Changes in vegetation greenness and its response to precipitation seasonality in Central Asia from 1982 to 2022

Central Asia is the world’s largest azonal arid region, with strong seasonal precipitation patterns. Vegetation in this region is relatively sparse and extremely sensitive to climatic changes. However, long-term trends in vegetation in Central Asia are still unclear or even controversially recognized, hindering the assessment of climate change’s impact on regional sustainability. Here, we present the longest time series of vegetation index in Central Asia and investigated its response to precipitation seasonality from 1982 to 2022 by integrating normalized difference vegetation index data from the Global Inventory Monitoring and Modeling Studies and the Moderate Resolution Imaging Spectroradiometer. The results indicate a greening trend during 1982–2000 and a browning trend during 2000–2008. In contrast to previous studies, we detected a rapid greening trend during 2008–2022, largely resulted from a continuous warm-wet trend in Central Asia. In addition, strong spatial variation in vegetation is uncovered within the region, suggesting spatial differences in vegetation responding to contrasting precipitation seasonality. Under CMIP6 climate scenarios, spring wetting and summer drying are projected to prompt Central Asian vegetation change to a simultaneous greening south and browning north.


Introduction
Vegetation plays a crucial role in ecosystems by regulating the exchange of water, carbon, and energy between the land and the atmosphere (Piao et al 2020, Zhu et al 2022).Over recent decades, global and regional changes in vegetation greenness have been studied through field surveys and satellite observations (Zhu et al 2016, Chen et al 2019a).It is generally accepted that the Earth has been greening as a result of climate change, rising atmospheric CO 2 concentrations, and human activities (Piao et al 2019, Zhang et al 2022).Notably, there are strong spatial differences in vegetation greening, emphasizing the need to investigate regional vegetation change patterns.
Central Asia is the largest azonal arid region in the world, with fragile and sensitive ecosystems that are vulnerable to human activities and climate change (Bi et al 2020, Chen et al 2020).The Global Inventory Monitoring and Modeling Studies (GIMMS) data have been primarily used to study vegetation changes in Central Asia, but the findings have been inconsistent (Jiang et al 2017, Fan et al 2020Luo et al 2020, Yang et al 2022b).For example, Yang et al (2022b) concluded that mountain vegetation in Central Asia browned from 1982 to 2015, while (Luo et al 2020) suggested an opposite greening trend in the same period.Differently, (Fan et al 2020, Li et al 2021) suggested a greening trend during 1982-1998 and a browning trend during 1998-2015.Additionally, given that GIMMS data extend only until 2015, their time series is relatively short to discuss the long-term trend of vegetation change in Central Asia at present.A recent study extended the analysis of vegetation changes to 2020 by combining the Moderate Resolution Imaging Spectroradiometer (MODIS) data and GIMMS data, revealing a greening stagnation after 1994 (Hao et al 2022).However, numerous studies have found a continuous warmwet climate in Central Asia (Jiang et al 2021, Ren et al 2022, Yan et al 2022), which largely contributes to vegetation growth in arid regions, contradicting the previously described browning or stagnation phenomena.
Water condition is the most important environmental restriction factor of ecological environment in arid regions, and vegetation growth tends to be highly vulnerable to wet/dry transitions (Deng et al 2020, Yang et al 2022a).Previous research has shown that precipitation is a critical factor in controlling vegetation change in Central Asia (Luo et al 2020, Tai et al 2020), but the region exhibits a high degree of heterogeneity, with significant seasonal differences between the north and the south.Summer precipitation dominates in northern Central Asia, while winter and spring (WS) precipitation account for up to 80% of the annual precipitation in southern Central Asia (Chen et al 2011, Dilinuer et al 2021, Yao et al 2021).However, the heterogeneity of seasonal precipitation has been rarely considered, leaving a knowledge gap on how vegetation dynamics responds to contrasting precipitation seasonality in this arid region.
In this study, we present the longest temporal record of vegetation changes in Central Asia and effectively resolved the controversy surrounding its long-term trend by integrating the GIMMS and MODIS data.Furthermore, we investigated the impacts of precipitation seasonality on vegetation change in Central Asia for the first time, filling a regional gap in understanding the region's vegetation dynamics.The results provide a scientific basis for understanding, managing, and predicting the evolution of vegetation under the context of global climate change and its impact to regional sustainability.

Physical setting of the study area
Central Asia is situated in the core area of the Eurasian continent (45 The region is located far from the ocean, with a total area of 4 million km 2 , covering Kazakhstan, Uzbekistan, Kyrgyzstan, Tajikistan, and Turkmenistan (Su et al 2021).The population of the region is nearly 75 million people and is mainly concentrated around the Fergana Basin (Jing et al 2020).The region experiences a temperate continental climate, and its precipitation is primarily influenced by a westerly circulation (Chen et al 2019b(Chen et al , 2022)).The elevation increases from the lowlands in the northwest to the Tianshan Mountains in the southeast (Pechenkin and Petrov 2022), with a diverse vertical zonation including deserts, grasslands, forests, oases, alpine meadows, and ice and snow (figures 1(a) and (b)).
Based on eco-regional categorization (Dinerstein et al 2017), landscape type (Yao et al 2021), and altitude as guidelines, Central Asia can be divided into three subregions: northern Central Asia (NCA), southwestern Central Asia (SWCA), and southeastern Central Asia (SECA) (figure 1(a)).Together, SWCA and SECA can be referred to as southern Central Asia (SCA).The main focus of this paper is on the long-term vegetation features and the effect of precipitation patterns on vegetation within these three subregions.
Cropland and forest dominate NCA with an average multiyear normalized difference vegetation index (NDVI) of 0.35 (figures 1(c) and (d)).SWCA is characterized by sparse, poorly maintained vegetation and bare land.Most of the SECA is mountainous grassland, and its multiyear average NDVI is 0.3.Despite the differences in vegetation among the three subregions, they share a common growth season from April to October (figures 1(e)-(g)).
The average multiyear temperature in Central Asia is approximately 10 • C, and the average annual precipitation is approximately 250 mm.The temperature patterns are consistent throughout the region (figures 2(c)-(f)).However, the precipitation seasonality in Central Asia has significant spatial differences.Both southern subregions of Central Asia are dominated by cold season (winter-spring) precipitation, which accounts for approximately 80% of the annual precipitation (figure 2(b)).In contrast, the precipitation in NCA is concentrated in the warm season (mainly in summer) (figure 2(a)).

Data sources 2.2.1. Vegetation data
The NDVI is a commonly used proxy for vegetation greenness and photosynthetic activity (Yuan et al 2019).To evaluate the long-term changes in vegetation, we utilised NDVI data from two sets of sensors: the GIMMS third-generation NDVI product (NDVI3g) based on retrievals from the Advanced Very High-Resolution Radiometer (AVHRR), and the MODIS NDVI.The GIMMS NDVI3g dataset has a spatial resolution of 1/12 • and a half-month temporal resolution, covering the period from 1982 to 2015 (Pinzon and Tucker 2014).We selected three datasets from the MODIS products: MOD13Q1, MOD13A2, and MCD43A4, with temporal resolutions of 16 d (MOD13Q1 and MOD13A2) and 1 d (MCD43A4), to assess vegetation changes from 2000 to 2022.The MOD13Q1, MOD13A2, and MCD43A4 datasets have geographical resolutions of 250 m, 1 km, and 500 m, respectively (Beck et al 2006).We synthesized the half month NDVI to monthly values using the maximum value composite method (Holben 1986).To fill up missing values, the nearest neighbour interpolation is used.All MODIS products were processed on the Google Earth Engine platform.GIMMS data were used to assess vegetation changes prior to 2000, while MODIS data (mainly MOD13Q1) were used for changes after 2000.The original values of the two datasets were extracted, and their trend consistency was assessed from temporal (long-term change) and spatial perspectives (linear fitting of the data at each grid).
Because NDVI may be saturated in densely vegetated areas, we also included other satellite-based products to independently verify the robustness of the NDVI-based findings (Cortés et al 2021): the leaf area index (LAI), fractional vegetation coverage (FVC), gross primary production (GPP), solarinduced chlorophyll fluorescence (SIF) and vegetation continuous fields (VCF).The global land LAI and FVC maps for the period from 1982 to 2018 were derived from AVHRR with a temporal resolution of 8 d and a spatial resolution of 0.05 degrees (Liu et al 2012, Xiao et al 2016).The global OCO-2-based SIF and GPP datasets (0.05 • , annual) from 2000 to 2020 were generated using discrete OCO-2 soundings, MODIS remote sensing data, and meteorological reanalysis data (Li and Xiao 2019).MODIS was used to produce the VCF data for each year from 2000 to 2020 with a 250 m spatial resolution.The product provides a continuous, quantitative assessment of land surface cover based on three surface cover components: percent tree cover, percent nontree cover, and percent bare land (Beck et al 2006).Areas with bare land as the primary land cover type were not included in the analysis since they have little vegetation.The study focused on vegetation during the growing season from April to October.

Observed climatological data
The monthly temperatures from 1982 to 2022 were extracted from the Climate Research Unit (CRU) dataset at a resolution of 0.5 • × 0.5 • (Harris et al 2014).Additionally, we used the gridded monthly precipitation dataset from the Global Precipitation Climatology Centre (GPCC) for the same period, with a spatial resolution of 0.25 • × 0.25 • (Schamm et al 2014).The GPCC dataset was selected because it contains data from a larger number of stations and is more suitable for detecting long-term precipitation trends across Central Asia.We accounted for interannual and seasonal variations in precipitation, analysing spring (March-May), summer (June-August), autumn (September-November), and winter (December-February).

CMIP6 future precipitation simulations
For future precipitation simulations, we used the Community Earth System Model 2 (CESM2) series of model simulations and projections from the CMIP6 archive, which are provided by National Center for Atmospheric Research, USA.Based on the general performance considering annual precipitation pattern, annual cycle precipitation, categorical performance and long-term precipitation trend, CESM2 demonstrate performances superior to other precipitation models in Central Asia (Guo et al 2021, Jiang et al 2021).Based on monthly data availability, we analysed the historical simulations and projections for four scenarios (SSP1-2.6,SSP2-4.5,SSP3-7.0, and SSP5-8.5),accessed through the Earth System Grid Federation data portal (O'Neill et al 2017).Only the first realization (r1i1p1) was used.The original spatial resolution of each global circulation model was resampled to 0.5 • × 0.5 • using the bilinear method (Eyring et al 2016).The historical simulations' baseline period from 1995 to 2014 was referred to as the 'Present' , and the periods from 2021 to 2040, 2041-2060, and 2081-2100 in the various emission scenarios were labelled the 'near-term' , 'mid-term' , and 'long-term' , respectively (Jiang et al 2020).
Y Su et al

Other supplementary data
The land cover map was extracted from the FROM-GLC10 dataset (Gong et al 2019), which is a global land cover dataset with a 10 m resolution.The Shuttle Radar Topography Mission (SRTM) provided digital elevation models (DEMs) obtained on a near-global scale by an international research effort (Berry et al 2007).The SRTM V3 product (SRTM Plus) has a resolution of 1 arc-second (∼30 m).

Methods
A least-squares linear regression approach was used to detect changes in the long-term trends of various variables.Linear trend estimation was used to establish a linear regression relationship between variables (x i ) and time (t i ).The regression coefficient (y) represents the trend of variable (x i ), and n represents the number of samples The nonparametric Mann-Kendall (M-K) test is broadly used to explore the change trends of environmental monitoring data and possible abrupt change points (Yue andWang 2004, Yao et al 2019).In this study, the M-K test was employed to identify the change-point of vegetation variables.For the sequence X = (x 1 , x 2 , …, x i ), the detailed calculation steps are given as follows: (2) where n is the length of the dataset.The S k series is cumulative calculation in which the value of the ith time series is larger than the jth.The statistical variable UF k is defined as follows: where UF 1 = 0. Considering that the time series are independently reciprocal and have the same continuous distribution, the statistic S k is in an approximately normal distribution, with the mean and the variance as follows: where UF k satisfies the standardized normal distribution and is calculated according to the order x 1 , x 2 , …, x n .Given the significance level α and threshold value U α , if |UF k | > U α , this suggests that the X time series has a significant tendency.The range beyond the U α is identified as the time horizon where the mutation occurs.
Reverse the order x n , x n−1 , …, x 1 of X and repeat the above process.Meanwhile, let UB k = −UF k (k = n, n −1, …, 1) and UB 1 = 0.An abrupt change point is expected if an intersection exists between the sequence curves UF k and UB k .The abrupt change point is considered to pass the significance test when it falls between the 2 straight lines of U α and −U α .In this study, α and U α were 0.05 and 1.96, respectively.
To identify potential turning points in the NDVI time series, we also used a piecewise regression model to determine whether the trends in the NDVI changed during the study period (Yuan et al 2019).The formula is as follows: where y is the vegetation variable; t is the year; α is the estimated turning point of time series that defines the timing of a trend change; β 0 , β 1 , and β 2 are the regression coefficients; and ε is the residual of the fit.Least squares linear regression was used to estimate α and the other coefficients and p values less than 0.05 were considered significant.Additionally, the ensemble empirical mode decomposition (EEMD) method was utilized to investigate the spatial and temporal patterns of nonlinear vegetation change in Central Asia (Xu et al 2020).Empirical mode decomposition (EMD) is a signal decomposition method created by (Huang et al 1998) that aims to decompose original time series data (D(t) (t = 1, 2, 3, …, N − 1, N, and N is the length of the time series data)) into diverse components, including n intrinsic mode functions (IMF1, IMF2,…, IMFn) and a residual (Rn(t)).Each IMF is considered the result of divergent driving factors on different time scales.The specific steps of the EEMD algorithm are as follows: Step 1: In the nth trial, a new time series is generated by adding a white noise time series u n (t) to a given signal x(t), y n (t) = x (t) + u n (t) for n = 1, 2, …, N, where N is the ensemble number.
Step 2: Based on the original EMD, the noisecontaminated signal y n (t) is decomposed into a set of IMFs and a residual where M − 1 is the total number of IMFs resulting in each decomposition of y n (t), IMF (n) m is the mth IMF and r (n) m is the residual obtained in the nth trial.To obtain an equal number of IMFs in each decomposition, we used a fixed sifting number of 10 is used.
Step 3: Steps (1) and ( 2) are repeated for N trials, with a different white noise series u n (t) added to the original signal in each trial.
Step 4: The final IMF of the EEMD (IMF ave m ) is obtained by averaging the total m IMFs related to the N trials: The results of the EEMD depend on the choice of the ensemble number (N) and the amplitude of added noise (A).The relation ε = A √ N should be satisfied, where ε is the final standard deviation of the error calculated as the difference between the original signal and the sum of the IMFs resulting from the EEMD.
In order to investigate vegetation responding to precipitation seasonality, the Pearson Correlation was used to examine the correlations between vegetation and different seasonal precipitation values.All correlation coefficients are between −1 and 1, and negative correlation values indicate that there is an inverse relationship between the variables (Morris and Udry 1980).Double-sided p values were used to assess the significance of these correlations.The formula used is: where n represents the number of samples; x i and y i represent the observed values at point i corresponding to variables x and y, respectively; and x and ȳ represent the sample averages of x and y, respectively.

Vegetation greenness changes based on GIMMS and MODIS
GIMMS NDVI accurately reflects the changes in vegetation greenness from 1982 to 2015 (figure 3(a)).Using the M-K mutation test method, we assessed the long-term trend and turning points of NDVI for Central Asia (figure 3(b)).The UF parameter from the M-K test, which measures the shift in NDVI accumulation, indicates that GIMMS NDVI began to increase in 1982, peaked in 1994, and remained high until around 2000.It then decreased until 2008, before reversing (figure 3(b)).We show the spatial distribution of NDVI changes in the Central Asia growing season during 1982-1994, 1982-2000, 2000-2008, and 2008-2015, using the results of the trend test in figures 3(c)-(f), respectively.From 1982 to 1994, the greening trend was evident, with an area percentage reaching 95% (figure 3(c)).The area of greening between 1982 and 2000 also exceeded 80%, primarily in NCA and SWCA, with an increasing rate greater than 0.002 yr −1 (figure 3(d)).
Correspondingly, more than 90% of the regions displayed obvious browning during 2000-2008, with a browning rate greater than 0.002 yr −1 (figure 3(e)).Subsequently, from 2008 to 2015, NCA started to turn green again, primarily in NCA (figure 3(f)).In contrast, in the two southern subregions, there was both greening and browning, with no discernible patterns (figure 3(f)).
Since GIMMS NDVI data are only available until 2015, we used higher-resolution MODIS NDVI datasets to track changes in vegetation greenness between 2000 and 2022.To enhance the reliability of the trend analysis, we employed EEMD method, which successfully eliminated background noise and provided accurate assessments of the long-term trend of vegetation change.We primarily examined the change trends of MOD13Q1, MCD43A4, FVC, and LAI using EEMD method (figure 4).The four vegetation indicators in the EEMD exhibited good consistency in their modal shifts, displaying an overall parabolic trend of first decreasing and then increasing.The NDVI reached its lowest value around 2008 and then began to rise until it peaked in 2022 (figures 4(a) and (b)).The slopes of the NDVI changes derived from MOD13Q1 during 2000-2008 and 2008-2022 were −0.0017 yr −1 and 0.0022 yr −1 , respectively.
Based on the lowest value of the EEMD trend decomposition, we separated the entire time period into two sections: 2000-2008 and 2008-2022.We examined the spatial change patterns of several vegetation metrics during these two time periods (figure 5).The geographical patterns of five vegetation metrics (MOD13Q1, MCD43A4, FVC, SIF, and GPP) consistently revealed a continuous browning trend from 2000 to 2008.The vegetation in NCA underwent significant browning, with trends surpassing −0.005 yr −1 in MOD13Q1 and MCD43A4 (figures 5(a) and (c)).The SIF and GPP also captured clear signs of vegetation browning in the mountains in the SECA, indicating that from 2000 to 2008, most of the vegetation browning occurred in NCA and SWCA, where there was better vegetation cover (figures 5(g) and (i)).However, the browning trend in SWCA was relatively insignificant due to the low vegetation cover.From 2008 to 2022, all vegetation products exhibited a clear greening trend.Similarly, the vegetation tended to be greener in NCA and SECA, where vegetation cover was relatively good, especially in MOD13Q1 and MCD43A4, which have relatively high resolutions.The increases in FVC (>0.005 yr −1 ) in NCA were also significant (figures 5(b), (d) and (f)).

Characteristics of long-term vegetation greenness changes from 1982 to 2022
The long-term change trend of vegetation greenness in Central Asia during 1982-2022 was derived by integrating GIMMS and MODIS data (figure 6(a)).NDVI data from the two products on the same  1982-1994, 1982-2000, 2000-2008, and 2008-2015, respectively.monthly scale from 2000 to 2015 (figure 6(b)) show that they have the same fluctuation patterns (figure 6(a)) with a slope of 1.09, R 2 = 0.9 (figure 6(b)).The consistency of the two datasets in Central Asia is also supported by (Hao et al 2022).When NDVI is less than 0.15, there is a little bias between GIMMS and MODIS data due to the low vegetation cover.Notably, the integration of GIMMS and MODIS NDVI in Central Asia indicates a significant greening (0.0018 yr −1 ) from 1982 to 1994 that persisted until around 2000, followed by a significant browning from 2000 to 2008 (−0.0017 yr −1 ), and then a greening trend from 2008 to 2022 (0.0022 yr −1 ).
We further calculated the change slope of NDVI for Central Asia and its three subregions over four time periods based on the long-term trend of NDVI, and used other vegetation variables to verify this trend.First, we found that the interannual variation in vegetation greenness in the three subregions was consistent with that in the whole region (figure 7).They all showed early gains (1982-1994), sustained high values (1994-2000), followed by declines (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008), and then a subsequent increase (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020)(2021)(2022).In all three subregions, greening before 2000 and after 2008 was highly significant (p < 0.05).Second, SIF and FVC trends were in agreement with the long-term NDVI trend, indicating the consistency of various vegetation variables in Central Asia and attesting to the reliability of NDVI trend data.For example, FVC showed a prominent positive slope before 2000, particularly in NCA and mountainous regions of SECA that had comparatively lush vegetation.

Effects of seasonal distribution of precipitation on vegetation greenness
To investigate the influence of precipitation seasonality on vegetation, we divided vegetation changes during the growing season into early growing season (EGS) (April and May) and late growing season (LGS) (June-August).We further analysed changes in vegetation in the entire region and the three subregions on a seasonal scale during 1982-1994, 1982-2000, 2000-2008, and 2008-2022 (figure 8).The results showed that there was no uniform trend of seasonal vegetation change in the whole Central Asia, as vegetation change was more prominent in the EGS during 1982-2000and 2008-2022, and dominated in the LGS during 2000-2008. .However, the three subregions exhibited distinct seasonal variations.In NCA, the vegetation change was more noticeable in the LGS than in the EGS (figure 8(b)).In contrast, in SCA, the change was more evident in the EGS than in the LGS (figures 8(c) and (d)).This suggests that there is a significant seasonal difference in vegetation change in SCA and NCA and that the seasonal changes are not synchronized.
Previous research has shown that precipitation is the most effective factor controlling vegetation change in Central Asia (Luo et al 2020, Tai et al 2020).
In this study, we first investigated the impact of seasonality on vegetation.We used correlation analysis to explore the relationship between vegetation greenness and precipitation seasonality in the Central Asia subregions over four time periods.Specifically, we examined the correlations between vegetation in the growing season and spring, summer, autumn, and winter, as well as precipitation in winter and spring (WS) and summer and autumn (SA) (figure 9).The results indicate that vegetation changes throughout Central Asia were more related to WS precipitation, while autumn precipitation was not significantly correlated with vegetation change (figure 9(a)).Notably, summer precipitation in NCA was most significantly correlated with vegetation change, with a correlation greater than 0.5 in all four time periods (figure 9(b)).In the two SCA subregions, vegetation change was more highly correlated with WS precipitation.Despite this, summer precipitation in the mountainous region of SECA was also correlated with vegetation change (figures 9(c) and (d)).
In summary, the impact of precipitation seasonality on vegetation greenness in Central Asia varies by seasons.Vegetation greenness changes in SCA are more noticeable in spring, and closely related to WS precipitation.Notably, the influences of winter precipitation (pre-growing season precipitation) on vegetation is complex and exhibits a certain time lag, which can affect vegetation in growing season through soil water (Luo et al 2020, Yang et al 2022b).In contrast, in NCA, vegetation greenness mainly occurs in summer as it is dominated by summer   1982-1994, 1982-2000, 2000-2008 and 2008-2022.Statistically significant trends with p < 0.05 and p < 0.01 are indicated by one and two asterisks, respectively.precipitation.Precipitation in NCA and SCA has a distinct yearly cycle that is mainly modulated by the subtropical high and the meridional shift of the westerly jet stream (Schiemann et al 2008, Bothe et al 2012).Nevertheless, further research is still needed to understand the physical mechanism and ecohydrological process underlying this effect.: 1982: -1994: , 1982: -2000: , 2000: -2008: , and 2008: -2022. EGS and . EGS and LGS represent the early (April and May) and late (June-August) growing seasons, respectively.: 1982-1994, 1982-2000, 2000-2008, and 2008-2022.WS and SA represent the cumulated precipitation in winter and spring and the cumulated precipitation in summer and autumn, respectively.Statistically significant trends of p < 0.05 and p < 0.01 are indicated by one and two asterisks, respectively.Y Su et al SIF, providing reliable and comprehensive data on vegetation dynamics.Compared to global vegetation research conducted by (Zhu et al 2016), our study focuses on fine-scale changes in regional vegetation, emphasizing regional vegetation greenness changes that can better capture specific extreme events.Our findings are consistent with previous research, which revealed that vegetation greenness peaked in 1994 (Hao et al 2022) and maintained a greening trend until 2000 (Fan et al 2020).However, unlike the previous studies that found that the vegetation in Central Asia browning or greening stagnation after 1994 (Fan et al 2020, Li et al 2021, Hao et al 2022, Yang et al 2022b), this study identified a significant greening trend of vegetation after 2008.This change is consistent with the results of continuous warming and wetting in Central Asia after 2008 (Jiang et al 2021, Ren et al 2022, Yan et al 2022), which is further supported by the greenness change in cultivated land (Aitekeyeva et al 2020).
The reasons for these findings are twofold.First, previous studies mainly used only GIMMS data to investigate vegetation changes in Central Asia (Fan et al 2020, Yang et al 2022b).However, GIMMS data only go up to 2015, failing to capture the significant greening trend that has occurred in recent years.As shown by the yellow line in figure 6(a), integrating MODIS data reveals significant greening.Second, previous research found that precipitation is the primary controlling factor for vegetation in Central Asia (Luo et al 2020, Tai et al 2020), and a severe drought occurred in Central Asia in 2008 (Aitekeyeva et al 2020), causing NDVI to reach its lowest level since the 1980s.Both the M-K mutation test and the EEMD method found that 2008 was a turning point for vegetation changes in Central Asia (figures 3(b) and 4), starting a subsequent greening trend in the region.
Greening and browning trends in arid Central Asia, where vegetation is sparse and dominated by grassland and shrubs, reflect changes in vegetation productivity, biodiversity, and the degree of desertification, which further affects, for example, the frequency of dust storms and the transformation of carbon sources and sinks (Piao et al 2019, Luo et al 2020).Furthermore, vegetation activity is significantly constrained by water conditions, showing high sensitivity to precipitation anomalies, which can capture the occurrences of extreme events such as drought, and provide a critical information for understanding the region's vegetation feedback to future climate change.

Prediction of future precipitation seasonality and its possible effect on vegetation greenness
Based on the comprehensive analysis of the influence of precipitation seasonal distribution on vegetation greenness, we conducted further investigations on the future precipitation seasonality in NCA, SWCA, and SECA under four scenarios (SSP1-2.6,SSP 2-4.5, SSP3-7.0, and SSP5-8.5)derived from the CMIP6 models (figure 10).Compared to the historical simulations between 1994 and 2015, we observed a projected increasing trend of precipitation in WS and a decreasing trend in summer across all three subregions under different future emission scenarios (figure 10).The present data showed two peaks in the periods from March to May and from October to December in SCA and a peak in June in NCA (figure 10, black lines).The changes are more significant in the long term based on the higher emission scenarios.In addition to the changes in the amount of precipitation, a shift in the first peak was found in NCA for SSP2-4.5,SSP3-7.0, and SSP5-8.5.The peak during May-July gradually shifted to a March-May peak in the near-term, mid-term, and long-term periods.This resulted in precipitation seasonality in NCA becoming similar to that of SCA (Jiang et al 2020).Therefore, in the future, under different emission scenarios, unlike the current summer precipitation dominant pattern, spring precipitation in NCA will significantly increase and become the dominant season for precipitation.For SWCA and SECA, the changes in precipitation in different seasons do not alter the previous precipitation distribution.
Studies have shown that under the lowest emission (SSP-2.6)and the highest emission (SSP-8.5)scenarios, the future vegetation LAI of Central Asia would be higher than the historical simulations level, which is mainly affected by regional climate change (Luo et al 2020, Piao et al 2020).However, changes in the growth cycle of vegetation such as phenology may be affected by seasonal variation in precipitation (Gao et al 2019).According to the results presented in section 3.3, vegetation greenness changes are mainly controlled by summer (WS) precipitation in NCA (SCA).Therefore, a significant increase in WS precipitation in the future will promote the simultaneous greening of vegetation in the two southern subregions.Conversely, as NCA transitions from summer rainfall-dominated to spring rainfall-dominated precipitation, we expect vegetation in NCA to brown in the summer, but there also may be a gradual shift in peak vegetation greening from summer to spring.
Although the simulation in future scenarios provides likely trajectories of vegetation changes in Central Asia, we have to realize the high uncertainty in the simulation sourced in assumption conditions, model parameterizations, and uncertainty rooted in the emission scenarios (Jiang et al 2020).Additionally, the factors considered in the models may not be exhaustive, and other factors, such as human activities, could also affect the simulation.Therefore, despite the preliminary vegetation changes revealed by the study, it is necessary to further improve the model and incorporate future factors to provide a more accurate projection, contributing to a more comprehensive understanding of the region's sustainable future.
As a result, the previous controversy regarding the vegetation greenness was largely due to the short duration of the research period.The combination of multiple vegetation parameters can effectively describe long-term changes in vegetation greenness in arid regions, and this approach may also be applicable to other regions.This study has significant scientific value for understanding vegetation growth and its relationship with variations in precipitation, as well as for predicting the future development of Central Asia.

Figure 1 .
Figure 1.Overview of the study area.(a) Geographical location and elevation distribution.(b) Land cover types.(c) and (d) Show the multi-year average states of interannual and growing season (April-October) vegetation, respectively.(e)-(g) show the monthly normalized difference vegetation index (NDVI) distributions in northern Central Asia (NCA), southwestern Central Asia (SWCA), and southeastern Central Asia (SECA), respectively.

Figure 2 .
Figure 2. Characteristics of the climate in Central Asia.(a) Ratios of summer and autumn precipitation to annual precipitation; (b) Ratios of winter and spring precipitation to annual precipitation.(c)-(f) show monthly temperature and precipitation distributions for Central Asia, NCA, SWCA, and SECA, respectively.

Figure 3 .
Figure 3. GIMMS NDVI trend (a) and MK test of the trend (b) in Central Asia from 1982 to 2015.UF represents the statistical series of the standard normal distribution, and UB represents the reverse statistical series.The green dashed line represents p values equal to 0.05.(c)-(f) Show the spatial distributions of NDVI trends in Central Asia during the growing seasons of1982-1994, 1982-2000, 2000-2008, and 2008-2015, respectively.

Figure
Figure EEMD decomposition process and trend test of four vegetation parameters: MOD13Q1 NDVI (a), MCD43A4 NDVI (b), FVC (c), and LAI (d).The red lines represent the final trends after EEMD decomposition.

Figure 6 .
Figure 6.Changes of NDVI in Central Asia obtained from GIMMS (1982-2015) and MODIS (2000-2022); the grey line represents the trend-smoothing curve of the two standardized datasets (a).Comparison of NDVI data consistency at the same time each month from 2000 to 2015 between GIMMS and MODIS (b).The density of the point is indicated by its colour; the denser the data, the deeper the red colour.

Figure 9 .
Figure 9. Correlation analysis between seasonal precipitation and growing season vegetation over all of Central Asia (a), NCA (b), SWCA (c), and SECA (d) during four periods:1982-1994, 1982-2000, 2000-2008, and 2008-2022.WS and SA represent the cumulated precipitation in winter and spring and the cumulated precipitation in summer and autumn, respectively.Statistically significant trends of p < 0.05 and p < 0.01 are indicated by one and two asterisks, respectively.