Concurrent and Lagged E ﬀ ects of Extreme Drought Induce Net Reduction in Vegetation Carbon Uptake on Tibetan Plateau

: Climatic extremes have adverse concurrent and lagged e ﬀ ects on terrestrial carbon cycles. Here, a concurrent e ﬀ ect refers of grids ≥ 6 years) to recover their net carbon uptakes compared to grasslands. Therefore, our results emphasise that extreme drought events have stronger lagged e ﬀ ects on forests than on grasslands on the TP. The improved resilience of forests in coping with extreme drought should also be considered in future research.

and potential evapotranspiration, combining the Palmer Drought Severity Index (with changes in evaporation demand) and Standardised Precipitation Index (with sufficient spatial comparability and multitemporal nature) and considering drought sensitivity and global warming scenarios [39]. Currently, the meteorology sites on the TP, which are scattered because of the special terrain and complex ecosystems, together with the coarse spatial resolution of the Global SPEI database, have led to uncertainties in the TP drought characteristics; hence, drought monitoring data of improved resolution are urgently needed to capture drought characteristics on the TP.
In this study, we calculated the SPEI based on precipitation and temperature datasets with improved spatial resolutions, applied the Carnegie-Ames-Stanford approach (CASA) model to develop a net primary production (NPP) dataset across the TP during 1982-2015, and quantised the net reduction in vegetation carbon uptake (NRVCU) based on the NPP dataset. Our primary aims are to (1) investigate the spatiotemporal variabilities in the NPP, NRVCU, and SPEI; (2) identify the differences in NRVCU induced by the concurrent and lagged effects of extreme drought in forests, alpine steppe, and alpine meadow ecosystems; and (3) analyse the recovery time of vegetation net carbon uptake from extreme drought in different ecosystems.

Study Area
The TP ( Figure 1) is composed of a series of magnificent mountains on all sides. Encompassing approximately 25% of China's mainland, the TP primarily includes the regions of north-western Sichuan, Qinghai, and Tibet. As 'Asia's water tower' [41], the Plateau possesses a typical thermal/moisture gradient from the southeast to northwest due to the elevation and monsoonal winds, and it includes a range of climate-sensitive vegetation along the climate change characteristics, from subtropical and tropical forests to alpine deserts. Specifically, the annual mean total precipitation increases from 16 mm in the northwest to 1764 mm in the southeast, as observed during the period of 1981-2011, with a corresponding variation gradient in the annual mean air temperature increasing from −5.0 • C in the northwest to 15.5 • C in the southeast [42]. The climate type spatial pattern significantly affects vegetation productivity and is a pivotal element in determining the carbon cycle.
Remote Sens. 2020, 12, x FOR PEER REVIEW 3 of 19 a water balance model for precipitation and potential evapotranspiration, combining the Palmer Drought Severity Index (with changes in evaporation demand) and Standardised Precipitation Index (with sufficient spatial comparability and multitemporal nature) and considering drought sensitivity and global warming scenarios [39]. Currently, the meteorology sites on the TP, which are scattered because of the special terrain and complex ecosystems, together with the coarse spatial resolution of the Global SPEI database, have led to uncertainties in the TP drought characteristics; hence, drought monitoring data of improved resolution are urgently needed to capture drought characteristics on the TP. In this study, we calculated the SPEI based on precipitation and temperature datasets with improved spatial resolutions, applied the Carnegie-Ames-Stanford approach (CASA) model to develop a net primary production (NPP) dataset across the TP during 1982-2015, and quantised the net reduction in vegetation carbon uptake (NRVCU) based on the NPP dataset. Our primary aims are to (1) investigate the spatiotemporal variabilities in the NPP, NRVCU, and SPEI; (2) identify the differences in NRVCU induced by the concurrent and lagged effects of extreme drought in forests, alpine steppe, and alpine meadow ecosystems; and (3) analyse the recovery time of vegetation net carbon uptake from extreme drought in different ecosystems.

Study Area
The TP ( Figure 1) is composed of a series of magnificent mountains on all sides. Encompassing approximately 25% of China's mainland, the TP primarily includes the regions of north-western Sichuan, Qinghai, and Tibet. As 'Asia's water tower' [41], the Plateau possesses a typical thermal/moisture gradient from the southeast to northwest due to the elevation and monsoonal winds, and it includes a range of climate-sensitive vegetation along the climate change characteristics, from subtropical and tropical forests to alpine deserts. Specifically, the annual mean total precipitation increases from 16 mm in the northwest to 1764 mm in the southeast, as observed during the period of 1981-2011, with a corresponding variation gradient in the annual mean air temperature increasing from −5.0 °C in the northwest to 15.5 °C in the southeast [42]. The climate type spatial pattern significantly affects vegetation productivity and is a pivotal element in determining the carbon cycle.

Climate Datasets
In this study, we used the monthly precipitation sum and monthly mean air temperature to calculate the 12-month SPEI. Anusplin 4.2 (Centre for Resource and Environmental Studies, Australian National University, Canberra) was utilised to process the spatial interpolations of precipitation and air temperature at a 1/12 • resolution for the period 1982-2015. Furthermore, the precipitation, air temperature, and solar radiation data were utilised to estimate NPP; however, few stations contain measurement of solar radiation on the TP. We used the daily air pressure, air temperature, relative humidity, and sunshine duration to estimate the monthly solar radiation, and the specific methods and details were based on those of previous studies [43,44]. The grid data of the monthly solar radiation were generated based on the Kriging interpolation method in ArcGIS 10.2 (ESRI, Inc., Redlands, CA, USA); the data resolution and period were consistent with those of the precipitation and air temperature data. These meteorological datasets were acquired from the China Meteorological Administration (http://cdc.cma.gov.cn), and the number of meteorological stations used in this study is 107 ( Figure 1).

Aboveground Site Biomasses
During the summer (from the end of July to mid-August) in 2015, we clipped the aboveground parts of plants at 115 sampling sites on the TP (Figure 1). Of the sampling sites, 55 sites were from alpine meadow, 55 sites were from alpine steppe, and 5 sites were from alpine desert steppe ecosystems. At each site (10 m × 10 m), we set up three 50 cm × 50 cm paired plots to harvest aboveground plant samples. To obtain a constant mass for the aboveground biomass in the laboratory, the aboveground parts of the plants were oven-dried at 65 • C after sun-drying them in a field and weighing them to the nearest 0.1 g.

Satellite-Derived Datasets
Normalised difference vegetation index (NDVI 3g , 1982(NDVI 3g , -2015 data were used in this study to estimate the gridded TP NPP at a 1/12 • × 1/12 • resolution. The data were acquired from the Advanced Very High Resolution Radiometer observation instruments created by the Global Inventory Modeling and Mapping Studies group [45]. In addition, one satellite and atmospheric data-driven global annual NPP product, with a 0.05 • spatial resolution for the period 2000 to 2015, was obtained from the Numerical Terradynamic Simulation Group, University of Montana (http://files.ntsg.umt.edu/ data/NTSG_Products/MOD17/GeoTIFF/MOD17A3/GeoTIFF_30arcsec/) and used as a comparison for the NPP produced in this study. Here, the Moderate Resolution Imaging Spectroradiometer (MODIS)-derived NPP product was resampled (via the bilinear interpolation method) in a 1/12 • × 1/12 • grid to match the estimated NPP.

Soil Characteristics Database
To investigate the effects of soil physical and chemical attributes on the vegetation net carbon uptake, we used the soil characteristics database including alkali-hydrolysable N (mg/kg), available P (mg/kg), available K (mg/kg), cation exchange capacity (me/100 g), soil organic matter (g/100 g), pH (H 2 O), total N (g/100 g), and total P (g/100 g), which was developed for a land surface model that included eight vertical variation layers with a 30 arc seconds resolution and was produced by the Land-Atmosphere Interaction Research Group at Sun Yat-sen University (http://globalchange.bnu.edu.cn/research/soil2). In this study, we used the first layer (0-0.045 m). The soil moisture data were obtained from the National Earth System Science Data Center (http://www.geodata.cn/data/datadetails.html?dataguid=5864857&docId=1865) at a 5-km resolution, and this data was rebuilt by downscaling based on soil moisture data from the European Space Agency Climate Change Initiative.

Vegetation Distribution Map
The TP vegetation distribution map used in this study was extracted from the 1:1,000,000 China vegetation map, which was obtained from the Resource and Environment Data Cloud Platform (http://www.resdc.cn/data.aspx?DATAID=122) and reclassified into nine categories: grassland, shrubland, cropland, evergreen needle-leaf forest, deciduous needle-leaf forest, evergreen broad leaf forest, deciduous broadleaf forest, mixed forest, and other (water bodies, snow and ice, etc.). Furthermore, the spatial patterns of these major vegetation types (alpine meadow, alpine steppe, and forests) are displayed in Figure 1.

Drought Indices
The SPEI is a standardised drought index that represents a simple climatic water balance (precipitation and potential evapotranspiration) with comparable results for multifarious climate regimes [40]. Furthermore, the SPEI possesses a flexible timescale and allows the monitoring of different drought types including hydrological, agricultural, and meteorological droughts [46]. Hence, the SPEI is applicable for TP drought monitoring. Unfortunately, there is the disadvantage that the SPEI requires at least 30 years of data to be employed [47], whereas some developing countries do not yet possesses >30 years of ground observations. Additionally, the resolution of the existing SPEI dataset (provided by the Global SPEI database at a 0.5 • resolution) cannot satisfy the requirements for regional research. In this study, precipitation and potential evapotranspiration datasets from 1982 to 2015 were used to calculate the 12-month SPEI, and the detailed potential evapotranspiration calculation was based on previous studies using the air temperature data [39,48]. Here, we intend to explore the concurrent and lagged effects of extreme drought on the vegetation net carbon uptake. Therefore, we selected a 12-month time-scale because long-term extreme drought has a significant lagged effect on plants, and the potential evapotranspiration can be better captured through long-term observation [39]. The SPEI drought severity categories are displayed in Table 1, and the comparison between the existing and calculating SPEI is supplemented in Figure S1.

NPP Data Reconstruction and Accuracy Validation
In this study, the CASA model was used to estimate the gridded TP NPP at a 1/12 • × 1/12 • resolution [49][50][51]. The NPP is the product of absorbed photosynthesis active radiation (APAR) by the green vegetation and actual light use efficiency (ε).
where NPP(x,t) (gC m −2 ) denotes the vegetation NPP at pixel x at time t, and the APAR(x,t) (MJ m −2 ) and ε(x,t) (gC MJ −1 ) represent the APAR by the green vegetation and actual light use efficiency, respectively. The specific and detailed parameter calculation method was discussed in previous studies [50,52]. The validation can be executed either by comparing the estimated NPP with the gauged data NPP or with other modelling results. Here, the gauged NPP is obtained from the conversion of aboveground biomass in 2015: (1) the belowground biomass is calculated based on the ratio of belowground to aboveground biomass for alpine meadow (7.92), alpine steppe (4.25), and alpine desert steppe (7.89) ecosystems [53]; (2) the sum of belowground and aboveground biomass is the total biomass, which is multiplied by a coefficient of 0.45 to get gauged NPP with carbon unit [54]. To ensure the availability of the estimated NPP, several validation statistical indices were used between the gauged and estimated NPP (or MODIS-derived NPP) in 2015, and the specific formulas derived from previous research [55].

The Quantification of NRVCU
Drought events have a stronger inhibition on gross photosynthesis than total ecosystem respiration, which leads to an NRVCU. Here, the NPP was applied to denote the vegetation net carbon uptake, and the multi-year average of NPP was used as the normal net carbon uptake level of vegetation. Then, the multi-year average NPP was subtracted from the annual NPP in each grid, the negative value is the NRVCU [6], the zero and positive value denote no NRVCU and are assigned to 0. Furthermore, the net reduction rate of vegetation carbon uptake (NRRVCU) is captured by the ratio of NRVCU to NPP.

Grid Trends, Interannual Variabilities, and Anomalies of NPP, NRVCU, and SPEI
The temporal trends (from 1982 to 2015) of NPP, NRVCU, and SPEI in each grid are computed using a linear least square regression method according to that of previous research [56]. Additionally, the standard deviations of the annual NPP and NRVCU during the studied period were calculated for each grid using the methodology in Anav et al. (2015) [57]. The counts of extreme drought (SPEI < −2) for each pixel during the study period were processed based on ArcGIS 10.2 software. The interannual variabilities of NPP, NRVCU, and SPEI for each grid were based on the following equations from Ahlström et al. (2015) [58,59]: where a gt denotes the NPP (or NRVCU or SPEI) anomaly (long-term deviation) for the grid g at time t, and A t denotes the total anomaly of NPP (or NRVCU or SPEI) in the TP. Here, f g is the averaged relative value a gt /A t for grid g weighted by the absolute TP NPP (or NRVCU or SPEI) total anomaly |A t |, and it includes both the sign and extent of the local contribution with a range from −1 to 1.

The Response of Vegetation Net Carbon Uptake to Extreme Drought
Extreme drought complexly impacts the vegetation net carbon uptake; thus, an understanding of the probable mechanisms and processes involved in extreme effects is crucial. As Figure 2 illustrates, the vegetation net carbon uptake decreases during the extreme drought period (e.g., reduced productivity or increased mortality), which is called the concurrent effect. However, there can be no suppression in a region with strong drought resilience (such as this region with abundant soil water and nutrient substance). Afterwards, the vegetation net carbon uptake gradually recovers as precipitation returns over time. The duration of the recovery process for vegetation net carbon uptake is the recovery time (∆t in Figure 2). The damage magnitude caused by extreme drought determines the disparate recovery time and lagged regrowth. The NRVCU over the time following the year of an extreme drought was primarily due to the lagged effect (e.g., reduced bud development). In this study, the difference in NRVCU induced by the concurrent (during extreme drought period) and lagged effects (in the year following the extreme drought) and the recovery time were explored in alpine steppe, forest, and alpine meadow ecosystems, and the selected area exhibited no extreme drought events in the 10 years before/after the extreme drought year.
Remote Sens. 2020, 12, x FOR PEER REVIEW 7 of 19 time were explored in alpine steppe, forest, and alpine meadow ecosystems, and the selected area exhibited no extreme drought events in the 10 years before/after the extreme drought year.

Spatiotemporal Patterns of NPP, NRVCU, and SPEI on the TP
In this study, the regional NPP dataset was developed during 1982-2015, and its reliability is apparent. Validation statistical indices for the dataset were calculated between the gauged and estimated (or MODIS-derived) NPP. The mean error (155.72), relative bias (-48.77), Nash-Sutcliffe coefficient efficiency (0.24), root mean square error (278.01), and correlation coefficient (0.80) of the estimated NPP have a better performance than that of the MODIS-derived NPP (-205.26, 64.29, 0.03, 314.09, and 0.74) when compared to the gauged NPP ( Figure 3a). According to the regional NPP dataset, the NPP demonstrates a descendant gradient in the southeast-northwest direction, with the lowest value (< 100 gC m -2 year -1 ) in the northwest and the highest value (> 1400 gC m -2 year -1 ) in the southeast ( Figure 3b). The NRVCU gradient variation is consistent with that of NPP, with the highest values (> 90 gC m -2 year -1 ) principally occurring in the southeast and lowest values (< 10 gC m -2 year -1 ) being primarily distributed in the northwest (Figure 3c). During the study period, the annual total net reduction in vegetation carbon uptake of the TP decreased from 0.13 PgC in 1982 to 0.05 PgC in 2015, with a significant downtrend of −0.0018 ± 0.0002 PgC year -1 (p < 0.001, Figure 4a). Inversely, the annual total precipitation and annual mean

Spatiotemporal Patterns of NPP, NRVCU, and SPEI on the TP
In this study, the regional NPP dataset was developed during 1982-2015, and its reliability is apparent. Validation statistical indices for the dataset were calculated between the gauged and estimated time were explored in alpine steppe, forest, and alpine meadow ecosystems, and the selected area exhibited no extreme drought events in the 10 years before/after the extreme drought year.

Spatiotemporal Patterns of NPP, NRVCU, and SPEI on the TP
In this study, the regional NPP dataset was developed during 1982-2015, and its reliability is apparent. Validation statistical indices for the dataset were calculated between the gauged and estimated (or MODIS-derived) NPP. The mean error (155.72), relative bias (-48.77), Nash-Sutcliffe coefficient efficiency (0.24), root mean square error (278.01), and correlation coefficient (0.80) of the estimated NPP have a better performance than that of the MODIS-derived NPP (-205.26, 64.29, 0.03, 314.09, and 0.74) when compared to the gauged NPP (Figure 3a). According to the regional NPP dataset, the NPP demonstrates a descendant gradient in the southeast-northwest direction, with the lowest value (< 100 gC m -2 year -1 ) in the northwest and the highest value (> 1400 gC m -2 year -1 ) in the southeast (Figure 3b). The NRVCU gradient variation is consistent with that of NPP, with the highest values (> 90 gC m -2 year -1 ) principally occurring in the southeast and lowest values (< 10 gC m -2 year -1 ) being primarily distributed in the northwest (Figure 3c). During the study period, the annual total net reduction in vegetation carbon uptake of the TP decreased from 0.13 PgC in 1982 to 0.05 PgC in 2015, with a significant downtrend of −0.0018 ± 0.0002 PgC year -1 (p < 0.001, Figure 4a). Inversely, the annual total precipitation and annual mean During the study period, the annual total net reduction in vegetation carbon uptake of the TP decreased from 0.13 PgC in 1982 to 0.05 PgC in 2015, with a significant downtrend of −0.0018 ± 0.0002 PgC year −1 (P < 0.001, Figure 4a). Inversely, the annual total precipitation and annual mean temperature significantly increased at a rate of 1.0952 ± 0.3577 mm year −1 (P = 0.004, Figure 4c) and 0.0546 ± 0.0069 • C year −1 (P < 0.001, Figure 4d in vegetation carbon uptake was negatively correlated with the annual total precipitation (R 2 = 0.19, P = 0.006, Figure 4e) and annual mean temperature (R 2 = 0.60, P < 0.001, Figure 4f). Furthermore, precipitation and temperature demonstrated a significant positive correlation (R 2 = 0.16, P = 0.01, Figure 4g).
Remote Sens. 2020, 12, x FOR PEER REVIEW 8 of 19 temperature significantly increased at a rate of 1.0952 ± 0.3577 mm year -1 (p = 0.004, Figure 4c) and 0.0546 ± 0.0069 °C year -1 (p < 0.001, Figure 4d), respectively. During the studied period, the largest extreme drought area occurred in 2009 (2.8 × 10 5 km 2 , Figure 4b), whereas in 1996, 2005 and 2012, there were no extreme drought events on the TP. The interannual variation of the annual total net reduction in vegetation carbon uptake was negatively correlated with the annual total precipitation (R 2 = 0.19, p = 0.006, Figure 4e) and annual mean temperature (R 2 = 0.60, p < 0.001, Figure 4f). Furthermore, precipitation and temperature demonstrated a significant positive correlation (R 2 = 0.16, p = 0.01, Figure 4g). The NPP temporal trend, interannual variability (IAV), and standard deviation demonstrated a heterogeneous geographical distribution for the 1982-2015 period. Most of the area (70.86%) presented an increasing trend, and approximately 35% of the study region displayed a significantly increasing trend (p < 0.05, Figure 5a), with a high positive trend mainly in the southeast and low negative value chiefly in the central region. In addition, the alpine steppe ecosystem mainly demonstrated the positive NPP trend (75.7%, Slope > 0; Figure 5a and S2a). The IAV of most regions (91.03%, IAV > 0) was consistent with the TP NPP, with the maximum mainly distributed in the forest ecosystem (the average value is 0.007%; Figure 5b and S2b). Similarly, high values (> 180 gC m -2 year -1 ) of the NPP standard deviation principally occurred in the southeast, with the lowest values (< 20 gC m -2 year -1 ) occurring in the northwest (Figure 5c). We also found that approximately 55% of the NRVCU grids experienced a downward trend, and a significant region (p < 0.05)  (Figure 5c). We also found that approximately 55% of the NRVCU grids experienced a downward trend, and a significant region (P < 0.05) accounted for 17.86%, and the sharpest declining trend (Slope ≤ −2) was primarily concentrated in the southeast (Figure 5d). The alpine meadow and alpine steppe ecosystems appeared to contribute the most to the IAV of the NRVCU (45.66% and 15.89%, respectively; Figure S2e), which was predominately distributed in the eastern and central parts of the study area (Figure 5e). The NRVCU standard deviation exhibited the same pattern as that of the NPP (Figure 5f). Furthermore, we also measured the SPEI temporal trend, which exhibited a significant trend towards drought that primarily occurred in the Qaidam Basin and most forest ecosystems (P < 0.05, Slope < 0, Figure 5g). Most of the study region (87.92%) experienced a consistent IAV in the TP SPEI (positive value in Figure 5h). Of the extreme drought events in the TP (SPEI < −2), the Qaidam Basin exhibited the highest event frequency (count ≥ 3, Figure 5i).
Remote Sens. 2020, 12, x FOR PEER REVIEW 9 of 19 accounted for 17.86%, and the sharpest declining trend (Slope ≤ -2) was primarily concentrated in the southeast (Figure 5d). The alpine meadow and alpine steppe ecosystems appeared to contribute the most to the IAV of the NRVCU (45.66% and 15.89%, respectively; Figure S2e), which was predominately distributed in the eastern and central parts of the study area (Figure 5e). The NRVCU standard deviation exhibited the same pattern as that of the NPP (Figure 5f). Furthermore, we also measured the SPEI temporal trend, which exhibited a significant trend towards drought that primarily occurred in the Qaidam Basin and most forest ecosystems (p < 0.05, Slope < 0, Figure  5g). Most of the study region (87.92%) experienced a consistent IAV in the TP SPEI (positive value in Figure 5h). Of the extreme drought events in the TP (SPEI < -2), the Qaidam Basin exhibited the highest event frequency (count ≥ 3, Figure 5i).

The Concurrent and Lagged Effects of Extreme Drought Leading to an NRVCU
We further examined the NRVCU induced by the concurrent and lagged effects in the alpine steppe, alpine meadow, and forest ecosystems. A dominant influencing difference was discovered in the alpine steppe and alpine meadow ecosystems (p < 0.05). Of the grids, 83.01% and 55.43% experienced an NRVCU that was induced by the concurrent effect in the alpine steppe and alpine meadow ecosystems, respectively (Figure 6a,b). However, an NRVCU caused by the lagged effect

The Concurrent and Lagged Effects of Extreme Drought Leading to an NRVCU
We further examined the NRVCU induced by the concurrent and lagged effects in the alpine steppe, alpine meadow, and forest ecosystems. A dominant influencing difference was discovered in the alpine steppe and alpine meadow ecosystems (P < 0.05). Of the grids, 83.01% and 55.43% experienced an NRVCU that was induced by the concurrent effect in the alpine steppe and alpine meadow ecosystems, respectively (Figure 6a,b). However, an NRVCU caused by the lagged effect only accounted for 8.25% and 12.90% of grids in the alpine steppe and alpine meadow ecosystems, respectively (Figure 6a,b). For the concurrent and lagged effects in forests, there was no significant difference (P > 0.05) in the NRVCU. Approximately 67% of grids experienced an NRVCU caused by the concurrent effect. Similarly, the lagged effect led to an NRVCU for 64.03% of the grids (Figure 6c). Furthermore, the NRRVCU possessed the same characteristics as those of the NRVCU (Figure 6d-f).
only accounted for 8.25% and 12.90% of grids in the alpine steppe and alpine meadow ecosystems, respectively (Figure 6a,b). For the concurrent and lagged effects in forests, there was no significant difference (p > 0.05) in the NRVCU. Approximately 67% of grids experienced an NRVCU caused by the concurrent effect. Similarly, the lagged effect led to an NRVCU for 64.03% of the grids ( Figure  6c). Furthermore, the NRRVCU possessed the same characteristics as those of the NRVCU (Figure  6d-f). , and (f) denote the net reduction rate of vegetation carbon uptake (NRRVCU) of alpine steppe, alpine meadow, and forest ecosystems, respectively. Red and blue bars (lines) represent the concurrent and lagged effects, respectively. The asterisk (*) indicates a significant difference between the concurrent and lagged effects at p < 0.05.

The Vegetation Net Carbon Uptake Recovery from Extreme Drought
After the extreme drought event, the vegetation net carbon uptake gradually recovered, with different recovery times across vegetation types (Figure 7). The alpine meadow ecosystem possessed a higher proportion of grids (44.57%) with no suppression when faced with extreme drought, followed by forests (32.76%) and alpine steppe (16.99%). The alpine steppe ecosystem's net carbon uptake recovered faster than the other two types (alpine meadows and forests), with more than 75% of grids recovering from the extreme drought within 1 year (Figure 7a). Compared to the alpine steppe ecosystem, the alpine meadow required a longer time period to recover the vegetation net carbon uptake (Figure 7b), with more than 50% of grids returning to normal levels within 4 years. Regarding forests, only 39.61% of grids recovered from the extreme drought within 5 years (Figure 7c), and more than 25% of grids require at least 6 years to recover. (d,e,f) denote the net reduction rate of vegetation carbon uptake (NRRVCU) of alpine steppe, alpine meadow, and forest ecosystems, respectively. Red and blue bars (lines) represent the concurrent and lagged effects, respectively. The asterisk (*) indicates a significant difference between the concurrent and lagged effects at P < 0.05.

The Vegetation Net Carbon Uptake Recovery from Extreme Drought
After the extreme drought event, the vegetation net carbon uptake gradually recovered, with different recovery times across vegetation types (Figure 7). The alpine meadow ecosystem possessed a higher proportion of grids (44.57%) with no suppression when faced with extreme drought, followed by forests (32.76%) and alpine steppe (16.99%). The alpine steppe ecosystem's net carbon uptake recovered faster than the other two types (alpine meadows and forests), with more than 75% of grids recovering from the extreme drought within 1 year (Figure 7a). Compared to the alpine steppe ecosystem, the alpine meadow required a longer time period to recover the vegetation net carbon uptake (Figure 7b), with more than 50% of grids returning to normal levels within 4 years. Regarding forests, only 39.61% of grids recovered from the extreme drought within 5 years (Figure 7c), and more than 25% of grids require at least 6 years to recover. Remote Sens. 2020, 12, x FOR PEER REVIEW 11 of 19

NRVCU Spatiotemporal Patterns
The interannual changes in annual total net reduction in vegetation carbon uptake experienced a significant downward trend during the 1982-2015 period (Slope < 0, p < 0.001, Figure 4a). A correlation analysis demonstrated that the NRVCU gradually decreased as the annual total precipitation and annual mean temperature increased (Figure 4c,d). The favourable water and heat conditions, such as abundant soil moisture and mineral elements that are salutary for the vegetation net carbon uptake, provided important resources for vegetation growth [4,[60][61][62][63][64]. The soil water supply determines whether the vegetation photosynthesis operates normally when the CO2 concentration and light are sufficient [4,21,60]; thus, increased precipitation played a crucial role in ameliorating the vegetation net carbon uptake (Figures 4e and S3a). Additionally, soil water is an indispensable intermediary that guarantees the transportation of nutrients [60], which seems to support our findings regarding the NRVCU on the TP (Figure 4a). In addition to water, temperature is another climate variable affecting the vegetation net carbon uptake (Figures 4f and  S3b). The increase in temperature facilitates vegetation growth without encountering water limitations [59]. In addition, climate warming can stimulate the enzymatic activities of photosynthesis [60,65], with an evident promotion in the vegetation net carbon uptake. The increase in soil respiration as the climate warms is suitable for plant productivity due to the acceleration of mineralisation and the decomposition of organic matter [60,66,67], which explained why the NRVCU gradually declined (Figure 4a). Finally, an increased temperature may extend the length of the vegetative growing season [60,68], which is responsible for an increase in plant productivity, thereby restraining the NRVCU. In general, although a few studies have shown a downward trend of precipitation at individual meteorological station [69], the general improvements in the water and heat conditions visibly enhanced the capacity of the vegetation net carbon uptake on the TP (Figure 4a,c-g, Figure S3a-c).
The multi-year (1982-2015) average NRVCU on the TP presented a distinct geographical pattern (Figure 3c), with a gradient variation from the southeast (> 90 gC m -2 year -1 ) to northwest (< 10 gC m -2 year -1 ) that was highly consistent with the multi-year average NPP (Figure 3b). The variation pattern potentially reflects the climate variability affecting plant productivity ( Figure S4 and S5). The TP climate is primarily controlled by the East Asia monsoon [59,70] because the Himalaya mountains obstruct the Indian monsoon, together with the route of westerlies (Figure 1),

NRVCU Spatiotemporal Patterns
The interannual changes in annual total net reduction in vegetation carbon uptake experienced a significant downward trend during the 1982-2015 period (Slope < 0, P < 0.001, Figure 4a). A correlation analysis demonstrated that the NRVCU gradually decreased as the annual total precipitation and annual mean temperature increased (Figure 4c,d). The favourable water and heat conditions, such as abundant soil moisture and mineral elements that are salutary for the vegetation net carbon uptake, provided important resources for vegetation growth [4,[60][61][62][63][64]. The soil water supply determines whether the vegetation photosynthesis operates normally when the CO 2 concentration and light are sufficient [4,21,60]; thus, increased precipitation played a crucial role in ameliorating the vegetation net carbon uptake (Figures 4e and S3a). Additionally, soil water is an indispensable intermediary that guarantees the transportation of nutrients [60], which seems to support our findings regarding the NRVCU on the TP (Figure 4a). In addition to water, temperature is another climate variable affecting the vegetation net carbon uptake (Figures 4f and S3b). The increase in temperature facilitates vegetation growth without encountering water limitations [59]. In addition, climate warming can stimulate the enzymatic activities of photosynthesis [60,65], with an evident promotion in the vegetation net carbon uptake. The increase in soil respiration as the climate warms is suitable for plant productivity due to the acceleration of mineralisation and the decomposition of organic matter [60,66,67], which explained why the NRVCU gradually declined (Figure 4a). Finally, an increased temperature may extend the length of the vegetative growing season [60,68], which is responsible for an increase in plant productivity, thereby restraining the NRVCU. In general, although a few studies have shown a downward trend of precipitation at individual meteorological station [69], the general improvements in the water and heat conditions visibly enhanced the capacity of the vegetation net carbon uptake on the TP (Figure 4a,c-g, Figure S3a-c).
The multi-year (1982-2015) average NRVCU on the TP presented a distinct geographical pattern (Figure 3c), with a gradient variation from the southeast (>90 gC m −2 year −1 ) to northwest (<10 gC m −2 year −1 ) that was highly consistent with the multi-year average NPP (Figure 3b). The variation pattern potentially reflects the climate variability affecting plant productivity ( Figures S4  and S5). The TP climate is primarily controlled by the East Asia monsoon [59,70] because the Himalaya mountains obstruct the Indian monsoon, together with the route of westerlies (Figure 1), which determined the gradually reduced gradient of precipitation ( Figure S4a) from the southeast to northwest and thereby dominated the plant productivity patterns related to the climate gradient (Figure 3b). In this context, the vegetation net carbon uptake may possess a greater reduction in the southeastern TP when normal ecosystem functioning is restrained by external factors, which seems to support our findings (Figure 3c). Similarly, the finding that vegetation growth is easily affected by the TP climate is also supported by previous studies [70,71]. The tendency towards drought (Slope < 0 of SPEI, Figure 5g) is a determinant that led to the greater NRVCU in forests than in other ecosystem types (Figure 3c), while the same SPEI trend (Slope < 0, Figure 5g) in the Qaidam Basin did not induce the large NRVCU that can be attributed to lower plant productivity in this alpine desert region (Figures 1 and 3b). In terms of arid regions, plants in the northwestern TP possessed a lower photosynthesis rate, leaf area, species diversity, nitrogen content, growth rate, and the growing season length, which resulted in a weak net carbon uptake capacity of vegetation [72][73][74]. However, plant is equipped with an adaptive ability to grow in poor environmental conditions ( Figures S4, S6, and S7); thus, it experiences a lower NRVCU (Figure 3c) than that in a nutrient-rich environment (e.g., southeastern TP) when a drought occurs. Furthermore, the NRVCU variation trends presented an increase in forests and eastern regions but a decrease in the northwestern and southeastern TP ( Figure 5d); this was supported by the SPEI variation trends (Figure 5g). The IAV of NRVCU on the TP was primarily dominated by alpine meadow and alpine steppe ecosystems (Figures 5e and S2e), suggesting that the pattern of precipitation and temperature variability are driving the NRVCU for alpine steppe and alpine meadow ecosystems, respectively ( Figures S4a,b,d,e and S5a-d).

Response of the Vegetation Net Carbon Uptake to Extreme Drought
The ecosystems experienced different negative feedbacks to the extreme drought event. We found that the differences in NRVCU (and NRRVCU) between the concurrent and lagged effects induced by extreme drought were significant (P < 0.05) for the alpine steppe and meadow ecosystems, while forests exhibited the opposite trend ( Figure 6). Grasslands generally have a simpler structure and fewer species than forests [6]. The aboveground portion of the ecosystem is more prone to blights when encountering extreme drought [6,[75][76][77], but it can quickly recover as precipitation occurs and soil moisture replenishes [6,76,78]. There are also many annual plants in grasslands [79] that can typically develop without nutrient restrictions in the year following an extreme drought. These characteristics determined the lower NRVCU for the lagged effects and thus the significant reduction difference for vegetation net carbon uptake between the concurrent and lagged effects caused by extreme drought (Figure 6a,b,d,e). In comparison, the forests possessed a more complex structure and higher productivity [80], which increased their resilience to extreme drought because they can capture deeper underground water [4,81]. Although forests have a more resilient system than grasslands, their mortality can significantly increase when an extreme drought imposes stress that is greater than the acceptable threshold [4], thereby inducing a serious lagged effect after the extreme drought year. Therefore, the vegetation net carbon uptake may not quickly recover to the previous level due to forest dieback [4], fire disturbances [6], pest and pathogen breakouts [82], etc., which may explain why the discrepancy in NRVCU between the concurrent and lagged effects induced by extreme drought was inconspicuous (P > 0.05) in forests (Figure 6c,f).
Not only do the difference in NRVCU (and NRRVCU) between the concurrent and lagged effects induced by extreme drought but also the recovery time distinctly differed in alpine steppe, alpine meadow, and forest ecosystems, highlighting ecosystem characteristics related to the soil moisture [60], soil physical and chemical properties [4], nutrients [83], light, non-structural carbohydrates [6], etc. We found that the alpine meadow ecosystem possessed a greater proportion (44.57%) of the study region that exhibited no suppression during the extreme drought year (Figures 6 and 7), followed by the forest (32.76%) and alpine steppe (16.99%) ecosystems. The alpine meadow ecosystem is closely connected with the prevailing grassland one where climatic conditions are unfit for woody plants to exist and with the comparatively higher turnover rates than forests [4] that can be more resilient to extreme drought without nutrient restrictions (Figures S6 and S7). For the alpine steppe ecosystem in arid regions (e.g., the northwestern TP), vegetation encountered an insufficient nutrition and soil moisture supply [84] and the vegetation net carbon uptake was easily disturbed by the extreme drought, but it quickly returned to the previous level within 1 year (more than 75% of grids, Figure 7a) due to the its adaptive capacity for drought [4]. When comparing the recovery times of forests and grasslands, we found that forests required a longer time to recover their net carbon uptakes, and more than 25% of grids experienced the lagged maintain (recovery time ≥6 years, Figure 7c). The longer recovery time of vegetation net carbon uptake for forests is primarily attributed to the lagged effect caused by extreme drought; for example, the forest dieback during the extreme drought year led to decreased photosynthesis and lagged carbon emissions following the extreme drought [3,4,13]. Forests were also more assaulted by pest and pathogen breakouts than grasslands were [4,82], which severely restrained vegetation growth and caused an NRVCU [6]. In addition, the extreme drought altered carbon allocation, with more carbon being assigned to roots (or leaves) than to bole growth [82], which led to growth decline. The lagged effect of extreme drought can also cause the remnants of xylem cavitation that impair the water transport process and thereby restrict normal net carbon uptake rates of vegetation [82]. These lagged effects are related to plant growth, and they extend the recovery time of vegetation net carbon uptake.

Importance and Uncertainties
In our study, with the improvement in spatial resolution, the SPEI was considered to be the water balance between precipitation and potential evapotranspiration, and it measured the drought sensitivity magnitude across the TP during 1982-2015, which captured a credible drought spatiotemporal variability when compared with the sparse and uneven station distribution and insufficient ground-based observations [40,85,86]. For the SPEI quantisation process, the potential evapotranspiration was introduced to calculate the SPEI, and global warming scenarios were considered [39]. Additionally, the complicated terrain also affected the climate, especially the precipitation and temperature, which are indispensable datasets for calculating the SPEI. Fortunately, the Anusplin 4.2 was used to process their spatial interpolations related to terrain. Moreover, we adopted a light use efficiency model (i.e., the CASA model) to estimate NPP across the TP during 1982-2015, which can decrease uncertainties owing to extrapolations and interpolations and to mirror the vegetation growth statuses and ecosystem function changes and health conditions as the interference of the external environment [87]. Based on the NPP dataset, the NRVCU was quantified at the spatiotemporal scale, which revealed the net carbon flux variations of vegetation on the TP, with a reliable interannual variation being related to climate change. Finally, the difference between the concurrent and lagged effects induced by extreme drought on the NRVCU highlighted the different response mechanisms of vegetation both during and after the extreme drought events, which is salutary for future studies regarding the effect of climate change on vegetation net carbon uptake.
Despite progress being made via this study, large uncertainties remain. The SPEI calculation requires a long-term meteorological dataset of at least 30 years, and the optimal duration is 50-60 years [40]. We adopted a duration of 34 years from 1982 to 2015, which is shorter than the optimal duration and may lead to decreased accuracy. The SPEI possesses multiple timescale characteristics, and this study employed a 12-month timescale because the lagged effect requires a long-term interference of extreme drought and potential evapotranspiration that is well-monitored via a long timescale [39] but can ignore the replenishment of soil water in the growing season, which is important for vegetation growth across the TP. In terms of NPP, our product has greater NPP estimates for forests compared with the gauged (or MODIS-derived) NPP, and it may overestimate the NRVCU; meanwhile, the latest version of MODIS-derived NPP (v006) and more sampling sites of forest should be adopted to verify the feasibility of estimated NPP. In fact, human activities and natural mechanisms are vital factors to NPP [51]; thus, future research can consider the influences of these factors on NPP that could be an interesting topic. Additionally, the spatial interpolation method of solar radiation adopted the Kriging method, which could induce the indeterminacy to estimate NPP, and a machine learning algorithm should be considered in the future research. Moreover, the selected lagged effect primarily occurred in the year following the extreme drought event. A longer lagged effect on vegetation net carbon uptake and respiration should be explored for forests owing to their complicated structures and ecosystem functions. Finally, many factors could affect drought resilience, and future research should capture such information in detail.

Conclusions
In the present study, we adopted the SPEI based on meteorological data with ameliorative spatial resolution to measure drought magnitude across the TP during 1982-2015. The estimated NPP, based on a light use efficiency model (i.e., the CASA model), was employed to quantise the NRVCU, and the response of the vegetation net carbon uptake to extreme drought was investigated in different ecosystems. The SPEI spatiotemporal patterns indicated that the Qaidam Basin and most forest regions on the TP demonstrated drought trends. The improvement in the water regime and thermal conditions restrained the NRVCU on the TP during the 1982-2015 period. The concurrent and lagged effects of extreme drought on the NRVCU exhibited significant differences in the alpine steppe and alpine meadow ecosystems, but no discrepancy was found for forests, which was primarily due to their vegetation properties and environmental conditions. Forests requires a longer time period to recover their net carbon uptakes than that of alpine steppe and alpine meadow ecosystems because of their complex ecosystem structure. This study provides information regarding the lagged effect of extreme drought on the vegetation net carbon uptake on the TP. In future studies, incorporating the process mechanisms such as the lagged effect of extreme drought on the vegetation net carbon uptake may be helpful in managing net carbon fluxes.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-4292/12/15/2347/s1, Figure S1. The comparison (ME: mean error, BIAS: relative bias, NSE: Nash-Sutcliffe coefficient efficiency, RMSE: root mean square error, and r: correlation coefficient) between the Global SPEI database and calculating SPEI in 1985, 1995, and 2005. Figure S2. (a)-(h) represent the temporal trends; interannual variability (IAV); and standard deviations for the net primary production (NPP), net reduction in vegetation carbon uptake (NRVCU), and Standardised Precipitation-Evapotranspiration Index (SPEI) in the (A) alpine steppe, (B) alpine meadow, and (C) forest ecosystems. Horizontal lines in box plots denote the 95th, 75th, average value, 25th, and 5th percentiles from top to bottom. Figure S3. The correlation analysis between (a) monthly mean net reduction in vegetation carbon uptake (NRVCU) and monthly mean precipitation, (b) monthly mean NRVCU and monthly mean temperature, and (c) monthly mean precipitation and temperature. Figure S4. Spatial patterns of the multi-year mean for (a) precipitation, (b) temperature, and (c) relative humidity in the Tibetan Plateau during the 1982-2015 period and temporal trends of (d) precipitation, (e) temperature, and (f) relative humidity. Figure S5. Spatial correlation between (a) net reduction in vegetation carbon uptake (NRVCU) and precipitation, (b) NRVCU and temperature. The significance level between (c) NRVCU and precipitation, (d) NRVCU and temperature. Figure S6. Correlation analysis between vegetation net carbon uptake and (a) alkali-hydrolysable N, (b) available P, (c) available K, (d) cation exchange capacity, (e) soil organic matter, (f) PH, (g) soil moisture, (h) total N, and (i) total P. Figure S7. (a) Alkali-hydrolysable N, (b) available P, (c) available K, (d) cation exchange capacity, (e) soil organic matter, (f) PH, (g) soil moisture, (h) total N, and (i) total P in (A) alpine steppe, (B) alpine meadow, and (C) forest ecosystems. Horizontal lines in the box plots denote the 95th, 75th, average value, 25th, and 5th percentiles from top to bottom.