Warming-induced unprecedented high-elevation forest growth over the monsoonal Tibetan Plateau

Growth of high-elevation forests is generally temperature-limited and thus sensitive to warming. The Tibetan Plateau has experienced fast warming rates associated with decreased summer monsoon rainfall over the last century. However, whether such warming and monsoon-induced drought could offset a potential warming-driven enhancement of forest growth has not been examined. Here, we have compiled high-elevation forest growth data at 40 sites over the monsoonal Tibetan Plateau (MTP), and combined them in a high-elevation forest growth index (HEFGI) spanning 1567–2010. Tree growth in this region was significantly and positively correlated with July–October minimum temperatures during 1950–2010 (R2 = 0.53 P < 0.001), and insignificantly coherent with soil moisture and precipitation. The HEFGI of MTP reaches its highest values from the 2000s onwards. This result suggests that the mean HEFGI of MTP has not been negatively affected by the current drying trend and responded positively to increased temperatures.


Introduction:
Annual tree growth at high latitudes and altitudes is generally limited by temperature (Korner 1998, Rossi et al 2007. Therefore, warming is expected to increase growth in such forests. However, increased dry condition associated with global warming has been reported in both observations and model simulations (Dai 2013). This 'global-change-type drought' has been profound since the beginning of the 20th century (Marvel et al 2019), and has led to a worldwide acceleration of forest mortality and decreases in growth rate (Adams et al 2009, Carnicer et al 2011, Choat et al 2012, Allen et al 2015.
Forest growth at high elevations was assumed to benefit from climate warming, as low growing season temperature has been the main limitation (Rossi et al 2007). Enhanced high-elevation forest growth was observed (Salzer et al 2009, Qi et al 2015, Dulamsuren et al 2017, Wang et al 2017. However, constant, even declined alpine forest growth in association with warming induce moisture limitations was increasingly reported, mainly in the dry areas, such as northwest China, Himalayas and north-eastern TP (Liang et al 2014, 2016. The monsoonal Tibetan Plateau (MTP), which received abundant precipitation from South Asian monsoon, is experiencing both fast warming and drying trends (Liu and Chen 2000, Pepin and Lundquist 2008, Wang et al 2014. Lots of studies have been conducted for climatic response of alpine forests over the MTP (Borgaonkar et al 2011, Liang et al 2014, Huang et al 2017, Guo et al 2019. However, the diverse results based on a few species, or at very local spatial scale are unable to represent the MTP regional mean response pattern. In order to understand whether the mean growth rate of high-elevation forests over the MTP is stimulated by warming or suppressed by moisture limitation, we developed a tree-ring width (TRW) dataset using 40 forest sites with 6 genera at elevations ⩾3700 m a.s.l. The TRW data were processed following standard dendrochronology methods (Fritts 1976) to compute the local high-elevation forest growth index (HEFGI, see section 2). All local HEF-GIs were averaged to estimate the mean HEFGI of the MTP. Local and MTP HEFGIs were calibrated to climate variables to test the climate-growth patterns at local scale and across the MTP.

Study region and climate data
The Tibetan Plateau, the highest and most extensive plateau in the world, has experienced a much faster warming rate than the global mean and regions at similar latitudes (Liu and Chen 2000). This trend has continued during the warming hiatus period of 1998-2012 (Duan and Xiao 2015) with continued warming at high elevations (Pepin andLundquist 2008, Wang et al 2014). Meanwhile, the South Asian summer monsoon, the main source of summer precipitation to most of the Tibetan Plateau south of 33 • N (hereafter monsoonal Tibetan Plateau, MTP, defined as 75-104 • E, 27-33 • N, figure 1) (Tian et al 2001), has weakened over the last century (Yao et al 2008, Salzmann andCherian 2015).
Summer precipitation on the Tibetan Plateau is dominated by the Asian summer monsoon south of 33 • N and by westerlies further north (Tian et al 2001, Yao et al 2013. Our study domain is bounded by 75-104 • E, 27-33 • N and is thus under the monsoonal regime. This area is also the most forested region of the Tibetan Plateau. Monthly climate data, including precipitation, self-calibrated Palmer drought severity index (scPDSI), minimum and mean temperature (MinT and MeanT, respectively) covering 1950-2010 were obtained from the Climate Research Unit (CRU) TS4.02 dataset (Harris et al 2014), which is a 0.5 • by 0.5 • gridded interpolation of existing instrumental data from meteorological stations. CRU data before 1950 is interpolated from distant meteorological stations and was not used.

Tree ring width data and HEFGI construction
We collected tree-ring width (TRW) measurements from 159 sites over the MTP (Filled circles in figure 1). Among them, 71 are from our own dataset, and 88 are from the International Tree Ring Data Bank (https://www.ncdc.noaa.gov/data-access/ paleoclimatologydata/datasets/tree-ring). Sampling sites with elevations ⩾3700 m a.s.l were defined as high-elevation forests (red circles in figure 1), as 3700 m a.s.l corresponds to the lowest alpine treeline elevation recorded over the MTP (Singh 2018, Shi et al 2019. This resulted in 40 out of the 159 sites being selected as high-elevation forests. The trees at high elevations of MTP are almost all conifers, mainly Abies and Juniper (28 out of 40 sites); Picea, Larix, Tsuga and Sabina can also be found (12 out of 40 sites). The TRW data from each site were detrended and standardized into site-chronologies using the signal-free regional curve standardization (sf-RCS) with the CRUST software (Melvin and Briffa 2014a, 2014b). On average 18% and 35% of trees' start year fall in the first 50 and 100 years of each chronology, respectively (Table S1). After selecting the periods with EPS (Expressed Population Signal) values >0.85 (Wigley et al 1984), the chronologies of six different genera were Z-score normalized (subtracted by the mean value and divided by standard deviation) and averaged into one record representing the highelevation forest growth index (HEFGI) of the entire MTP. By pooling data from 40 high-elevation forest sites with six genera, we assume that the local nonclimatic signals inherent in the TRW records, such as species competitions, growth release and topography changes are largely cancelled out, and thus that the common climate signal is strengthened. Sections of the MTP HEFGI with replications of less than eight sites (equivalent to an EPS value of 0.90) were removed to ensure a well representative of the whole MTP.

HEFGI vs. climate
To determine the key climate factors controlling highelevation forest growth, correlation and response functions between the HEFGI and 1950-2010 monthly CRU climate indices averaged over the MTP were calculated using Dendroclim2002 (Biondi and Waikul 2004). Both HEFGI and climate data were linearly detrended to ensure the reliability of the statistics. In order to test the temporal coherence stability, a 30-year moving correlation analysis was conducted between the 1950-2010 MTP HEFGI and key climate variables high-pass filtered by their 30-year loess smoothing trends. At a local scale, each of the high-elevation TRW chronologies was regressed to the key climate variables of the corresponding CRU grid point over 1950-2010.

High-elevation forest growth index (HEFGI) of the monsoonal Tibetan Plateau (MTP)
The mean correlation coefficient among HEFGIs is 0.46, suggesting a large common variability embedded in these records. The mean HEFGI in figure 1 depicts an above-average growth rate since the 1820s with an unprecedented growth enhancement since the 1980s, with highest values from 2000s onwards. The lowest growth rate was found in 1817-1819, two years after the 1815 Tambora volcanic eruption (Stothers 1984). HEFGIs composited by genera showed Abies, Juniper and Larix have experienced an above-average growth rate and an unprecedented fast growth since the 1820s and the 1980s as well (supplementary figure 1 (stacks.iop.org/ERL/15/054011/mmedia).

Tree growth vs. climate indices
Significant positive correlation was found between the HEFGI and July-October minimum temperature (MinT) from instrumental records (figure 2(a), 1950-2010) suggesting that the HEFGI was mainly dominated by thermal variability in the warm season. No significant correlation or response was found between HEFGI and moisture variability (here the self-calibrated Palmer drought severity index scPDSI and precipitation) (figures 2(a) and (b)). A total of 53% of the HEFGI variability on the MTP between 1950 and 2010 can be explained by the July-October averaged MinT (figure 1). Moreover, the 30-year moving correlation coefficients between the highpass filtered HEFGI and MinT are significant above 95% level over 1950-2010 in spite of a general decreasing trend (figure 2(c)). Thesame analysis for HEFGI and scPDSI shows the significance level is below 95% within the entire instrumental period ( figure 2(c)).
The Pearson correlation coefficients of individual site-level HEFGI time series with July-October MinT, precipitation, and scPDSI derived from the

Discussion and conclusion
A weakening of the South Asian summer monsoon intensity over the last century has been recorded in instrumental observations (Roxy et (Liang et al 2014, 2016). However, site-specific results analyses over a large elevation belt show a dominant and constant relation between forest growth and July-October MinT, and a much smaller effect of water availability variables on tree growth. The low correlations observed with moisture availability could also due to the inaccuracy of gridded datasets to represent the high spatial variability in moisture availability. Furthermore, the mean HEFGI is closely related to the MTP averaged MinT at interannual and decadal time scales, despite the increasing temperature, decreasing precipitation, and accelerating drought stress shown by the instrumental records in recent decades (supplementary figure 3). We thus conclude that the drying and warming trends over the MTP have not reached the level beyond which overall high-elevation tree growth show signs of hydrological stress, although warming-induced drought stress and associated growth declines were reported at some alpine treeline forests in the Himalayan region (Liang et al 2014).
The 1815 Tambora Eruption led to the coldest period over the last millennium (Ahmed et al 2013, Although increasing CO 2 has been projected to increase plant water-use efficiency and could stimulate forest growth (Keenan et al 2013), the HEFGI of the MTP is unlikely to be enhanced by increasing CO 2 concentrations since high-elevation forest growth is largely thought to be limited by low thermal availability in the growing season (Korner 1998, 2003, Hoch et al 2002, Korner and Paulsen 2004 rather than by scarcity of photosynthetic products, as higher concentrations of photosynthetic products (non-structural carbohydrates) are observed for trees growing at higher elevations on the TP (Shi et al 2006). Moreover, decreasing HEFGI values from the 1950s to the 1970s closely followed the cooling trend detected in the instrumental record (figure 1), while the air CO 2 concentration continued to increase, although such finding does not preclude per se the possibility of CO 2 concentrations affecting tree growth in future, non-analogous conditions.
The warming-enhanced forest growth is not limited to the alpine treeline ecotone, as observed in western north America (Salzer et al 2009), but was found over a wide elevation range in the sub-alpine forests of the MTP in this study and alpine treeline forests on the south-eastern TP (Shi et al 2019). Synchronous growth enhancement of treeline and subalpine forests was also reported in the European Alps (Paulsen et al 2000). However, with a lasting warming and drying trend, there are still large uncertainties for future hydrological conditions and high-elevation forest growth trends.

Data availability statement
The climate data used are available at http://www. cru.uea.ac.uk/data. Part of tree-ring width data are available at the International Tree-Ring Data Bank (ITRDB, https://www.ncdc.noaa.gov/data-access/ paleoclimatology-data/datasets/tree-ring). The rest tree-ring width data that support the findings of this study are available from the corresponding author upon reasonable request.