Nonlinear response of ecosystem respiration to multiple levels of temperature increases

Abstract Global warming exerts profound impacts on terrestrial carbon cycles and feedback to climates. Ecosystem respiration (ER) is one of the main components of biosphere CO2 fluxes. However, knowledge regarding how ER responds to warming is still lacking. In this study, a manipulative experiment with five simulated temperature increases (+0℃ [Control], +2.1℃ [warming 1, W1], +2.7℃ [warming 2, W2], +3.2℃ [warming 3, W3], +3.9℃ [warming 4, W4]) was conducted to investigate ER responses to warming in an alpine meadow on the Tibetan Plateau. The results showed that ER was suppressed by warming both in dry and wet years. The responses of ER to warming all followed a nonlinear pattern. The nonlinear processes can be divided into three stages, the quick‐response stage (W1), stable stage (W1–W3), and transition stage (W4). Compared with the nonlinear model, the linear model maximally overestimated the response ratios of ER to warming 2.2% and 3.2% in 2015 and 2016, respectively, and maximally underestimated the ratio 7.0% and 2.7%. The annual differences in ER responding to warming were mainly attributed to the distinct seasonal distribution of precipitation. Specially, we found that the abrupt shift response of ER to warming under W4 treatment in 2015, which might be regulated by the excitatory effect of precipitation after long‐term drought in the mid‐growing season. This study highlights the importance of the nonlinearity of warming effects on ER, which should be taken into the global‐C‐cycling models for better predicting future carbon–climate feedbacks.


| INTRODUC TI ON
Under global warming, the earth's surface temperature has increased 0.76°C since the industrial revolution and is expected to increase to 1.1-3.1°C by the end of this century (Stocker et al., 2013). Global warming affects terrestrial carbon cycles, which can cause positive or negative feedbacks to future climates (Brient & Bony, 2013;Luo, 2007;Melillo et al., 2002). Ecosystem respiration (ER) is one of the largest terrestrial carbon fluxes (Luo, 2007). The model simulation and field observations showed that the annual variation of CO 2 concentration in atmosphere is closely related to the ER fluctuation (Cox, Betts, Jones, Spall, & Totterdell, 2000;Kato et al., 2004;Luo, 2007;Niu, Sherry, Zhou, & Luo, 2013). Therefore, understanding how ER responds to climatic change is critical for predicting the carbon-climate feedbacks at regional to global scales.
Warming could stimulate ecosystem carbon release across various terrestrial biomes in simulated warming experiments (Niu et al., 2013;Wan, Hui, Wallace, & Luo, 2005). This is largely attributable to that elevated temperature could directly stimulate root and microbial respiration (Niu et al., 2008). However, warming does not necessary result in increasing in ER, because other biotic and abiotic factors could regulate their responses (Wan, Norby, Ledford, & Weltzin, 2007). Water availability may play a predominant role in regulating ER responses to warming, especially in arid and semiarid regions (Xia, Niu, & Wan, 2009). Distinct effects of warming on ER under a soil water gradient were reported in tundra ecosystem (Welker, Fahnestock, Henry, O'Dea, & Chimner, 2004). Lower soil water availability related with warming will exacerbate water limitations, offsetting parts of positive warming effects (Niu et al., 2008).
A growing body of evidences demonstrated that climate warming could alter plant community structure and composition (Botkin et al., 2007;Keryn & Mark, 2009). Warming effects on ER vary with plant species (Xia et al., 2009) and functional groups (Niu et al., 2013).
Except for these factors, low-and high-level warming induce different changes in soil water availability, water use efficiency (Quan et al., 2018), and community composition  and may lead to distinct ER responses to warming magnitudes.
However, warming effects on ER were largely studied in two level warming (control and warming), and consequently reveal simple linear increasing (Niu et al., 2013), linear decreasing (Fu et al., 2013), F I G U R E 1 Conceptual diagram of the response of ecosystem respiration (ER) to warming. Dashed line and solid line represent regression equations (solid line: nonlinear regression equations; dashed line: linear regression equations), which evaluate the response of ER to warming or no change (Chen et al., 2016;Lin et al., 2011;Xia et al., 2009) of warming effects. To improve our understanding about responses of ER to warming, we need experiments with multiple levels of temperature increases to investigate nonlinear responses (Luo, 2007).
To date, related studies for alpine ecosystem distributed in extreme environments are in severe shortage.
There are six types of possibilities in terms of ER response to a particular temperature range (warming 1 to warming 4 [W1-W4]; Figure 1). When ER nonlinearly increases or decreases with temperature increases, linear models may underestimate ER response to warming under W1-a and b-W4 (Figure 1a,b). Warming effects on ER were overestimated by linear models in a-b (Figure 1a Therefore, nonlinear models may reasonably reveal responses of ER to warming in multiple levels of temperature increases. There is growing evidence at global, regional, and local scales that interannual precipitation regimes have already become more extreme (Knapp et al., 2008), particularly in arid and semiarid regions. Importantly, transient CO 2 release ascribable to the "Birch" effect in response to precipitation pulses is a notable property of arid and semiarid ecosystems (Birch, 1958). This indicates carbon fluxes could respond quickly to precipitation events (Huxman et al., 2004). Previous studies have examined priming effects of precipitation pulses on soil respiration after drought in arid or semiarid ecosystems (Austin et al., 2004;Liu, Wan, Su, Hui, & Luo, 2002;Smart & Peñuelas, 2005). The increased soil respiration caused by priming effect contributes 16%-21% of annual total soil respiration (Lee, Nakane, Nakatsubo, Mo, & Koizumi, 2002). It is important to note that "Birch" effect is closely related to soil water condition, showing that precipitation pulses stimulate soil respiration more strongly in drier soil than that in wetter soil (Wang et al., 2010). The above-mentioned studies all reported priming effects of precipitation pulses on soil respiration, particularly for dry soils. However, knowledge on what role this stimulating effect plays in regulating responses of ER F I G U R E 2 Air temperature (°C, a: 2015; b: 2016), soil temperature at 5 cm depth (°C, c: 2015; d: 2016), soil moisture (%, e: 2015; f: 2016) at 5 cm depth and daily precipitation (mm, bars) during the growing season under different warming treatments to warming, and how the effects would influence the nonlinear response of ER to warming is even more unclear.
Studies on effects of climate warming on alpine ecosystems have been plentifully conducted on the Tibetan Plateau (Chen et al., 2016;Fu et al., 2013;Lin et al., 2011 (Zhu, Zhang, & Jiang, 2017). The long-term mean annual temperature and precipitation is −1. 28℃ and 430 mm (1955-2016), respectively. The growing season normally starts in mid-May and lasts until mid-September.

| Experimental design
Open-top chambers (OTCs) were used as passive warming devices based on the International Tundra Experiment design (Marion et al., 1997). The OTCs used in the current study were similar to those in other studies (Chen et al., 2016;Dabros & Fyles, 2010;De Frenne et al., 2010). Warming effects were regulated through changing the heights of OTCs. The treatments in this study include control (C), W1, W2, W3, and W4 (n = 3 per treatment). The OTCs were set up in October 2013 and made of 6 mm thick solar transmitting material. They are conical in shape, and are 40 cm (W1), 60 cm (W2), 80 cm (W3), 100 cm (W4) in height, respectively. The top sides of each OTC are 80 cm in order to maintain the same size. The bottom sides are 100 cm (W1), 110 cm (W2), 120 cm (W3), and 130 cm (W4) and cover an area of 2.60 m 2 (W1), 3.14 m 2 (W2), 3.74 m 2 (W3), and 4.39 m 2 (W4) at the ground, respectively. The 15 plots are separated by a 3.5-m buffer and arranged following a randomized block design.

| Data collection
In October 2014, cylindrical PVC rings (Diameter 9 cm and Height 5 cm) were inserted into soils to a depth of approximately 3 cm and emerge aboveground 2 cm. The ER were measured with an infrared gas analyzer (LI-6400; LiCor Inc., Lincoln, NE, USA) attached to a respiration chamber. The measurements were implemented from early June to early September, with an interval of approximately 5 days. In each measurement, we firstly obtained stabilized CO 2 S_ml in the natural state, and then set it as the target value. Second, after steady-state conditions, we set the delta value as 10 ppm. After instrument reduces CO 2 concentration to below target value of 10 (target value −10) within the chamber, it starts to work, and above target value of 10 (target value +10), it stops to work. This processes F I G U R E 3 Seasonal dynamics (mean ± 1 SE) of ecosystem respiration (ER) in 2015 (a) and 2016 (b) cycle three times in each plot. Each measurement was conducted during 9:00 and 12:00 a.m. of sunny days. Totally, 15 times of measurements were accomplished in 2015 and 2016, respectively.
Aboveground biomass was collected by clipping vegetation samples from 0.25 × 0.25 m sections (adjacent to the aluminum frame) at the peak growing season (Aug 22, 2015 andAug 15, 2016). After clipping, all aboveground plant matter was oven dried at 65℃ for 72 hr before being weighed. Three soil columns with a diameter of 7.0 cm were drilled at depths of 0-10, 10-20, and 20-30 cm.
A 1 × 1 m frame with 100 equally distributed grids (0.1 × 0.1 m) was placed above the vegetation canopy to measure vegetation coverage (1 × 1 m). Grids with plants appearing over 1/2 of the grid were marked as 1, otherwise marked as 0. The total number of grids within the frame is the actual coverage value. The cover was mainly measured in mid-growth season and late growth season and was accomplished 2-3 times in both growing seasons.
Air temperature and moisture at 10 cm aboveground were measured using the Vaisala HMP155A sensor (Vaisala, Helsinki, Finland).
Soil temperature and moisture at 5 cm belowground were measured at the centers of each plot using Campbell CS655 sensors (Campbell Scientific, Logan, UT, USA). Measurements of air temperature, soil

F I G U R E 4 Monthly precipitation (lines) and ecosystem respiration (ER) (bars) in 2015 and 2016
F I G U R E 5 Relationships between the response ratios of ecosystem respiration (ER) and air temperature changes at early season (a, b), late growing season (c, d), and the whole growing season (e, f) in 2015 (left) and 2016 (right). Dashed line and solid line represent regression equations (dashed line: linear regression equations; solid line: nonlinear regression equations) between the response ratios of ER to warming and air temperature changes temperature, and soil moisture were taken with 30-min intervals, and averages of the two measurements were stored as the hourly averages, and averages of the 48 measurements were the day averages. In each warming treatment (three plots), we installed air and soil sensors (soil temperature and soil moisture) in two of them, and used average of the two measurements (Zhu et al. 2017).

| Statistical analysis
Repeated-measures ANOVA (RMANOVA) were used to examine warming effects on ER over the growing seasons in 2015 and 2016.
The between-subject effects were treated as warming effect and the within-subject effects were time-of-season. To analyze the seasonal variations of ER response, the whole growing season was divided into two stages, early-growing season and late growing season. Then, one-way ANOVA was applied to analyze the treatment difference for ER at two stages and the whole growing season in 2015 and 2016.
The curve estimation was employed to analyze the relationship between ER and soil temperature and soil moisture. All statistical analyses were conducted with SPSS software (SPSS 20.0 for windows).
To examine the nonlinear responses, we calculated the response ratios of treatment (W1, W 2, W3, W4) to control (no temperature TA B L E 1 Coefficient of determination (Adjusted R 2 ), Akaike's information criterion (AIC) and Bayesian information criterion (BIC) from two different models predicting the effects of warming on ecosystem respiration in 2015 and 2016 Note. Compared to nonlinear models, the range of linear models underestimate or overestimate ecosystem respiration was displayed by underestimate (℃) and overestimate (℃), respectively, and the maximum overestimate or underestimate are represented overestimate (%) and underestimate (%), respectively. Precipitation also exhibited strong seasonal patterns in 3 years. In spring and summer, precipitations were less than the MAP (Figure 2e).
Particularly, in July of 2015, precipitation was 63.2% lower than the MAP (Figure 2e). In growing season of 2016, precipitation exhibited a single peak patterns, and the summer total was 39.1% higher than that of the MAP (Figure 2f). were lower in summer and higher in autumn of 2015, and were higher in summer and lower in spring and autumn of 2016 ( Figure 3). We further found that monthly mean ER between growing seasons coincided with monthly precipitation (Figure 4). These results may suggest that controls of precipitation pattern on seasonal variations in ER.

| Model fitting of the response of ER to air temperature changes
R 2 , AIC, and BIC values showed nonlinear models performed better than linear models for ER response to warming in 2015 and 2016, except for the early-growing season in 2015 (Figure 5a;  (Figure 5c; Table 2). However, the nonlinear model overestimated the response ratio 2.2% at 2.8-3.8°C warming (Figure 5c; at 2.0-3.6°C warming (Figure 5d; Table 2). During the growing season of 2016, the linear model overestimated the response ratios of ER by 2.8% and 3.0% at 1.7-1.9°C and 3.3-3.8°C warming, respectively, and underestimated by 2.5% at 1.9-3.3°C warming (Figure 5f; Table 2).

| Nonlinear responses of ER to warming
Nonlinear response processes of ER to warming are composed of sensitive response, stable response, and transition response. ER response to warming was more sensitive to low temperature increase  (Table 3; p < 0.1).
This indicated that transition response was observed under W4.
However, such a transition was not identified in 2016 (Table 3,

| Impacts of abiotic factors on ER
The exponential model indicated that ER negatively correlated with soil temperature, which explained approximately 5% of variations in ER over 2015 growing season (Figure 8a). Similar to 2015, the quadratic model revealed that ER also negatively correlated with soil temperature, and 16% variations in ER were explained over 2016 growing season (Figure 8c). The ER was positively correlated with soil temperature when it was lower than 10.12℃, but negatively correlated with soil temperature when it was higher than 10.12℃ (Figure 8c). The power model and quadratic model indicated that ER positively correlated with mean soil moisture in both years (Figure 8b,d). What is more, ER positively correlated with soil moisture when it was lower than 18.13%, but negatively with soil moisture when it was higher than 18.13% in 2016 ( Figure 8d).

| Negative nonlinear response of ER to warming
This study explicitly revealed that response of ER to warming all followed a nonlinear pattern, a feature that previous studies failed to Note. In this table, we only list the situations that compared W1, W2 or W3 (at least one of the three), W4 treatments significantly increased ER. Because only in this context, W4 may exist to break the stable state that composition of the W1, W2, W3 treatments (between the three were not significant).
Warming can exert negative effects on ER primarily through limiting soil water availability, especially in arid and semiarid regions (Xia et al., 2009). Lower soil water availability would restrict root, microbial activities, translating into reduced ER (Niu et al., 2008).
These passive effects have reported in same study area (Zhu et al., 2017). Resent study reveals water availability more than temperature drives carbon fluxes of alpine meadow (Zhu et al., 2017). In this study, positively linear dependence of ER upon soil moisture further supports these findings.
Plant growth effects on carbon fluxes are modulated by soil water availability in growing season (Liu, Cieraad, Li, & Ma, 2016).
For the alpine meadow ecosystem, K. pygmaea, as a dominant species, is shallow-rooted, and mostly utilize shallow soil water (Dorji et al., 2013). More importantly, it is a drought vulnerable species (Li, Wang, Yang, Gao, Liu, & Liu, 2011), and warming further exacerbate the vulnerability. Thus, warming significantly decreases K. pygmaea coverage (Supporting Information Figure S1).
Decreased plant cover could decrease canopy cover and increase bare soil evaporation, and consequently decrease aboveground plant respiration (Verburg et al., 2004; Supporting Information Figure S2). In addition, K. pygmaea belongs to dense bush fibrous root perennial plant, which could form a huge underground biomass (Liu, Sun, Zhang, Pu, & Xu, 2008). The negative effects of warming on K. pygmaea could decrease underground biomass , translating into reduced below-ground plant respiration. For alpine ecosystems, ER variations are controlled by plant respiration (Chen et al., 2016). In addition, nonlinear relationships between abiotic factors and ER may cause the nonlinear response of ER to warming in this study.

| Sensitive response to warming under low temperature increase
The ER was more sensitive to warming under low temperature increase (W1) for the alpine ecosystem (Supporting Information Figure   S3). The temperature sensitivity of ER is mainly related to temperature range (Lin et al., 2011;Tjoelker, Oleksyn, & Reich, 2001) and soil moisture (Flanagan & Johnson, 2005;Reichstein et al., 2002;Wen et al., 2006). The temperature sensitivity of ER was significantly affected by soil temperature, and weakened with increased temperature (Supporting Information Figure S4). This result is in accord with previous studies (Bekku, Nakatsubo, Kume, Adachi, & Koizumi, 2003;Lin et al., 2011;Zheng et al., 2009;Zhou, Wan, & Luo, 2007

| Stable and transition response to warming under high temperature increase
Our results demonstrated that ER exhibited stable and transition response to warming as temperature increases in 2015. But the transition response was not found in 2016. These contrasting responses may be related to the distinct seasonal precipitation distribution during the growing season. Previous studies have shown that the fluctuation of ER is regulated by the seasonal distribution of precipitation (Marcolla et al., 2011;Nijp et al., 2014;Ryan et al., 2015). 85% of ER in this study (data not shown). Consequently, warming had no significant effects on CO 2 emission of ecosystem (Bontti et al., 2009), and ER exhibited a flat response to warming. In contrast, adequate precipitation in growing season of 2016 resulted in sufficient soil moisture. Further increase in precipitation did not continue to rise in respiration (Liu, Zhang, Zhen-Zhu, Zhou, & Hou, 2012), even decreased it (Cavelier & Penuela, 1990). ER negatively correlated with soil moisture when soil moisture was higher than 18.13% in this study. The standardized major axis estimation regression showed that no significant change in regression slopes between ER and soil moisture were found from W2 to W4 (Supporting   Information Table S1). This could further supply an additional explanation for the stable response of ER to warming in 2016.
In addition, the lower soil moisture content was, the stronger the excitatory effect was (Liu et al., 2002;Shi et al., 2006). Therefore, the increase in soil respiration triggered by precipitation pulse is proportional to drought time and inversely proportional to soil respiration rate before precipitation (Xu et al., 2004

| CON CLUS IONS
By conducting a warming experiment on an alpine meadow over two growing seasons, this study showed that ER displayed a nonlinear pattern with temperature increases. Further, the nonlinear processes could be divided into three stages during the dry growing season. First, ER was more sensitive to low temperature increase, which may be attributed to the quick changes in soil conditions and vegetation coverage. Second, ER displayed a flat response to warming due to the combined effect of biotic and abiotic factors. Finally, precipitation at late growing season could rapidly stimulate soil respiration. The flat state was broken by the excitatory effect of precipitation under high temperature increase. This study demonstrated the nonlinearity is likely a general feature for ER in response to warming, and these nonlinear processes and regimes should be taken into the global-C-cycling models for better predicting future carbon-climate feed backs.

ACK N OWLED G M ENTS
This study was financially supported by the National Key

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest