Vegetation Index Reconstruction and Linkage with Drought for the Source Region of the Yangtze River Based on Tree-ring Data

Variations in vegetation are closely related to climate change, but understanding of their characteristics and causes remains limited. As a typical semi-humid and semi-arid cold plateau region, it is important to understand the knowledge of long term Normalized Difference Vegetation Index (NDVI) variations and find the potential causes in the source region of the Yangtze River. Based on four tree-ring width chronologies, the regional mean NDVI for July and August spanning the period 1665–2013 was reconstructed using a regression model, and it explained 43.9% of the total variance during the period 1981–2013. In decadal, the reconstructed NDVI showed eight growth stages (1754–1764, 1766–1783, 1794–1811, 1828–1838, 1843–1855, 1862–1873, 1897–1909, and 1932–1945) and four degradation stages (1679–1698, 1726–1753, 1910–1923, and 1988–2000). And based on wavelet analysis, significant cycles of 2–3 yr and 3–8 yr were identified. In additional, there was a significant positive correlation between the NDVI and the Palmer Drought Severity Index (PDSI) during the past 349 yr, and they were mainly in phase. However, according to the results of correlation analysis between different grades of drought/wet and NDVI, there was significant asymmetry in extreme drought years and extreme wet years. In extreme drought years, NDVI was positively correlated with PDSI, and in extreme wet years they were negatively correlated.


Introduction
As the main biomass of terrestrial ecosystems, vegetation plays a very important role in the energy conversion and material circulation of the Earth.Vegetation not only responds to climate change but also has feedback effects on it and is an indicator of global change (Piao et al., 2006;Jong et al., 2011).In recent decades, remote sensing data have been extensively used to study vegetation changes, with the development of remote sensing technologies (Tucker et al., 1986;Zhou et al., 2001;Hmimina et al., 2013).The Normalized Difference Vegetation Index (NDVI), determined by the red/near-infrared reflectance of remote sensing images, is a commonly used indicator of the growth status of surface vegetation, which can reflect the coverage, growth conditions, biomass, and photosynthesis of vegetation in a relatively accurate manner (Sellers et al., 1995;Tucker et al., 2001;Ju and Masek, 2016).In recent years, many scholars have conducted extensive studies on the variation characteristics and influencing factors of the NDVI around the world (Hlásny et al., 2015;Liu and Menzel, 2016;Mariano et al., 2018;Zhao et al., 2018;Butterfield et al., 2020).However, because the observation time of the NDVI data is relatively short (since the 1980s), our understanding of the long-term characteristics and influencing factors of vegetation growth and change remains limited.
As a climate proxy, tree-ring data has the characteristics of high accuracy, high resolution, strong continuity, and applicability over a wide geographical range (Cook et al., 2010;Büntgen et al., 2013;PAGES 2k Consortium, 2013;Ljungqvist et al., 2020).It is widely applied in studies of historical climate change, including for the reconstruction of temperature (Anchukaitis et al., 2013;Wilson et al., 2016;Chen et al., 2019a), precipitation (Fritts et al., 1971;Shao et al., 2010;Yang et al., 2014), runoff (Li et al., 2018;Chen et al., 2019b;2019c), and drought (Cook et al., 2010;Gaire et al., 2019).Recently, researchers found that there was a general positive relationship between annual growth rings and NDVI (Vicente-Serrano et al., 2016;Brehaut and Danby, 2018), and progress has been made in the reconstruction of NDVI data from tree rings.Malmström et al. (1997) first pointed out that NDVI-based Net Primary Productivity (NPP) was significantly correlated with treering width indices collected from Alaska.D' Arrigo et al. (2010) subsequently found a significant correlation between the NDVI and the maximum latewood density of tree rings in coniferous forests in the northern United States.Shishov et al. (2002) focused on the spatial correlation between the tree-ring index and NDVI in Siberia.Furthermore, Leavitt et al. (2008) found a significant positive correlation between the NDVI and δ 13 C in tree rings in the southwestern United States.There is evidence that different types of tree-ring measurement can reflect changes in NDVI (Kaufmann et al., 2004;Wang et al., 2004;He and Shao, 2006;Lopatin et al., 2006;Liang et al., 2009;Berner et al., 2011;Wang et al., 2017;Seftigen et al., 2018;Zhang et al., 2018).
The Yangtze River is one of the most important rivers in China.Its source region is located on the southeastern part of the Tibetan Plateau, which is a very important ecological barrier, and a region with sensitive and fragile ecological environments (Yao et al., 2004).Studies have obtained significant results regarding vegetation changes in the Source Region of the Yangtze River (SRYR) and surrounding regions (Zhang et al., 2010;Yang et al., 2011;Bai et al., 2012;Wang et al., 2016;Xia et al., 2019).However, there are no reports on longterm NDVI changes and their causes in the SRYR.Meanwhile, tree rings in this region have been widely used to reconstruct climate and runoff (Qin et al., 2003;Liang et al., 2008;Li et al., 2018;Xiao et al., 2018).In order to understanding the knowledge of NDVI variations of SRYR and finding the potential causes, treering width indices obtained from four tree-ring sampling points in the SRYR and its vicinity were used to explore their relationship with the NDVI, and NDVI variation sequences for July and August spanning the past 349 years were reconstructed to further analyze the variation in vegetation growth, and its correlations with drought and wet years, to predict ecological changes in the SRYR.This study provides a reference for predicting changes in the ecological environment of the SRYR.

Study area
The source region of the Yangtze River (Fig. 1) in the southwestern Qinghai Province is approximately 738 km from the east to west, and 406 km from the south to north, with an area of approximately 26.2 × 10 4 km 2 (Qian et al., 2006).The landform is dominated by high plains and hills, with an average elevation of more than 4000 m above sea level (a.s.l.).The monthly average temperature is above 0℃ only from June to September, with the lowest temperature in January (-16.5℃) and the highest temperature in July (6.0℃).The total monthly precipitation ranges from 2 mm (December) to 94 mm (July).Precipitation from May to September (months with relatively high temperatures) accounts for 88.7% of the annual precipitation, and the concurrent warm and wet climatic conditions are favorable for the growth and development of vegetation.The ecosystems in the source region of the Yangtze River principally include alpine steppe, alpine meadow, alpine marsh and wetland, localized and sparsely distributed alpine shrubs in some river valleys, and cushion plants and sparse vegetation on the rock slopes of the high mountains.The NDVI exhibits unimodal pattern, changes only slightly from January to March, and in November and December; therefore, almost no vegetation grows during these periods.It begins to increase in April, when the vegetation enters the rapid growth season, and peaks from July to August (0.37-0.38), when vegetation growth is most active.In September, the NDVI starts to decrease as the vegetation withers.It is evident that the vegetation growth in SRYR is concentrated from May to September, and this period is the main vegetation growth season in this region.

Tree-ring data and chronology development
We sampled tree-rings at four sites in SRYR in September 2013 (Fig. 1 and Table 1).All the trees were living in remote areas that are rarely influenced by human activities.We used growth cores with a diameter of 5.15 mm and obtained two or three cores from each tree.A total of 203 cores from four sites were used in this study.The COFECHA program (Holmes, 1983) was used after the collected samples were processed by dry-ing, pasting, polishing, and measuring.Cross-dating and verification of the results were performed to ensure accurate dating results.The data were detrended by a negative exponential curve using ARSTAN (Cook, 1985; http://www.ldeo.columbia.edu/tree-ring-laboratory/resources/software).The autoregressive model was then used to eliminate low-frequency variation caused by competition among trees within the forest.Eventually, we obtained three types of chronologies: Standard (STD) chronologies, Residual (RES) chronologies, and ARSTAN (ARS) chronologies.The RES chronologies remove the continuous influence of the special early physiological conditions on the following growth years, and therefore retain more high-frequency signals (Fritts, 1976).Meanwhile, NDVI is sensitive to climate conditions (Hlásny et al., 2015;Mariano et al., 2018;Butterfield et al., 2020), which leads to significant high-frequency variation.Therefore, the RES chronologies were used for reconstruction in this study.
The Mean Sensitivity (MS) and Signal-to-Noise Ratios (SNR) were employed to test whether the chronologies could reveal regional climate and environmental changes (Wang et al., 2016), the higher values suggest  the more climatic signals in tree-ring width.In general, MS > 0.1 and SNR > 5, which suggest the chronology could be used to reconstruct.Meanwhile, the Subsample Signal Strength (SSS) was used to evaluate the reliable starting year, an SSS > 0.85 was used to ensure a reliable chronology length (Liu et al., 2010;Bao et al., 2012).According to Table 1, the MS values of the four chronologies were 0.33, 0.31, 0.31, and 0.29, and the SNR values were 11.98, 18.89, 13.11, and 7.19, suggesting that the radial growth was sensitive to climate and environmental changes.And the reliable period of the four chronologies were 1665-2013, 1617-2013, 1568-2013 and 1553-2013, respectively.As shown in Fig. 2, although the longest distance between the sample sites is nearly 150 km, their variation characteristics are still highly consistent.The correlation coefficients between the chronologies in the common period (1665-2013, n = 349) were 0.395 (AS and BG, P < 0.0001), 0.437 (AS and YG, P < 0.0001), 0.518 (AS and XRS, P < 0.0001), 0.593 (BG and YG, P < 0.0001), 0.674 (BG and XRS, P < 0.0001), and 0.682 (YG and XRS, P < 0.0001).

Meteorological and NDVI data
Meteorological data, including the mean maximum temperature, mean minimum temperature, mean temperat-ure, precipitation, were provided by China Meteorological Administration, which is a monthly gridded dataset with a spatial resolution of 0.25° × 0.25°.The dataset is based on the observed data from over 2400 weather stations in China.The method of 'anomaly approach', which considering the change of climate factors with altitude, was applied in the dataset production.The details could be found in the reference Xu et al. (2009) and Wu and Gao (2013).The dataset covers the period 1961 to 2018.The Palmer Drought Severity Index (PDSI) data were from the Monsoon Asia Drought Atlas (MADA), which was reconstructed by Cook et al. (2010) using tree-ring data.The dataset covers the period 1300 to 2005, and has a total of 534 grid points and a spatial resolution of 2.5° × 2.5°.
The NDVI dataset provided by Global Inventory Modeling and Mapping Studies (GIMMS) is currently one of the most commonly used vegetation indices.This dataset eliminates the influence of temporal changes on volcanic eruptions, solar elevation angle, and sensor sensitivity (Fensholt et al., 2009), and can effectively represent regional changes in vegetation growth.GIMMS NDVI dataset 3 g.v1(https://ecocast.arc.nasa.gov/data/pub/gimms/3g.v1/) is a 15 d synthetic dataset from the National Aeronautics and Space Administration's (NASA's) Goddard Space Flight Center, with a spatial resolution of 1/12°, which is about 8 km.It is the longest NDVI dataset, spanning from 1981 to 2015.Monthly NDVI data were obtained using the maximum value composite (MVC) method (Holben, 1986).To exclude the effect of non-vegetation factors, pixels with a multiyear average NDVI smaller than 0.1 were excluded.The NDVI variation dataset of the SRYR can be obtained through preprocessing, such as format conversion, projection transformation, subset extraction, and cropping.

Methods
The relationships between tree-ring width, climate and then NDVI were analyzed using the Pearson correlations in the SPSS, and then a multivarate linear regression was employed to reconstruct the NDVI.The leaveone-out method (Michaelsen, 1987)  To analyze the characteristics of NDVI changes, we also used some mathematical statistical diagnosis methods including wavelet analysis (Torrence, 1998) and cross-wavelet analysis (Grinsted et al., 2004).

Relationship between tree-ring width, climate, and NDVI
According to Fig. 3, the relationship between the treering width chronology and climatic factors exhibits good consistency at the four sampling points, and the correlation between the tree-ring and precipitation was most significantly positive in May (r = 0.580; P < 0.05).
In terms of temperature, it exhibited significant negative correlations with the mean maximum temperature in May (the highest correlation coefficient was -0.543, P < 0.05), meanwhile, the chronology of AS and YG was significant negative correlated with the mean temperature in May as well.And, chronologies had a significant positive correlation with both the mean minimum temperature (the highest correlation coefficient was 0.435, P < 0.05) and mean temperature (the highest cor-relation coefficient was 0.436; P < 0.05) in July (except the tree-ring of AS).
Rather than a single climatic factor, the NDVI is jointly affected by various factors, such as precipitation, temperature, and evaporation (Peng et al., 2010;Sun et al., 2013;Wang et al., 2015).Table 2 shows that there were significant correlations (P < 0.05) between NDVI and climatic factors (mainly include the monthly average maximum temperature in March to July and September, the monthly average minimum temperature in June and September, the monthly average temperature in March, May to July and September, monthly precipitation in May) in the SRYR.Thus, it exhibited a better correlation with tree-ring, especially during the growth season.Table 3 shows the correlation between the tree-ring and NDVI for each month from 1982 to 2013.The tree-ring was significantly positively correlated with the NDVI of July and August at all four sampling sites with the highest correlation coefficient 0.612.The correlations reached significance at a level of 0.01, indicating that the tree-ring mirrored changes in the NDVI.Therefore, it is feasible to restore the NDVI data using the tree-ring chronology.

NDVI reconstruction
Based on the above results, because of the significant correlation between tree-ring width and the NDVI of July to August, the RES chronology established from the AS, XRS, YG, and BG tree-ring sampling sites over the period 1982-2013 was selected as the independent variable, while the average NDVI from July to August in SRYR was used as the dependent variable, to establish the following multiple regression equation: where P NDVI represents the regional average of the NDVI from July to August in SRYR, C AS , C BG , C YG and C XRS represents RES chronology from tree-ring sites AS, BG, YG, and XRS, respectively.The explained variance (R 2 ) in the reconstruction equation was 43.9%, and the adjusted explained variance was 35.9%.The correlation coefficient between the reconstructed sequence and the original measured sequence was 0.663 (n = 33, P < 0.01).
To verify the stability and reliability of the reconstruction equation, it was validated using the leave-one-  4. The statistical characteristics, including the correlation coefficient (r = 0.663), and average value of the product (t = 3.797), reached significance at a level of 0.01.Reduced Error (RE = 0.521) is positive, which indicates that the reconstruction equation is stable and reliable.The first-order sign test reached significance at a level of 0.05, while the sign test failed to reach significance at a level of 0.05, indicating that the reconstruction equation performed better in the reconstruction of high-frequency variations than low-frequency variations.

Tree-ring growth and climate
The change of NDVI is controlled by many factors, and climate is one of the main factors.Most studies show that temperature and precipitation was the most important factors (Ding et al., 2007;Wang et al., 2015).However, in different regions, the effects of temperat-ure and precipitation are different.The results of previous indicated that the temperature played more important roles (Tang et al., 2006) in three-river source region.
In our research, it also exhibited a better correlation between NDVI and temperature.This may be related to the vegetation, climate and landform of the study area (Tang et al., 2006).Our samples were taken from SRYR, an area located in the eastern part of the Tibetan Plateau.During the early growth season, the significant positive correlation between tree radial growth and precipitation, and the significant negative correlation between tree radial growth and temperature, may indicate that an increase in temperature can result in enhanced evaporation and plant transpiration, thereby exacerbating the water stress effect and inhibiting the growth of trees, while an increase in precipitation could be conducive to tree growth.Similar conclusions have been confirmed in previous studies (Qin et al., 2003;Gou et al., 2008;Shi et al., 2010).However, during July, the positive correlation between tree radial growth, precipitation, and temperature, suggests that concurrent rain and heat are conducive to photosynthesis, resulting in wide rings (Liang et al., 2008).There is a high statistical correlation between the tree ring width chronology and NDVI.The radial growth of trees is related to the climatic conditions in the growing season.NDVI reflects the growth status of vegetation and is also related to the climatic and environmental conditions, indicating that there is consistency in the response of the two to the limiting factors.

Linkage between the NDVI and PDSI
The growth conditions of vegetation are constrained by a variety of factors, among which climate is one of the most important.Among all climatic factors, drought is considered to be of key influence on vegetation.The PDSI was calculated based on precipitation, temperature, soil moisture, etc., and is representative of the regional extent of drought and wet conditions (Palmer, 1965;Dai et al., 2004).Increasing evidence suggests that NDVI series could capture the signals of drought events, especially extreme drought events (Wang et al., 2017;Zhang et al., 2018).As shown in Fig. 6, there were significantly positively correlated in SRYR and its vicinity between the reconstructed NDVI and PDSI during the period 1665-2005, with the correlation coefficient between the mean PDSI and the reconstructed NDVI of the SRYR was 0.611, which reached significance at a level of 0.001.The comparison with other hydro-climatic reconstructions suggests low NDVI period often matched drought period.For example, the low NDVI period 1910-1923 was matched the 1910s-1920s drought periods which widely occurred in central and western China (Bao et al., 2012;Yang et al., 2014;Li et al., 2018).To specifically examine the relationship between drought and NDVI, cross-wavelet analysis was used to show the coherence and phase lag between the NDVI and PDSI (Fig. 7).The result indicates significant resonance between the NDVI and PDSI at scales of quasi 4 yr, quasi 8 yr, and quasi 32 yr.The resonance of quasi 4 yr was found in 1650-1700, 1750-1800, 1860-1980, and 1950-1970.The resonance of quasi 8 yr occurred in 1740-1760, 1830-1850, 1940-1960, and 1980-2000.There was a quasi 32 yr resonance in 1900-1950.Furthermore, a nearly horizontal rightward arrow was in resonance of quasi 4 yr and quasi 8 yr, suggesting that the NDVI and PDSI signals were in phase.A right-downward arrow with an approximately 30° phase angle in the quasi 32 yr resonance indicated that the PDSI led the NDVI change by 1/8 cycle, which is approximately 4 yr.This result indicates that overall, the degree of wet and drought in the SRYR will significantly positively impact the growth of local vegetation, with wetter conditions leading to better vegetation and vice versa.However, an ensuing question is what the impact of different levels of drought and wetness on the NDVI is?To answer this question, the wet and dry years were firstly defined according to the mean and Standard Deviation (SD) of the PDSI series (Table 5).During the period 1665-2005, the number of normal years was the highest, 252 yr, accounting for 73.9% of the total; the  number of drought (wet) years was 20 yr (27 yr), accounting for 5.9% (7.9%) of the total; the number of extreme drought (extreme wet) years was 25 yr (17 yr), accounting for 7.3% (5.0%) of the total.The frequency of extreme drought years was higher than that of extreme wet years, while the frequency of wet years was higher than that of drought years.Then, the NDVI and PDSI in years corresponding to different drought and wetness grades were extracted for correlation analysis, and the results (Table 6) show that different grades of drought and wet exhibited distinctly different impacts on the NDVI: in normal years, the PDSI and NDVI had significant positive correlations, and the correlation coefficient reached 0.465, indicating that better water conditions resulted in better vegetation, and vice versa.In the dry and wet years, there was still a positive correlation, although it failed to pass the significance test, suggesting that in drought or wet years, the water condition was no longer the key limiting factor.The cause of this phenomenon requires further investigation and discussion.
In the extreme drought years, the PDSI and NDVI were positively correlated, with a correlation coefficient of 0.429, and the correlation reached significance at a level of 0.05, indicating that the water condition was one of the most important limiting factors in extreme drought years.However, in extreme wet years, the PDSI and NDVI exhibited significant negative correlations, and the correlation coefficient reached -0.681, reaching significance at a level of 0.01, which was possibly because the study region has a cold plateau climate; the average temperature in summer is less than 10℃, and the temperature difference between sunny days and cloudy days is typically large.Generally, extreme wetness means more precipitation and long-term rainy days, which may lead to lower temperatures and less sunshine, thus limiting vegetation growth.Excessive precipitation can lead to waterlogging, reducing the oxygen content of the soil and thus affecting photosynthesis, causing vegetation to wither (Nakano, 2007;Ashraf and Harris, 2013).

Conclusions
The analysis of correlations between the tree-ring width chronology, NDVI, and climatic factors of the SRYR, indicates that the tree-ring width chronology is most significantly correlated with the NDVI of July and August.Thus, we developed a stable and reliable multiple regression equation to reconstruct the summer NDVI in the SRYR over the past 349 yr.During this period, there were eight growth stages and four degradation stages.The growth stages were 1754-1764, 1766-1783, 1794-1811, 1828-1838, 1843-1855, 1862-1873, 1897-1909, and 1932-1945; the degradation stages were 1679-1698, 1726-1753, 1910-1923, and 1988-2000.There were significant cycles of 2-3 yr and 3-8 yr, indicating that vegetation growth might be affected by large-scale ocean-atmospheric circulations, such as PDO and ENSO The reconstructed NDVI and the PDSI of the SRYR and its vicinity were significantly positively correlated, but the correlation showed significant asymmetry in extreme drought years and extreme wet years.In extreme drought years, the PDSI and NDVI were positively correlated, whereas in extreme wet years, the PDSI and NDVI had a significant negative correlation.This paper provides a long time scale data of local vegetation changes, and reveals that the vegetation growth in SRYR is restricted by the large spatial climate.Also, it was found that there was significant asymmetry in extreme drought years and extreme wet years.This will provide support for future study on the dynamic consistency of vegetation growth.However, this study is still limited because out reconstruction is only based on four sampling sites.Therefore, it is critical to develop a wide tree-ring network in China, even in the whole northern hemisphere.It could help to verify the relative study results and to better understand the impact of climate change on ecological environment.

Fig. 1
Fig. 1 Location of the tree-ring sample sites at the source region of the Yangtze River.AG, XRS, YG, BG represent the tree-ring sites' ID

Fig. 2
Fig. 2 Tree-ring width Residual (RES) chronologies of the source region of the Yangtze River, and number of cores.The year with Subsample Signal Strength (SSS) > 0.85 is marked by an arrow.AG, XRS, YG, BG represent the tree-ring sites' ID

Fig. 3
Fig.3Correlation analysis between the four tree-ring chronologies and climatic factors from October of the previous year to September of the current year during 1962-2013 in the source region of the Yangtze River.AG, XRS, YG, BG represent the treering sites' ID

Fig. 4 Fig. 5
Fig. 4 The changes of reconstructed Normalized Difference Vegetation Index (NDVI) during 1665-2013 and comparison between the observed and reconstructed NDVI during 1982-2013.The period with grey (white) rectangle was degradation (growth) periods

Fig. 6 Fig. 7
Fig. 6 The correlation coefficient between Normalized difference Vegetation Index (NDVI) and Palmer Drought Severity Index (PDSI) from 1665-2005 around the source region of the Yangtze River

Table 1
Information of the four tree-ring sampling sites in the source region of the Yangtze River

Table 2
Correlation between monthly average or growth season average normalized difference vegetation index and climate factors during 1982-2015 in the source region of the Yangtze River

Table 3
The relationship between the four tree-ring chronologies and monthly normalized difference vegetation index during 1982-2013 in the source region of the Yangtze River * , ** suggests significance at a level of 0.05 and 0.01, respectively

Table 4
Statistical parameters of the transfer function and cross-validation for the Normalized Difference Vegetation Index (NDVI) reconstruction during 1982-2013 in source region of the Yangtze River ** the significance at a level of 0.01; a, b: Numbers of identical signs required to reach significance at a level of 0.05 and 0.01, respectively teenth century, late 18 century, and early 19 century.

Table 5
The standard of wetness and dryness according to the mean and Standard Deviation (SD)

Table 6
Statistics for graded drought and wet years, and correlations between Palmer Drought Severity Index (PDSI) of drought/wet years and the corresponding Normalized Difference Vegetation Index (NDVI) during 1665-2005 in source region of the Yangtze River Notes: * , ** suggests significance at a level of 0.05 and 0.01, respectively