Evaluating tree survival and modeling initial growth for Atlantic Forest restoration

ABSTRACT: Ecological restoration has become an important complementary practice to protect natural resources and preserve biodiversity. However, native species may be used in restoration programs in ways that do not optimize their performance. This research evaluated the survival and to model the initial growth of 15 native tree species planted in “filling” and “diversity” lines in the post-planting phase of a restoration experiment in the subtropics of the Brazilian Atlantic Forest. We measured survival rate (%) one year after planting and collar diameter (mm), total height (m), crown projection area (m²) and crown volume (m³) in the first 48 months after planting. Growth modeling for each variable and species was based on the non-linear mathematical Logistic, Gompertz, and Chapman-Richards models. Model selection for each variable/species was supported by the Akaike Information Criterion, standard error of the estimate, and coefficient of determination. The highest survival rates were reported for Cordia americana, Gochnatia polymorpha, Inga uruguensis, Peltophorum dubium, Prunus sellowii e Zanthoxylum rhoifolium (91.7%) and for Solanum mauritianum (90.3%). The species with faster growth were, by increasing order, Mimosa scabrella, Trema micrantha, Solanum mauritianum and Croton urucurana. With a better understanding of the initial developmental potential of tree species, it is possible to increase the species and functional diversity of the filling group. There was no single model capable of describing the variables analyzed and different models were needed to describe different characteristics and species.


INTRODUCTION
The United Nations established 2021 to 2030 as the decade for ecosystem restoration to halt degradation and restore ecosystems to achieve global goals (UN, 2021).Ecosystem restoration becomes thus a political priority to "enhance people's livelihoods, counteract climate change, and stop the collapse of biodiversity" (UN, 2021).
Deforestation in the Atlantic Forest in the last decades has caused habitat loss and degradation increasing fragmentation and biodiversity loss, in particular endemic species (SCHNEIDER et al., 2018).In recent years, there is an apparent stabilization in the Atlantic Forest which hides; however, increasing loss and fragmentation of mature forests and expansion of secondary growth forests in marginal farmland (ROSA et al., 2021).To respond to increasing degradation in Atlantic Forest ecosystems and landscapes, planting native species in "filling" and "diversity" lines is currently the most used forest restoration technique in Brazil (RODRIGUES et al., 2009;RODRIGUES et al., 2011;TRENTIN et al., 2018;TOPANOTTI et al., 2019;BRANCALION & HOLL, 2020)."Filling" lines have the goal of covering the ground to control the establishment of invasive weeds, providing at the same time crown heterogeneity and shade favoring the growth of the species from the "diversity" group.The "diversity" group is composed of a much higher number of tree species which is crucial to restoring forest dynamics (NAVE & RODRIGUES, 2007;RODRIGUES et al., 2009).
Tree species of low economic interest are usually left out in research (PAIVA- SOBRINHO & SIQUEIRA, 2008).However, to better understand how forest restoration processes occur and how possible it is to implement successful restoration programs, it is necessary to acquire information about the initial silvicultural development of native tree species used in restoration projects (STOLARSKI et al., 2018;BECHARA et al., 2016;TRENTIN et al., 2018).Growth modeling is an important tool in this regard since it allows the integration of observed data in mathematical or statistical models to describe how species are expected to grow over time contributing to increasing the current knowledge on tree species performance which is key in the selection of species to use in forest ecological restoration.
This study evaluated survival and model the initial development of 15 native forest species in the post-plant phase of a controlled restoration experiment in the Iguazu River basin, Parana State, South Brazil.With this study, we aimed to better understand species characteristics that can support their selection in successful restoration programs.

Study area
The study was conducted in a 7.2 ha restoration experiment located on the campus of the Federal University of Technology -Paraná (UTFPR), Dois Vizinhos, Southwest of Paraná State, Brazil (25º41'44" S; 53º06'07" W) (Figure 1).
The study area is in the Araucaria Forest region in the transition zone of the Semideciduous Seasonal Forest.The climate is humid subtropical (Cfa), without a defined dry season and with an average temperature of the warmest month of 22 °C, according to Köppen's classification (ALVARES et al., 2018).According to IAPAR (2021), frost occurs frequently from June to September in the southwest region of Paraná.Here, the absolute minimum temperature (-5.0 ºC) was observed in July.Altitude ranges from 475 to 510 m and soils are Nitisols.
The landscape where the restoration area is located is dominated by agricultural production areas, pastureland, and small patches of remnant forest.In the north of the experimental area, there is a secondary forest area (40 ha) approximately 20-30 years old.In the south, there is an area of annual crop production (oats, wheat, beans, corn, and soybean).From 2006 to 2008 the restoration area was a pasture of Cynodon nlemfuensis and Arachis pintoi.In 2009 the area was used for annual crops until the last harvest on October 15, 2009.On October 20, 2010, the area was isolated from disturbances and the restoration.
The experiment was established in December 2010 (BECHARA et al., 2016).For that the area was divided into four blocks where three ecological restoration treatments were implemented.The experimental design included 10 filling species and 60 diversity species, based on the minimum number of plants per species according to the size of the experimental area (RODRIGUES et al., 2009).All trees were planted at a spacing of 3 x 2 m in four plots of 54 x 40 m (one in each block) (Figure 1).In each plot, 18 filling and 3 diversity seedlings were planted.
At the time of plantation, seedlings, 30 to 50 cm in height, were fertilized with 36 g of NPK (5-20-10) + 40 g of urea, and three liters of hydrogel was applied in the holes.The control of leaf-cutting ants was carried out using baits applied every six months until 30 months after initial planting.The soil was protected with cardboard mulching and biannually, weeds were controlled by mechanical cutting followed by chemical weeding until the third year (GERBER et al., 2020).Dead plants were replanted twice until 12 months after initial planting.

Data collection
Data was collected for survival once on December 2010, 12 months after planting, and biannually for growth variables, such as collar diameter (cd), total height (h), crown height (ch) and crown diameter (dl and de), up to 48 months after planting.
Survival rate was calculated as the percentage of individuals in each species that survived one year after planting.Growth was assessed based on variation over time of the growth variables measured (cd, h, ch, dl and de).Collar diameter was measured using a digital pachymeter, positioned at the collar of the plant, near the soil surface.Total height was measured with a graduated scale at 0.05 cm intervals positioned vertically as close as possible to the tree.Total height was the distance between the base of the tree at the soil level and its uppermost point in the crown.The crown projection area was obtained by measuring the diameter of the crown with a tape measure based on two measurements, the first in the direction of the length of the plot (X) and the second (Y) perpendicular to it.
The crown projection area was later estimated through the ellipse area formula (SANTOS et al., 2015;STOLARSKI et al., 2018) (Equation 1): ca (m²) = dl.de.π/4 (1) where ca is the crown projection area, and dl and de are the crown diameters (m) measured in two perpendicular directions.
Crown volume was estimated, assuming that it is well described by an elliptical cylinder, multiplying the crown projection area by its height (KOIZUMI & HIRAI, 2006;SPINELLI et al., 2010;STOLARSKI et al., 2018) (Equation 2): cv (m 3 ) = ca.ch (2) where cv is the crown volume, ca is the crown projection area, and ch is the crown height.Ciência Rural, v.53, n.7, 2023.
Gerber et al.

Growth models
The species' growth in collar diameter, total height, crown projection area, and crown volume was evaluated through three growth functions, which are normally used in forest modeling, such as Logistic, Gompertz, and Chapman-Richards.These have great explanatory power and are commonly applied in forest science.In forestry, they have been widely applied to describe the growth of dendrometric variables of species with timber potential (PÖDÖR et al., 2014;VENDRUSCOLO et al., 2017;SILVA et al., 2018) as well as to describe hypsometric relations (ALVES et al., 2017;ANDRADE, 2017;MACHADO et al., 2019).
These models are described by equations 3 to 5 (BURKHART & TOMÉ, 2012): Logistic: y = β 0 / [1 + e ((β1 -x)/β2) ] (3) where y is the predictable variable, β 0 is the asymptote, β 1 is the x value in the inflection point of the curve, β 2 is a numerical scale parameter in the input axis and x is the independent variable (age).Gompertz: y = β 0 * e ( -β1 * β2^x)  (4) where y is the predictable variable, β 0 is the asymptote, β 1 the function at x = 0, β 2 is a numerical parameter related to the scale in x-axis, and x is the numerical vector of input values (age).Chapman-Richards: y = β 0 * [(1 -e (-β1 * x) ) β2 ] (5) where y is the predictable variable, β 0 is the asymptote, β 1 is an empirical parameter, β 2 is related to the plant ś biology, and x represents a numeric vector of values at which to evaluate the model (age).
Model selection for each species and variable was based on criteria such as coefficient of determination (R²), standard error of the estimate (S yx %), and Akaike Information Criteria (AIC) in addition to the biological interpretation of the parameters.The selected model must present the lowest AIC and S yx and the highest R².These information criteria are usually applied in regression model selection (AKAIKE, 1973;SCHWARZ, 1978).All statistical analyses were performed in R (R CORE TEAM, 2020).

Cluster analysis
A cluster analysis was conducted to group tree species according to growth characteristics, in particular, mean increment in collar diameter (cd), total height (h), crown projection area (ca), and crown volume (cv), for the whole evaluation period (48 months).We used the nearest neighbor grouping method with the Mahalanobis distance as a dissimilarity measure.Cluster analyses were performed in R (R CORE TEAM, 2020).

Survival
The survival rate of the species evaluated 12 months after planting was in general high (>70%) or very high (>90%) (Table 1), according to levels of survival proposed by Carvalho (1982) when evaluating the silvicultural performance of native forest species in a subtropical forest in South Brazil, at 84 months of age.

Growth assessment
In ascending order, M. scabrella, T. micrantha, S. mauritianum, and C. urucurana showed the highest growth rates among all species for all variables (Figures 2 and 3; Supplementary Material).Differences between top-ranked species were often small.For instance, the mean height of M. scabrella, the species with the highest growth (7.56 ± 1.02 m), was just 15% higher than S. mauritianum, the species with the lowest (6.39 ± 0.76 m) growth.For crown projection area and volume, differences between the species with the largest and the smallest growth were also small: T. micrantha (34.65±8.55m 2 and 135.65±38.29 m 3 ) and Z. rhoifolium (7.51±1.80m 2 and 27.31±4.21m 3 ), respectively.

Growth models
Based on the evaluation criteria applied for all species, the Logistic model showed the best fit for collar diameter for 8 of the 15 assessed species (C.urucurana, B. forficata, I. uruguensis, G. ulmifolia, G. polymorpha, T. micrantha, P. dubium, and C. trichotoma) (Table 2).The Gompertz model was the best to describe collar diameter growth for C. floribundus, M. scabrella, Z. rhoifolium, S. mauritianum, P. sellowii, and C. americana, and the Chapman-Richards model showed the best performance only for S. terebinthifolius.
For the crown projection area (Table 4), the Gompertz model performed better for 8 of the 15 species (C.urucurana, M. scabrella, Z. rhoifolium, S. mauritianum, G. polymorpha, T. micrantha, P. sellowii, and C. americana) and the Chapman-Richard for S. terebinthifolius, C. floribundus, B. forficata, I. uruguensis, G. ulmifolia, P. dubium, and C. trichotoma.The Logistic model was not the best-fit one for any of the species in this variable.
The Chapman-Richards model presented the best fitting statistics for crown volume in 12 species, as follows: S. terebinthifolius, C. urucurana, B. forficata, M. scabrella, I. uruguensis, G. ulmifolia, Z. rhoifolium, G. polymopha, T. micrantha, P. dubium, C. americana, and C. trichotoma (Table 5).The Gompertz model was the best to describe crown volume for S. mauritianum and P. sellowii while the Logistic model showed the best fitting performance only for C. floribundus.
The model fitting criteria scores for crown volume are associated with the high standard error of the estimates, which varied from 51.1 to 101.6% (Logistic), 50.9 to 101.6% (Gompertz), and 61.4% to 103.6% (Chapman-Richards), due to the high variability of the dataset.

Cluster grouping
The hierarchical relationship of the 15 species plotted in the dendrogram of figure 4, enabled the establishment of three major groups of species based on their average growth in collar diameter, total height, crown projection area and volume during the evaluation period.As mentioned before and confirmed by the cluster analysis, the species M. scabrella showed the highest growth performance, and it compounded one group by itself.The other species were divided into two different groups.

Survival
Cordia americana, G. polymorpha, I. uruguensis, P. dubium, P. sellowii, Z. rhoifolium, and S. mauritianum, showed a survival rate greater than 90%.This indicated that these species were tolerant to stress caused by frost.According to the criteria   established by HIGA et al., (2000) and CARON et al., (2011), tolerant species present 25-100% damage to leaf area and <25% of damage to stem, conditions found in the study.These species presented more rusticity and the ability to endure stressful conditions within bearable boundaries, as they are colonizing species of degraded environments (STOLARSKI et al., 2018).The survival rate (91.7%) of I. uruguensis was relatively high in this study.VIEIRA et al. (2003), reported a survival rate of 94% when assessing the species' survival after frost in Florianópolis.
Gerber et al.Evaluating tree survival and modeling initial growth for Atlantic Forest restoration.

Species
Our results suggested that frost might negatively affect the establishment of species in restoration initiatives, especially when only a few individuals are planted.The occurrence of less severe frost causes several physiological damages to the plants, which leads to a delay in their growth, particularly in the initial stages of development.The mortality rates for C. floribundus and S. terebinthifolius in this study were classified as high.This can indicate that the tolerance of these species frost in reforestation is low.

Growth assessment
Twelve and 36 months after planting, growth in total height, collar diameter, and crown projection area and volume dropped when compared to the previous period.The reduction in growth at month 12 occurred, probably, because of frosts in the winter season (Figure 5).In 2013, the frosts occurred in July and August, with minimum absolute air temperature of -2.4 and -1.8 ºC, respectively (GERBER et al., 2020).This is clearly visible in the case of collar diameter (Figure 2).However, at month 18, G. polymorpha, C. trichotoma, T. micrantha, C. floribundus, C. urucurana, B. forficata, I. uruguensis, P. dubium, G. ulmifolia, and Z. rhoifolium presented a considerable increase in collar diameter growth (Figure 2).
The height growth of the species at 12 and 36 months after planting also expresses the damages caused by frost in the apex of some individuals, reducing the mean height per species (Figure 3).By losing apical dominance, trees developed more than one lower branch but lost height.This emphasizes that these species are not so tolerant, but they are resilient to frosts due to their high sprouting capacity, as observed by GERBER et al. (2020).
Mimosa scabrella showed outstanding growth over the initial 48 months.The decrease in mean height increment at month 36 also happened due to winter frosts, but the species increased its height again 6 months later.This species is not considered tolerant to frosts (CARON et al., 2011), and damage caused by frosts can affect the apical meristem, which leads to bifurcation and, consequently, to a delay in the main tree axial growth (KOZLOWSKI et al., 1991).
For T. micrantha, a filling species, the crown projection area was five times larger than the planting spacing (6 m²), implying that it has great potential in controlling invasive grasses, an essential role in the establishment of new species in the under story (STOLARSKI et al., 2018).Ciência Rural, v.53, n.7, 2023.

13
In general, crown projection area and volume increased gradually over time (Supplementary material).There was an increase in the amplitude of the observed values due to the different levels of the species' development.Boxplots at 12 months showed less dispersion making the median to be lower than in other periods.Physical and physiological damage in cold-susceptible individuals during the period from May to September reduced their vegetative growth, probably resulting in the death of the aerial part of some plants, which was retaken after this period due to anatomical changes in the cambium region (SOUZA et al., 1991;GERBER et al., 2020).
In 2012, there was no frost in the region and plants grew normally, but in the months from June to August 2013, frost occurred again; consequently, causing a further reduction in growth in total height, crown projection area, and crown volume at 36 months.Even susceptible to frost, individuals of the 15 studied species managed to overcome competition with other tree species, occupying the upper layer of the canopy at month 42.

Growth modeling
Best models for each variable and species were selected based on the criteria AIC, R 2 and S yx .It was possible to find a specific growth model fitted to each species and variable, noting that each species has different behavior in their growth.This pattern was observed especially in the first years of age when they are susceptible to different growth stimuli in the ecosystem undergoing restoration.However, it was not possible to fit the Logistical and Gompertz models to crown projection area for P. dubium.For crown volume, it was not possible to fit the Logistical and Gompertz models for C. urucurana, B. forficata, Z. rhoifolium, T. micrantha, P. dubium and C. americana.
There are few studies that analyze the long-term silvicultural performance of species of the Atlantic Forest using growth models.This is mainly due to the lack of long-term research on native Brazilian flora species.One of the biggest restrictions in this context concerns the lack of information on the growth of species in local restoration projects.Some of such studies are presented below.
TOPANOTTI et al. ( 2019) applied nonlinear regression models to estimate growth in collar diameter for three forest species with timber potential planted in a restoration area at 36 months.
Here, the coefficient of determination was 0.   The logistic model was, in general, the best to describe collar diameter growth while the Gompertz model was better fitted for total height and crown projection area.The Chapman-Richards model showed the best performance for crown volume growth.
Models can be used in practice to describe the growth behavior of species and dendrometric variables and make short -to mediumterm projections, aiming to point out their likely growth in the coming years.This is useful for the application of indicators and monitoring of forest restoration projects.

Cluster grouping
Three distinct groups were established for intermediate levels of similarity (or distance) in the cluster analysis (Figure 4).Due to its much faster growth in height, diameter and crown projection area, M. scabrella constituted an isolated group.The second group was formed by C. americana, B. forficata, G. polymorpha, C. trichotoma, and P. sellowii, species that showed the lowest performance of all in terms of growth of collar diameter, total height, and crown projection area and volume.The third group, the largest, included species with intermediate performance in the evaluated dendrometric variables and was formed into two clusters.The species I. uruguensis, Z. rhoifolium, C. urucurana, S. terebinthifolius and P. dubium, formed the first one and T. micrantha, C. floribundus, S. mauritianum and G. ulmifolia, were grouped in the second.
The fact that species of the Fabaceae family (M.scabrella, P. dubium and I. uruguensis) showed much higher growth than the others support the results from CHADA et al. (2004), in which the authors stated that species of this family show rapid growth in adverse environments mainly due to their ability to associate with mycorrhizal fungi and Rhizobium bacteria.Therefore, the species of this family are essential to ecological restoration projects.MACHADO et al. (2015) and BALIEIRO et al. (2017) suggest using fast-growing species that can generate N and C input to the degraded soil and consequently, increase the availability of other nutrients, organic matter, and promote nutrient recycling.
C. americana, B. forficata, G. polymorpha, C. trichotoma, and P. sellowii presented a slower growth, but their high percentage of survival added to the capability of most of these species to produce fruits favoring animals, make them of paramount importance in the ecosystem restoration process.According to ALMEIDA (2016) andBECHARA et al. (2016), the use of native fruit species in degraded areas is an interesting restoration tool to provide food and attraction to wildlife.This can be applied in areas close to forest fragments, where fruit production can encourage the movement of animals from surrounding forest areas to the area under restoration.These movements aim to bring propagules from the original forest to the recovering ecosystem.
From the filling species group, those that fulfill the role of presenting fast growth and high capacity as shading species were M. scabrella, T. micrantha, S. mauritianum and C. urucurana.However, Z. rhoifolium and I. uruguensis, classified as diversity species, presented desirable characteristics for filling species, that is, rapid growth in crown area and density.
It is expected that the selected species, especially in the filling group, will initially present a rapid growth that will help them to overcome the competition with grasses and to survive (and develop) even in face of abiotic adversity.Some species that were initially in the diversity group showed growth that suggests using them as filling species.
There was no single model capable of describing all the growth variables analyzed.Different models are needed to describe different characteristics of different species.

Figure 1 -
Figure 1 -Experimental restoration site at the Federal University of Technology -Paraná -Dois Vizinhos, Brazil.Plots marked in black refer to the filling and diversity planting lines treatment considered in this study.

Figure 2 -
Figure 2 -Boxplots of collar diameter (mm) of 15 tree species over 48 months in a restoration experiment in Dois Vizinhos, Paraná, Brazil.Boxplots show median, quartiles, and outlier values.

Figure 3 -
Figure 3 -Boxplots of total height (m) of 15 tree species over 48 months in a restoration experiment in Dois Vizinhos, Paraná, Brazil.Boxplots show median, quartiles, and outlier values.

Figure 4 -
Figure 4 -Dendrogram of the cluster analysis (hierarchical, Euclidian distance) of dendrometric variables of 15 tree species in a restoration experiment, Dois Vizinhos, PR, Brazil.

Figure 5 -
Figure 5 -Monthly rainfall (mm), mean temperature (ºC), and minimum absolute temperature (ºC) registered at the study area from October 2010 to November 2014.

Table 1 -
Survival rate (%) after 12 months, and average growth in collar diameter (mm), total height (m), crown projection area (m 2 ) and crown volume (m 3 ) after 48 months of 15 tree species in a restoration experiment in Dois Vizinhos, Paraná, Brazil.

Table 2 -
Models' parameters and fitting statistics for collar diameter (mm) by species.

Table 4 -
Models' parameters and fitting statistics for crown projection area (m²) by species. -

Table 5 -
Models' parameters and fitting statistics for crown volume (m³) by species.
TONINI et al. (2003)lethra scabra, and Ilex paraguariensis, respectively.The standard error of estimate varied between 15.23 and 30.57%.Similar results were reported for the Chapman-Richards model in our study, which ranged from 0.47 to 0.79 for the coefficient of determination, and from 24.99 to 52.80% for the standard error of the estimate.TONINI et al. (2003)used the Chapman-Richards model to describe the initial (nine years) growth of six Amazon Forest species, obtaining coefficient of determination values ranging from 0.63 to 0.79.