The Kosñipata Valley (13°03′S, 71°33′W) lies along the eastern slope of the Andes in southern Peru. Elevations range from about 800 m to over 4000 m, and vegetation changes from pre-montane rainforest at the lowest elevations to tropical subalpine forest and puna (alpine grassland) at the highest elevations ³⁹ . The experiment was installed along a single forested ridge, with three transplant sites at 1500, 1650, and 1800 m. We chose these elevations because the large increase in vascular epiphyte and bryophyte biomass ⁴⁰ , and step changes in soil properties ⁴¹ and biomass carbon stocks ⁴² between 1500 and 2000 m elevation in the Kosñipata Valley are likely associated with higher cloud incidence and lower temperatures, as seen on other tropical mountains ^{43, 44} . The bedrock underlying the ridge is Permian granite, and soils are classified as umbric Gleysols ⁴⁵ .

The Kosñipata Valley has a perhumid climate described in detail in Rapp and Silman ⁴⁶ . Temperature decreases linearly with altitude and annual rainfall is high across the gradient, with a distinct, but weak dry season (monthly rainfall > monthly potential evapotranspiration except during drought). Vapor pressure deficit (VPD) typically decreases with altitude across the east Andean slope above 1500 m for all months except June ⁴⁶ . While differences in VPD between 1500 m (lowest experimental elevation) and 1800 m (highest experimental elevation) are typically small, excursions to higher VPD (greater desiccation) are more severe at 1500 m ⁴⁶ . The dry season (May–August) is when cloud immersion is expected to be most important for cloud forest plants. In July, the driest month, high relative humidity associated with cloud immersion (>95%) is more common at 1840 m than at 1500 m at climate stations approximately 1 km from the study site, while vapor pressure deficits greater than 1.0 kPa are more common at the higher elevation ( Table 1). Vapor pressure deficits greater than 1.0 kPa are associated with moisture stress in cloud forest plants ^{47, 48} . We took daily photographs at 16:00 local time from a fixed location at 1400 m facing up-valley towards the ridge were the experiment was installed during June and July 2005 and July 2006. We categorized cloud base height using these photographs as being less than 1500 m (below experimental elevations), 1500–1800 m (within experimental elevations), or greater than 1800 m (above the experimental elevations). This analysis confirmed that cloud base height did not differ between years (Χ ² = 3.32, p = 0.19), and that cloud frequency increased with elevation (cloud base <1500 m in 9% of observations; 1500–1800 m in 30% of observations; >1800 in 61% of observations).

In the 2005 austral winter (dry season: June and July), we selected five Alzatea verticillata Ruiz & Pav. trees at each of three elevations, 1500 meters, 1650 meters, and 1800 meters elevation. Alzatea was an appropriate choice for a host tree because: (1) it was common at all three elevations; (2) it attains large size and has strong wood suitable for supporting climbers working in the trees; and (3) its unique architecture resulted in many large horizontal branches that supported sizeable epiphyte mats. We accessed trees using roped arborist techniques ⁴⁹ . In each tree, we chose four sections of epiphyte mat that were at least 25 cm wide and 30–40 cm long. Within each of the 60 mats, we marked an area of 25 × 25 cm with wire, and marked all ramets of vascular epiphytes within it, identified them to morpho-species, and recorded the length of shoots and number of leaves of each ramet.

Most taxa, and all of the focal taxa (see below), were non-reproductive when surveyed. This, combined with the fact that the epiphytic flora of the Andes is relatively poorly known, made it impossible to identify all taxa to species. We therefore used morpho-species designations in our analysis. Taxonomic uncertainty therefore, could affect our results if individuals of multiple cryptic species were combined in the analysis. Nonvascular epiphytes were present, but not considered in this experiment.

Most mat dwelling epiphytes are clonal, with individual ramets connected by subsurface stems, but capable of surviving without connection to other ramets. It was difficult to determine individual genets without excavating the plants, so we identified and measured individual ramets rather than genetically distinct plants. For strap-leafed ferns in the genus Elaphoglossum, each ramet was a single leaf, while ramets for other taxa consisted of one or more stems with multiple leaves.

On each tree, one of the mats was left in place to serve as an undisturbed control. We cut each of the other three mats from the tree, lowered them to the ground, and then transplanted one each to a random tree at each of the three elevations. After all transplants were complete, each tree had one undisturbed mat, one mat that had been removed and then replaced at the same elevation, and one mat from each of the other two elevations. We tied each mat in place using wire, and then watered it with one liter of water to minimize any desiccation effect that handling may have had. Supplementary Figure 1– Supplementary Figure 4 illustrate the process of transplanting the epiphyte mats.

We left the mats undisturbed for one year, and resurveyed them the following year in June and July 2006 ( Dataset 1). We searched for all marked ramets and counted the new ramets of each morpho-species. We assumed ramets obviously more than a year old (28 out of 1400+ original ramets) had lost their tag if previously marked ramets of the same morpho-species were not found in the same mat. Eight ‘old’ ramets were still not accounted for; we assumed these were missed during the first census. Any other missing ramets were assumed to be dead.

With these data, we defined three measures of population performance: 1) survival, 2) recruitment, and 3) population change. Survival was defined as:

            Survival = (N ₂₀₀₅ – D ₂₀₀₆)/N ₂₀₀₅,

where N ₂₀₀₅ was the number of ramets surveyed in a mat in each year, and D ₂₀₀₆ was the number of ramets surveyed in 2005 that had died by 2006. Recruitment was defined as:

            Recruitment = n ₂₀₀₆/N ₂₀₀₅,

where n ₂₀₀₆ was the number of new ramets surveyed in 2006, which were not present in 2005. Population change was defined as:

            Population change = (N ₂₀₀₆ – N ₂₀₀₅)/N ₂₀₀₅.

We conducted analyses at the community level and for the most common morpho-species individually. The common morpho-species occurred in at least half (10) of all epiphyte mats transplanted from at least one elevation. These included four common morpho-species identified to genus, by which we will refer to them: a strap-leaf fern ( Elaphoglossum Schott ex J. Sm.), two orchid morpho-species ( Maxillaria Ruiz & Pav.; Scaphyglottis Poepp. & Endl), and an ericaceous shrub ( Cavendishia Lindl.). Collectively, these morpho-species accounted for 78% (1127/1452) of the ramets surveyed in the initial 2005 survey ( Table 2). Elaphoglossum was abundant across the gradient, Maxillaria and Scaphyglottis were most abundant at upper elevations, and Cavendishia was most common at the central elevation ( Table 2).

All analyses were performed using the mat as the experimental unit to account for within-mat correlations between ramets, i.e. to avoid pseudo replication. We fitted models to data that included all ramets irrespective of morpho-species to explore the overall community patterns of ramet survival, recruitment, and population change, and then modeled common morpho-species separately to look at individual morpho-species responses. We tested whether ramet recruitment was different than mortality in transplanted mats using a two-sided t-test. We then analyzed ramet survival, recruitment, and population change with respect to experimental manipulations using generalized linear mixed-effects models (GLMMs). Survival was modeled as a binomial distribution with a logit link function to account for the binary nature of the response (alive, dead). Recruitment (new ramets in 2006) and population change (total ramets in 2006) were modeled as a rate relative to the initial ramets per mat by using a Poisson distribution with a log link, and adding an offset of the log of the number of initial ramets in 2005. To account for the natural blocking by tree in our experimental design, models that included multiple source elevations and transplant elevations also included random effects for source tree and transplant tree. Models including only one source elevation included a random effect for source tree only. Likelihood ratio tests were used to assess the fixed effects, while Wald z-tests were used to evaluate differences between levels of fixed effects. We did not evaluate the significance of random effects because they were a required part of our experimental design. Finally, we confirmed that the residuals of the final model were not overdispersed ⁵⁰ using code from Bolker et al. ⁵¹ . All analyses were done in R [Version 2.15.2; 52]. In all analyses we considered an effect significant if the P-value was less than 0.05.

First, we tested for an effect of manipulating mats using data for undisturbed control mats and mats transplanted within elevation. Source elevation and treatment (transplant versus undisturbed) were modeled as fixed effects in this analysis. Then, we tested for effects of source and transplant elevation on the response variable, using data from just the transplanted mats. We took this two-tiered approach because a full model including all mats was unbalanced (e.g. there could not be a control mat that moved between elevations) and statistical models accounting for this would not converge computationally. For analysis of individual morpho-species, we used only source elevations for which the morpho-species was present in at least half (10) of the source mats from that elevation (see Table 2).

To investigate patterns in mat species composition we used Detrended Correspondance Analysis (DCA) because our compositional data collected across a directional gradient matched the assumptions of DCA. First, we investigated the change in composition versus elevation using the pre-transplantation composition of all mats. We then investigated compositional change due to experimental treatments by ordinating the composition of all mats during 2005 before transplantation, with the composition of mats in 2006, one year after transplantation. Permutation Multivariate Analysis of Variance using distance matrices [function adonis in the vegan R package; 53] was used to test for compositional changes with altitude and among years due to the transplantation.

First, we tested for an effect of transplantation independent of elevational distance moved by asking whether epiphytes in mats transplanted to the same elevation had different ramet survival, recruitment, and population change or turn-over than those in intact mats. Across all morpho-species, there was no significant effect of transplant, elevation, or their interaction on survival, recruitment, or population change of epiphyte ramets ( Table 3, Figure 1). However, individual morpho-species were affected by transplantation. For Elaphoglossum, survival was lower in mats transplanted to another site at the same elevation than in undisturbed controls, but there was no effect of elevation on survival or any interaction between elevation and transplantation ( Table 4, Figure 1). There was an interaction between elevation and transplantation for recruitment and population change in Elaphoglossum, however ( Table 4); both were lower for transplanted mats at 1500 m and 1650 m, but higher for mats transplanted at 1800 m ( Figure 1). For Maxillaria, recruitment was lower in transplanted mats, but not affected by elevation, and neither survival nor population change was affected by either transplanting or elevation ( Table 4, Figure 1). For Cavendishia, recruitment and population change were lower for transplanted mats ( Figure 1), but only significantly so for recruitment; survival was unaffected by transplantation ( Table 4). Transplanting did not affect survival, recruitment, or population change in Scaphyglottis ( Table 4, Figure 1).

Across all morpho-species, there were no significant effects on survival of any of the treatments for mats transplanted across elevations ( Table 5 and Figure 2). For recruitment and population change, there was a significant interaction between source and transplant elevation ( Table 5), with both positively associated with elevation for mats transplanted from 1500 and 1800 m, but negatively associated with altitude for mats from 1650 m ( Figure 2). Overall for transplanted mats, more ramets died than were recruited (mean change number of ramets per mat between years = -1.38; two-sided t-test, P = 0.01).

Elaphoglossum ramets in mats originating at 1500 m had consistently and significantly higher survival than those originating at 1650 m or 1800 m, but there was no effect of transplant elevation on survival ( Table 6, Figure 2). For both recruitment and population change, however, there was a significant interaction between source elevation and transplant elevation, with both recruitment and population change declining in mats transplanted at lower elevations for mats originating at 1500 m and 1800 m, but for mats originating at 1650 m recruitment was greater and population change more positive for mats transplanted to 1500 m than for mats transplanted to higher elevation ( Table 6, Figure 2).

For Maxillaria, the only significant effect for transplanted mats was that for transplant elevation on recruitment ( Table 6); recruitment was low in all transplanted mats, but there was zero recruitment in mats transplanted to 1500 m ( Figure 2). There were no significant effects of source elevation or transplant elevation on survival or population change ( Table 6), but survival was lower and population change more negative for ramets transplanted to 1500 m ( Figure 2).

All three measures of performance were unaffected by transplant elevation in Cavendishia ( Table 6, Figure 2). For Scaphyglottis, survival, recruitment, and population change were all progressively lower in mats transplanted to lower elevations ( Figure 2), but the difference was significant only for survival ( Table 6).

Prior to transplanting mats, the epiphyte community composition showed significant differences across the elevational gradient, although relatively little of the variation could be explained by elevation ( Table 7); most of the compositional separation was between mats at 1500 m and the other two elevations ( Figure 3). Morpho-species richness increased with elevation (Poisson regression, Z = 2.446, P = 0.0144), while the number of ramets per mat declined (Poisson regression, Z = -2.281, P = 0.0225; Table 8). Comparison of pre- and post-treatment species compositions in mats revealed no directional shift in community composition due to transplantation ( Table 7). A few mats did show large changes ( Figure 3), likely because of large changes in abundance in Elaphoglossum, either through high ramet mortality or recruitment ( Figure 2).

While our results for epiphytes transplanted from the highest elevation are consistent with the hypothesized altitudinal gradient in moisture stress, this gradient had less of an effect on epiphyte performance than the one in Monteverde, Costa Rica ²⁹ . The relative importance of cloud immersion for the distribution of epiphytes in this system may account for the difference. A consistent cloud base is a significant feature of many tropical montane forests ^{2, 58} , and regular low cloud is assumed to maintain the diversity and abundance of cloud forest epiphytes, and control many of the unique structural and functional features of cloud forests ^{43, 44} . Indeed, we chose the elevations for this experiment because of a suite of changes in ecosystem structure and function that occur at these elevations, including a step-change in bryophyte and vascular epiphyte biomass ⁴⁰ , tree height, above ground biomass, and forest productivity declining ⁴² , and soil organic matter increasing ⁴¹ above 1500 m. Tree diversity also begins to decline above 1500 m in the study region ^{59, 60} mimicking the general pattern in the Andes ^{61, 62} . These clear changes in forest structure, diversity and productivity contrast with smoother changes in climate. Mean temperature, precipitation, and VPD, a measure of moisture stress on plants, all decrease linearly with elevation above 1000 m ⁴⁶ .

High rainfall in this part of the Andes may mean that epiphytes here are less dependent on cloud immersion to maintain their water balance than their counterparts in other cloud forests. Even in 2005 under drought conditions, total precipitation for the year was 3273 mm. In this pluvial system, cloud base may be less important in determining epiphyte distributions than in other systems. It is noteworthy that the Nadkarni and Solano ²⁹ experiment was carried out on the leeward Pacific slope of Monteverde, which is drier than the Caribbean slope ^{63, 64} . Mean annual precipitation on the Pacific slope is 2155 mm at 1480 m in the cloud forest ⁶⁵ , and declines at lower elevations ⁶⁴ , and there is a 5–6 month dry season where much of the hydrologic balance is maintained by cloud immersion ⁶⁵ . This steep moisture gradient between cloud forest and lower elevations probably leads to a greater dependence of epiphytes on cloud immersion. If this previous study had been carried out on the Caribbean slope, where precipitation is higher at lower elevation ⁶⁴ , the results may have been similar to our study. On leeward slopes rainfall compensation may occur, in which epiphyte survival is enhanced by high rainfall even when there is less frequent cloud immersion.

Even though high rainfall may maintain epiphytes under normal conditions in the eastern Andes, droughts do occur, and epiphytes may be adapted to infrequent drought, especially at the lower fringe of the cloud forest. Drought in the Amazon basin during 2005 ^{66, 67} resulted in lower precipitation in the cloud forest. Although microclimate data were not available at the experimental elevations during the study, rainfall at the Peruvian SENAMHI meterological station at Rocotal (13°06′41″S, 71°34′14″, approximately 7 km from the transplant site at 2010 m elevation) for May–August in 2005 was the lowest for any year measured (mean May–August precipitation for 2000–2008: 601 mm; 2005: 175 mm; Figure 4). There was no recorded rainfall in July 2005, the only month during the nine-year measurement period with no recorded precipitation (mean July precipitation: 112 mm). In addition, actual cloud water interception based on fog collectors in place during the experiment did not show a gradient of increasing moisture with elevations during the 2005 dry season (four week total weight of water collected: 1500 m, 1109 g m ^-2; 1750 m, 35 g m ^-2; 1900 m, 72 g m ^-2). All elevations were very dry, and desiccation was evident in bryophytes and non-succulent vascular epiphytes in the study area. However, ramet mortality in undisturbed control mats was not significantly greater than recruitment at any elevation ( Figure 1). In addition, mat species composition did not change directionally between years ( Figure 3). Thus, undisturbed epiphytes between 1500 and 1800 m in this Andean cloud forest appeared resistant to drought over the one-year time scale of our experiment.

This resistance to drought may be related to the normally variable and seasonal rainfall at the study site. Annual rainfall totals ranged between 3 and 6 m per year in a five year period not including the 2005 Amazonian drought. Precipitation was lowest in June and July ⁴⁶ , when temporary drought is possible, though for no month did potential evapotranspiration exceed precipitation at these elevations in most years. However, prolonged (days-to-weeks) periods of direct sun can induce drought stress, and epiphyte species that live in this part of the Andes may possess adaptations for surviving drought, similar to those in lowland dry or seasonal forests ^{36, 37, 68} . Epiphyte drought tolerance is higher in areas where drought occurs more frequently ^{19, 69} , and many epiphyte species have adaptations for surviving drought – crassulacean acid metabolism, desiccation tolerance, pseudobulbs, succulent leaves and other water-storing organs. Consistent with this idea, Elaphoglossum ramets transplanted from 1500 m had higher survival than those from higher elevations, regardless of the transplant elevation ( Figure 2). The stronger response of epiphyte mats transplanted from normally cloud immersed elevations (i.e. 1800 m) compared to those transplanted from lower elevations suggests that lower elevation populations may be better adapted to drought stress due to less frequent cloud immersion. Another example of locally adapted epiphytes was observed in subtropical China, where bryophytes transplanted downslope lost biomass, while in situ measurements showed no change in biomass across the gradient ³⁸ .