Asymmetric warming significantly affects net primary production, but not ecosystem carbon balances of forest and grassland ecosystems in northern China

We combine the process-based ecosystem model (Biome-BGC) with climate change-scenarios based on both RegCM3 model outputs and historic observed trends to quantify differential effects of symmetric and asymmetric warming on ecosystem net primary productivity (NPP), heterotrophic respiration (Rh) and net ecosystem productivity (NEP) of six ecosystem types representing different climatic zones of northern China. Analysis of covariance shows that NPP is significant greater at most ecosystems under the various environmental change scenarios once temperature asymmetries are taken into consideration. However, these differences do not lead to significant differences in NEP, which indicates that asymmetry in climate change does not result in significant alterations of the overall carbon balance in the dominating forest or grassland ecosystems. Overall, NPP, Rh and NEP are regulated by highly interrelated effects of increases in temperature and atmospheric CO2 concentrations and precipitation changes, while the magnitude of these effects strongly varies across the six sites. Further studies underpinned by suitable experiments are nonetheless required to further improve the performance of ecosystem models and confirm the validity of these model predictions. This is crucial for a sound understanding of the mechanisms controlling the variability in asymmetric warming effects on ecosystem structure and functioning.

H istorical observations over a large section of the earth's land area suggest that minimum temperatures (T min ) have increased significantly faster than maximum temperatures (T max ) since 1950 -a phenomenon commonly referred to as asymmetric warming [1][2][3][4][5] . These observations are further supported by climate change scenarios predicting faster increases in T min than T max particularly in mid to high northern latitudes and in arid regions 6,7 . At the same time, a growing body of evidence from long-term observations [8][9][10][11][12][13] , manipulation experiments [14][15][16][17][18] and model simulations [19][20][21] has demonstrated differential impacts of increases in minimum and maximum daily temperatures on plant productivity and terrestrial ecosystems carbon budgets. However, most experiments have been conducted under diurnal constant (symmetric) warming simulations 22,23 , and many models only use daily, monthly, or even annual mean temperatures for the temperature parameterizations when simulating and predicting the responses and feedbacks of terrestrial ecosystems to global warming 24,25 . In contrast, few studies have to date been conducted where the differential warming has been explicitly incorporated to examine the impact of the observed asymmetries on terrestrial ecosystem behaviors 20,26,27 .
The effects of warming on plants and entire ecosystems also depend on interactions with other environmental factors such as precipitation, atmospheric CO 2 concentration, nitrogen depositions and general nutrient availability [28][29][30] . In addition, ecosystems located in different climatic zones are likely to respond differently to changes in these factors 12,13,31,32 . It is therefore important to understand the combined effects of asymmetric warming and changes in other environmental variables impact on fundamental metabolic ecosystem processes like photosynthesis and respiration.
Manipulative experiments are key tools to understand the mechanisms of ecosystem responses to climate change 22 . Nonetheless, establishing the impact of asymmetric warming on terrestrial carbon cycling in the field is a key challenge 32 , and it is very difficult to simultaneously simulate the interactive effects of precipitation, elevated CO 2 and temperature 33 . Ecosystem modeling is therefore highly instrumental to stimulate hypotheses formula-tion and to extrapolate results from very limited, selected ecosystem settings across ecosystems, wider geographic areas and into the future 28 .
In this study, we use a well-established process-based ecosystem model, Biome-BGC (BioGeochemical Cycles) 34 , to compare the differential effects of symmetric and asymmetric warming on net primary productivity (NPP) and resulting carbon balances of six contrasting ecosystems in northern China. Our main objectives are to determine how plant productivity and ecosystem carbon sequestration are affected by temperature change asymmetries under various environmental change scenarios, and how these responses relate to variations in precipitation and atmospheric CO 2 concentrations.

Methods
Ecosystem processes are modelled using the Biome-BGC, which can simulate biogeochemical and hydrological processes of multiple biomes, using daily meteorological data including maximum, minimum and average temperature, precipitation, vapor pressure deficit, daylight average shortwave radiant flux density, and length of the day between sunrise and sunset 34 . Several further variables like the average daytime temperature (T day ) and average night-time temperature (T night ) are calculated from recorded maximum and minimum temperatures and meteorological principles 35 , allowing for sunlight-dependent processes like photosynthesis to be driven by T day , while processes such as decomposition are driven by 24 h averages. At the same time, maintenance respiration (R m ) of all living tissues is driven by changing temperature conditions throughout the day. R m is calculated separately for sun and shade leaves and partitioned into night-and daytime respiration, with daytime respiration also needed to calculate net assimilation. R m of sapwood is calculated separately for night and day respiration based on T night and T day , respectively. R m of the root system finally is calculated based on the soil temperature, which is assumed to be the 11-day running weighted average of T day . Overall, the simulated photosynthesis and respiration processes are sensitive to asymmetric temperature patterns and form the basis for the subsequent model outputs including Net primary productivity (NPP), heterotrophic respiration (R h ) and net ecosystem production (NEP 5 NPP 2 R h ). We selected a total of three forest and three grassland ecosystems varying in their temperature and precipitation regimes on the north sections of the North-South Transect of Eastern China and the east sections of the China Grassland Transect, respectively 36 (SI: Figure S1). The Biome-BGC model was adjusted for the six selected sites with a set of site-specific parameters (Table 1). Plant eco-physiological parameters were used according to White et al. (2000) 37 , except where detailed sitespecific data were available (SI : Table S1). Since the model does not currently simulate mixed forest stands, we divided the temperate mixed forests (TMF) site into evergreen needle-leaf forest (ENF) and deciduous broadleaf forest and simulated them separately 38 . The results were then added given different weights according to the basal area fraction covered by the respective plant functional types 39 (0.35 for the ENF and 0.65 for the deciduous broadleaf forest, respectively).
Our initial analytical focus was on the differences in ecosystem carbon budgets when comparing symmetric versus asymmetric climate change. For this, we used four different scenarios 20 : ambient scenario corresponding to the historical recorded temperature data during the period of 1961-1990 (T amb ), symmetric warming (T sym ), double asymmetric warming (T asy2 ) and triple asymmetric warming (T asy3 ). The three scenarios for temperature increases were based on a combination of recorded recent temperature increases (SI: Figure S2) and the predicted future magnitude of temperature increases simulated by a regional climate model (RegCM3) under the A2 IPCC CO 2 emission scenarios (SRES A2) 40 (SI: Figure S3). In the second step, the interactive effects of changes in temperature, precipitation, and atmospheric CO 2 concentrations were investigated. The precipitation treatment had two levels: an ambient level corresponding to the historical mean precipitation amounts recorded during the period of 1961-1990 (P amb ), and precipitation change based on the 2071-2100 predictions from the RegCM3 (P cha ) 40 . The model MT-CLIM (Version 4.3) was used to compute meteorological variables not included in the standard weather station records and required by the Biome-BGC model 41 . The CO 2 treatment also had two levels: an ambient level corresponding to the historical concentrations recorded during the period of 1961-1990 (C amb ) based on the Mauna Loa measurements (http://co2now.org/), and a scenario taking into account the gradual predicted increase in atmospheric CO 2 concentrations from 626 ppm v in 2071 to 836 ppm v in 2100 (C inc ) as predicted by the SRES A2 emission scenario data 42 .
Analysis of covariance was used to assess the effects of the different temperature treatments on NPP, R h and NEP under the four scenarios, respectively. To avoid overinterpretation of modeled values, rigorous significance tests for the interactive effects of the three factors temperature, precipitation and atmospheric CO 2 concentration were not attempted. Instead, response patterns of each ecosystem were identified using the method outlined by Luo et al. (2008) 28 .

Results
Net primary productivity (NPP). At most of our study sites, asymmetric warming is predicted to have a significant impact on NPP under the various environmental change scenarios (Figure 1). Under the control scenario, a significantly lower NPP is predicted for T sym than for both T asy2 and T asy3 scenarios for all ecosystems except for BCF and MStp. Furthermore, significant differences in NPP are computed between T asy2 and T asy3 scenarios for the two forest ecosystems TMF and DBF.
In scenarios taking into account predicted changes in precipitation (P cha ), NPP is significantly higher in both T asy2 and T asy3 in comparison to the T sym scenario for TMF, DBF and TStp, while no significant differences are predicted between T asy2 and T asy3 . For BCF and DStp, NPP predictions are significantly higher for T asy3 in compar- www.nature.com/scientificreports ison to the T sym scenario. By contrast, the different warming treatments has no significant effect on NPP for the MStp.
When increases in CO 2 concentrations are taken into account (C inc scenarios), NPP shows significant differences between all three warming scenarios for TMF and DBF in the rank order T asy3 . T asy2 . T sym . NPP in T asy3 is also significantly higher than in T sym for all three steppe ecosystems. In contrast, no significant changes in NPP for any of the three warming treatments are predicted for BCF.
Under the P cha 3 C inc scenarios, NPP shows significant differences between all three warming treatments for DBF in the order T asy3 . T asy2 . T sym . NPP is also significantly higher under the T asy3 scenario in comparison to T sym for TMF, TStp and DStp. No significant differences between scenarios are recorded for BCF and MStp.
Interactive effects of warming with C inc on NPP are positive at all sites, while the magnitude of these effects varies (Table 2). However, the interactive effects of warming and P cha are negative for DBF, MStp and DStp. The three-way interactions of warming with P cha 3 C inc are positive for BCF, TMF and DBF. The effects of the remaining two-way and three-way interactions are small in magnitude and not consistent among the three treatments at each site.
Heterotrophic respiration (R h ) and Net ecosystem productivity (NEP). Differences between simulated NPP and R h are small when seen in relation to their overall magnitude, and the overall response pattern of modeled R h in the different treatments ( Figure 2) is similar to that for NPP. The three-factor combinations of T, C inc and P cha consistently stimulates R h , whereas joining temperature regimes individually with either C inc or P cha does not cause consistent response patterns amongst the sites ( Table 2).
The overall response patterns of NEP to the three warming treatments differs strongly to that modelled for NPP and R h , with no significant differences resulting for the different temperature treatments under any of the various environmental change scenarios (control, P cha , C inc , or P cha 3 C inc ) (Figure 3).
Similar to the patterns of NPP, the interactive effects of temperature increases with C inc are generally positive for NEP ( Table 2). The interactive effects of warming and P cha are positive for BCF, but negative for MStp. The three-way interactions of warming with P cha 3 C inc are chiefly negative for TMF and MStp. The other twoway and three-way interactions effects on NEP are small in mag-nitude and highly variable amongst warming treatments for each ecosystem.

Discussion
In agreement with reports based on historical data analyses 12 and local experimental observations from the TStp 16 , our model suggests that NPP is significant larger when asymmetries are taken into consideration under various environmental change scenarios at the majority of our study sites. In the BIOME-BGC, day-and nighttime warming could have different impacts on the NPP induced by the bias of climate forcing both directly via alterations of leaf processes and indirectly via changes in soil water availability and soil nutrient mineralization rates 34 . This pattern is underpinned by previous modeling simulations [19][20][21] . All these studies report that asymmetries in climate change patterns have a significant impact on ecosystem productivity, highlighting the great importance to include temperature change asymmetries in future experimental and model studies to realistically project responses and feedbacks of an ecosystem's carbon cycle to climate change [32][33] . With photosynthesis occurring during daylight hours and plant and microbial respiration occurring continuously, it could be expected that the latter is much more strongly affected by the strength of asymmetries 8,12,13 . Nonetheless, our model outputs indicate that NPP and R h show fairly similar response patterns to temperature increases under the various environmental change scenarios at most of the study sites. As a consequence, NEP remains widely unaffected by the degree of asymmetric temperature change in the investigated ecosystems. This result indicates that increases in NPP cannot simply be equated to more carbon sequestration, as other ecosystem processes appear to counter-balance any NEP changes. More importantly, it also strongly suggests that processes of photosynthesis, respiration and carbon sequestration are considered as tightly linked, with photosynthesis and respiration appearing as entities closely coupled through carbon and nutrient supply and demand feedbacks 16 . Daytime warming alters net photosynthesis, which supplies the ecosystem with substrates for respiration at night. Night warming, however, does not only affect night-time ecosystem respiration, but may also stimulate plant compensatory photosynthesis during the following day by the depletion of leaf carbohydrates at night 14,16,43 . However, like most Table 2 | Relative strength of two-or three-way interactive effects on net primary production (NPP), heterotrophic respiration (R h ) and net ecosystem production (NEP)  [44][45][46] , that further impact ecosystem productivity and carbon sequestration 47,48 . We therefore suggest that more attention should be paid to the structural and functional responses of carbon-related processes to changes in maximum and minimum day and night temperatures in the current generation of ecosystem models.  Our results indicate simple additive effects of the interactive effects of temperature, CO 2 concentrations and precipitation are rare, which is consistent with reports based on experiments manipulating temperature and atmospheric CO 2 concentrations 30 . Overall, single-factor response models may be misleading, creating unreliable predictions of ecosystem responses to multifactorial global change patterns, a trend already observed in temperaturefocused experimental studies 16,27 . Our study further supports the need for more multifactorial experiments including not only the asymmetric shifts in temperature, but also the influence of precipitation regimes, nutrient availability and atmospheric CO 2 concentrations to improve predictions of ecosystems responses to global   29,49 and allow an improved model parameterization and validity.
Models based on the interactions of all three factors considered in our study reveal substantial differences in the magnitude of effects between sites, which somewhat contradicts reports from earlier investigations 28 . This outcome highlights the importance of the local environment and ecosystem structure for the assessment of ecosystem carbon budgets and their response to asymmetric warming 11,13,32 . While the present analysis was restricted to a limited number of sites focused only on boreal and temperate ecosystems, we acknowledge that particularly the response of tropical and subtropical ecosystems to asymmetric warming is not well researched at present and merits further investigation.