Stem growth of Amazonian species is driven by intra-annual variability in rainfall, vapor pressure and evapotranspiration

ABSTRACT Intra-annual distribution of precipitation in central Amazonia often leads to a short mild dry season and an increase in irradiance, temperature, and vapor pressure deficit; however, the accurate effect of intra-annual microclimatic variability on stem growth is still under investigation. The objective of this study was to determine how stem growth responds to monthly variations in microclimatic factors in the central Amazon. During five years (2008-2012) we measured diameter stem growth of 109 trees (26 species) and used principal component regression to evaluate the effect of microclimatic variability on stem growth in diameter. We found that the mean stem growth in diameter across species increased in response to an increase in rainfall and reference evapotranspiration, but it decreased with a rise in mean and minimum vapor pressure deficit. A contribution of this study is to show that even when irradiance and temperature had no significant effect on stem growth, small changes in vapor pressure deficit significantly affect stem growth. If the dry season becomes longer, as predicted by models, trees currently more sensitive to microclimatic variability associated with droughts would be the most affected by climate changes.


Introduction
The Amazon rainforest is of paramount importance for the global carbon cycle because of the large amount of carbon stored in the forest biomass -a total of about 86 Pg of carbon over the Amazon basin, including dead and belowground biomass (Saatchi et al. 2007).Tree growth (defined as biomass gain) is a major component of net primary production, and hence it has been used to infer forest productivity.Tree growth is the result of a myriad of biochemical reactions and processes of which photosynthesis, a light-and water dependent process, is of special importance (Kozlowski & Pallardy 1997).
Several factors have been associated with variations in stem growth, photosynthesis, and vegetation greenness in the Amazon region including precipitation, solar radiation, and temperature (Zhao et al. 2017;Yang et al. 2018;Elias et al. 2020).Studies that aim to assess the effect of the dry season on stem growth in the Amazon have led to contradictory conclusions.For instance, Silva et al. (2003) and Dias et al. (2022) found no effect of precipitation on stem growth in the central Amazon, while Yang et al. (2018) found that photosynthesis, over the entire Amazon region, can decline in the dry season.Likewise, Méndez (2018) and Souza & Marenco (2022) reported a decline in stem growth during the dry season in a terra-firme rainforest of the central Amazon.
In this context, Principal Component Regression has proved to be a valuable tool for dealing with the collinearity problem, whereby a new set of independent variables can be extracted from the original data by principal component analysis (Montgomery et al. 2012).Principal component regression has been used for a long time to assess the climate-tree growth relationship (e.g.Fritts et al. 1971;Marquardt et al. 2019;Antezana-Vera & Marenco 2021), and although much research has been carried out to assess the effect of environmental conditions on tree growth of tropical rainforests, there is still no consensus about the relative importance of individual microclimatic variables on stem growth.Understanding the effect of microclimatic variability on stem growth of Amazonian trees is particularly important because of the large influence of the Amazon forest on the global carbon balance and regional climate.
Thus, the main objective of this study was to determine the effect of intra-annual variation in precipitation, temperature, and irradiance on stem growth at a terra-firme rainforest site in the central Amazon.We hypothesized that stem growth would vary reflecting the availability of water and changes in irradiance and temperature, and expected that the trees would grow faster following an increase in irradiance and temperature.Because in the central Amazon the dry season is often mild and the roots can extract water from deeper layers of the soil (Broedel et al. 2017), we did not predict an effect of intra-annual precipitation variability on stem growth.

Study site
The research was conducted at the Tropical Silviculture Experiment Station (ZF2 Reserve), located about 60 km north of Manaus.The study area is a terra-firme forest on a plateau (centered at 02° 36´ 21" S, 60° 08' 11" W, 110-120 m above sea level).In this region tree density can reach up to 637 trees per hectare (Rankin-De-Merona et al. 1992), and species diversity is high -up to 179 species ha -1 (Prance et al. 1976).However, it is not uncommon that trees of the same species are hundreds of meters apart.Thus, for this study, tree species were selected based on the availability of at least three trees of the same species, each of them with stem diameter (at 1.3 m above the ground -diameter at breast height, DBH) of at least 10 cm.In the experimental site, the annual rainfall is about 2,540 mm with a mild dry season from June through October (Antezana- Vera & Marenco 2021;Dias et al. 2022), being July-September the driest months (INMET 2021 -climate data for the nearby city of Manaus).During 2013-2017, reference evapotranspiration (ET o ) was 120.8 mm month -1 , mean temperature 26.5 °C, and mean relative humidity (RH) 78.9% (Antezana-Vera & Marenco 2021).In this area, the predominant soil type is a clayed Oxisol (Yellow Latosol in the Brazilian classification) with low fertility, and pH (in water) of 4.0-4.3 at 0-20 cm depth (Chauvel 1982;Magalhães et al., 2014).

Physical environment, plant material and stem growth
During the period of January 2008 to December 2012, air temperature (T), photosynthetically active radiation (PAR), RH, and rainfall data were daily recorded above the forest canopy, at the top of a 40-m-tall observation tower (02°35´21"S, 60°06´53"W).Temperature and RH were measured with specific sensors (Humitter 50y, Oy Vaisala, Finland) and PAR with a quantum sensor (Li-190SA, Li-Cor, NE, USA) connected to a data logger (Li-1400, Li-Cor, Lincoln, NE).Data were logged at 15 min (PAR) or 30 min intervals (T and RH).PAR data were integrated over time to obtain daily PAR values (mol m -2 day -1 ).Rainfall data were recorded using a rain gauge (Em5b, Decagon, WA, USA).We used RH (%) and temperature (T °C) to calculate vapor pressure deficit (D) as: D (hPa) = e s -(e s × RH), where e s is the saturated vapor pressure in hectopascal (Buck 1981).The daily reference evapotranspiration (ET o , mm day -1 ) was computed as (Hargreaves & Samani 1985): ET o = 0.0023R a (T mean + 17.8)(T max -T min ) 0.5 (1) where R a represents the extraterrestrial radiation (expressed in mm day -1 ), and T mean , T min and T max (in °C) the mean, minimum and maximum temperature, respectively.Then, from daily data, the monthly ET o was obtained.
In this study we collected data from 109 trees of 26 evergreen species (located on the above-mentioned plateau), which are described in the Supplementary Material (Tab.S1-A).In these trees we measured the breast-height stem growth in diameter (S G ) at monthly intervals during 60 months (January 2008 -December 2012), as previously described (Dias et al. 2022).The S G was measured using stainless steel dendrometer bands, which were installed at least two years before the beginning of data collection.

Statistical analyses
The effect of microclimatic variability (PAR, rainfall, temperature, vapor pressure deficit, and ET o ) on the mean diameter stem growth across species (S G-mean ) was evaluated using Principal Component Regression (PCR).Prior to conducting PCR, the S G-mean data were centered (observed value minus the mean) and the microclimatic variables standardized (observed value minus the mean divided by standard deviation), and then microclimatic data subjected to principal component analysis (PCA) to extract orthogonal factors.The PCR is performed in several steps (Montgomery et al. 2012), and briefly summarized as follows (equations 2-9): These equations describe the standard multiple linear regression (MLR, Eq. 2) and the PCR model (Eq.3), and following the classic notation, Y denotes a vector of observations (dependent variable), X a matrix of regressors, b and α vectors of coefficients, and ϵ the random errors.In equation 4, the columns of Z represent a new set of orthogonal components -z i (hereafter termed principal components), while T is a matrix whose columns represent eigenvectors.The computation of α (coefficients of the PCR model) is described in equation 5, and that of b pc in equation 6.The values of a ^ (estimator of α) are obtained after regressing Y on the principal components (z i ).In equation 6, the "pc" subscript indicates that only a reduced k number of z i components has been retained in the model (i.e.PCR reduced model).The variance (var) of b pc , its standard error (SE) and the t values are computed as described in Eq. 7-9.To determine the number of principal components to be used in the PCR reduced model, we used the adjusted coefficient of determination (R 2 ajd ), after Jolliffe (2002).While the significance of b pc was tested on individual coefficients using t-test, and n -k -1 degree of freedom (DF), where n is number of observations (i.e. 60 months in this study) and k the number of principal components in the reduced model (the DF in the PCR analysis of variance).In addition, we also described the relationship between S G-mean and rainfall, temperature and vapor pressure deficit to illustrate the trend of microclimatic variables.For further information, we used the Random Forest machine learning technique to rank the microclimatic variables (predictors) according to their importance in predicting stem growth (S G-mean ).The analyses were performed using R v.4.0.5 (R Core Team, 2021).The PCR was performed using PLS (Liland et al. 2021), while the randomForest package was used for computing random forest algorithm (Liaw & Wiener 2002).In all analyses, we used p ≤ 0.05 to define statistical significance.

Microclimatic variability and its relationship with stem growth: a heuristic approach
We found no correlation between the stem growth of species (S G ) and wood density (p = 0.44, Tab.1), and although the largest trees tended to grow faster, the correlation between stem diameter (DBH) and S G was not significant (p = 0.06, Tab. 1).Thus, data were pooled to assess the effect of microclimatic variability on stem growth over species (S G-mean ).
Table 1.The relative importance of climatic variables computed by random forests (RIRF).It is also shown the mean stem growth over species (SG-mean) during the study period, as well as the correlation between stem growth of species (SG) and wood density (WD) and between SG and stem diameter of trees at breast height.Abbreviations: PAR: photosynthetically active radiation, T: temperature, Tmax: mean maximum T, Tmin: mean minimum T, Tmean: mean T, D: vapor pressure deficit, Dmax: mean maximum D, Dmin: mean minimum D, Dmean: D mean, and ETo: reference evapotranspiration.

Climatic variable RIRF (%)
Rainfall (mm month -1 ) 100.0 Dmax (hPa) 21.9 ETo (mm month -1 ) 20.9 Tmean The PCR showed that the maximum R 2 adj value (0.1458) was found when the fifth principal component was added to the model (Tab.2).Hence, the first five components (z 1 -z 5 ) were used for PCR analysis.Furthermore, it is shown in Tab. 2 that although the fifth component (z 5 ) was associated with a rather low-magnitude eigenvalue (λ 5 = 0.40, Fig. 1) it had a significant effect on stem growth Acta Botanica Brasilica, 2023, 37: e20220219 Table 2. Principal component regression of the relationship between SG-mean and the principal components z1 -z5.Abbreviations: DF: degree of freedom, R 2 : coefficient of determination (with increasing zi) , R 2 adj: adjusted R 2 , SG-mean: mean tree growth across species, α: regression coefficient, SE(α): standard error of α.Also, the R 2 ajd for Z6 is also shown.Microclimatic data were standardized and tree growth data centered prior to statistical analysis.(p = 0.0172), whose p value was even larger than that associated with the first component -z 1 (p = 0.0752).As the first and the fifth components were more closely associated with stem growth (inferred from p values), only Factor 1 and Factor 5 are shown in Fig. 1.It is worth noting, that the first five factors extracted by PCA from microclimatic data combined accounted for 97.1 % of the total variance [i.e.100×(6.14+ … 0.40)/9.0] in microclimatic data (Fig. 1).That is, by discarding very small eigenvalues (λ < 0.4) only a small fraction (3%) of the total microclimatic variance was disregarded.

Source of variation
It is displayed in Fig. 1 that S G-mean (indicated by the square symbol) shares the same quadrant with rainfall.Therefore, it can be expected that precipitation positively affects stem growth.Because both S G-mean and ET o (in Fig. 1) are negatively correlated with Factor 5, ET o may be positively correlated with stem growth.The Fig. 1 also shows that D min and S G-mean (square symbol) are in opposite quadrants indicating that they are negatively correlated.Although the other microclimatic variables (temperature, PAR and D mean and D max ) are closely related to Factor 1, it is difficult to infer (from Fig. 1) how these variables can affect stem growth, as the p value of z 1 did not reach a significant level (p = 0.0752).In the next section, by using PCR we examine the growth-microclimatic relationship in more detail.

Effect of microclimatic variability on stem growth based on PCR
During the study period mean temperature (T mean ) was 25.6 °C, D mean 5.15 hPa and monthly rainfall 242.1 mm month -1 (2,905 mm yr -1 , Fig. 2, 3).The PCR analysis showed that S G-mean responded positively to both rainfall and ET o (Fig. 2A, Tab.3), whereas D min and D mean had a negative effect on the mean stem growth (Fig. 3, Tab.3).Temperature, on the other hand, had a neutral effect on S G-mean (Fig. 2B, Tab. 3).Thus, using the coefficients shown in Tab. 3, the centered S G-mean as a function of the standardized microclimatic variables can be represented as follows (equation 10): S G-mean (mm month -1 ) = 0.000874PAR+ 0.020837Rainfall + 0.006953T min + 0.003356T mean + 0.005285T max -0.01976D min -0.005033D mean -0.002294D max + 0.016414ET o (10)

Comparison of PCR with the Random Forest model
The PCR model showed that the coefficient associated with rainfall has the largest beta (0.020837).In this respect, the prediction based on Random Forest concurs with the outcome of the PCR model, as Random Forest showed that rainfall was the most important variable (Tab.1).On the other hand, the Random Forest model predicted that D max , ET o and T mean performed similarly (importance  of 20-22%); whereas the PCR showed that D max and T mean had no significant effect on stem growth, which is an important aspect to consider when the performance of these models is compared.With respect to the effect of PAR on stem growth, both models converged, as PAR was ranked as non-important by Random Forest, and likewise the PCR model showed that PAR has no significant effect on stem growth.

Discussion
In this study we found that the minimum and mean vapor pressure deficit had a constraining effect on S G-mean , while stem growth increased with increasing ET o and rainfall.The reduced stem growth with a rise in D min indicates an effect of nocturnal atmospheric conditions on tree growth, as the lowest values of D min often occur at night when Acta Botanica Brasilica, 2023, 37: e20220219 relative humidity is high (INMET 2021).This is important, as it shows that stem growth can be affected not only by environmental factors that limit photosynthesis, but also by nocturnal conditions that tend to increase transpiration and thereby to lower leaf water potential.Indeed, it is widely accepted that changes in expansive growth are related to changes in leaf water potential (Bradford & Hsiao 1982;Kozlowski & Pallardy 1997).
The negative effect of D mean can be explained by considering the effect of vapor pressure deficit on stomatal conductance, as the most common response to a rise in vapor pressure deficit is a decrease in stomatal conductance and hence in carbon uptake by photosynthesis (McDowell & Allen 2015;Vinod et al. 2023).In fact, Yang et al. (2018) reported a decrease in ecosystem photosynthesis in the dry season (over the forests of the Amazon basin) when temperature and vapor pressure deficit often increase (Antezana-Vera & Marenco 2021).The constraining influence of an increase in vapor pressure deficit on stem growth is relevant in the face of global climate change, especially taking into account that in the Amazon region temperature has increased (around 0.16 °C per decade) and the rainfall pattern altered (e.g.decreased rainfall in the eastern and southern Amazon, Marengo et al. 2018), which combined may lead to changes in vapor pressure.In this regard, it is known that the dry season is associated with an increase in irradiance, temperature and vapor pressure deficit (Méndez 2018;Antezana-Vera & Marenco 2021;INMET 2021), which may lead to a decline in photosynthesis and tree growth (Yang et al. 2018;Antezana-Vera & Marenco 2021).
The positive effect of rainfall variability on stem growth suggests that even during a relative mild dry season, Amazonian trees tend to respond to variation in water availability, even when it has been found that in the central Amazon root water uptake can occur even below 4.8 m depth, during severe dry periods (Broedel et al. 2017), which eventually can help to withstand the effect of water stress in mild dry seasons.Even when the enhancing effect of T max on S G-mean did not reach a significant level effect (p = 0.07), S G-mean increased with increasing ET o (p = 0.02, Tab. 3).At first glance, it seems unexpected to record a positive effect of ET o and at the same time a neutral effect of temperature (T min , T mean and T max ) on stem growth (Tab.3).This can be explained by taking into account that ET o is a function of solar radiation and temperature, including T max (Eq.1), and T max as mentioned above tended to have a positive on stem growth (Tab.3).The direct effect of temperature on biomass gain can occur through its effect on photosynthesis, via the direct effect of temperature on enzyme activity, electron transport chain, and stomatal conductance (Lloyd & Farquhar 2008;Marenco et al. 2014;Vinod et al. 2023).In addition, rootzone temperature regulates photosynthesis via its influence on water absorption and stomatal conductance, because temperature affects water viscosity and root permeability to water (Kaufmann 1975;Delucia 1986).The neutral effect of T min on S G-mean suggests that variations in night temperatures did not alter significantly metabolic processes, such as root, leaf and stem respiration.This is important, as respiration provides energy and carbon intermediates for growth and maintenance of tissues (Kozlowski & Pallardy 1997).The absence of an effect of temperature on stem growth does not support our working hypothesis, as we had expected that trees would grow faster under warmer conditions, as reported by Elias et al. (2020).The optimum temperature for photosynthesis in tropical rainforests is about 29 °C (Liu 2020).Thus, the lack of an effect of temperature on S G-mean indicates that changes in temperatures over the year were not high enough to alter stem growth.The importance of irradiance on tree growth is indisputable via its effect on photosynthesis.Notwithstanding, we found no effect of PAR variability on stem growth, which indicates absence of adverse photochemical effects associated with the increase in irradiance which often occurs during the dry season (Antezana-Vera & Marenco 2021).In fact, leaf photochemistry tends to remain constant in plants subjected to moderate water stress as shown by Rascher et al. (2004).
We have shown that PCR proved to be a useful approach to separate the individual contribution of microclimatic variability on stem growth.Even though Random Forest was effective to identify rainfall as one of the most important variables influencing stem growth and PAR as the least, by using this technique it is rather difficult to assess the individual contribution of microclimatic variables due to the effect of collinearity.Indeed, Random Forest is a powerful algorithm for prediction, but when the primary objective is explanation, principal component regression seems to perform better than Random Forest.

Conclusions
In this study we show that the mean diameter stem growth decreased in response to an increase in mean and minimum vapor pressure deficit, while rainfall and the reference evapotranspiration had a positive effect on S G-mean .Irradiance and temperature, on the other hand, had no significant effect on the mean stem growth.These results do not support our working hypothesis, as we had expected that stem growth would increase with an increase in temperature and irradiance, and that a slight decrease in precipitation during the dry season would not affect stem growth.A contribution of this study is to show that even when temperature and irradiance intra-annual variability did not significantly affect stem growth, besides precipitation, environmental factors related to atmospheric dryness can influence stem growth.These results are important in the context of the current climate changes and enhance our understanding on the drivers of Amazonian trees' radial growth.

Figure 1 .
Figure 1.Principal component analysis of microclimatic variables, with the mean stem growth in diameter (SG-mean) as a supplementary variable.The eigenvalues (λi) of orthogonal factors are shown in the inset.Note that the first five factors account for 97.1% of total variance (100×8.74/9.0).Abbreviations as described in Tab. 1.

Figure 2 .Figure 3 .
Figure 2. Mean stem growth in diameter (SG-mean) as a function of rainfall (A) and mean temperature (Tmean, B).In panel A, the solid line shows the trend.Each symbol represents the mean stem growth across species (26 species) for a given month.Data were collected from January 2008 to December 2012.The means (± SE) of rainfall and Tmean were 242.1 ± 18.3 mm month -1 and 25.6 ± 0.14 °C, respectively.