Recent CO2 rise has modified the sensitivity of tropical tree growth to rainfall and temperature
1 INTRODUCTIONTropical forests account for a third of global gross and net primary productivity, store 25% of the carbon in terrestrial ecosystems (Beer et al., 2010; Bonan, 2008) and drive fluctuations of the land carbon sink (Friedlingstein et al., 2019). Their responses to future atmospheric CO2 (c a) rise and warming will thus influence the pace of climate change (Mitchard, 2018). A key challenge in model predictions of tropical forest responses to these changes is the uncertainty of the magnitude of effects of elevated CO2 levels, commonly referred to as ‘CO2 fertilization effects’ (Cernusak et al., 2013; Fatichi, Pappas, Zscheischler, & Leuzinger, 2019; Körner, 2009; Lewis, Edwards, & Galbraith, 2015; Settele et al., 2014; Terrer et al., 2019). Rise of c a increases photosynthetic efficiency (Lloyd & Farquhar, 2008), reduces water use (Cernusak et al., 2013) and may thus stimulate tropical tree growth (Cernusak et al., 2013). These effects can mitigate negative impact of warming on tropical forest productivity, as predicted by dynamic global vegetation models (DGVMs; Cox et al., 2013; Huntingford et al., 2013). Yet, most DGVMs likely overestimate effects of CO2 rise as they do not include nutrient or hydraulic limitations (Fatichi et al., 2019; Körner, 2009; Smith et al., 2014; Yang, Thornton, Ricciuto, & Hoffman, 2016). Model benchmarking using empirical studies on c a‐rise effects is therefore needed (Clark et al., 2017; Zuidema, Poulter, & Frank, 2018). The effects of c a rise on tropical tree photosynthesis and water use are modified by rainfall and temperature (Cernusak et al., 2013; Körner, 2009; Lloyd & Farquhar, 2008) and may thus vary along climatic gradients or with temporal climatic fluctuations. Such c a × climate interactions likely drive biome‐specific c a responses (Baig, Medlyn, Mercado, & Zaehle, 2015; Hickler et al., 2008; Norby et al., 2016). Reduced drought sensitivity under c a rise may be particularly important in regions with low precipitation (P) and during dry years (Cernusak et al., 2013; Fatichi et al., 2016; Zuidema et al., 2013). The effect of c a rise on photosynthesis is likely also modified by temperature (T ). A positive effect of c a rise on photosynthetic efficiency can be stronger in warmer regions or during warmer years, as photorespiration increases with temperature and stronger c a‐driven reduction in photorespiration is therefore expected at high temperature (Cernusak et al., 2013; Long, 1991). On the other hand, c a rise may also lead to growth reduction at high temperatures if stomatal closure reduces transpiration, leading to leaf warming beyond the optimum temperature for photosynthesis (Cernusak et al., 2013; Wood, Cavaleri, & Reed, 2012). Studies on c a × climate interactions of (sub‐)tropical trees have been limited to seedling experiments under controlled conditions (Cernusak et al., 2011; Fauset et al., 2019; Kelly, Duursma, Atwell, Tissue, & Medlyn, 2016; de Oliveira & Marenco, 2019a, 2019b; Quentin, Barton, Crous, & Ellsworth, 2015). Responses of canopy trees may be different, as these experience higher irradiance, temperature and vapour pressure deficit. Understanding tropical canopy tree responses to c a × climate interactions is crucial as they are responsible for the main share of photosynthesis and biomass in tropical forests (Slik et al., 2013) and are particularly sensitive to drought (Phillips et al., 2009). Tree‐ring analysis represents a powerful method to evaluate effects of c a rise on canopy tree physiology and growth (Zuidema et al., 2013). So far, tropical tree‐ring studies have consistently shown positive trends in intrinsic water‐use efficiency (iWUE) over the past century (Hietz, Wanek, & Dünisch, 2005; Loader et al., 2011; Nock et al., 2011; Rahman, Islam, Gebrekirstos, & Bräuning, 2019; van der Sleen et al., 2015), but no consistent growth stimulation (Groenendijk et al., 2015; Nock et al., 2011; Rahman et al., 2019; van der Sleen et al., 2015). These studies were conducted using trend analyses of growth or iWUE. While providing important information on changes in tree physiology and growth, such trend analyses cannot detect interactive effects of c a and climate on tree growth. Annual tree growth from dated tree rings allows evaluating modifications of sensitivity of tree growth to climate variation by c a rise. Such analyses can be considered as an observational version of c a × climate experiments in climate chambers, open‐top chambers or Free Air CO2 Enrichment experiments and may also be used to benchmark DGVMs (Baig et al., 2015). To our knowledge, tree‐ring‐based analyses of c a × climate interactions are very scarce (Voelker et al., 2017; Wyckoff & Bowers, 2010) and have not been conducted for any tropical tree species. Here we use tree‐ring‐width chronologies for a long‐lived tropical tree species (Toona ciliata ) from four sites with contrasting climate, to evaluate c a × climate interactions. Mean climate at these sites allowed comparing c a × climate interactions at cooler and warmer sites. In Figure 1, hypothesized interactions are represented by differences in slopes: if the slope of a relation between growth and climate is steeper for high c a than low c a, the interaction is positive; if the slope is less steep at high c a, the interaction is negative. We tested two alternative hypotheses on c a × P interactions (1a and 1b; Figure 1) and two on c a × T interactions (2a and 2b; Figure 1):
2 MATERIALS AND METHODS
2.1 Study speciesWe studied T. ciliata (Meliaceae), a large‐stature and long‐lived pioneer tree species distributed in South Asia, South‐East Asia and Australia (Figure 2b). T. ciliata is usually deciduous or semi‐deciduous but under optimal growing conditions saplings can show opportunistic non‐deciduous behaviour (Heinrich & Banks, 2006). This species is ideally suited to study c a × climate interactions as it forms growth rings that are reliably dated (Heinrich, Weidner, Helle, Vos, & Banks, 2008; Rahman, Islam, & Islam, 2017; Vlam, Baker, Bunyavejchewin, & Zuidema, 2014), produces high‐quality growth chronologies, exhibits clear climate–growth relations (Heinrich et al., 2008, 2009; Vlam et al., 2014), reaches >150 years in age (Heinrich et al., 2008; Vlam, van der Sleen, Groenendijk, & Zuidema, 2017) and occurs across a large climatic range (Figure S1). T. ciliata forms distinct (semi) ring‐porous growth rings, characterized by initial parenchyma and large early‐wood vessels (Heinrich & Banks, 2005; Islam, Rahman, & Bräuning, 2018).
2.2 Study sites and climate dataWe used tree‐ring‐width measurements from published chronologies of T. ciliata . Tree‐ring data were available from four sites in Australia and South and South‐East Asia which vary in rainfall and particularly in temperature (Table 1; Figure 2a). Characteristics of the original chronologies are included in Table S1 and show a high dating accuracy and a strong common response to environmental drivers at all sites. All sites are located in nature reserves with no signs of logging or silviculture. Annual rainfall distribution at all sites is unimodal (Figure 2a), with peaks during northern hemisphere (SE Asian sites) or austral summer (Australian sites). Annual temperature also fluctuates during the year: wet seasons are warmer and dry seasons cooler (Figure 2a).
|Site||Name||Country||Study period||Wet season||Leafless||N|
|HKK||Huai Kha Khaeng||Thailand||1953–2010||May–Oct||0.5 months||46|
- Sites are ordered with increasing mean annual temperature. ‘Leafless’: period during which trees were observed to be deciduous; ‘Wet season’: months with >100 mm rainfall; N : number of sampled trees. Original publications: ATH (Heinrich et al., 2008), LAM (Heinrich et al., 2009), HKK (Vlam et al., 2014) and RKW (Rahman et al., 2018).
2.3 Data collection and tree‐ring measurementsAt all sites, increment cores were collected from natural populations of T. ciliata in extensive forest areas (>500 ha in all cases). At ATH and LAM sites, sampled trees were selected to be sub‐dominant or dominant, with 30–40 m (ATH) or 40–50 m (LAM) in height and 80–120 (ATH) or 80–300 cm (LAM) in diameter at breast height (DBH). At HKK and RKW, trees of >5 cm DBH were selected and maximum DBH was 115 and 25 cm DBH, respectively. Coring height was 1.0 m at HKK and 1.3 m at other sites. While the selection of tree sizes differed between sites, chronology statistics were similar (Table S1). A total of 20–46 trees were sampled per site, with 1–2 (ATH, LAM and RKW) or 2–3 cores per tree (HKK). Sample preparation, scanning, microscopy, ring measurements and cross‐dating followed standard and fully exchangeable dendrochronological practices, resulting in ring‐width measurements to the nearest 0.01 mm. Dating of rings at the two Australian sites followed the Schulman convention (Schulman, 1956), implying that the ring is dated according to the year in which ring formation starts. So, a growth ring at the ATH site that started to be formed in end of 1980 (start of austral summer) is dated ‘1980’, but continues into 1981 and is influenced by climatic conditions during both 1980 and 1981: wet season (December 1980–March 1981) and dry season (October–November 1980 and April–May 1981). The raw ring‐width measurements of trees that were part of the chronology are the basis for the analyses in this study. By using only these trees, we have a high certainty that only correctly dated individual trees are included in the analyses. A total of 5,318 ring‐width measurements were included in our analyses.
2.4 Statistical analysesWe evaluated the significance and magnitude of c a × climate interactions on T. ciliata diameter growth using mixed‐effect models (MEMs; Zuur, Ieno, Walker, Saveliev, and Smith (2009)). We applied MEMs instead of more commonly used correlation analyses in tree‐ring research as MEMs allow (a) testing for effects of multiple and interactive explanatory variables of ring‐width variation at the same time, (b) accounting for the repeated measurement structure of the data (by including tree individual as a random factor) and (c) quantifying the degree to which variation in ring width is explained by intrinsic (within individual) and extrinsic (environmental factors and their interactions) sources of growth variation. MEMs have recently been used to study climate–growth relations of tree‐ring data (Galván, Camarero, & Gutiérrez, 2014; Gea‐Izquierdo, Cherubini, & Cañellas, 2011), also for our study species (Vlam et al., 2014). For all sites, tree‐ring width was log‐transformed to obtain a normal distribution of residuals. We chose not to detrend ring‐width series, as this may affect the detection of climate–c a interactions. Instead, we accounted for ontogenetic effects on ring width by adding tree age at ring formation (‘Age’) as explanatory variable in the MEM. Age was estimated as cambial age for HKK (Vlam et al., 2014) or estimated by the number of rings measured in each series, for the other sites. The former accounts for missing distance to the pith; the latter does not. Means of these age estimates were 80, 84, 46 and 58 years for ATH, LAM, RKW and HKK, respectively; maximum tree ages were 185, 144, 83 and 150 years. We also tested whether adding age2 to MEMs would improve model fit or alter model output, but this was not the case. Note that non‐linear ontogenetic effects on growth were also accounted for by log transformation of ring width. We accounted for growth heterogeneity within the population by adding TreeID as random variable (random intercept) in the MEM. We did not include TreeID as random slope because we did not aim to account for individual variation in climate–growth relations and as this would involve adding a large number of random slopes which results in highly complex models. We checked whether adding an autocorrelation structure to account for temporal autocorrelation in tree growth (Vlam et al., 2014) would change the MEM output. This was not the case, and we therefore did not include autocorrelation in the analyses presented here. Four groups of explanatory variables were included in our MEMs: (a) T max and P, for both the wet season and dry season; (b) c a; (c) interactions of c a and season‐specific T max and P and (d) age. All explanatory variables were scaled (mean = 0, SD = 1) prior to inclusion in MEMs, to obtain standardized coefficients and allow for direct comparison of the strength of effects. For each MEM, we evaluated collinearity of the explanatory seasonal climate variables; we did not find cases for which the variance inflation factor exceeded 5. Per site, we constructed three sets of MEMs. Set A: to evaluate general sensitivity of Toona growth to climate, we performed a climate–growth analyses using only climate and age variables (groups a and d). Set B: to test Hypotheses 1 of c a × climate interactions, we ran models including all variables (groups a–d), with age to account for ontogenetic effects on growth. A comparison of the model output of sets A and B allowed us to evaluate the extent to which climate–growth relations are modified by c a × climate interactions. Set C: as set B, but for a common period of 1953–1998 to verify robustness of model output to differences in duration and timing of study periods among the study sites. We ran one MEM per site and did not combine sites into one MEM, in order to prevent modelling three‐way interactions. We used backwards selection of explanatory variables, based on AIC change (ΔAIC ≥ 2), taking the simplest model in case of ΔAIC < 2. Analyses were performed using the lme function of the nlme package in R (R Core Team, 2019). We calculated conditional and marginal R 2 to evaluate variation explained by fixed effects alone and fixed and random effects together (MuMIn package).
3.1 Accounting for ontogeny and individual heterogeneityRing‐width data for all four sites show strong annual variation in growth of individual trees, with a considerable degree of common growth variation (Figure 3). The clear and synchronous occurrence of narrow and wide rings in these raw ring‐width series illustrate that our study species are climate sensitive. The raw ring‐width data also show that individual trees differ widely in growth.
3.2 Growth sensitivity to climatic variationThe set‐A mixed‐effects models, in which effects of climatic variation on ring width were tested (Table 2), showed positive effects of dry‐season rainfall on growth at two sites (ATH and HKK) and negative wet‐season rainfall effects at one site (HKK). The latter is possibly associated with increased cloud cover during wet years.
|Site||Model set||AIC||R 2 m||R 2 c||df||N|
|ATH||A. Climate effects only||3,662||.22||.35||1,541||1,580|
|B. With climate × c a interactions||3,635||.25||.38||1,537||1,580|
|C. As B, but for common period||3,416||.25||.38||1,441||1,482|
|LAM||A. Climate effects only||1,903||.16||.45||915||939|
|B. With climate × c a interactions||1,886||.18||.47||911||939|
|C. As B, but for common period||1,738||.20||.48||846||874|
|RKW||A. Climate effects only||2,646||.25||.37||1,051||1,079|
|B. With climate × c a interactions||2,647||.26||.37||1,050||1,079|
|C. As B, but for common period||124||.12||.22||663||692|
|HKK||A. Climate effects only||5,327||.11||.25||2,237||2,289|
|B. With climate × c a interactions||5,299||.13||.27||2,235||2,289|
|C. As B, but for common period||3,926||.11||.28||1,686||1,738|
- Sites ATH and LAM are cooler; RKW and HKK are warmer. For the selected models, the table includes AIC, marginal R 2 (R 2m, based on fixed effects only), conditional R 2 (R 2c, both fixed and random effects), degrees of freedom (df ) and sample size (N ). Coefficients are shown in Figure S6 (set A), Figure 4 (set B) and Figure S7 (set C).
3.3 Climate × c a interactions in MEMsThe effect of c a on the climate–growth relations of T. ciliata can be inferred by comparing results of MEM sets A and B (without or with c a × climate interactions). First, at three of the four sites, adding c a × climate interactions improved model fit and increased variance explained by fixed variables (R 2m; Table 2). Thus, adding c a × climate interactions increased explained variation in annual ring width of our study species. Yet, change in R 2m was small (0.01–0.03; Table 2), which indicates a reorganization of explained variance rather than the explanation of additional variance. Second, inclusion of c a × climate interactions caused shifts in the sets of climate effects that were significant (compare Figure S6 and main effects in Figure 4). While the sign of significant coefficients never shifted between sets A and B, significant climate effects in set A sometimes shifted to non‐significant in set B, and vice versa. For instance, at the ATH site, the model including interactions (Figure 4) contained significant negative effects of T max during the dry season and rainfall during the wet season, which were absent from the model without interactions (Figure S6).
3.4 Rainfall × c a interactionsRainfall × c a interactions in MEMs (set B) differed between sites (Figure 4). At one of the two cooler sites (LAM), we found a negative interactions for wet‐season precipitation. This is consistent with Hypothesis 1a: the expected response of a c a‐induced reduction in drought sensitivity. At the two warmer sites (HKK and RKW), we found positive rainfall × c a interactions. This implies that positive slopes of the ring‐width versus rainfall relations became steeper at higher c a levels or negative slopes became less negative. These interactions are in accordance with Hypothesis 1b and are expected at warmer sites. At these sites, the c a‐induced reduction of g s may cause T leaf during dry years to frequently exceed optimum temperature for photosynthesis. This effect may be enhanced if dry years also tend to be warm years, which was the case for both warmer sites (negative correlations between T max and P, Table S2). In addition, these two sites also experienced increase in T min over time (Figure S3), which may counteract the reduction in drought sensitivity occurring with gradual c a rise. In our MEMs, such warming‐induced change in drought sensitivity could become apparent as positive c a × rainfall interactions, even if the mechanism is warming. Thus, increase in seasonal T max and T min during the study period at RKW and HKK may have induced an increase in drought sensitivity that was stronger than the reduction in drought sensitivity induced by c a rise. Analyses for the common period (Set C, Table 2; Figure S7) yielded the same c a × rainfall interactions for two sites (ATH and LAM), a seasonal shift at one site (RKW) and loss of a temperature interaction at the HKK site. Thus, overall c a × temperature interactions were quite robust to the shift in period.
3.5 Temperature × c a interactionsTemperature × c a interactions in MEMs (set B) also differed between sites (Figure 4). We expected positive interactions at cooler sites (LAM & ATH) because of the stronger reduction in photorespiration during warm years and because at these sites of T leaf does not often exceed the optimum temperature for photosynthesis. The positive interactions of c a with T max during the dry (=cooler) season at LAM and the positive interaction of c a with T min in the wet season at ATH are consistent with such a response. Yet, a negative interaction (T max in wet season) was also found at LAM, which is not expected for a relatively cool site where T max during the wet season averages 25°C. The negative correlation between wet‐season T max and rainfall may help explaining these interactions: hotter wet seasons also tend to be drier (Table S2). Another possible explanation is that the incidence of days during which T max leads to leaf temperatures exceeding optimum for photosynthesis is similar for cooler and warmer sites during the wet season. This seems to be the case for the LAM site (Figure S4). At the two warmer sites, one negative interaction was found (HKK), in accordance with Hypothesis 2a. This may be explained by the c a‐induced reduction in g s and increase in T leaf, and is supported by the high frequency of hot days (T max > 30°C; Figure S4). Yet, negative interactions at HKK may also have resulted from the significant warming (Figure S2), which occurred alongside c a rise. Effects of c a rise and warming cannot be distinguished and warming may have aggravated negative effects of T max during the wet season (HKK). Models for the common period (set C) yielded the same c a × T max interactions for the cooler sites, a seasonal shift at one warmer site (HKK) and a new (positive) interaction at the other warm site (RKW). The latter change is not in accordance with the expected response for a warm site (and during the warmer season), and we do not have an explanation for this result. In general, the change from set B to set C did not cause a major shift in MEM results that could be explained by recent global warming and recent increase in hot years at the warmer sites (Figure S4).
4 DISCUSSIONOur analysis of ring‐width measurements of T. ciliata revealed that recent c a rise caused a significant change in the sensitivity of tree growth to climatic variation. This shift in sensitivity was evidenced by a better fit of statistical models that included c a × climate interactions (at three sites), the presence of robust c a × climate interactions in these models, and the changes in climate–growth relations between models with and without c a × climate interactions. We found that c a × climate interactive effects on tree growth varied across sites that differ in mean climate. For the two cooler sites in our dataset, results agree with hypothesized reduction in drought sensitivity (negative interaction, Hypotheses 1a). Temperature sensitivity shifted in different directions: the expected positive effect of c a rise on ring width was found during the cooler dry season (Hypothesis 2b), whereas negative interactions were found during the hotter wet season (Hypothesis 2a). For the two warmer sites, drought sensitivity increased under c a rise and temperature sensitivity shifted such that hot years caused a stronger growth reduction. Both responses were in accordance with hypotheses for warm sites (Hypotheses 1b and 2a).
4.1 Climate × c a interactive effects on Toona growthThe studied T. ciliata trees included in this study grew during a period when c a increased by 28%. For one of the sites (HKK), stable isotope analyses revealed that c a rise caused an increase in iWUE of approximately 35% for our study species during this period (Nock et al., 2011; van der Sleen et al., 2015). Thus, Toona physiology has responded to c a rise at that site, and similar responses are likely for the other sites, given the generic c a‐induced iWUE increases for tropical tree species worldwide (van der Sleen, Zuidema, & Pons, 2017). Increased iWUE response likely involved both a reduction in transpiration (due to a reduction in stomatal conductance) and increased photosynthesis (due to an increase in c i). Tree‐ring‐based analyses of growth trends for the RKW and HKK sites did not yield evidence that these physiological changes have stimulated growth (Groenendijk et al., 2015; Nock et al., 2011; Rahman et al., 2018; van der Sleen et al., 2015). This suggests that tree growth is more strongly limited by rainfall, heat and/or nutrient availability than by c a; or that trees respond to c a rise by increasing organ turn‐over rates, respiration or investments in other tree parts (van der Sleen et al., 2015). One of these factors—limitation of diameter growth by high temperature and low rainfall—has indeed been observed of T. ciliata , both in the analyses included in this study and elsewhere (Heinrich et al., 2008; Rahman et al., 2018; Shah & Mehrotra, 2017; Vlam et al., 2014). The above insights into growth‐determining factors and c a responses lead to the expectation that c a rise would modify climate–growth relations of Toona , or—put differently—that effects of c a rise on Toona growth depend on its sensitivity to climatic fluctuations. Our results confirm that this is the case. Interactive effects of c a and climate on Toona growth seem to reflect a balance of c a‐induced increase in water‐use efficiency and decrease in drought sensitivity on the one hand, and c a‐induced increase in leaf temperature beyond the temperature optimum for photosynthesis on the other hand. These results are in line with those from a mechanistic tree‐growth model, parameterized for T. ciliata at the HKK site and forced by observed annual temperature and rainfall (Schippers, Sterck, Vlam, & Zuidema, 2015). Simulations of annual wood production using this model yielded temporal fluctuations that showed high correlations with the HKK tree‐ring chronology, but this match did not improve when effects of c a rise were simulated. Overall, our results suggest that for our study species, climate × c a interactions lead to complex responses that do not necessarily result in c a‐induced growth stimulation and depend on season, site and time period considered. As a result, it seems that for our study species c a rise did not cause an overall ‘growth bonus’, but rather induced subtle and variable modifications of climate–growth relations (cf. Clark, Clark, & Oberbauer, 2013). We deliberately did not interpret the main c a effects in our MEMs because tree age and c a both increased over time, and hence it is difficult to partition variance driven by ontogeny or by c a in our models. Furthermore, we aimed to evaluate the effects of c a rise on climate–growth relations, and not on growth averaged over multiple years. We therefore did not conduct any detrending on the tree‐ring series and we assumed a linear effect of age on log‐transformed ring width (confirmed by significant age effects in models and clear ontogenetic relations, Figure 4; Figure S5). Thus, negative or positive effects of c a in our statistical models cannot be separated from those caused by ontogenetic effects and climatic trends. Can similar results be expected for other tropical forest species? T. ciliata exhibits climate–growth relations that are similar to those obtained from other tropical forest species (Rozendaal & Zuidema, 2011), has experienced comparable increases in iWUE compared to other tropical species (Nock et al., 2011; van der Sleen et al., 2015) and occurs along large climatic gradients (Figure S1). We therefore expect that studies on other tropical tree species would reveal c a × climate interactions that depend on mean climate in similar ways. The strength and abundance of these interactions may differ depending on climate envelops. For instance, c a × T interactions would probably be less common for specialists of warmer areas that possess a higher optimal leaf temperature for photosynthesis, and more common for specialists of cooler areas. Clearly, it is important to test this expectation and extend this analysis to other species. Possible extensions are to include other Toona species (e.g. T. sinensis , T. fargesii , T. sureni and T. calantas ) or setting up networks on other tropical tree genera with high potential for tree‐ring analysis (e.g. Cedrela and Entandophragma ). Such networks combine a large climatic variation with limited phylogenetically induced differences in climate responses.
4.2 Results of experimental c a × climate manipulationsTo put our results on c a × climate interactive effects on tree growth in context, we performed a qualitative review of c a × water and c a × T experiments. We focused our review on (sub‐)tropical species, implying that it is not complete for extra‐tropical studies. Our review only included studies on seedlings or small trees, as c a × climate experiments for canopy trees or forests are absent. We included 12 publications, in which six tropical, two sub‐tropical and four temperate tree species were studied (Table S3). All studies were conducted in a greenhouse except for one conducted in climate controlled field chambers (Quentin et al., 2015). A total of 19 species × study combinations were included (six c a × water studies, seven c a × T factorial treatments and six three‐way factorial experiments). For each study, we evaluated whether results are consistent with one of the four hypotheses (Figure 1) and summarized this in Table 3.
|Interaction tested||Studies supporting hypothesis||N|
|c a × water availability||1a||1b||NS|
|c a × temperature||2a||2b||NS|
- The review included studies from temperate, subtropical and tropical species, see full overview in Table S3. N is the number of study × species combinations.