Impact of vegetation leaf area change on global terrestrial water storage in recent 30 years based on GRACE and reconstruction data

Vegetation change has an important impact on land water cycle by changing transpiration and other water exchange between land and atmosphere. Terrestrial water storage (TWS) is an important component of global water cycle and freshwater resources. However, the impact of vegetation change on terrestrial water storage under the background of global climate change is still unclear. Based on the GRACE satellite observed and GRACE-REC reconstructed global terrestrial water storage data, this study investigated the impact of global vegetation leaf area change on terrestrial water storage in recent 30 years. The results show that there is a significant positive correlation between leaf area index (LAI) and terrestrial water storage in the demand-limited region. The sensitivity of TWS on LAI change is high mainly in Australia, central and southern Africa, South Asia, Mediterranean region, western United States, southern South America and other regions with high temperature and low precipitation, and the analysis of GRACE-REC shows the sensitivity in demand-limited region has an increasing trend. Compared with climate factors such as temperature and precipitation, the TWS trend caused by LAI is nearly the same, and has the same sign (all positive or all negative) as that of originally TWS in about 63.6% global land area, and the LAI-related TWS trend is high in the region with annual average precipitation of 500-1000mm. In the six different global land cover classes, the sensitivity of TWS to the LAI change is much higher in semi-arid, grass cop, sparsely vegetated regions, and LAI plays an important role in the interannual variations of TWS in semi-arid, grass cop regions. This study emphasizes the important role of vegetation change in the land water cycle, which is of great significance to the management and utilization of water resources in the future, especially in the arid and semi-arid regions.


Introduction
Land vegetation change alter CO2 absorption, evaporation and land surface albedo, and then the exchange of carbon, energy and water flux between land and atmosphere, has an important impact on climate (Chapin et al., 2008;Bonan et al., 2008). Vegetation transpiration, as the largest amount of land evaporation (Jasechko et al., 2013), plays an important role in regulating the global land water cycle, and the evaporation change caused by vegetation greening plays a fundamental role in the interaction between vegetation and land water cycle (Zeng et al., 2018a).
Vegetation change has an important impact on the components of land water cycle.
Studies have shown that the recent increase trend of global evaporation is mainly driven by vegetation greening (Zeng et al., 2018b). Observation studies have also pointed out that compared with climate change, the impact of large-scale vegetation change on global runoff change cannot be ignored (Wei et al., 2018), although precipitation is the main factor that determining runoff change, the extension of vegetation growing season also induced significant regional runoff change (Geng et al., 2020). Model studies also show that global vegetation greening has an important impact on global and regional water cycle components such as evaporation, precipitation, runoff and soil moisture (Zeng et al., 2018b;Yue Li et al., 2018).
In addition, the effect of increasing CO2 concentration on vegetation also alter the land surface water cycle. The increasing CO2 concentration increases the vegetation leaf area and transpiration, and also increases the water use efficiency of vegetation which may reduce the impact of drought caused by climate change, these two effects are offset. Studies show that the combined effect of the two effects has an important impact on the land surface water cycle (Lemordant et al., 2018;Cheng et al., 2017;Hong et al., 2019). Observational studies indicate that the response of vegetation in semi-arid and semi-humid areas of Australia to the increase of CO2 concentration in the past 30 years leads to decrease of evaporation, increase of runoff (Ukkola et al., 2016), and the decrease of base flow (Trancoso et al., 2017). The model results show that the vegetation response to rising CO2 concentration reduces transpiration and cloud cover, and the dry boundary layer reduces the possibility of rainfall and increases the frequency of drought (Skinner et al., 2017(Skinner et al., , 2018. The projection of the earth system model also shows that as the continuous increase of CO2 concentration, the vegetation leaf area and water use efficiency change may play an important role in evaporation and runoff change in the future (Lemordant et al., 2018;Mankin et al., 2018;Kooperman et al., 2018;Fowler et al., 2020;Cui et al., 2020).
Terrestrial water storage includes groundwater, soil moisture, surface water such as rivers and lakes, snow and ice, plays an important role in the global water cycle and freshwater resources, and the change of terrestrial water storage is increasingly concerned (Wang et al., 2020;Wang et al., 2018;Cao et al., 2019). The terrestrial water storage is mainly affected by precipitation, evaporation, runoff, snowmelt and other land surface water cycle components (Humphrey et al., 2017;Rodell et al., 2018;Mankin et al., 2019;Cuthbert et al., 2019;Wu et al., 2020;Zhang et al., 2019), some studies have assessed the significant impact of climate change or internal variability of large-scale climate systems such as ENSO on terrestrial water storage (Ni et al., 2018;Anyah et al., 2018;Thomas et al., 2019). Although terrestrial water storage change is dominated by precipitation, many studies show that glacier retreat, lake expansion, groundwater exploitation and some other human activities also have a great impact on terrestrial water storage (Xu et al., 2019;Tao et al., 2020;Ndehedehe et al., 2020;Chaudhari et al., 2019).
In addition, some studies have revealed the impact of vegetation change on terrestrial water storage, and found a interaction between groundwater and vegetation in global vegetation covered areas (Koirala et al., 2017), and the increase of vegetation leaf area reduces the tendency of dryland soil and groundwater drying (Deng et al., 2020;Han et al., 2020).
Considering the important impact of vegetation on land surface water cycle, and the facts that vegetation atmospheric feedback exists widely in the world (Green et al., 2017) and there has been vegetation greening in a large area of the global land in recent terrestrial water storage is worthy of further study. Due to the large uncertainty of land surface model and earth system model in the simulation of terrestrial water storage and vegetation atmosphere feedback (Jensen et al., 2019;Green et al., 2017), satellite observations of terrestrial water storage are very important to further constrain the response of the land water cycle to climate change. With the availability of GRACE (Gravity Recovery and Climate Experiment) global satellite observed terrestrial water storage data (Tapley et al., 2019), the impact of global vegetation leaf area change on terrestrial water storage can be analyzed.
This study aims to investigate the impact of vegetation changes on global terrestrial water storage revealed by satellite observations. The temporal and spatial responses of terrestrial water storage to global vegetation change were investigated by using the vegetation leaf area index (LAI), GRACE terrestrial water storage (TWS) data observed by satellite, and GRACE-REC the reconstructed long time series of terrestrial water storage data. This paper investigates the following issues: (1) spatial distribution and temporal variation of the sensitivity of terrestrial water storage to vegetation leaf area change.
(2) relative influence of change of vegetation leaf area on terrestrial water storage, compared with temperature and precipitation. (3) the influence of vegetation in different land cover areas on terrestrial water storage, find out the regions where vegetation has most significant impact.

Compute the sensitivity of TWS to LAI change
The sensitivity of TWS to LAI change was computed as the partial derivative that results from a multiple linear regression (1), relating the annual averaged terrestrial water storage change (∆ ) to annual averaged LAI (∆ ), annual averaged precipitation (∆ ), annual averaged temperature (∆ ), and the annual cycle of all variables has been removed. The calculation excludes the influence of temperature and precipitation on TWS, and the independent effect of LAI was obtained. In addition, the method assumes that each factor is independent of each other and does not consider the influence of other factors such as human activities.
In order to investigate the long-term change of TWS sensitivity to LAI change, (1) is used to compute the sensitivity of 7-year and 12-year sliding intervals of GRACE-REC to LAI change from 1982 to 2016. And to reduce the statistical error, the sensitivity of each grid point is calculated in the 3x3 grid area with the grid point as the center.
Using the sensitivity of TWS to LAI change and the local LAI trend, the TWS trend caused by LAI change ( ) is computed as (2),

Demand-limited, Supply-limited region and Land cover classes
Referring to Forzieri (2020), the correlation between latent heat flux (LE) and temperature (T), and correlation between LE and precipitation (P) are compared to divide global land into demand-limited region (cor(LE,P)<cor(LE,T)) or supplylimited region (cor(LE,P)>cor(LE,T)), and the latent heat flux, precipitation and temperature data are form ERA5.
Land cover maps are retrieved from MODIS(MCD12C1， https://lpdaac.usgs.gov/products/mcd12c1v006/), referring to Ahlström (2015), We defined six land cover classes covering the global land area: tropical forest, extra-tropical forest (boreal and temperate), semi-arid ecosystems, tundra and arctic shrub land, grasslands and crops, and sparsely vegetated. and Australia (Fig. 1a), while the TWS was significantly higher in most of the above regions, central and Eastern Europe and central and eastern North America when LAI was high (Fig. 1b), indicating that there may be a positive correlation between vegetation leaf area and terrestrial water storage in these regions, which is generally consistent with the spatial distribution of correlation between TWS and LAI (Suppl Further analysis the relationship between the sensitivity of TWS on LAI change and the median of local annual mean temperature and annual total precipitation, it shows that during 2002-2016 the sensitivity is high in low precipitation regions, especially in high temperature and low precipitation regions (Fig 4a), and there are some anomalous high sensitivity in low temperature regions of middle and high latitudes. In the GRACE-REC, the sensitivity is high in high temperature and low precipitation regions (Fig 4b), generally consistent with the spatial correlation distribution in Fig1, which shows the TWS and LAI in high temperature and low precipitation regions such as Australia, southern South America and southern Africa are highly correlated.

Relationship
Based on GRACE-REC the long time series of reconstructed TWS data from 1982 to 2016, the temporal variation of sensitivity of TWS to LAI was further investigated. The moving average sensitivity of TWS on LAI over 7 and 12 years sliding were computed, and the trend of the moving average sensitivity of supplylimited and demand-limited regions were investigated. The results show that the moving average sensitivity of 7-year sliding in demand-limited region has an increase trend (Fig. 5a), and has no significant trend in supply-limited region, the overall global average sensitivity has a slight increasing trend. The trend of 12-year sliding moving average sensitivity is similar to that of 7-year (Fig 5b). These results indicate that the sensitivity of TWS to LAI changes in demand-limited region has increased, that is, the impact of LAI on TWS changes in demand-limited region were enhanced over recent 30 years, which means LAI may play a more important role in the regional TWS changes.

The impact of LAI on TWS trend
The sensitivity of TWS to LAI change and LAI trend are integrated to derived the trend of TWS caused by LAI (2). Fig 6b shows (Fig 3b), the trend of LAI is low in these areas, so the trend of TWS caused by LAI is not as significant as the sensitivity.
Compare with the original TWS trend, the LAI-related TWS trend is small, and there are some areas with negative TWS trend, indicating that temperature and precipitation may have an impact in these regions, Supplemental Fig 3b shows Fig 3a). The above results show that the impact of LAI on TWS trend is mainly concentrated in Australia, South Africa and other demand-limited regions.
The average original TWS trend and the TWS trend caused by LAI are generally all positive in most latitudes, and the magnitude are close, indicating that LAI has a great influence on TWS trend in most latitudes. Analysis the TWS trend caused by LAI and the local annual average temperature and annual total precipitation, the results differ from that the sensitivity of TWS to LAI is high in regions where annual total precipitation is less than 500mm (Fig 4b), but the LAI-related TWS is high in regions where the annual total precipitation is 500-1000mm (Fig. 7a) , which indicates that the semi-arid region with annual precipitation of about 500mm-1000mm may be the region where LAI has the most significant impact.
The relative effects of precipitation, temperature and LAI on TWS trend were compared (Fig 8). The results show that the TWS trend caused by LAI is generally same magnitude to that caused by temperature and precipitation, and both the increased LAI and precipitation increases TWS in many regions, while the increased temperature mainly decreased TWS. In the demand-limited region, although the sensitivity of TWS to LAI change is high (Fig 8b), the TWS trend in the region is small (Fig 8c), so the overall trend of TWS caused by the LAI is small, while the LAI trend in the supply-limited region is large, thus the overall trend of TWS caused by the LAI is large. The above results indicate that the impact of LAI on TWS is comparable with that of temperature and precipitation. LAI-related TWS trend and original TWS trend have the same sign (all positive or all negative) in 63.6% global land area, which indicates that LAI has a significant impact on global TWS trend.

Impact of LAI on TWS in different regions
In order to further distinguish the impact of LAI in different regions, the global land is divided into six regions (Fig 11): tropical tree, extra tropical tree, tundra shrub, sparsely vegetated, grass cross and semi-arid (Ahlstrüm et al., 2015). The results show that the TWS trend caused by LAI is large in three regions: gray cross, semi-arid and tunda shrub, the TWS trend caused by LAI is small in the sparsely vegetated region because the LAI change is very small in the region (Fig 9a), and the TWS trend caused by temperature and precipitation is also small in this region, indicated that the TWS may be mainly affected by other factors. In the grass crop and semi-arid regions, LAI change increase TWS, and the LAI-caused TWS change is larger than that caused by temperature and precipitation, which indicated that LAI played a significant role in these two regions, that generally consistent with the conclusion that LAI had a great impact in demand-limited region. In the tropical tree, extra tropical tree, tundra shrub and other supply-limited regions, the TWS trend caused by LAI are smaller than that caused by temperature or precipitation. In the tropical tree region, the warming decrease TWS strongly, which offset the positive impact of LAI in the region.
The sensitivity of TWS to temperature, precipitation and LAI show that the sensitivity of TWS on LAI is high in semi-arid, gray cross and sparsely vegetated regions, much higher than that in tropical tree, extra tropical tree and tundra shrub regions (Fig 9b). The sensitivity of TWS to precipitation is nearly close in each region, and relative larger in sparsely vegetated, extra tropical tree regions (Fig 9c).
The regions with high negative sensitivity of TWS on temperature are mainly in tropical tree, semi-arid, and grade cross regions (Fig 9d) (Fig. 10a, b), with the correlation coefficients of 0.45, although the interannual variations of original TWS is larger. In the sparsely vegetated region, although the sensitivity of TWS on LAI is high, the interannual variations of LAI is too small, and partly offsets the effect of LAI, which lead the interannual variations of LAI-reconstructed TWS quite different from that of original TWS (Fig. 10c), indicating the interannual variation of TWS in sparsely vegetated region may be dominated by other factors, such as precipitation.

Conclusions
In this study, the correlation between LAI and TWS, the sensitivity of TWS to LAI change, and the region where LAI has most significant impact on TWS are investigated, by using the LAI and GRACE TWS data from satellite and the GRACE- shows that the sensitivity of TWS to LAI is high in semi-arid, gray cross and sparse vegetated regions, much higher than other regions. Moreover, the interannual variations of TWS reconstructed by LAI in semi-arid and gray cross regions is much close to that of original TWS, indicates that LAI in these two regions has an important impact on TWS interannual variations. In conclusion, we emphasize the significant impact of LAI on TWS, and cannot be ignored compared to temperature and precipitation, also pointed out the regions where LAI have most significant impact, and emphasize the important role of land vegetation in management of land water resources in these areas in the future.

Funding
This study was funded by the Key Laboratory of Civil Aviation Flight Technology and Flight Safety of CAFUC research projects (FZ2020ZZ05), and Sichuan Science and Technology Program (2021JDRC0083).