Rainfall partitioning in fragments . . . RAINFALL PARTITIONING IN FRAGMENTS OF CERRADO VEGETATION AT DIFFERENT STAGES OF CONDUCTION OF NATURAL REGENERATION

1 Received on 12.04.2017 accepted for publication on 07.03.2018. 2 Universidade Federal de São Carlos, Programa de Pós-Graduação em Ciências Ambientais, São Carlos, SP-Brasil. E-mail: <debora.bessi@gmail.com>. 3 Universidade Federal de Viçosa, Departamento de Engenharia Florestal, Viçosa, MG-Brasil. E-mail: <herly@ufv.br>. 4 Universidade Federal de São Carlos, Departamento de Ciências Ambientais, Sorocaba, SP-Brasil.. E-mail: <kellytonello@yahoo.com>. *Corresponding author.


INTRODUCTION
Forests play a major role as regulators of water balance by intercepting and distributing rainwater across forest compartments, thereby aiding in the regulation of water flow through the soil; and by recharging aquifers (Arcova et al., 2003).
The role played by vegetation as a water regulator begins with the water input in forest compartments.When rainwater precipitates, vegetation first catches it with the tree crowns, and the initial rain volume is then partitioned.Thereby, a portion of rainwater gets retained in the canopy, through a process called interception, while the other portion reaches the soil, through a process called net precipitation -which consists in the sum of throughfall and stemflow.All these interactions not only reduce the force with which the water reaches the soil, but also distributes the water across the forest compartments (Arcova et al., 2003;Lorenzon et al., 2013;Honda, 2013).
According to Lorenzon et al. (2013), values of net precipitation and rainfall interception may vary within a same ecosystem, among near fragments or even within a same fragment, depending on the stage of forest regeneration.These two variables allow for a comparative evaluation of areas with and without plant cover, which is crucial in view of the different effects that may occur in each of these situations.Thus, we aimed to characterize the rainfall partitioning in forest areas at different stages of natural regeneration conduction in a microbasin at the Floresta das Águas Perenes RPPN (Portuguese acronym for: 'Reserva Particular de Patrimônio Natural'; i.e.: Private Reserve of Natural Patrimony), Brotas municipality, São Paulo state, southeastern Brazil.By that, we also wished to contribute to the discussion on the forestwater relationship in forest areas undergoing regeneration.

MATERIAL AND METHODS
The study was conducted in the Floresta das Águas Perenes RPPN, located in Brotas municipality, São Paulo state, southeastern Brazil.In 2011, the area was recognized by the Forest Stewardship Council as a High Conservation Value Forest, since it provides basic environmental services, such as protecting river basins.The RPPN extends over 809.78 ha of Cerrado, and its phytophysiognomy is characterized as secondary vegetation of types Typical Cerrado and Cerradão.The area had been previously used for pasture, but in the last 43 years it has been used for eucalyptus cultivation.From 2006, following the harvest of some eucalyptus stands, natural regeneration started to be conducted in the area.
According to Köppen's classification, climate in the region is type Cwa, with wet summers and dry winters, and mean annual precipitation ranges between 1400 and 1500 mm (Cepagri, 2016).The predominant soil type is Quartzarenic Neosol (Embrapa, 1999).
Hydrological indicators were monitored from December 2015 through November 2016, after one or more rainfall events, in areas at three different stages of natural regeneration conduction: treatment 1 = 4 years of natural regeneration after eucalyptus harvesting; treatment 2 = 10 years of natural regeneration after eucalyptus harvesting; and treatment 3 = 43 years without human intervention.At each treatment, we established three plots (20 x 20 m) with a 10-m interplot spacing.
Rain gauges were made using PVC pipes and fittings with a 78.5-cm catchment area and 20-cm height, thus yielding a 1.6-L storage capacity.Gross precipitation (GP) was measured with a rain gauge installed near the plots, with a maximum distance of 30 m among treatments.We cared to eliminate any barrier that could influence rainwater catchment.GP was calculated by the equation: Where GP is gross precipitation (mm), V is the rain gauge volume (mL), and A is the rain gauge catchment area (cm 2 ).Throughfall (Tf) was obtained from 108 rain gauges distributed along the nine plots, totalizing 36 gauges per treatment.To calculate Tf, we used the equation: Where Tf is throughfall (mm), V is the volume of each rain gauge (mL), A is the catchment area of each gauge (cm 2 ), and ni is the number of gauges.Stemflow (Sf) was measured on all trees having diameter at breast height (DBH) > 5 cm.For that, trees were involved by a polyurethane gutter, which conducted the water that drained through the stem surface to Eq-1 Eq-2 Rainfall partitioning in fragments... 20-L collectors.Overall, we monitored 185 trees (4, 24 and 157 trees for treatments 1, 2 and 3, respectively), with a total 210 collectors.To assess water balance in the area, we followed the methodological survey conducted by Honda (2013), who recommended measuring Sf on all trees in the sampling plot, by dividing the total collected volume by the plot area.This procedure enabled us to avoid either underestimating the collected volume whenever tree crowns overlapped or overestimating it when individuals were sparsely distributed.To calculate Sf, we used the equation: Where Sf is stemflow (mm), V t is the total collected volume (L), and A is the plot area (m 2 ).
Net precipitation (NP) corresponded to the sum of stemflow and throughfall in each forest formation.Thus, NP was calculated by the equation: Where NP is net precipitation (mm), Tf is throughfall (mm), and Sf is stemflow (mm).
Lastly, rainfall interception (RI) was obtained by the difference between gross precipitation and net precipitation in each forest formation, according to the equation: Where RI is rainfall interception (mm), GP is gross precipitation (mm), and NP is net precipitation (mm).
In order to detect differences among treatments, we performed an analysis of covariance (ANCOVA) and Tukey's test at 95% confidence level, using software R 3.1.3.

RESULTS
We performed a total 21 field collections.Along the 12 months of monitoring, we recorded a total 1340 mm of rainfall.Tf decreased from treatments 1 to 3 (Table 1), which is coherent with the plant cover in each area.RI also increased in that same direction.The total Tf recorded at treatment 1 was similar to the GP recorded in the study.This can be justified by the presence of a vegetation composed of grasses and/or shrubs covering an extensive area of that treatment, which favors the occurrence of low interception or even of no interception at all thereat (Table 1).
Tf ranged from 98% to 102% of the total GP at treatment 1, from 87% to 102% at treatment 2, and from 68% to 98% at treatment 3 (Table 1), and the analysis of covariance on Tf data showed a significant difference among treatments.This suggests that increasing forest maturity is capable of causing significant differences in Tf (Table 2).The few rainfall events that yielded Tf values higher than GP ones were not able to interfere with the mean Tf of treatments to the point that they would exceed 100%.The Tf values of treatments were subjected to linear regression analysis (Figure 1-A,  B, C).All areas showed a strong relationship between Tf and GP, with coefficients of determination (R²) being as high as 0.99 for all treatments.Thus, 99% of the Tf was explained by GP.Additionally, this relationship was significant at the 95% confidence level at all treatments.From the linear equation of each treatment (Figure 1-A, B, C), we estimated the GP volume needed for Tf to occur.We found values higher than 0.99, 1.02, and 1.13 mm for treatments 1, 2, and 3, respectively (Table 2; Figure 1).
Sf contributed the least to water balance (Table 1).However, Sf values increased with increasing plant cover, and the analysis of covariance of this component revealed significant differences among treatments (Table 2).Linear regression of the GP x Sf relationship at the studied treatments yielded coefficients of determination ranging from 0.87 to 0.94 (Figure 1-D, E, F).The correlation index was 0.93, 0.96, and 0.97 at treatments 1, 2, and 3, respectively, which indicates a strong correlation between the tested variables, while also suggesting a possible interference of aspects other than just GP.The relationship was significant at all treatments.With the linear regression model equation of each treatment, we estimated the GP values needed for Sf to occur.We found values equal to or higher than 0.9, 9.35, and 5 mm for treatments 1, 2 and 3, respectively.
The NP collected at treatment 1 was the highest of all treatments (100%), followed by treatments 2 (96%) and 3 (88%) (Table 1).The linear regression between GP and NP yielded coefficients of determination of 0.99 at all treatments (Figure 1-G, H, I).The correlation coefficient was higher than 0.99 at all treatments, being significant in all of them.
When negative RI values recorded at treatments 1 and 2 were excluded from the dataset, we obtained mean RI values of 1% and 7% at treatments 1 and 2, respectively.The analysis of covariance performed Eq-3 Eq-4 Eq-5

BESSI D et al.
on all three treatments revealed significant differences in RI (Table 2).The coefficients of determination of the GP x RI relationship were the lowest ones compared with the relationships of GP with Tf, Sf, and NP.The

DISCUSSION
The Tf values that were higher than 100% at treatments 1 and 2 correspond to readings with mean Tf higher than GP.Moura et al. (2009), Togashi et al. (2012), and Lorenzon et al. (2013) have also reported such incoherence.The authors attributed it to the irregular spatial distribution of rainfall across an elongated river basin, as well as to the distance between the GP rain gauge and the plots, the altitudinal differences among plots, and the unevenness of plant cover.
Theoretically, it is impossible to obtain Tf values higher than GP, due to the portion of rainwater that gets intercepted by vegetation (Moura et al., 2009).However, it should be noted that the area at treatment 1 is practically devoid of trees, and thus the RI recorded therein was practically null.Another relevant aspect is that the Tf values that were higher than GP were recorded precisely in the month with highest rainwater volumes (January 2016), when rainfall events were characterized as storms with strong winds.
Therefore, these differences in Tf may be related to variations in the structure of plant cover, especially at treatment 2, where branches frequently overlapped, Rainfall partitioning in fragments... which might have contributed to the occurrence of dripping on some rain gauges (Camargo et al., 1999); to the irregular topography, size and spacing between trees (Neal et al., 1991); or even to the high heterogeneity of precipitations, due mainly to the turbulence of air currents (Togashi et al., 2012).Honda (2013) reported Tf values of 95%, 89%, and 80% to the phytophysiognomies Typical Cerrado, Dense Cerrado, and Cerradão, respectively.In another study conducted on a Cerrado area, Lima and Nicolielo (1983) reported a Tf of 73%.Leopoldo and Conte (1985), on the other hand, obtained a Tf value of 80% at São Manuel municipality.Therefore, Tf in Cerrado environments may range from 73% to 95%.In tropical forests, Bruinjzeel (1990) and Oliveira et al. (2008) reported Tf values within the range of 75% to 96%.Our Tf data are within the range regarded as reference values.In a study performed at Assis municipality, Honda (2013) concluded that the higher the rainwater volume and the lower the biomass in Cerrado gradients, the higher is the volume of rainwater that reaches the soil.In other words, there is a positive correlation between GP and plant biomass.We observed a similar trend in our study.The R² values found in the Tf x GP relationship at all areas were similar to the values obtained by Arcova et al. (2003), Diniz et al. (2013), Lorenzon et al. (2013), andTogashi et al. (2012).
The Sf values we obtained differed from those reported by Honda (2013) in a study performed using the same tree selection criterion we adopted for Sf quantification (i.e., DBH > 5 cm).Sf was 0.7% at the Typical Cerrado area, 1.5% at the Dense Cerrado, and 2.4% at the Cerradão (Honda, 2013).Leopoldo and Conte (1985), on the other hand, found even higher values in Cerrado areas at São Manuel municipality, where Sf was 2.9%.In other vegetation types, like Atlantic Forest environments, Sf has been reported to range from 0.2% to 1.1% of GP (Arcova et al., 2003;Moura et al., 2009;Freitas et al., 2013).
In light of the fact that the studied environments are at different stages of natural regeneration, it is important to compare them with studies that evaluated Sf through a similar perspective.In that sense, Diniz et al. (2013) evaluated precipitation in Atlantic Forest remnants at different stages of succession and found Sf values of 0.5% on trees with DBH > 10 cm.Oliveira Júnior and Dias (2005), on the other hand, conducted a study in an Atlantic Forest remnant at early stage of succession and reported a Sf value of 1.7% to trees with DBH > 5 cm.Both those values approach the ones we found at the studied treatments.
Several factors may be involved in causing the disparities we observed between the results we found and the ones reported in the literature.For instance, not all studies follow the same standards (like adopting the same tree selection criterion), and some of them may not even adopt the most recommended method, namely choosing a representative area and measuring Sf on all individuals in the plot (Hanchi and Rapp, 1997).These methodological differences usually occur due to the difficulty of measuring Sf on all arboreal individuals or even of accurately determining a representative area for the study (Aboal et al., 1999).
Besides the strong relationship between Sf and precipitation features, factors such as tree size; bark roughness; branch angle, inclination, and tortuosity; leaf density; and the relationship between neighboring trees are all relevant contributors to the volume of water that drains through stem surfaces (Aboal et al., 1999;Li et al., 2008;Stan Ii and Levia Jr, 2010;Bialkowski and Buttle, 2015;Honda et al., 2015;Terra et al, 2017).Thus, in areas of Cerrado, it might be possible that water not only takes different pathways but also reaches the soil in different amounts, since the Cerrado biome has different physiognomies with variable gradients of biomass, community structure and tree architecture, encompassing areas ranging from grasslands to forest formations (Durigan et al., 2002;Honda, 2013;Honda et al., 2015).GP values lower than the ones estimated by the equation obtained by the regression between Sf and GP would probably not suffice to produce Sf.Therefore, in that case, rainwater would be retained in tree crowns or on tree stems, ending up returning to the atmosphere through evaporation, or even dripping from leaves directly on the ground, thus leading to Tf formation.Analogously, Shinzato et al. (2011) reported that GP values equal to or higher than 11.0, 6.6, and 8.2 mm suffice for Sf to occur in a seasonal semideciduous forest and in populations of E. cloeziana and Pinus sp., respectively.
Interestingly, regarding the minimum GP needed to produce Sf, the value estimated for treatment 3 (5 mm) was lower than the one estimated for treatment 2 (9.3 mm).This could be explained by the presence BESSI D et al.
of palm trees in the former, which may facilitate rainwater catchment and its subsequent drainage through stem surfaces (Togashi et al., 2012;Lorenzon et al., 2015;Corrêa et al., 2016).
Considering the occurrences in which we obtained NP values higher than GP, we also recorded a large rainwater volume before and during those rainfall events.This could be due to the reduced capacity of tree crows to store water during events of very heavy rainfall (Tobon Marin et al., 2000).Literature reports indicate a high variation in NP among different vegetation types.In seasonal semideciduous Atlantic Forest environments, NP has been reported to range from 81.4% to 87.06% (Castro et al., 1983;Arcova et al., 2003;Oliveira Júnior and Dias, 2005;Alves et al., 2007).Diniz et al. (2003) studied that same vegetation type, but with plant formations of different ages.The authors observed that in a 20-year-old vegetation, NP was 77.89%, while in a 30-year-old vegetation it was 79.26% and in a 60-year-old vegetation it was 76.98%.Thus, comparing the 20-and 60-year-old areas, NP showed a decreasing trend with increasing forest age.
The linear regression results of the NP x GP relationship corroborate the fact that practically all rainwater in areas at early stage of regeneration is converted to NP, due to the low density of arboreal individuals therein.In the literature, there are reports of R 2 values for such relationship of 0.99, by Oliveira Júnior and Dias (2005); 0.96, by Diniz et al. (2013); and 0.99 and 0.98, by Lorenzon et al. (2013), all of which are compatible with the values found in our study.
RI at treatment 1 may also reflect evaporation of the collected water, in view of the exposure of rain gauges to full sun.In order to avoid evaporation, the equipment was built in such a way that it would reduce that loss to a minimum.Still, we recorded negative RI values after rainfall events in January at treatments 1 and 2. These records correspond to NP values higher than 100% of GP, which thus indicates that no RI occurred, but rather NP increased.These data may have been influenced by meteorological conditions regarding rainfall volume, intensity and duration; wind speed; and the interval between rainfall events (Crockford and Richardson, 2000).The recorded data may also have some relationship with vegetation conditions and forest-related features, such as plant architecture, canopy structure, community composition, species heterogeneity, and especially canopy density and leaf area (Ávila et al., 2014).
In less dense phytophysiognomies, i.e., those with smaller and more sparsely distributed trees, rainwater retention is minimal.Additionally, the transition to denser phytophysiognomies leads to an increase in community density, plant size, and leaf biomass, and thus to a consequent increase in rainfall interception capacity (Honda, 2013).The same processes can be observed in forests at different stages of natural regeneration, as indicated by the gradual increase in the volume of intercepted rainwater with increasing biomass along the study areas.
The determination and correlation coefficients we observed show that RI is little explained by GP.Thus, RI might be influenced by other aspects, like rainfall intensity and the occurrence of winds during rainfall events (Moura et al., 2009).Diniz et al. (2013) and Lorenzon et al. (2013) also found low values of those coefficients, which were even lower than 0.57 for the GP x RI relationship.
In terms of water management and in view of the hydrological processes observed in our study, one could expect that, since treatment 1 has the lowest regeneration time and receives the highest Tf volume, that would be the treatment with best conditions for soil water recharge.However, it is important to highlight that precisely because rainfall in that treatment reaches the soil directly, with vegetation playing no role in Rainfall partitioning in fragments... reducing the impact of raindrops, soil thereat is compacted, which thus hampers water infiltration and percolation through the soil profile, thereby propitiating the occurrence of surface runoff.On such a scenario, Sf is crucial for recharging soil water.In that sense, Sf has been shown by several studies performed on arid and semiarid areas (Garcia-Estringana et al. 2010;Yuan et al., 2016;Zhang et al., 2017) to be the only pathway of water maintenance enabling plant development.Much research is still necessary for us to fully understand the dynamics of water balance in the soil-plant-atmosphere system.Studies aiming to comprehend the development and management of vegetation with a focus on soil restoration for water recharge are urgently needed.Lastly, we conclude our study with the following question: From what moment is there an effective contribution to soil water recharge along the development of forest species in areas undergoing restoration?

CONCLUSION
Throughfall and net precipitation decreased with advancing natural regeneration, whereas stemflow and rainfall interception increased with increasing vegetation development.When data was analyzed for its linear trend, the throughfall, stemflow, and net precipitation showed higher coefficients of determination, throughfall being the variable that contributed the most to the volume of water that reached the soil.Rainfall interception, on the other hand, showed weak correlation with gross precipitation, thus indicating that it might be influenced by other factors.Besides, interception was null in events of very intense rainfall, which shows that the forest canopy has a maximum rainwater retention capacity.Based on the analysis of covariance conducted on each rainwater component at each treatment, we conclude that there are significant differences among treatments according to their stage of natural forest regeneration.