrinsic relationship between urbanization and vegetation growth in arid regions, isolating the urban interior and removing the impact of the natural conditions of arid regions are necessary [4,7].Long-term raster data series [8][9][10] are conducive to researching the response of vegetation to urban environmental changes against the background of urbanization.This is important for a correct understanding of the impact of urbanization in arid regions and local environmental improvement [7,11,12].
In the past few decades, the rate of urban expansion has been unprecedented [13], especially in arid regions, which is significantly higher than the global average [14].Natural climate conditions such as precipitation will seriously restrict the phenological cycle and distribution pattern of vegetation, which is also more sensitive to the fluctuations of climate change in arid regions than in humid regions.In traditional thinking, urbanization directly encroaches upon a large area of vegetation and has a negative impact on the local ecological environment [3,4,15,16].However, some studies have shown that with the gradual improvement in human maintenance and management, urbanization has a positive impact on urban vegetation [17][18][19].In China, although UAs are rapidly expanding, the urban gross primary productivity (GPP) is also increasing at the same time [20].Moreover, some studies have confirmed a positive impact of urbanization on vegetation growth and the annual trend of urban GPP is even slightly higher than that of non-urban GPP [21][22][23].In some African arid regions like Tunisia, Morocco, and Sudan, vegetation enhancement also has been reported [4].
Coarse-resolution data will obscure the dynamic changes that occur within a city during the longterm process of urban expansion [8,24,25].Several data points collected over time intervals will make it hard to capture the transitional changes and details of urban-related characteristics [26].Various studies have shown that fine-scale data are more suitable for monitoring dynamic changes in urban vegetation [27][28][29], while the uncertainty of coarse-resolution products in multiscale validation can significantly reduce the reliability of validation results [24,25].Most current studies use three or fewer time points to measure urban characteristic changes [24], but studies rely heavily on short-term, coarse-resolution datasets will neglect the need for continuous monitoring [30].Especially in oasis cities, the distribution and growth changes of vegetation are constrained by local natural and human conditions [31][32][33].And, coarse data will neglect spatial details in narrow and small oasis regions, emphasizing that it is necessary to use fine-resolution data for arid urban research [10,12].
Urban vegetation experiences the impacts of both human and natural factors [7].The impact of urbanization caused by human activities on vegetation can often be divided into direct and indirect impacts [23].Direct impacts are typically negative, referring to losses due to land cover conversion [14,15,34].Indirect impacts result from urban management and climate (UMC) [7,17,23].The impact of urbanization on vegetation in arid regions has been a subject of growing concern [14].However, the impact of urbanization on vegetation growth remains unclear in arid regions due to the special geographic patterns of these areas [11,17,22].
In this study, we analyzed the urban spatial domain in four arid cities (Urumqi, Lanzhou, Yinchuan, and Hohhot) of China from 1990 to 2021 using 30 m land cover and VG data.By comparing VG in four arid cities and considering the temporal trend of urban expansion and spatial differentiation, we analyzed the relationship between urbanization and VG.Moreover, we explored the positive or negative relationship between urbanization and VG in arid regions.Our study can reveal the crucial link between urbanization and VG, and provide vital insights for researching the impact of urbanization on the regional ecology and climate in arid regions of China.

Study area
We used the arid region map from the United Nations Environment Programme World Conservation Monitoring Centre dataset (www.unep-wcmc.org/en) to define arid system boundaries [35,36].We also referred to the arid region of Northwest China, one of the three major natural zones in China.Additionally, we used precipitation data from 1990 to 2021 with a spatial resolution of 1 km from the Tibetan Plateau Data Centre (http://data.tpdc.ac.cn/) to identify the 400 mm precipitation line in China [12,37].Then, we finally selected four provincial capital cities in arid regions of China (Urumqi, Lanzhou, Yinchuan, and Hohhot) within the arid region (figure 1).

Data
To meet the long-term span and effectively identify the vegetation within the city, we used the China land cover dataset (CLCD) with a 30 m spatial resolution from 1990 to 2021 to study rapid urban expansion in arid regions (https://zenodo.org/record/5816591) [38].Compared to other products such as GLC_FCS30, Global30, and ESRI10, CLCD has a longer and more continuous time series [16,39].The VG data were extracted from Landsat NDVI products in Google Earth Engine (https://earthengine.google.com/), with a 30 m spatial resolution.Landsat NDVI was monthly maximum processing in pixel scale, and the greenest pixel means the pixel with the highest NDVI [40].
To separate the influence of natural conditions on urban vegetation growth, we classified desert vegetation areas as the natural vegetation area (NA) according to the degree of human intervention.Farmland, forestry, shrubs, and grassland were defined as seminatural vegetation areas (SA) [41].Water bodies, wetlands, ice, and snow were excluded due to their minimal urbanization impacts.Finally, three types were classified in the study area, namely UA, NA, and SA.We also resampled the coverage of 30 m impervious layer pixels to 300 m and calculated the percentage of the pixel number of the 30 m impervious layer pixels within a 300 m pixel [21].We used a fixed threshold of 0.1 to define the vegetation growing season based on previous studies [42][43][44].The beginning of the vegetation growth season is defined as the first time when NDVI is greater than 0.1 in spring, and the ending time is defined as when NDVI is less than 0.1 in winter.

Mann-Kendall test
Mann-Kendall test is widely used to evaluate the trend of time series changes and whether such trends are statistically significant, without being affected by a small number of outliers [41,42].
The test sequence is performed at a specific signal level.When |Z| > Z 1−a/2 , the null hypothesis is rejected, indicating the presence of a significant trend in the time sequence.Z 1−a/2 is obtained from the normal distribution table.Z 1−a/2 > 2.58 means extremely significant, 1.96 < Z 1−a/2 ⩽ 2.58 means significant, 1.65 < Z 1−a/2 ⩽ 1.96 means moderately significant, and Z 1−a/2 ⩽ 1.65 means not significant.
where x i and x j are the observations corresponding to the i and j of time series (i < j), sgn () represents the sign function, Z refers to standardized test statistics, n is the number of the time series, and t m refers to the range of any given tie m.

Theil-Sen median estimator
Theil-Sen median estimator, as a non-parametric method, has been proved to be highly reliable in estimating changes over time [45].Therefore, we used it to assess the intensity of trends in urban expansion and VG [16,46], and it can be calculated below: where β is the slope of the pixel regression equation, A i is the median VG in the i year, and n is the duration of the study.β > 0 represents an upward trend, β < 0 represents a downward trend, β = 0 represents no significant change.

Direct and indirect impacts
To systematically quantify the impact of urbanization on vegetation, Zhao et al [7] proposed a theoretical framework that decomposed urban VG into contributions from vegetated and non-vegetated land.The zero-impact line is defined by two characteristic vegetation NDVI values [11], corresponding to fully vegetated and fully UA, and it can be expressed as follows: where NDVI zi represents the theoretical NDVI of 30 m resolution pixels, while NDVI v corresponds to the NDVI of fully vegetated areas (UDI = 0, NDVI = NDVI v ), and NDVI nv corresponds to the NDVI of fully urbanized areas (UDI = 1, NDVI = NDVI nv ).When considering the indirect urbanization impacts on vegetation (depicting by ω i ) the changes in VG can be formalized as: where NDVI obs represents the observed NDVI of each pixel under different urban density index (UDI).
The direct impact of VG loss from land cover conversion can be expressed as NDVI v -NDVI zi .
The direct impact of urbanization mainly means the direct loss of vegetation caused by urban land cover conversion.Indirect impacts mainly include the impacts of UMC on vegetation.Indirect impacts can be measured by the relative change in NDVI compared with the zero-impact line (direct impact, VG directly linear decreases after vegetation covered areas are invaded by urban impermeable layers) [8,12].The calculation formula for the indirect impact ω i is as follows:

Land cover change in four cities in arid regions of China
The UA of the study area increased year by year, resulting in a gradual decrease in NA (figure 2).For individual cities, In Urumqi, UA was mainly concentrated in Xinshi and Midong Districts, with an increase of 72.  3).For individual cities, the area difference of NA and SA conversion between Urumqi and Lanzhou from 1990 to 2021 was relatively small, at −12.43 km 2 and 12.23 km 2 , respectively (figure 3(a)).The difference of Yinchuan was −627.42 km 2 , and a large amount of NA was converted into SA during the urbanization process.For Hohhot, the area difference between NA and SA converted into SA was 382.62 km 2 , indicating that a large amount of SA was converted to NA. Urumqi and Lanzhou, which are located in the east, were all constrained by both urbanization and natural climate conditions, in the area changes of SA and NA was little difference.From west to east, the stage of urbanization development is gradually accelerating, and urban expansion is also relatively fast.Therefore, a large number of NA was converted into SA in Yinchuan, and a large number of SA was converted into UA in Hohhot (figures 3(b)-(g)).In addition, the land cover converted into UA was respectively SA and NA in the two cities, which is mainly determined by the land cover types around the city.
The rapid increase in UA from 1990 to 2021 indirectly led to a decrease in NA and SA (figure 4).Overall, UA in the study area increased rapidly at a rate of 39.2 km 2 /a, and UA increased by 1123 km 2 from 1990 to 2021, expanding by 3.39 times (figure 4(e1)).The decrease rate of NA and SA were 6.8 km 2 /a and 36.8 km 2 /a, respectively (figure 4(e2-3)).As for individual cities, the NA in Urumqi and Lanzhou showed a trend of first decreasing and then increasing, with turning points in 2000 and 2010, respectively, while the SA showed a gradually decreasing trend.The NA in Yinchuan decreased the fastest at a rate of 23.2 km 2 /a.From 1990 to 2021, the NA decreased by 71.24%, while the area of SA slightly increased (slope = 9 km 2 /a) (figure 4(c3)).The NA in Hohhot decreased by 70.95% from 1990 to 2002 and remained relatively stable after that, with an overall decrease of 80.60% from 1990 to 2021.The area of SA slightly decreased, by 2.38% over the past 32 years.This indicates that the urbanization process was rapid, and urban expansion led to the disappearance of a large number of SA and NA.

VG change in four cities of China
As urban expansion proceeded rapidly, the VG gradually increased with a significant trend of increasing from west to east (figure 5(a)).However, although the climate zone within a city is basically the same, vegetation distribution and human impact within a city are different.Therefore, there is not a significant trend of increasing from west to east within a city.The VG in the four cities showed an increasing trend, while decreased greenness was mainly found in some western parts like Xinqu, Chengguan, and Saihan Districts (figures S1(a)-(d)).The proportion of pixels with significantly increased VG was highest in Urumqi, Lanzhou, Yinchuan, and Hohhot from west to east, accounting for 62.34%, 64.44%, 36.40%, and 27.03%, respectively (figure S1).The VG increased significantly in constant UA, with more than 80% of the study area showing an increasing trend (figure 5(b)).In the areas where UA was converted to UA, the increase ratio of VG was 88.46% and 75.06% for Lanzhou and Yinchuan, respectively (figure 5(b)).This also implies that while ensuring that UA remains unchanged, urban management measures and the local environment promote the increase of VG in UA.
To facilitate the comparison of interannual variation characteristics in VG across various LCs, we utilized a linear trend line that provided a more visually intuitive representation.The study area's VG increased annually (slope = 2.16 × 10 3 ) from 1990 to 2021 (figure S2).The VG of the UA and NA was similar in the early stages, but in later stages, the UA showed significantly higher VG than the NA (figure 6).This also verifies that when urbanization reaches a certain period, the improvement of urban management measures will promote the increase of VG in UA.Overall, the VG gradually increased from west to east, with Urumqi in the far west showing a lower VG and growth rate compared to Hohhot in the far east (figure S2).The VG of the UA was similar to NA before 2000.However, as urban expansion, the VG gradually surpassed that in NA, and VG rate significantly increased.This was particularly evident in Urumqi and Yinchuan, where the VG rate in UA exceeded that of both NA and SA, with an increasing positive impact from urban expansion on VG (figures 6(a)-(c)).Although the greening rate of Lanzhou and Hohhot was slightly lower than the rate of VG in the SA, at 2.30 × 10 −3 and 2.00 × 10 −3 , respectively, it was still higher than that of NA (slope = 1.37 × 10 −3 , slope = 0.38 × 10 −3 ) (figures 6(b) and (d)).Thus, all these indicate that the optimization of management measures in the later period of urbanization promoted the positive development of urban vegetation to a certain extent.

The impact of urbanization on VG 3.3.1. Comparative analysis of the growing season of UA and NA
The urban vegetation had a longer and earlier growing season, resulting in a significant increase in VG (figure 7 and table S2).The vegetation growing season had advanced from May to July in 1990 to March and earlier in 2021 (figures 7 and S3).Vegetation growth in UA started earlier and lasted longer than in NA (figure 8).From 2019 to 2021, Urumqi's vegetation growth period extended by 47% compared to 1990-1992, and SA's extended by 25% (table S3).In Lanzhou, the urban vegetation growth period from 2019 to 2021 extended by 33% compared with 1990-1992, which was also higher than the NA at 8% and the SA at 25%.In Yinchuan and Hohhot, the UA vegetation growth period was extended by 31% and 31% from 2019 to 2021, respectively, and the NA vegetation growth period was extended by 25% and 25% respectively (table S3).

The impact of urbanization on VG
VG in UA decreased with the increase of UDI, but the non-linear relationship was significant, and the positive impact of urbanization on VG gradually emerged (figure 9).The average VG of UA in the study area was 83% and 87% above the 'zero-impact line' , respectively (table S4).In Urumqi, the VG of UA exhibited a negative correlation with UDI in 1990, however, in 2021, it showed an inverted U-shaped relationship with the increase of UDI, indicating that due to the improvement of management measures after urbanization, densely populated UAs not only have no interference with vegetation transition, but also to some extent show that their environment is more conducive to vegetation growth (figure 9(a)).In Lanzhou and Yinchuan, the fitted dashed line in 1990 closely resembled the 'zero-impact line' , but in 2021, it gradually diverged from and surpassed the 'zero-impact line' , indicating that urbanization had a discernible positive impact on VG (figures 9(b) and (c)).In Hohhot, the VG is above the 'zero-impact line' when UDI is greater than 20% in 1990.However, by 2021, the same area showed a significantly higher VG compared to the 'zero-impact line' , indicating that this also showed that with the increase of UDI, VG did not decrease, but showed a non-linear increase (figure 9(d)).
Compared with the direct impact of urban expansion on VG, the indirect impact of effective vegetation reconstruction and management has a more significant positive impact on vegetation in arid regions  (figure 10).Overall, the indirect impact in the study area increased from 5.09% in 1990 to 9.49% in 2021, with a positive impact on vegetation rising by 4.40% (figure S4).The indirect impact in Urumqi and Lanzhou in 1990 was found to exhibit a superlinear relationship with the UDI.When the UDI was  below 70%, the impact was weak.However, as the UDI gradually increased, the positive impact rapidly increased, indicating that urbanization had a stronger positive impact on vegetation in densely populated UA than in sparsely populated ones (figures 10(a) and (b)).In Lanzhou and Hohhot, this positive impact increased with the increase of UDI and gradually stabilized.This positive impact peaked when Lanzhou's UDI reached 84% in 1990 and Hohhot's UDI reached 82% in 2021, respectively (figures 10(b) and (d)).This can be attributed to the steadily increasing positive impact of urban management.However, once a certain threshold is reached, this impact gradually becomes relatively stable.In Yinchuan, the positive impact gradually diminishes as UDI increases, but it still surpasses that of 1990 in 2021 (figure 10(c)).This indicates that in areas with better urbanization indirect impact can offset part of the direct impact   caused by UE, and there is a tendency to gradually increase.

Fine-resolution data can better reveal the impact of urbanization on vegetation in arid regions
The improvement and growth of urban vegetation is a common phenomenon [15,44,47].Urban expansion and urban vegetation growth in arid regions have undergone unique changes and more attention should be paid to cities in arid regions with fragile ecological environments [15,19,38].With the intensification of climate change in recent years, vegetation in arid regions is more sensitive to climate change than in other areas [48].However, the processes of global cities in urban expansion, population agglomeration, and urban greening are not consistent [49].There is a lack of research on small and medium-sized oasis cities in arid regions, especially in terms of refined small-scale and long-term series [32,33].Furthermore, continuous long-term time series data is a necessary condition for monitoring urban expansion [50][51][52].The limited time span may ignore some of the annual and intra-annual changing characteristics of urban vegetation in the ongoing urbanization process [26,30].Therefore, extending the time series span means that the dynamic changes of urban vegetation under human interference can be better observed [30].
Urban internal research requires obtaining more detailed information within a city [27,50,53].Coarse-resolution data can blur dynamic changes [54,55], especially in small to medium-sized developing cities [26].However, using coarse-resolution land cover data may lead to larger biases in smaller-scale research due to accuracy issues [29,56].Because the distribution of urban vegetation is relatively scattered and the area is relatively small, mixed pixels with coarser resolution tend to ignore limited urban vegetation patches, and the vegetation signals are not clearly recognized.Li et al [8] discovered that coarseresolution data can distort the spatial details of urban features such as the green infrastructure within or surrounded by urban land, leading to a significant underestimation of urban land areas, with 30 m resolution being the critical point.Based on our findings, the urban expansion in Yinchuan maintained a rate of 2.6 km 2 /a from 1990 to 2021, illustrating that land cover data with a coarse resolution like more than 1 km can lead to greater bias in small-scale studies [29, W Li et al 56].In this study, we choose land cover data with a spatial resolution of 30 m and the resolution is also the critical point [8].Furthermore, our results are generally consistent with ground experiments [57], which also showed that VG in UAs has been obviously enhanced [7,23].More consistently, our results on four arid cities are also like that of Cui and Liu's studies in China and global scale [21,47].Similarly, Zhang et al [11] also found that fine-resolution data are needed to observe the process of vegetation phenology models.Therefore, fine spatial resolution data with more spatial details is more suitable for monitoring urban vegetation [29,[58][59][60].
In addition, by comparing the results of the significance test, we also found that the statistical significance gradually decreased from west to east.In other words, from west to east, the VG of various land cover types gradually showed no significant linear trend, which may be due to natural and human factors like precipitation, radiation, and ecological engineering.Therefore, the phenomenon may indicate that vegetation change in arid areas is more easily monitored to a certain extent, and Nonlinear patterns may be more suitable for humid areas [47].Although urban vegetation is in the same climate zone, this spatial law from west to east has not been found within the city.From another perspective, this may be related to the fact that most vegetation is cultivated species and distribution is fragmented.Further study should be conducted with the support of more information and high-resolution data.

Comparing the VG in purely natural areas and UAs can better reveal the positive and negative impacts of UE
Vegetation in arid regions is usually limited by natural conditions like local climate [36, 61], especially rainfall [48].However, urbanization and vegetation are not simply linear relationships [7,47], and are affected by the superimposed effects of natural and human activities [58,59].Studies have shown that the response of vegetation productivity to environmental changes is now determined by multiple interactive factors, rather than just one dominant factor as in the past [46,48,62].Therefore, separating the natural vegetation growth in the context of climate change background can more intuitively analyze the impact of urbanization on vegetation [7,63].It is crucial to distinguish the impacts of natural conditions and urbanization on vegetation [63].However, commonly used control experiments may simulate and observe vegetation responses at the ecosystem scale, but their environmental conditions are often discontinuous and singular [63], which differ from natural growth environments of vegetation [64] and can interfere with understanding the positive or negative impacts of urbanization on vegetation [7,46].
If large-scale urbanization does not occur in arid regions without human activities, dense vegetation succession is in a natural state [12,32].Therefore, we divided the study area into UA, NA, and SA based on land cover types, and compared VG between UA and NA to isolate the impact of natural conditions on urban vegetation.Some studies have shown that NDVI gradually increased in global cities from 2000 to 2015, especially in arid regions of Asia (in Inner Mongolia, China, and northern India) [36].Our findings showed that urbanization had a positive impact on VG in UA, with significantly higher VG and enhancement trends than in NA.This is mainly because, in recent years, the construction of a large number of urban parks and green spaces and the continuous improvement of vegetation management and protection measures have made urbanization have a positive impact on promoting the growth of vegetation [23, 42,47].Our results also confirmed that the positive impact can offset part of the direct loss of urban vegetation caused by urban expansion, making arid UAs a suitable environment for vegetation growth (figure 9).In the past 30 years, the positive impact of urbanization on vegetation in the study area has also increased by 4.40% (figure S2).All these indicate that people will tend to seek a better living environment and thus perform good management of vegetation in the city.

Conclusion
Using continuous 30 m land cover and VG data from four arid cities (Urumqi, Lanzhou, Yinchuan, and Hohhot) in China from 1990 and 2021, this study examined the inherent impacts of urbanization on vegetation by separating natural influences.The results showed that the UA in the study area expanded rapidly at a rate of 39.2 km 2 /a, increasing 3.39 times from 1990 to 2021.During the same period, VG in the four arid cities increased year by year, with an annual slope of 2.16 × 10 −3 .Especially, VG in UA was significantly higher than that in NA, implying the positive impact of urbanization on vegetation growth to a certain extent.In addition, the duration of the urban vegetation growth period was longer than that of the NA, which also confirmed the positive impact of urbanization on the vegetation growth in arid cities.By analyzing the intrinsic relationship between urbanization and vegetation in arid cities of China, this study further demonstrated over time and 83% and 87% of average VG in UA were above the 'zero-impact line' (directly negative impact of urbanization on VG) in 1990 and 2021, respectively.Our findings enhance the understanding of the response of urban vegetation to urbanization process in arid regions.However, given this study area is in four arid cities in China, future studies should extend to wider regions to further extract the impact of urbanization on urban vegetation.

Figure 1 .
Figure 1.Study areas.Yellow represents the area within the 400 mm equivalent precipitation line.

Figure 2 .
Figure 2. Spatial changes in land cover in four arid cities of China from 1990 to 2021.(a)-(d) correspond to Urumqi, Lanzhou, Yinchuan, and Hohhot, respectively.

Figure 3 .
Figure 3. Area changes of urban expansion and other land cover type conversions in four arid cities of China from 1990 to 2021.(a) is the area of different land cover type conversions between 1990 and 2021.(b)-(g) are the converted area of different land cover types during 5 years, namely from 1990 to 1995, 1995-2000, 2000-2005, 2005-2010, 2010-2015, and 2015-2021, respectively.

Figure 4 .
Figure 4. Land cover change trends in four arid cities of China from 1990 to 2021 (a)-(e) represent Urumqi, Lanzhou, Yinchuan, Hohhot, and the whole study area (average of all the four cities), respectively.The numbers 1, 2, and 3 correspond to UA, NA, and SA, respectively.

Figure 5 .
Figure 5. Spatial distribution of average VG and land cover changes in four arid cities of China from 1990 to 2021.(a) is the annual average VG from 1990 to 2021, and (b) is the changes percentage of VG pixels in different land cover types.A1-A3 represents the conversion of UA, NA, SA to UA, B1-B3 represents the conversion of UA, NA, SA to SA, and C1-C3 represents the conversion of UA, NA, SA to NA, respectively.From left to right, there are Urumqi, Lanzhou, Yinchuan, and Hohhot, respectively.

Figure 6 .
Figure 6.Trends of VG changes in UA, NA, and SA in four arid cities of China from 1990 to 2021.(a)-(d) correspond to Urumqi, Lanzhou, Yinchuan, and Hohhot, respectively.

Figure 7 .
Figure 7. Changing trend of vegetation growth period in four arid cities of China from 1990 to 2021 and from 2019 to 2021.(a)-(d) correspond to the vegetation growth period changes in Urumqi, Lanzhou, Yinchuan, and Hohhot, respectively.

Figure 8 .
Figure 8. Changing of the growth periods in UA, NA, and SA in four arid cities of China from 1990 to 2021 and from 2019 to 2021.(a)-(d) correspond to Urumqi, Lanzhou, Yinchuan, and Hohhot, respectively.

Figure 9 .
Figure 9. Changing trends of VG and urban density index (UDI) in four arid cities of China in 1990 and 2021.(a)-(d) correspond to Urumqi, Lanzhou, Yinchuan, and Hohhot, respectively.The red and purple dashed lines represent the 'zero-impact line' , respectively, while the black and blue dashed lines represent the fitted curves in 1990 and 2021, respectively.The gray and green circles represent the NDVI values in 1990 and 2021, respectively.The UDI was divided into intervals of 2% to calculate the corresponding changes in VG.

Figure 10 .
Figure 10.Changing trends of the relative indirect impact in four arid cities of China, with respect to the UDI in 1990 and 2021.(a)-(d) correspond to Urumqi, Lanzhou, Yinchuan, and Hohhot, respectively.