Data was taken in a pure Nothofagus pumilio forest harvested through variable retention (Martínez Pastur et al., 2009) located in the Isla Grande of Tierra del Fuego (Argentina). This harvesting method retains a percentage of the original forest structure, leaving aggregates (a circular patch of 30 m radius per ha) and dispersed dominant trees (10-15 m².ha^–1 of basal area) evenly distributed between the aggregates. The forest of medium site quality (dominant height of 22-24 m) is located in Estancia Los Cerros (54°18’ S, 67°49’ W), and before harvesting it presented a range of 700-900 m³.ha^–1 of timber volume and 70-80 m².ha^–1 of basal area (Martínez Pastur et al., 2009). This forest belongs to a long-term permanent plot of a network in Patagonia Sur (PEBANPA, Parcelas de Ecología y Biodiversidad de Ambientes Naturales en Patagonia Austral, INTA-CONICET, Argentina) (Martínez Pastur et al., 2010). Regeneration permanent plots (500 x 20 cm) established in primary unmanaged forests (PF) (n = 18) and harvested stands 10 years ago with variable retention (n = 54) were used. In the harvested forests, the plots where located covering 3 different conditions according to the retention levels (see Martínez Pastur et al., 2011b): (i) within-aggregate retention (AR) (n = 18), in the dispersed retention under the influence of the aggregated retention (<20 m from the aggregates edges) (DRI) (n = 18), and in the dispersed retention far away the influence of the aggregates (DR) (n = 18).

Seedling recruitment was recorded during summer from 2007 to 2014. Moreover, litter traps were placed close to each plot (50 x 30 cm) to collect seeds during May of each year from 2006 to 2013. Dominant height (DH) was taken in each plot by Impulse Laser Rangefinder (Laser Technology, US), basal area (BA) using a Criterion RD-1000 (Laser Technology, US) with a K-coefficient between 1 and 6. Total over bark volume (TOBV) was estimated using equations proposed by Martínez Pastur (2006) with stand basal area and site quality as independent variables. Hemispherical photos were taken to estimate canopy cover (CC) and relative leaf area index (RLAI) integrated over the zenith considering an angle between 0° and 60° (Stenburg et al., 1994). The photos were taken at a height of 1 m above the ground with 8-mm fisheye lens (Sigma, Japan) mounted on a 35-mm digital camera (Nikon, Japan), and were analyzed with Gap Light Analyzer v.2.0 software (Robison & McCarthy, 1999; Frazer et al., 2001). The program was supplemented with a distortion projection provided by the lens manufacturer, and a division of the sky composed of 20 azimuths and zenith 4 regions was used.

There is a lack of local climate information for the study area, for this we use data from regional climate using satellite images. Different data sources were used to characterize this regional climate between years 2006 and 2013: (i) For the Argentine sector of Isla Grande de Tierra del Fuego, data from the re-analysis NCEP (National Center for Environmental Prediction) of NOAA (National Oceanic and Atmospheric Administration) were used: average monthly air temperature data near earth surface (0.995 sigma level) with a spatial resolution of 2.5°; and average monthly precipitation (precipitable water) was obtained from the same source and resolution, since this data source provides more consistent and continuous information for areas with little climate information (Kalnay et al., 1996). (ii) For the sea surrounding Isla Grande de Tierra del Fuego (48°55’ to 59°24’ S and 54°52’ to 76°16’ W), data of Sea Surface Temperature (SST) were obtained from NASA Ocean Color. This service provides information on monthly composite SST re-sampled to 4-km from version 3 of SST product of nighttime images of the original MODIS (1-km resolution and 3 scenes daily). Climate data were averaged by month (PP = precipitation, T = temperature, S = temperature of sea surface), and were analyzed considering minimum (MIN), maximum (MAX) or annual mean values (A) (April to March of the following year); and according to summer (SUM = October to March of the following year) or winter (WIN = April to September) seasons. For seed recruitment and efficiency data (SEE, REC, EF), values were correlated with the climate of summer of the corresponding season either with the previous winter or with the sum of both in the case of annual data.

Three analysis of variance (ANOVA) were performed: (i) for different forest types (PF, AR, DRI, DR) considering forest structure and canopy variables (DH, BA, TOBV, CC and RLAI); (ii) for different forest types (PF, AR, DRI, DR) considering seed production and recruitment variables (SEE, REC, EF); and (iii) years after harvest (3 to 10 years) considering seed production and recruitment variables (SEE, REC, EF). Beside this, multiple ANOVA considering forest types (PF, AR, DRI, DR) and relative seed production according to magnitude of seed rain (1 = low, 2 = medium, 3 = high) was performed to analyze data of seed production and recruitment variables (SEE, REC, EF). Differences between factor means were compared using Tukey test (p <0.05).

A canonical correspondence analysis (CCA) was performed (terBraak & Šmilauer, 2002) with seed production, recruitment (SEE, REC, EF), and regional climate data. The objective of this analysis was to: (i) estimate how much the regional climate values influenced the variance of seed production and recruitment data, and (ii) identify those regional climate variables with the highest influence for each group. The significance of the CCA analysis was quantified by Monte Carlo test, where 499 permutations were performed for each model. The most significant variables in each multivariate analysis were used to adjust linear regressions without origin coefficient for the different seed production and recruitment variables (SEE, REC, EF) and forest types (PF, AR, DRI, DR). To evaluate the adjustment of each equation, the significance of the coefficients, the r²-adj., the standard error of the estimation (SEE) and the mean absolute error (MAE) were considered.

Significant differences were detected in all the studied variables (Table 1). Dominant height was slightly higher in the harvested forests (aprox. 24.0 m) compared to primary unmanaged forests (22.2 m). Basal area and volume was higher in PF, intermediate in AR, and significantly lower in harvested areas (DRI and DR). Canopy cover also presented significant differences among all the treatments, with a gradient from PF and from within the aggregates to the farthest areas in the dispersed retention. Finally, relative leaf area index did not present any differences between PF and AR, which were significantly higher than in the harvested areas (DRI and DR).

When all the studied years were analyzed together, significant differences were found in the seed production and seedling recruitment for the different forest types (Table 2). The higher values were recorded in PF, followed for the AR and the harvested areas (DRI and DR). However, in the seedling recruitment the AR and dispersed retention areas did not present significant differences. Finally, no significant differences were found for the recruitment efficiency among forest types. However, when considering the inter-annual variations for all the harvested treatments (AR, DRI and DR) significant differences were detected among years for the three studied variables (Table 3). These differences did not follow an uniform pattern, due to years of high (e.g. year 4 after cutting) or low (e.g. year 9 after cutting) seed production were recorded.

When recruitment variables were considered (REC and EF) to compare relative production of seeds in the different forest types (in PF: 1 = 0-3, 2 = 3-12, and 3 = > 12 mill.ha^–1.year^–1; in AR: 1 = 0-3, 2 = 3-10, and 3 = > 10 mill.ha^–1.year^–1; in DRI: 1 = 0.0-0.8, 2 = 0.8-2.5, and 3 = > 2.5 mill.ha^–1.year^–1; in DR: 1 = 0.0-0.5, 2 = 0.5-1.5, and 3 = > 1.5 mill.ha^–1.year^–1) significant differences were detected for both variables in PF, increasing REC and EF as SEE increased. No differences were found in the harvested forests, except for the REC in DR, where it significantly increased with SEE (Table 4).

The canonical correspondence analysis (CCA) for seed production (Table 5 and Fig. 1) showed a total inertia of 0.21 (eigenvalue of axis 1 = 0.09 and of axis 2 = 0.02) explaining with both axes 53% of the variation in seed production by forest types, and 98% of the variation in seed production by regional climate. The Monte Carlo test presented a F(p) = 2.38(0.048). The climatic variables with higher significance were the average precipitation during winter (PP-A-WIN), the minimum temperature over sea surface during winter (S-MIN-WIN), the average temperature over land surface during summer (T-A-SUM), the maximum temperature over land surface during summer (T-MAX-SUM), and the minimum precipitation during winter (PP-MIN-WIN). The first axis separates the treatments in two groups, primary forests to the left and harvested stands to the right, while the second axis generates a relative gradient of canopy cover.

The CCA for seedling recruitment (Table 5 and Figure 1) showed a total inertia of 0.38 (eigenvalue of axis 1 = 0.11 and of axis 2 = 0.02) explaining with both axes 34% of the variation in seedling recruitment by forest types, and 99% of the variation of seedling recruitment by regional climate. The Monte Carlo test presented a F(p) = 1.32(0.236). The climatic variables of higher significance were the minimum temperature of land surface during winter (T-MIN-WIN), the maximum temperature of land surface during summer (T-MAX-SUM), the mean precipitation (PP-A), the average temperature of sea surface during summer (S-A-SUM), and the average temperature of land surface during summer (T-A-SUM). As was observed for seed production, the first axis separates the treatments in primary forests to the right and harvested forests to the left, while the second axis generates a relative gradient of canopy cover.

The CCA for recruitment efficiency (Table 5 and Figure 1) showed a total inertia of 0.43 (eigenvalue of axis 1 = 0.14 and of axis 2 = 0.02) explaining with both axes 39% of the variation in recruitment efficiency by forest types, and 98% of the variation of recruitment efficiency by regional climate. The Monte Carlo test presented a F(p) = 1.30(0.198). The climatic variables of higher significance were the minimum precipitation during summer (PP-MIN-SUM), the minimum temperature of sea surface during winter (S-MIN-WIN), the maximum temperature of land surface during winter (T-MAX-WIN), the minimum temperature of sea surface during summer (S-MIN-SUM) and the minimum average temperature of sea surface (S-MIN). The first axis separates PF and AR (areas with higher canopy cover) from the harvested areas (DRI and DR), whereas the second axis separates AR from the rest of the treatments.

The linear regression fitted for seed production (SEE) explained between 62% and 97% of the variation (Table 6), using average precipitation during winter (PP-A-WIN) and minimum temperature of sea surface during winter (S-MIN-WIN) for PF and DRI, while AR also included average temperature of land surface during summer (T-A-SUM). The equation for DR also included PP-A-WIN, and another two different values: minimum precipitation during winter (PP-MIN-WIN) and maximum temperature of land surface during summer (T-MAX-SUM). The fitted models were significant for the four forest types (p = < 0.001 to 0.035).

The linear regression fitted for seedling recruitment (REC) explained between 55% and 68% of the variation (Table 7), using only minimum temperature of land surface during winter (T-MIN-WIN) for the harvested areas (DR and DRI), and included average temperature of land surface during summer (T-A-SUM) and average precipitation along the year (PP-A) for the areas with higher canopy cover (PF and AR). The fitted models were significant for RDI and RD (p = 0.023 and 0.017, respectively), and marginally significantly for PF and AR (p = 0.095 and 0.078, respectively).

The linear regression fitted for recruitment efficiency (EF) explained between 59% and 95% of the variation (Table 8), using minimum temperature of sea surface during winter (S-MIN-WIN) and summer (S-MIN-SUM) for the primary forests, while the harvested stands using minimum precipitation during summer (PP-MIN-SUM) as main variable, and minimum temperature of sea surface (S-MIN) for AR and DRI, and minimum temperature of sea surface during winter (S-MIN-WIN) for DR. The fitted models were significant for the harvested stands (p = < 0.001 to 0.006), and marginally significant for PF (p = 0.109).

Harvesting significantly modifies the forest structure, and the magnitude of these changes depends on the cutting method (Martínez Pastur et al., 2009; Parker et al., 2013). Variable retention method generates a gradient of changes from the center of the aggregates, where the structure is similar to the primary forest, to the farthest areas in the dispersed retention (Cellini, 2010; Martínez Pastur et al., 2011b). The aggregates maintain some of the natural characteristics of the primary forest (e.g. RLAI) but mostly present intermediate values with the harvested areas. Also, the harvested areas under the influence of the aggregates presented higher values of forest structure than the harvested areas without influence of the aggregates.

Retention levels are directly related with the microclimatic conditions generated into the forests that were harvested (Chen et al., 1993, 1995), e.g. amount of rainfall that reaches to the forest ground, available radiation under the canopy, wind exposure and temperature (Richter & Frangi, 1992; Caldentey et al., 2005, 2009). These factors had the biggest influences on growth and survival of natural regeneration in Nothofagus pumilio forests (Lieffers et al., 1999; Heinemann et al., 2000; Martínez Pastur et al., 2007, 2011b). However, the influence of the forest structure over regeneration is not linear, e.g. intermediate values of forest cover (45-55%) generate the best conditions for the seedlings growth (Martínez Pastur et al., 2011a). Also, low levels of forest cover generate high light availability, but also generate high contents of soil moisture due to the amount of rainfall, affecting the physiological performance and biomass allocation of seedlings (Lencinas et al., 2007; Martinez Pastur et al., 2007).

Seed production and recruitment are directly related to the forest cover (Cellini, 2010; Parker et al., 2013), but not to the recruitment efficiency. Some studies suggest that the new stand conditions in the harvested forests were adverse for the seedling recruitment (Collado et al., 2008) but other studies suggest that those conditions improve the establishment and growth (Rosenfeld et al., 2006; Cellini, 2010). In this study, no significant differences were detected in the recruitment efficiency among the studied forest types, where lower values were found in the extreme forest covers (1.39% and 1.46% in DR and PF, respectively) rather than in intermediate covers. This results can indicate better ecological conditions for seedlings, as was previously cited for height growth and eco-physiological performances in plants successfully established (Lencinas et al., 2007; Martinez Pastur et al., 2007, 2011a).

Seed production greatly changed along the years, where the existence of extraordinary seed-falls were observed (e.g. years 4 and 8 after harvesting). This cyclic production of seeds has been cited for many forests species (Kelly, 1994; Kelly et al., 2000; Koenig & Knops, 2000) and was specially described for the Nothofagus genera (Monks & Kelly, 2006) due to the high variation levels among years (Kelly, 1994; Kelly & Sork, 2002). However, seedling recruitment and recruitment efficiency values decrease along the years. This decay could be due to changes in the biotic and the abiotic factors of the harvested forests after the cuttings. Among the biotic changes, the dynamic of the understory (e.g. Lencinas et al., 2011; Gallo et al., 2013) and the herbivore grazing (Soler et al., 2012), are the factors with greater influence; and among the abiotic changes, the climate extreme events could be considered as the most influent (e.g. Curran et al., 1999; Schauber et al., 2002; James, 2011). Some studies explored the influence of these factors at stand scale (microclimate), the seed foraging (e.g. insects, birds and mice) or the relation with seed production (Soler et al., 2013; Martínez Pastur et al., 2013) in Nothofagus forests.

There are different hypothesis that relate some of the seeding processes and climate (Isagi et al., 1997; Liebhold et al., 2004; Bahamonde et al., 2011), however the absence of local climate data in Patagonia did not allow us to determine these relationships. One alternative can be exploring the influence of these factors with the climate at regional scale that was presented and discussed here.

In southern Patagonia Sur, and specially Tierra del Fuego Island, regional climate is controlled by Atlantic Ocean, the influence of the Antarctic climate and Los Andes Mountains, by generating strong climate gradients at regional levels (Kreps et al., 2012). Growing season at these latitudes is restrained to the summer months (Massaccesi et al., 2008; Villalba et al., 2010) and regional climate determined its duration. Also, climate events that occurred during the long winter season influence over summer season, influencing in the ecological conditions of the stands, e.g. higher snow accumulation during winter enlarge ice melting period, increasing the amount of water available in rivers and streams and consequently the availability of water in the forest ground during the summer season (Bales et al., 2011).

The multivariate analyses showed the importance of rainfall over the seed production and recruitment, e.g. the amount of winter rainfall over seed production in the following year, the amount of annual rainfall over the recruitment, and the summer precipitation over the recruitment efficiency. Also, different variables related to winter and summer temperatures were detected as important to predict the observed changes of these variables. For example, minimum winter temperature and summer temperature were related to seed production, summer temperature and minimum winter temperature were related to seedling recruitment in primary and harvested forests, respectively, and minimum temperatures of both seasons were important for the recruitment efficiency. Other studies found similar correlations, e.g. Richardson et al. (2005) related seed production of Nothofagus solandri with temperature, finding that cold and humid summers produce a higher number of fruits, whereas Smaill et al. (2011) also determined the influence of rainfall for the same species. Kon et al. (2005) found correlations between minimum temperatures of the previous spring and seed production of Fagus crenata forests, as well as Masaki et al. (2008). James (2011) found correlations between growth and recruitment of Larix sibirica and Picea obovata with fires, land use climate (temperature during different seasons). Finally, Selås et al. (2002) found correlations between growth, seed production and climate (temperature and rainfall) in Picea abies forests.

Early regeneration of Nothofagus pumilio is very sensitive to the lack of soil moisture, mainly during summer season (Heinemann et al., 2000), but when plants growth, the system changes from soil moisture sensitive to light sensitive (Heinemann & Kitzberger, 2006; Martínez Pastur et al., 2011a, 2011b). For this reason, rainfall accumulated during winter or summer rainfall plays a determinant role in seedling recruitment. Also, seedlings are sensitive to ice damage, mainly in environments with low levels of forest cover (e.g. harvested areas) where the occurrence of frost can occur during all the summer season (Martínez Pastur et al., 2009; Kreps et al., 2012).

Some of the studied variables (e.g. seeding patterns) presented cycles that were described for many forest species including Nothofagus genera (Monks & Kelly, 2006). This phenomenon occurred both, at regional scale (Kelly, 1994; Kelly & Sork, 2002) and at stand scale (Martínez Pastur et al., 2013; Soler et al., 2013), and probably the first one is correlated with the regional climate, and the second one can be the responsible of the observed variation that can not be explained with the models adjusted here. Beside this, some correlation among the regional climate variables exists, e.g. years with cold winters leads to obtain lower average annual temperatures. These outputs can be taken in account when these models were used.