3.1. Meteorological Conditions
The meteorological station of Mocambinho, which was close to the MSSP, registered an annual rainfall of 602.2 mm yr−1 during the hydrological year 2013–2014, and it was distributed among three periods of convective rains. Daily temperatures oscillated between 21.1 °C to 30.5 °C, with a mean annual temperature of 25.8 ± 1.8 °C. Mean annual relative humidity was 62.3% (monthly range: September (45.2%) and March (72.7%)) and was strongly influenced by rain events; humidity decreased almost automatically with the absence of rains. During the sampling period in MSSP, total rainfall was 294.4 mm for the early successional stage, 229.3 mm for the intermediate stage, and 174.4 mm for the late stage; there was a maximum daily rainfall of 70.5 mm d−1 (from November 2012 to January 2013).
SRNP had an annual rainfall during the hydrological year 2013–2014 of 1118.0 mm yr−1. Daily temperature fluctuated between 22.4 °C and 27.4 °C, with a mean annual temperature of 25.1 ± 0.9 °C, and temperature did not exhibit strong changes throughout the year. The relative humidity increased from a minimum of 51.4% in March to a maximum of 98.4% in the rainy season, with small daily changes that were contrary to the Brazil site. During the sampling period, SRNP recorded 372.7 mm of rain in the early successional stage, 385.2 mm in the intermediate stage, and 327.8 mm in the late stage, with a maximum daily rainfall of 83.5 mm d−1.
In general, temperatures at both sites showed comparable ranges of values during this study. However, relative humidity was notably drier in MSSP than in SRMP as a direct consequence of total rainfall measured during the hydrological year: 602.2 mm yr−1
for MSSP and 1118.0 mm yr−1
for SRSP. It is important to mention that despite the geographical proximity among plots in each site, there was significant heterogeneity in gross rainfall among stages and plots (Table 2
), which reinforced the need to measure gross rainfall in each plot to reduce the estimation error. In addition, total sampled days during this study were greater for SRNP (89 to 78 days) than for MSSP (34 to –36 days), which is an expected result due to the lower annual rainfall in MSSP (Table 3
3.3. Water Fluxes and Forest Structure
Rates of throughfall for SRNP followed the expected decline with increasing complexity of forest structure from early successional stage (87.3%) to intermediate (72.7%) to late (63.2%) (Table 2
). This implies that as the forest structure became more complex, the capacity to intercept water increased (Table 3
). At MSSP, there was no such trend: early (85.2%), intermediate (79.2%), and late (84.3%). One possible explanation is that the structural differences in forests between intermediate and late stages were not as strong as at SRNP. For example, the difference in average tree height (Table 4
) between intermediate and late stages in Brazil was only 0.3 m (15.9 m, 16.2 m), but at SRNP the difference of 5.9 m (16.5, 22.4) was remarkably higher. In addition, a high density of lianas could affect interception capacity in MSSP. Madeira et al. [17
] reported for MSSP in plots of 1000 m2
that there were no lianas in the early stage and an outstanding density of lianas in the intermediate stage (32 lianas plot−1
) compared with the late stage (15 lianas plot−1
). In addition, the basal area of lianas in the intermediate stage was 55% greater than in the late stage (0.56 m2
compared with 0.36 m2
). These explanations justify also the same trend differences observed for net rainfall and interception between both sites (Table 2
) and stemflow (PSF
) fluxes increased with gross rainfall (PGr
) in all stages at both sites (Figure 2
). As expected, forest interception (PInt
) was higher when the gross rainfall events were low, and it was lower when the gross rainfall events were higher, due to a decline in the interception capacity as the foliage and bark became saturated during the rainfall event. It is important to note that the dispersion of points for PTF
have a nearly linear trend in all stages at both sites. However, the dispersion of the points increased as it progressed from the early to the late stages, especially at SRNP. This dispersion reflected the high variability in forest structure among the SRNP plots, particularly in the late stage.
The trend of stemflow with respect to successional stages also differed between sites. At SRNP, stemflow was lowest in the early stage (0.2%) and highest in the intermediate and late stages (0.3%). These results were expected due to the dominance of larger trees in the advanced stages that had higher collecting capacity due to more developed crowns, branches, and trunks. At MSSP, the trend was reversed, with the highest collecting capacity in the early stage (0.2%) and the lowest in intermediate and late stages (0.1%). These results contradict previous studies developed in Brazil in TDF that showed the same stemflow trend as the SRNP site [11
]. Therefore, the results from MSSP were unforeseen and difficult to interpret in this study. One possible explanation for the low rates of stemflow in the intermediate and late stages in MSSP was the outstanding abundance of lianas found in both stages, which could alter the flow of water from tree crowns and branches to the stems. Madeira et al. [17
] indicated that the intermediate and late stages at MSSP had a high density of lianas, but in the early stage, lianas were almost absent.
3.4. Regression Models for Forest Interception Fluxes
The coefficients of determination (R2
) of throughfall were > 0.93 in all stages at each site, using the gross daily rainfall as the independent variable (Table 3
). These good linear regression adjustments were the result of the sampling technique and the selected sample size for this study. Because throughfall represented the highest proportion of net rainfall, the coefficients of determination for equations for net rainfall had also almost identical values of R2
and levels of statistical significance compared with the equations for throughfall.
Coefficients of determination (R2
) for stemflow were >0.70 in all stages at each site, except for the late stage at SRNP, which had the lowest R2
(0.56) (Table 3
). Hence, in comparison to equations for throughfall, results for stemflow had lower linear regression adjustments (R2
) because of the small sample size and higher dispersion of data (Figure 2
). The lower R2
for the late stage at SRNP was due to the high variability in structural characteristics among the late stage plots. Plot CR1L3 (Table 4
) was the oldest, and it included several remnant trees of outstanding dimensions (height, diameter, crowns).
In general terms, most of the intercepts (α) in all the equations were not significantly different from zero with p < 0.05 and p < 0.10. Regardless of these results and as stated in the methodology, the models were not forced through the origin, so that all equations were equal and allowed a fair comparison of the slopes (β) among stages and between sites. In all equations, the slopes (β) were statistically significant at p < 0.05, hence, these models provided a tool to estimate the water fluxes in successional TDF of Costa Rica and Brazil.
Because the study sites have different climates and tree species composition, we tested if it was feasible to generate generalized equations to estimate net rainfall for successional TDF that would be applicable for any site within the ranges of climate and forest structural characteristics of this study. Following the test indicated in the methodology to compare equations between sites, we found that the equations for early and intermediate stages were statistically the same (p < 0.05). Therefore, the two generalized equations to estimate net rainfall for the early and intermediate successional stages of TDF are included (Table 3
). In the case of late successional stages, the equations were not equal; therefore, we recommend that the investigator choose one equation that best represents the biophysical characteristics (climate and forest structure) of the site of interest. It is important to point out that according to literature review, the slopes (β) of the equations for net rainfall with gross rainfall are the best estimation of the rainfall interception of the forest stages that we analyzed [23
3.5. Relationship between Net Rainfall and Forest Structural Parameters
According to Table 4
net rainfall resulted in significant correlations at regional level only with the variables height (H, r = −0.491, p < 0.05) and plant area index (PAI, r = –0.717, p < 0.01). However, when analyzing Figure 3
and Table 4
, it is clearly observed that the linear relationship for PAI is only valid for SRNP and not so for MSSP. In Figure 3
it can also be observed how the plot CR1L2 in SRNP has a determining effect on the correlations between net rainfall and height, so that if this plot is removed from Figure 3
the relationship is insignificant for both sites (Table 5
). Hence the only valid correlation is found in SRNP with PAI (r = –0.755, p < 0.05), proving that more complex TDF structures in Costa Rica can retain more water than simpler forest stands. On the other hand, none of the structural variables in Brazil correlates significantly with net rainfall.
There are few studies in the literature that analyze the partitioning of rainfall by forest interception in TDF, and it is even more difficult to find information that considers the effect of successional stages. However, we found two relevant studies from Brazil, which were from the same state of Minas Gerais in Mata do Paraíso-Viçosa, with an annual average rainfall of 1500–2000 mm yr−1
(more similar than the Costa Rica site). Oliveira et al. [11
] found that throughfall rates in early stages of 79.3% and in intermediate stages of 72.6%, and Lorenzon et al. [22
] reported 84.39% and 73.04% for throughfall for the same region in early and intermediate successional stages. These findings differ slightly from our site in MSSP with 85.2% and 79.2% for intermediate stage (Table 2
Oliveira et al. [11
] found stemflow rates of 0.44% and 1.52% for the early and intermediate stages, whereas Lorenzon et al. [22
] found 0.68% and 1.8%, respectively. These results differ strongly from our stemflow values for MSSP of 0.2% (early) and 0.1% (intermediate and late). According to a review by Cavelier and Vargas [8
], stemflow rates ranged from 0.3% to 1.8% in 15 studies conducted in Brazil and Colombia in dry to moist tropical forests (1500 mm yr−1
to 3140 mm yr−1
). Hence, our estimates for stemflow for SRNP were within the reported range of estimates in the literature, but not for MSSP; this was probably due to the effect of high density of lianas in the intermediate and late stages and because MSSP is far drier (840 mm yr−1
) than any other reported study site.
This study used 20 funnel-type gauges, which provided acceptable precision for measuring throughfall. If the sample size is increased in search of higher precision, then researchers must consider the time required for one person to measure the entire trial from dawn to 8:00 am. For both MSSP and SRNP, the time available to measure the trial was rather tight for one person.
For the applied sampling design in this study, we concluded that the number of sampled trees to evaluate stemflow must be greater in future studies and, to the extent possible, tree sampling must consider the effect of tree bark types of the most dominant species. Several studies had pointed out that specific stand characteristic such as tree morphology influence the interception capacity [24
A practical objective of this study was to generate linear regression equations to estimate water fluxes such as throughfall, stemflow, interception, and net rainfall. These equations are required to complement studies of nutrient dynamics in forest ecosystems, to study the impact of climate change, and to determine the influence of changes in forest cover in the hydrological cycle at the landscape and ecosystem levels, among other applications. Given the lack of equations to estimate net rainfall in almost all countries, we developed two generalized equations for the early and intermediate stages. These equations must be applied with caution in other sites, and they should be based on the similarity in forest structure and climate characteristics. Because it was not possible to develop a generalized equation for the late stage, it is recommended to select one of the two equations, SRNP or MSSP, that best corresponds to the biophysical characteristics of the site of interest.
Finally, this study makes an important contribution to understanding the impact of successional stages of TDF on the hydrology of the landscape, especially considering the alteration of rainfall and temperature regimes due to climate change. This study provides robust data to encourage the use of hydrological models to simulate the effects of changes in forest cover and climate change in the hydrology of the landscape.