Winter Daytime Warming and Shift in Summer Monsoon Increase Plant Cover and Net CO2 Uptake in a Central Tibetan Alpine Steppe Ecosystem

Over the past decades, human‐induced climate change has led to a widespread wetting and warming of the Tibetan Plateau (TP), affecting both ecosystems and the carbon cycling therein. Whether the previously observed climate changes stimulate carbon uptake via enhanced photosynthesis or carbon loss via enhanced soil respiration remains unclear. Here we present 14 years of observations of carbon fluxes, meteorological variables and remotely sensed plant cover estimations from a central Tibetan alpine steppe ecosystem at Nam Co, the third largest lake on the TP. Using modified Mann‐Kendall trend tests, we found a significant increasing daily net carbon uptake of 0.5 g C m−2 decade−1, which can be explained by a widespread greening at the southern shore of lake Nam Co. The Plateau‐wide changes in temperature and precipitation are locally expressed as an increasing diurnal temperature range during winter, higher water availability during spring, higher cloud cover during early summer and less water availability during late summer. While these changes differ over the course of the year, they tend to stimulate plant growth more than microbial respiration, leading to an increased carbon uptake during all seasons. This study indicates that during the 14 years study period, a higher amplitude in winter temperatures and an earlier summer monsoon promote carbon uptake in a central Tibetan alpine steppe ecosystem.

The TP is characterized by strong diurnal and seasonal gradients of radiation, temperature and moisture availability. The majority of the observed warming trend, at about twice the mean annual trend, occurs during the winter months (Yao et al., 2019). Corresponding to the increase of the mean temperatures, the maximum and minimum temperatures increase as well. The minimum temperature increases more than the maximum temperature, which leads to a decrease in the daily temperature range (DTR) of 0.2°C per decade (You et al., 2016). The narrower DTR is explained by increasing low level cloud cover, which intercepts outgoing longwave radiation during nighttime and, to a smaller extend, blocks incoming solar radiation during the day (Duan & Wu, 2006). Another cause is the increased evaporation due to the higher soil moisture, which in turn is a consequence of increasing precipitation (Yang & Ren, 2017;Yang et al., 2011;You et al., 2016). As the surface becomes darker due to the increasing loss of snow and ice covered areas, it can absorb more solar radiation. The additional energy input heats up the surface, thus amplifying the role of the TP as mid-tropospheric heat source, which plays a crucial role in the timing and intensity of the Indian summer monsoon (ISM) (Ge et al., 2017). About 80% of the annual precipitation occur during the summer months from June to August and only about 10% during the winter months from December to February (Wang et al., 2018). Since the 1960s, precipitation has shown an average increasing trend of about 5.1 mm per decade (Wang et al., 2018). However, the increase in precipitation is unevenly distributed spatially and temporally. While the precipitation is increasing in the central, northern and southern TP, it is strongly decreasing in the eastern plateau and on the south-eastern periphery (Yang et al., 2014). The increase in precipitation takes place mainly in spring, while the other seasons only show slight but not significant increasing trends (Wang et al., 2018). The earlier onset of summer monsoon precipitation was found to play a major role in increasing lake areas of the endorheic inner TP (Liu et al., 2019), as well as in increasing soil moisture and a widespread greening of the ecosystems during early summer (Zhang et al., 2017). Several authors highlight that the observed changes in temperature and precipitation are responsible for advancing phenological trends, as well as summer greening in most regions of the TP (e.g., H. Li et al., 2019;Shen et al., 2011Shen et al., , 2016Zhang et al., 2013).
The identification of the main drivers of ecosystem respiration and soil organic carbon stocks have been extensively investigated in the K. pygmaea pastures, which occur mostly on the central and eastern TP (e.g., Babel et al., 2014;Gu et al., 2003;Kato et al., 2004Kato et al., , 2006Zhang et al., 2018;Zhao et al., 2017Zhao et al., , 2018. With decreasing precipitation to the west, the K. pygmaea pastures are replaced by alpine steppe vegetation. The alpine steppe dominates wherever the K. pygmaea plants cannot form their typical mono-specific closed root mats, either due to lack of precipitation or overgrazing by livestock. Although the transition is associated with the loss of up to 70% of the SOC stocks (Miehe et al., 2019), at least the eastern alpine steppe The few studies which exist on ecosystem-atmosphere carbon cycling in the Tibetan alpine steppe ecosystems assign the inter-annual flux variability mainly to varying monsoonal precipitation Zhu et al., 2015).  and Zhu et al. (2015) found the alpine steppe to act as a carbon sink during the growing seasons of dry and wet years but Wang et al. (2016) found a small net carbon source on an annual basis. It remains uncertain if the Tibetan alpine steppe ecosystem is currently a net sink or a net source of CO 2 and in which way the observed climate changes influence the carbon cycling. On the one hand, the warmer and wetter climate could enhance the autotrophic and heterotrophic soil respiration, thus increasing loss of carbon to the atmosphere. On the other hand, increasing temperature and precipitation are stimulating plant growth, thus enhancing carbon uptake through increased primary production . The difficult accessibility and harsh environmental conditions provide a challenging environment for conducting (micro-) meteorological measurements. Hence, most studies draw their conclusions only from observations during the growing seasons of a few years at most (e.g., Pei et al., 2009;Wei et al., 2012Wei et al., , 2014Zhu et al., 2015). The aim of this study is to disentangle the complex interplay of climatic variations and the ecosystem-atmosphere exchange of CO 2 , from daily to monthly and annual to inter-annual time spans. The analyses of long term carbon fluxes, associated meteorological observations and plant cover estimates from a central Tibetan alpine steppe ecosystem allow the investigation of the following research questions: 1. Can the general trend of increasing temperatures and precipitation also be observed in the alpine steppe ecosystem at Nam Co? 2. How is the ecosystem-atmosphere exchange of CO 2 characterized on daily to monthly and annual to inter-annual time scales? 3. Do the observed changes in environmental variables promote or hamper plant productivity and soil respiration in the long run?

Study Site
The study site is located at the Nam Co Station for Multi-sphere Observation and Research (NAMORS, Chinese Academy of Sciences [CAS]), about 220 km north of the Tibetan capital Lhasa (30 46'N, 90 57'E). It is situated on a plain, sandy terrace between lake Nam Co (1 km NW) and the Nyainqêntanglha range (15 km SSE) at 4,730 m a.s.l. The climate is controlled by large-scale circulations and is therefore subject to strong annual variation. During winter, the prevailing westerlies provide cold air masses from central Asia. From December to February, the daily mean air temperatures stay below 0°C (minima below −20°C) with mostly clear and dry conditions. Usually, no closed snow cover is achieved for longer time periods, but in some years there can be heavy snowfalls (Wei et al., 2017). Therefore, the spring snow melt supplies only a small amount of moisture to the soil, which percolates quickly to deeper soil layers. The soil moisture then decreases until the ISM arrives, typically between May and June. From May to September, the humid air masses from the Bay of Bengal and the Arabian Sea provide about 80% of the annual precipitation, which ranges from 291 to 568 mm (mean = 403 mm). Between September and October, the weather gets colder and dryer again (Yao et al., 2013). The study site is situated in the transition zone between the K. pygmaea pastures and the dryer, less productive alpine steppe ecosystem. The alpine steppe at the NAM-ORS is characterized by Stipa purpurea but comprises further species from the families Artemisia, Stipa, Poa, Festuca, and Carex (Li, 2018;Miehe et al., 2011). Due to heavy grazing by yak, sheep and goat, the vegetation does not exceed 5 cm in height and a surface coverage of 50%. However, it is to note that the whole station was fenced in 2005, thus excluding the otherwise ubiquitous livestock grazing. Also in 2005, the Institute of Tibetan Plateau Research (ITP), CAS installed (micro-) meteorological measurements at the NAMORS (Ma et al., 2009). A 3 m Eddy Covariance tower measures the 3D wind vector with a CSAT3 ultrasonic anemometer and the carbon and water vapor fluxes using a Li-7500 open path infrared gas analyzer (IRGA). A 52 m planetary boundary layer tower (PBL) is equipped with air temperature and relative humidity measurements at 5 different heights (1.5, 2, 4, 10, 20 m) and measurements of wind speed and wind direction at 3 different heights (1.5, 10, 20 m). Additionally, the site is equipped with soil temperature 4 of 20 and soil moisture measurements (0, −10, −20, −40, −80, −160 cm). The soil moisture content is measured using time-domain reflectrometry (TDR) sensors. Because ice has a lower permittivity than liquid water, soil moisture measurements become unrealistically low during frozen conditions. Therefore, soil moisture measurements with soil temperature  E 0 C were discarded before aggregating to daily, monthly and annual averages. The station is furthermore equipped with measurements of soil heat flux (−10, −20 cm) and radiation (short and long wave downward and upward radiation, global radiation), precipitation and air pressure measurements. Since 2013, the station comprises a sensor measuring photosynthetic active radiation (PAR) (Zhu et al., 2015). For detailed information about the equipment and infrastructure, please see Ma et al. (2009Ma et al. ( , 2017.

Previous Work
In an earlier study, the authors of this manuscript published the fully processed and quality controlled EC flux data set from Nam Co (Nieberding et al., 2020). The 30-min averaged fluxes of CO 2 , water vapor (H 2 O), sensible and latent heat flux (H and LE, respectively) were calculated from the original 10 Hz EC measurements using the raw data processing software EddyPro (v7.0.6, LI-COR Inc.). Due to the remote location and harsh environmental conditions on the TP, sensor maintenance and calibration were performed irregularly, leading to a drift in mean CO 2 and H 2 O concentration measurements. This drift was corrected by applying the offset between the measured CO 2 and H 2 O concentrations and a reference time series to the raw 10 Hz measurements following Fratini et al. (2014). This way, the drift in concentration measurements was efficiently removed which prevented an overestimation of the mean carbon uptake by roughly 28%. During the course of the measurements, several buildings and scientific infrastructure were constructed close to the EC tower. Analyses of the wind direction distributions of wind speed and turbulent kinetic energy (TKE) suggest that the buildings do have some influence on the flow regime. Hence, fluxes originating from possibly disturbed wind sectors were completely excluded from further analyses. To get a more complete picture, the CO 2 flux data gaps were filled using the look-up table (LUT) approach provided in the REddyProc R package (Wutzler et al., 2018). Due to the harsh environmental conditions and the rigorous exclusion of possibly disturbed wind sectors, the overall data availability for Net Ecosystem CO 2 Exchange (NEE) after QC was only 28% (Figure 1).

Data Screening, Gap Filling and Flux Partitioning
The LUT gap filling procedure relies heavily on the availability of meteorological measurements (mainly air temperature, radiation and vapor pressure deficit). These variables are narrowly binned within certain time windows around the data gap in order to average the missing CO 2 flux values from available values during similar meteorological conditions and times (Wutzler et al., 2018). If not enough data points are available, the time window is successively widened until averaging is possible. Thus, the LUT procedure replicates the existing flux values before and after the data gap causing a greater uncertainty the larger the data gap is. Hence, we decided to modify the quality filtering of our previous study to make the gap filling procedure more robust by obtaining as much physiologically plausible data as possible.
First, we relaxed the wind direction filtering. In our previous study, we preemptively excluded all fluxes from possibly disturbed wind sectors, even though only the main building exerts a clear influence on the quality of the fluxes. This led to the exclusion of roughly 20% of all measured data points. Now we exclude only those wind sectors that are clearly disturbed, namely the two-storey main building and the back of the CSAT3 ultrasonic anemometer. Furthermore, an analytic approach was applied to exclude implausible measurements of night time carbon uptake while accounting for the seasonal variation of fluxes and flux noise. Originally, we rigorously excluded night time fluxes  E −0.1 mol m −2 s −1 to discard unreasonable fluxes while allowing some variability in the data. Now, the cut-off threshold was estimated for every month individually by fitting a tangent to the density curve of night-time CO 2 fluxes at the point where the first derivative of the density curve has its maximum. The intercept of this tangent with the x-axis is then taken as cut-off threshold. The threshold is set to 0 mol m −2 s −1 if it was found to be  E 0 mol m −2 s −1 . This way, physiologically implausible measurements were discarded while allowing the typical flux variability around the 10.1029/2021JG006441 5 of 20 mean. The calculations can be found in the R scripts and monthly plots illustrating the process are provided in the supporting information (Figures S1-S12).
Building on our previous work, we also enhanced the gap filling of meteorological variables. As mentioned in Section 2.1, there are usually several measurements of the same variable available. For air temperature (Tair) and relative humidity (rH) there are measurements from 1.5, 2, 3, 4, 10, and 20 m available, which originate from the two different towers (EC and PBL). Similarly, soil temperature (Tsoil) and soil moisture (SMC) measurements exist from 0, 10, 20, 40, 80, and 160 cm depth. Incoming solar radiation (Rg) data exists from 3 independent devices which measure shortwave incoming radiation (Kipp & Zonen CNR1 and CMP11) and PAR (Apogee Original Quantum Sensor). Based on annual linear regression, the different measurements were combined to one overall time series. The most reliable time series closest to the ground was used as starting variable which was then iteratively gap-filled with the ones having the highest linear regression coefficient R 2 . Vapor pressure deficit (VPD) was calculated from the gap-filled time series of Tair and rH. Remaining short data gaps of up to two time steps were filled using linear interpolation. Following the procedure from our previous manuscript, we applied the quality control scheme for CO 2 and H 2 O fluxes also to the flux measurements of H and LE. Finally, the LUT method from REddyProc was used for gap filling of the fluxes and meteorological variables.
The measured CO 2 fluxes represent the NEE, which we partitioned into its components gross primary production (GPP) and ecosystem respiration (Reco), representing the photosynthetic uptake of CO 2 and the autotrophic and heterotrophic soil respiration, respectively. We use the day time partitioning approach within REddyProc to model GPP and Reco (Lasslop et al., 2010;Wutzler et al., 2018). Please note that GPP and Reco are estimated individually and therefore do not necessarily add up exactly to NEE. The final 30min time series of the fluxes and meteorological variables were aggregated to daily, monthly and annual estimates (Reis & Ribeiro, 2020).

Resource Use Efficiency
To get a wider picture of the mass-energy exchange between the alpine steppe ecosystem and the atmosphere, we calculated different resource use efficiencies. As we are using Eddy Covariance measurements, our estimates of water use efficiency (WUE) and light use efficiency (LUE) represent whole ecosystem estimates, rather than the oftentimes used resource use efficiencies at the plant level. Ecosystem LUE (g C MJ −1 ) was calculated as the ratio of E GPP (g C m −2 time −1 ) to incoming shortwave radiation (MJ m −2 time −1 ): where E ET was calculated from latent energy flux (LE in W m −2 ) and air temperature (Tair in °C) following Equation 65 in Reis and Ribeiro (2020): ET was calculated for every half hourly time step and then summed up to daily or monthly values.

Trend Analysis
When analyzing such a long data set, the question arises as to whether a trend exists and how this might be related to the profound environmental changes of the recent decades. The non-parametric Mann-Kendall (MK) test is widely applied in the detection of monotonic trends in hydro-meteorological time series because it has no requirements of homoscedasticity or prior assumptions of the sample distribution (F. . The MK test statistic is determined by the ranks and sequences of a time series rather than the original values. It is robust against outliers, missing values and non-normally distributed data. However, the MK test assumes that the data are independent and randomly ordered. Hence, it is not robust against serial auto-correlation as it is frequently encountered in hydro-meteorological time series. A positive auto-correlation in the time series may lead to Type I errors, i.e., detecting a trend where none exists. Hamed and Rao (1998) developed an empirically determined correction factor to modify the variance of the MK test statistic (hereafter called MMKH). While this method can reduce the effect of serial correlation, it will lead to an increase in Type II errors (i.e., rejection of a trend where one exists) if a trend is present in the data (Yue & Wang, 2002;. To overcome these shortcomings, Yue and Wang (2004) proposed to remove an existing trend prior to the estimation of the auto-correlation coefficients from the the original data, rather than from the ranks of sample data (hereafter called MMKY). As pointed out by several studies (e.g., Blain, 2013;, neither the original Mann Kendall test, nor its variance correction approaches can be used as a "silver bullet" for trend detection. The proposed methods have either problems preserving the adopted significance level (Type I error) or they are less powerful (Type II error). Hence, we performed both modified MK tests using "modifiedmk" library in R programming environment (Patakamuri & O'Brien, 2020).
Because the data gaps are irregularly distributed within the individual years, any trend test may lead to erroneous results when applied to the whole time series. While there are only few years without data gaps, there are usually more complete months available. Hence, the modified MK tests were performed using daily fluxes, meteorological variables and resource use efficiencies which were pooled on a monthly basis.
Only months with  E 30% measured values after quality filtering and 100% data availability after gap filling were used. This ensures comparability between the years and accounts for uncertainties which may be 7 of 20 introduced when filling long data gaps. We report the E p -values of both modified MK tests and treat them equally in our analysis. However, exceeding a certain significance level should not be used alone for binary hypothesis testing or model interpretation (Wasserstein & Lazar, 2016;Wasserstein et al., 2019). Hence, we also calculate Sen's slope, which is the median of the slopes of all linear regression lines through each pair of points (Sen, 1968). Since this slope estimation is stable against outliers, we get a more robust estimate of the effect size and can ultimately draw conclusions about the relevance of our results. The exact E p -values of both modified MK tests and the Sen's slope estimations are provided in the supporting information (Tables S1-S3).

Plant Cover
We estimated annual plant coverage for the area close to the NAMORS using the pre-trained support vector machine regression (SVM) models published in Lehnert et al. (2015). These are fully validated against field samples of the five major grassland types found on the TP: (1) Kobresia pygmaea pastures, (2) Kobresia humilis pastures, (3) swamps and salt marshes, (4) montane, and (5) alpine steppes. With an overall RMSE of  E 7%, SVM showed the best performance compared to three other regression models. For our study site, the SVM models were applied to multispectral satellite data acquired by Landsat 5, 7, and 8 at 30 m spatial resolution (downloaded from USGS EarthExplorer). The Landsat scenes were processed using an extended version of the 6S code for atmospheric correction (Curatola Fernández et al., 2015;Vermote et al., 1997) and the Minnaert model for topographic corrections (Riano et al., 2003). The data was filtered for obstructions (e.g., clouds) using the quality bands shipped with the Landsat data. Therefore, a scene listed in the Earth-Explorer with a high cloud cover percentage that had clouds distributed over lake Nam Co rather than the mountain ranges to the south was chosen over a scene with a lower listed cloud cover with non-optimal cloud distribution. In addition, if both Landsat TM and Landsat ETM + images were available for a year, TM scenes were chosen over ETM + scenes if there was no remarkable difference in cloud distribution. To ensure full development of plant coverage and hence comparability between the years, only Landsat scenes acquired during the summer months (JJA) were used. During the study period from 2005 to 2020, a total of 14 catchment-wide acquisitions were available for the plant cover assessment. If multiple cloud-free scenes acquired during the summer months were available for one year, the scene acquired closest to fifteenth of July has been selected. Pixel-wise mean and standard deviations of plant coverage were calculated and the final product was cross-verified by post hoc assessment of the field investigation. For the trend analysis, the same modified non-parametric Mann-Kendall correlation tests following Hamed and Rao (1998) and Yue and Wang (2004) were applied on a pixel based approach. Only pixels with at least 9 valid plant cover estimates were used to detect changes of plant cover over time (at 95% significance level). In the following, we will use the term greening to refer to a significant increase in plant cover.

Climate Variability
Daily and monthly variation in the major environmental variables are shown in Figure A1 and Table 1. All variables reflect the strong annual variation which not only follows the insolation but is also subject to the large scale circulation patterns, namely the ISM and the westerlies during winter. Solar radiation, representing the main heat source on the Plateau peaks in May, just before the arrival of moist air masses and the associated increase in cloud cover. While Tair and surface Tsoil follow the basic annual insolation pattern, there are remarkable differences between them. The average annual soil temperature is more than 5 C higher than the average annual air temperature. This is true for every month but especially pronounced during summer. The largest differences occur in May and June, where the average Tsoil is more than 8 C warmer than the average Tair. The lowest values occur between December and February with mean temperatures well below freezing point. The mean annual precipitation is roughly 400 mm but the inter-and intra-annual variability is high. Annual precipitation sums range from 295 mm in 2006 to 569 mm in 2010. More than 80% of the mean annual precipitation occur between May and September and less than 4% occur between November and March. While the relative air humidity follows the annual precipitation distribution quite fast, there is a delay in soil moisture, with highest values between July and October. However, with a maximum daily average water content of 37%, soil moisture remains quite low throughout the year. Vapor pres-10.1029/2021JG006441 8 of 20 sure deficit peaks during early summer and stays relatively high until September. The lowest VPD values of 2 hPa occur during January and February. More detailed tables including minimum and maximum air and soil temperatures, as well as the respective standard deviations for the average values shown in Table 1 can be found in the supporting information (Tables S4 and S5).

Seasonal Course of Fluxes
Following the usual micrometeorological convention, negative values for NEE, H and LE represent fluxes toward the surface and positive values represent fluxes away from the surface. Note that while GPP and Reco are both positive, GPP represents carbon uptake and Reco represents carbon release. Table 1 shows the monthly average, and Figure 2 shows the daily average carbon and heat fluxes throughout the whole study period. Between November and January, daily NEE remains close to zero. Beginning in February, NEE becomes more and more negative and reaches a maximum monthly carbon uptake of  E 19 g C m −2 month −1 during June, July, and August. Between September and October, carbon uptake decreases rapidly back to values close to zero. GPP and Reco show a somewhat different pattern with a maximum during August and rapid decline afterward. Interestingly, both variables indicate biological activity even during the cold and dry winter months December to February. The sensible and latent heat fluxes show a complimentary behavior, following the strong annual seasonality on the TP. The sensible heat flux peaks already in May, right before the onset of summer monsoon precipitation and cloud cover. Correspondingly, the strongest latent heat fluxes occur from July to September, when water availability is highest.

Trend Analysis
The results of the two different modified MK tests (MMKH and MMKY, Figure 3) show a significant increase of the CO 2 sink strength during all months. In September only MMKY indicates a significant trend whereas during all other months, both tests yield statistically significant trends. The months of May, June, and July exhibit the strongest increase in daily carbon uptake with −1.3, −0.7, and −1.0 g C m −2 decade −1 , respectively. The smallest changes of NEE were found for the months August, September and February. When averaged over the whole year, the daily carbon sink strength increased by 0.5 g C m −2 decade −1 . When looking at the partitioned fluxes, a distinct pattern becomes visible: During the months November to July, GPP and Reco generally increased, whereas during August to October GPP and Reco generally decreased. In the first case, GPP increased stronger than Reco while in the second case the decrease in GPP was less than that of Reco, leading to the overall increasing carbon sink strength. On average, the daily GPP is increasing by 0.4 g C m −2 decade −1 but no overall trend was found for Reco. As the carbon fluxes are governed by the availability of light, heat and moisture, any changes in these variables can have diverse effects. The main changes during different periods of the year are as follows: 1. During November to February, we found a decrease of rH and LE and an increase of VPD which indicates a lower water availability. This very likely led to a change of the partitioning of available energy, away from LE and towards an observed increase of H. A strongly increasing diurnal soil temperature range can be observed ( E 10°C), with increasing maximum and decreasing minimum soil temperatures. The air temperatures do not exhibit such a clear trend but when averaged over the four months, we still see slightly increasing Tair mean, max and DTR and slightly decreasing Tair min. Figure 5 indicates that, during November to February, increasing maximum and minimum soil temperatures have opposite effects on NEE. Higher maximum soil temperatures increase carbon uptake while higher minimum soil temperatures increase carbon release. 2. During early summer, daily light availability is strongly decreasing with −2.3 and −1.3 MJ m −2 decade −1 during June and July, respectively. With significantly decreasing H and Tsoil (mean, max and DTR), heat availability is following this trend, especially at the surface. At the same time, higher moisture availability is indicated by increasing SMC and rH and lower VPD. 3. Especially during August but also during September to October, lower moisture availability is indicated by decreasing LE, and rH while VPD exhibits a positive trend. At the same time, Tair mean starts to in- crease and also air DTR is increasing, mostly due to stronger increases in Tair max than in Tair min. The drying trend is also supported by Figure A2, which shows decreasing monthly precipitation sums during August (−6.26 mm yr −1 ) and September (−2.15 mm yr −1 ).
The other months do not show any trends in precipitation, but it is to note that variability is quite high between the years, especially during the summer months. Because precipitation is a discontinuous phenomenon, it was not included in the modified MK trend test. The many days with zero precipitation and the broadly distributed precipitation events would not be covered adequately by the test statistics.
The water use efficiencies show a consistent increase throughout all months (Figure 3), with the strongest effect from December to March. The LUE is also increasing during most times of the year with the highest increases from May to July whereas a negative, yet not always significant trend can be observed between August and October. For the summer months it becomes clear, that the LUE is negatively correlated with H, Rg and VPD and positively correlated with LE, rH and SMC (6). The highest LUEs are observed at intermediate temperatures of 5-10 C and 10-15 C for Tair and Tsoil, respectively.

Plant Cover
The alpine steppe ecosystem around the NAMORS station exhibits mean plant covers of 30%-50% (Figure 4a). The area that is covered by the footprint of the EC system has a mean plant coverage of less than  Hamed and Rao (1998, MMKH) and Yue and Wang (2004, MMKY). The numbers represent the effect size in native units per decade (Sen's slope). They are written in bold if both, MMKH and MMKY yielded a significant trend ( E p -value  E 0.05), in italics if only MMKY yielded a significant trend and no numbers are printed when both tests were insignificant. The relative effect strength (Kendall's Tau) is color coded from blue to red (−1-1). The column Avg represents the mean Sen's slope and Tau of all significant and non-significant months.
40%. There are areas to the south east and south west that show higher plant coverage up to 80%. These areas are mostly situated in depressions with continuous water availability from streams. The trend analysis ( Figure 4b) shows a large scale greening of the alpine steppe ecosystem that is not only evident for the area close to the NAMORS but also for larger areas of alpine steppe ecosystems more to the south. The more productive areas at lower elevation exhibit a rather decreasing plant cover. As expected from theory (Section 2.4), MMKH indicates less significant trends than MMKY. If there is a significant trend in MMKH, it is almost always significant in MMKY as well. The tests only differ in the number of significant trends, not in the relative effect strength, nor the direction of the trend.

Cold-Season Carbon Uptake
As already stated in our previous work, we find a diurnal pattern of carbon uptake and release even during the coldest months when air temperatures remain continuously below freezing point ( Figure A3). However, due to the strong radiation and clear sky conditions, the soil heats up well above zero degrees, facilitating plant and soil microbial activity (Nieberding et al., 2020). While the temperature is sufficiently high, the availability of water for the biological activity remains unclear. The measured soil moisture content during December to February is typically below 5%. However, the dominant plants in the Tibetan alpine steppe such as Stipa purpurea are highly adapted to drought stress, for example, by regulating stomatal conductance (Yang et al., 2015). This enables the plants to maintain photosynthesis even under prolonged drought conditions. A possible, as yet unquantified, source of water could be dew and hoarfrost. This could form regularly at night during clear sky conditions on the TP and be taken up efficiently by the plants in the morning, before entering the soil and being detected by the soil moisture sensors. Groh et al. (2018) found dew and hoarfrost contributing substantially to the water budget at two humid low mountain grasslands in Germany, accounting for up to 38% of the total monthly precipitation amount, especially during winter months. Guo et al. (2016) estimated that dew formation contributed 7.2% of the corresponding rainfall amount to the growing season water availability in a cold desert-shrub ecosystem in Ningxia, NW China (1,530 m a.s.l).  Hamed and Rao (1998) and Yue and Wang (2004) for the same area. Only statistically significant ( E p -value  E 0.05) increasing or decreasing pixels are shown. Nearly all pixels that are significant for Hamed and Rao (1998) are also significant for Yue and Wang (2004). The SE to NW oriented "stripes" result from the failure of the Landsat 7 sensor in 2003, which led to missing data and consequently causes fewer samples to be available in the areas affected by the sensor failure. The decreased sample size yields slightly different trends and less significant pixels in these areas.
To further test our hypotheses, in situ measurements of dew and hoarfrost formation are required by which their relevance for plant metabolic activities can be assessed, especially during the winter months.

Differential Effect of Soil Temperature on GPP and Reco
Between November and February, a strong increase of the diurnal soil temperature range of  E 10 C per decade was measured, caused by the combination of increasing daily maximum Tsoil and decreasing daily minimum Tsoil. Soil respiration is happening at night and during daytime, while carbon uptake is only possible at daytime through photosynthesis. An increase of daytime soil temperatures (Tsoil max) will increase both GPP and Reco while a decreasing night time temperature (Tsoil min) will only decrease soil respiration (Figure 5). Hence, both affect the carbon sequestration capacity toward higher carbon uptake, which is reflected by increasing GPP but a more or less steady Reco (Figure 3). An increasing DTR during December-January  was also reported by Yang and Ren (2017) for our study site at 30°N, 90°E while most other parts of the TP exhibit decreasing DTRs. The overall decreasing DTR was also found by other studies, which assign a stronger increase in minimum temperatures than in maximum temperatures to enhanced nighttime cloud cover (e.g., Duan & Wu, 2006;Yao et al., 2019). Hence, these seemingly contradictory findings actually highlight that the great spatial and temporal climatic variability throughout the TP has to be taken into account for studies on carbon, energy and water cycling.

Greening and Water Availability
Over the past decades, an advancing spring green-up date (GUD) has been reported for the TP grasslands by many authors (e.g., Zhang et al., 2013;Zheng et al., 2016). For the months March and April, we observed an increase in carbon uptake and a slightly increasing Reco, which agrees well with Li et al. (2019), who found that decreasing minimum temperatures in January and increasing minimum temperatures in spring fulfill the winter chilling and spring heat accumulation requirements earlier, thus promoting an earlier GUD. In the meantime, we found that both WUE and LUE in March increased. This can be a consequence of earlier GUD indicating that plants can use the available water and light resources more efficiently if they sprout earlier.
For the months May, June, and July, we found the strongest increase in carbon uptake and partly also in ecosystem respiration. For June and July it becomes clear that lower incoming shortwave radiation decreases soil temperatures during daytime, which also decreases Tsoil mean, Tsoil DTR and the sensible heat flux. While the increasing rH and decreasing VPD could be a result of the lower air temperatures, SMC clearly indicates higher water availability. However, the monthly precipitation sums do not show any increasing trend ( Figure A2). Together, these changes are a consequence of increased cloud cover during the early summer months reducing the heat availability at the surface. No long term change is observed in LE during May to July, indicating that the effects of reduced energy availability and increased water availability roughly balance each other out.
Unlike during the other months, the increase in carbon uptake during late summer (August and September) is mainly caused by a decreasing daily Reco of up to −1 g C m −2 decade −1 , which roughly doubles the amount of the GPP decrease. August and September are also the only months with decreasing precipitation sums (−6.26 and −2.15 mm yr −1 , respectively). Together with increasing Tair and VPD and decreasing LE and rH, these variables indicate an enhanced drought situation during the late growing season.
The strongly increasing LUE between May and July and the decreasing LUE between August and September indicates that radiation, as well as heat and drought stress influence the ability of the alpine steppe to efficiently use the available light for photosynthesis. The increasing cloud cover during June and July is expected to decrease the amount of direct radiation and increase the amount of diffuse radiation (Huang et al., 2014). Diffuse radiation has a lower tendency to cause canopy photosynthetic saturation, resulting in higher light use efficiencies by the plant canopy (Gu et al., 2002;Zhu et al., 2015). On the contrary, increasing drought conditions during late summer may trigger stomatal regulation of the S. purpurea steppe, especially during noon when VPD and temperature are high (Körner, 1995;Yang et al., 2015;Zhu et al., 2015). These findings are corroborated by the positive correlation between LE, rH, and SMC with LUE during May to September (Figure 6). At the same time, H, Rg, and VPD exhibit a negative correlation with LUE, highlighting the sensitivity of plant productivity to shifting climatic patterns.
The special location of the study site just south of the large water masses of Nam Co leads to the formation of a regular lake-land breeze system Dai et al., 2020;Gerken et al., 2014). When the moist air masses from Nam Co get uplifted along the Nyainqêntanglha range to the south, deep convection gets triggered frequently due to adiabatic cooling and mesoscale circulation patterns (Gerken et al., 2015). Hence, the plain and the mountains south of Nam Co are preferential areas for cloud formation and precipitation. Any changes affecting either the land (like greening or pasture degradation) or the lake (e.g., warming, expansion) may affect local circulation patterns and hence cloud formation Dai et al., 2020). However, Ma et al. (2016) found that evaporation of Nam Co exhibits a decreasing trend of −12 mm yr −1 between 1998 and 2008. These findings are consistent with our estimations, which exhibit an average −0.37 mm decade −1 decrease of the daily ET for the alpine steppe ecosystem.
The NAMORS station was fenced in 2005, thus excluding the otherwise ubiquitous grazing by yak, sheep and goat. The absence of large grazers is expected to increase the above-and below-ground organic carbon content, which could also lead to the observed increase in CO 2 uptake . For our study site, we estimated that about 50%-60% of the fluxes originate from within the fenced area. Hence, the reported increase in carbon uptake is representative for the steppe ecosystem inside, as well as outside the fenced area. When taking the trends in plant cover ( Figure 4) into account, it becomes obvious that the observed greening trend is not only happening within the fenced area close to the NAMORS but all over the alpine steppe ecosystem close to the lake, as well as on the plains to the south.
We conclude that the earlier onset of summer monsoon cloud cover ameliorates drought conditions during early summer (June and July), thus increasing plant productivity and consequent carbon uptake of the alpine steppe ecosystem. On the contrary, the earlier reduction of monsoon precipitation during late summer (August and September) increases drought conditions, decreasing soil respiration more than plant productivity. Both effects lead to a net increase in ecosystem CO 2 uptake.

Conclusions
The TP is characterized by strong diurnal and seasonal fluctuations in radiation, heat and moisture availability. To our knowledge, this is the first study to assess the long term trends in carbon fluxes and meteorological variables alongside plant cover estimations for the Tibetan alpine steppe ecosystem. During winter (November to February) higher maximum and lower minimum temperatures cause an increased diurnal soil temperature range of more than 10°C decade −1 . Decreasing solar radiation during early summer (June and July) and decreasing precipitation during late summer (August and September) indicate a shift in the summer convection and precipitation dynamics toward earlier in the year. Using remotely sensed plant cover estimations, the general greening trend over the TP was also confirmed for the alpine steppe at the  (rH), (e) soil moisture (SMC), (f) vapor pressure deficit (VPD), (g) air temperature (Tair) and (h) soil temperature (Tsoil) with LUE between May and September, using only months with  E 30% data availability. The blue line represents the second order polynomial regression between the variables.