Forest patch symmetry depends on direction of limiting resource delivery 1,2,3,5, 2,4 2,4 1 DANIEL E. STANTON, BEATRIZ SALGADO NEGRET, JUAN J. ARMESTO, AND LARS O. HEDIN

1Department of Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey 08544 USA 2Departamento de Ecologı´a, Pontificia Universidad Cato´lica de Chile, Santiago, Chile 3Ecology, Evolution and Behavior Department, University of Minnesota, St. Paul, Minnesota 55108 USA 4Institute of Ecology and Biodiversity (IEB), Santiago, Chile

Citation: Stanton, D. E., B. Salgado Negret, J. J. Armesto, and L. O. Hedin. 2013. Forest patch symmetry depends on direction of limiting resource delivery. Ecosphere 4(5):65. http://dx.doi.org/10.1890/ES13-00064.1

Abstract. Edge effects are a major concern in the study and conservation of forest patches. The traditional perspective, derived from patches formed by fragmentation, considers forest edges as intermediates in a gradient between interior and exterior conditions, symmetrically distributed around the core of the patch. We present a more general conceptual model that shows that this perspective is only one of several possible environmental gradients across forest patches. When resources are delivered horizontally (e.g., fog, surface runoff ), environmental parameters and species composition are expected to have very different, asymmetric, distributions within forest patches. We conducted transect surveys characterizing environmental conditions (light, soil moisture, soil nutrients), vegetation structure and species composition in fog-fed patches of relict temperate forest in northern Chile. Windward edges differed most from the surrounding scrubland, whereas the core merely represented an intermediate between windward and leeward edges. Community composition changed drastically from temperate forest specialists on the windward edge to mediterranean shrub species leeward. The simple edge-core model is shown to be inadequate for describing spatial patterns in fog-influenced forests: a more universal model including the directionality of external resource inputs and internal dynamics must be considered when evaluating forest patch dynamics.

Key words: asymmetry; community composition; ecosystems; fog; forest patch; matorral; temperate rainforest; vegetation banding.

Received 25 February 2013; revised 23 April 2013; accepted 1 May 2013; published 29 May 2013. Corresponding Editor: M. Anand. Copyright: Ó 2013 Stanton et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. http://creativecommons.org/licenses/by/3.0/ 5 Present address: Division of Sciences, Research School of Biology, the Australian National University, Acton, ACT 0200, Australia. E-mail: [email protected]

INTRODUCTION been to evaluate how far into a forest these ‘edge effects’ penetrate. Within this context, forest Concerns over increasing forest fragmentation patches have often been portrayed as an edge have drawn attention to the particularities of (bearing the influence of the area outside of the forest patches. The edge of a forest will be patch) surrounding a core unaffected by the affected by the surrounding matrix outside the external matrix (Murcia 1995). Often implicit in forest, and thus differ considerably from the this understanding of forest patches is the interior. A consistent focus of the literature has assumption that small patches were formerly

v www.esajournals.org 1 May 2013 v Volume 4(5) v Article 65 STANTON ET AL. parts of a larger continuous forested area. situation is evidently implied. This is qualitative- This idealization is inherently radially sym- ly different from the reported differences be- metrical when viewed from above. The width of tween north- and south-facing edges of forests the edge may be variable but the nucleus is (Wales 1972, Matlack 1993, 1994, Chen et al. 1996, conceived as a core around which approximately Hylander 2005), which are attributed to effects of symmetrical sides extend (Fig. 1A). Intrinsic to insolation (i.e., response to microclimatic differ- this concept of the patch is an assumption that ences rather than growth towards light). The resources are delivered either vertically or diffuse difference between vertically and horizontally horizontally from all sides equally. Edge effects delivered resources is therefore best considered such as slanted light, wind and non-forest in terms of directionality relative to the axis of animals diffuse inwards towards the core from plant growth, either parallel (Fig. 1A) or perpen- all sides. Vertically delivered resources (light and dicular (Fig. 1C). Although light competition rain) are delivered approximately evenly to the may be the most commonly known form vertical entire top layer of the forest. Ecosystem proper- competition, rainwater and nitrogen can also be ties that are tied to these resources, such as soil subject to ‘vertical processing’ (Ewing et al. 2009). moisture, soil nutrients and plant height can Many resources are effectively co-limiting at therefore be expected to also show a radial the ecosystem scale. For example, if we consider symmetry around the core (Fig. 1B), as compe- banded forests in semi-arid environments, sur- tition for them will occur along the vertical axis. face run-off water (perpendicular) determines Although often the case in antropogenically the presence and scale of forest patches, but light modified landscapes, patchiness is not necessar- may structure vegetation within the patch ily formed by fragmentation (Rietkerk and Van (parallel). As such, while some ecosystem prop- De Koppel 2008). Vegetation patches can also erties (e.g., soil moisture) may be horizontally arise by self-organization through local facilita- asymmetrical, others may be horizontally sym- tion. For example, Klausmeier (1999) showed metrical (e.g., vegetation structure) (Fig. 1E). theoretically that forest patches can arise from Spatial patterns in plant communities composi- directional surface-runoff in semi-arid ecosys- tion, which are driven by co-limitation and trade- tems, and similar patterns have been empirically off between both parallel and perpendicular been demonstrated to occur in a wide range of resources, will reflect an intersection of these ecosystems, from fog-fed bromeliad fields in the bidirectional effects (Fig. 1F). Atacama (Borthagaray et al. 2010) through semi- The consideration of directionality of resource arid shrublands (Klausmeier 1999, Van De delivery additionally challenges the static view of Koppel and Rietkerk 2004, Saco et al. 2007, Ke´fi forest patches. In a vertically structured forest et al. 2010) to Sphagnum aggregations in fens patch the dynamics of light competition will lead (Eppinga et al. 2008, 2009, Manor and Shnerb to upward growth and traditional forest succes- 2008) and tree islands in the Everglades (Wetzel sion (Horn 1971). If critical resources are deliv- et al. 2008). In all of these cases limiting resources ered horizontally, we expect competition to occur such as water and nutrients are delivered in a along the horizontal axis, for example upslope horizontally asymmetric manner (Fig. 1C). This for water and nutrients from surface run-off strong directionality of resource delivery is likely (Saco et al. 2007), windward for resources from to lead to asymmetrical distribution of related fog (Borthagaray et al. 2010) or leeward when ecosystem properties (Fig. 1D). wind is a cause of mortality (Watt 1947, Sprugel We propose a single general conceptual model and Bormann 1981, Sato¯ and Iwasa 1993). Since that unites symmetrical and asymmetrical forest horizontal expansion is not constrained by the patch structures. These two models are not so biomechanical costs of overcoming gravity and different when we consider that light is also a retaining access to soil resources, it should directional limiting resource in forests. Although become apparent as a directional progression of light competition is considered a fundamental forest patches across a landscape, with consider- aspect of plant community structuring, it is rarely able differences in community composition and considered in the context of generating asymme- ecosystem processes between leading and lag- try along the axis of delivery, even if such a ging edges.

v www.esajournals.org 2 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

Fig. 1. Hypothetical resource distributions predicted according the directionality of resource input: (A, B) vertical inputs only (e.g., rainfall, parallel to the direction of plant growth); (C, D) horizontal only (e.g., fog or slope runoff, perpendicular to the direction of plant growth), and (E, F) both vertical and horizontal (e.g., fog and rainwater inputs, bidirectional). The principal axis of plant growth is illustrated by the dotted line. Soil based resources (e.g., %C, %N) are controlled by water availability, and thus indirectly controlled by water input direction (upward arrows in panels A, C and E).

v www.esajournals.org 3 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

To test this conceptual model of spatial Matrix; Fig. 1F). Furthermore, we predicted distribution of resources within forest patches, asymmetries to be stronger in the small patches, we measured a number of above- and below- in which horizontal competition is expected to be ground environmental variables across fog-de- stronger than in the larger patches. pendent forest patches in northern Chile. These forest patches contain temperate rainforest trees METHODS far outside of their main climatic range in the midst of Mediterranean semi-arid matorral Site description (Squeo et al. 2004). The fog-water inputs are Research was conducted in Fray Jorge Nation- 0 0 strongly directional, and lead to large differences al Park, IVth Regio´n, Chile (30840 S, 71830 W). A in tree recruitment and mortality between wind- large number (370) of small patches of forest ward and leeward patch edges (del Val et al. form a mosaic embedded in a xerophytic 2006). This directionality makes these patches an matorral scrubland (Squeo et al. 2004, del Val et ideal system in which to test whether ecosystem al. 2006, Gutie´rrez et al. 2008, Barbosa et al. 2010). properties are more strongly associated with The persistence of these forest patches, whose directionality of resource delivery or simply species composition closely resembles Valdivian symmetrically determined by distance from temperate rainforest (Villagra´n et al. 2004), forest edge. despite very low rainfall (147 mm annually) at The spacing and width of fog-created banding Fray Jorge has been attributed to fogwater inputs is also dependent on slope: steep slopes decrease (del Val et al. 2006). Forest patches span a wide the strength of horizontal competition for fog, range of sizes, from 0.1 to 36 ha (Barbosa et al. leading to broader bands or even continuous 2010) and in some areas form bands perpendic- plant cover, whereas flatter ground encourages ular to the predominant wind direction (Fig. 2). the formation of narrow, widely-spaced bands (Borthagaray et al. 2010). Fog forest relics occur Sampling design in areas of highly variable topography, and larger Transects were established perpendicular to patches tend to be associated with steeper slopes forest patch edges and parallel to the dominant (Barbosa et al. 2010), and so the effects of wind direction. The length of each transect directionality on resource distribution might be depended on the width of the forest patch expected to weaken with increasing slope and crossed, and was chosen to extend at least 3 patch size. sampling points beyond both lee- and windward Considering that the primary source of water border. The ‘borders’ of the patch were deter- is horizontally driven fog, we predicted below- mined as the first and last point along each ground ecosystem properties controlled by water transect at which at least one plant exceeded 3 m availability (soil moisture and nutrient availabil- in height. Ten transects were conducted crossing ity) to be horizontally asymmetrical (Fig. 1D), a total of 14 patches, with several transects whereas above ground properties (plant height, crossing more than one patch. The patches understory light availability, litter depth) to be sampled had been identified as representative driven by light competition, and therefore of the range of patch sizes by previous studies horizontally symmetrical (Fig. 1B). Plant com- (Barbosa et al. 2010), and can be roughly munity composition, which is expected to be categorized as small (width , 30 m, area , 1 driven by competition for light, water and ha), medium (30 m , width , 100 m, 1 ha , area nutrients, was predicted to reflect both vertical , 10 ha) and large (width . 100, area . 10 ha). and horizontal influences (Fig. 1F). We hypoth- Light environments, vegetation structure and esized that the fog influenced forest patches composition was assessed at 2-m intervals along would not show a symmetrical resource distri- each transect (4-m intervals in medium and large bution (Core . Windward Edge ¼ Leeward Edge patches). Measurements of photosynthetically . Matrix; Fig. 1B) but rather an entirely active radiation (PAR) were made using a asymmetrical (Windward Edge . Core . Lee- quantum sensor (LI-1905B; LI-COR, Lincoln, ward Edge . Matrix; Fig. 1D) or mixed Nebraska, USA) under uniformly cloudy condi- (Windward Edge ¼ Core . Leeward Edge . tions, and expressed as a percentage of the

v www.esajournals.org 4 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

Fig. 2. Small forest patches in Fray Jorge National Park, IVth Regio´n, Chile (308400 S, 718300 W), as seen from the leeward side. The temperate forest patches are easily distinguished from the surrounding arid matorral The asymmetry of the patches is also clearly visible, with leeward primarily dead and windward plants with full foliage. The arrow indicates the primary direction of fog entering from the nearby coast. Photo by D. Stanton. above-canopy PAR, which was simultaneously weight in the Biogeochemistry lab of the Pontif- recorded using a second quantum sensor that icia Universidad Cato´lica de Chile, Santiago, had been placed in a nearby forest clearing. PAR Chile. Subsamples (;3 g) were sieved through measurements along the transects were taken 1m 1-mm mesh and sent to the Hedin Lab, Princeton above the ground to represent the light environ- University, New Jersey, USA, for analysis. Sam- ment of small saplings. The area surrounding the ples were ground by mortar and pestle and oven- sampling point was divided into 4 equal quad- dried at 608C for 3 days prior to carbon and rats. The height and species of the canopy nitrogen analysis using an elemental analyzer overlying the sampling point, as well as the (Carlo Erba 4500, Costech, California, USA). height and identity of the nearest woody plant species in each quadrat were recorded. Data analysis Volumetric soil moisture at 2-m intervals was The points within each transect were parti- recorded in situ using a hand-held TDR probe tioned according to location within the transect (Fieldscout TDR 100, Spectrum Technologies, as one of four zones: core, leeward edge, Illinois, USA). Five measurements were recorded windward edge, matrix. The core of each patch for each sampling point, after clearing away leaf was defined as the region in which average plant litter and subaerial roots. The depth of leaf litter height at each sampling point .3 m. Edges were was recorded when present. defined as those points points within the patch Soil samples for soil nutrient content were (at least one plant .3 m tall) but not contained in collected at intervals of 2 m (small patches), 4 m the core. The matrix was considered to be all (medium patches) or 8 m (large patches). points within a transect in which no plants Approximately 20 g of soil were collected, exceeded .3 m in height, corresponding to homogenized and oven-dried at 608C to constant scrubland rather than forest.

v www.esajournals.org 5 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

Table 1. Effects of forest patch size and location with patch (zone) on soil moisture, plant height, leaf litter depth, understory light availability, soil carbon, soil nitrogen and woody plant community composition in linear mixed effects models (LME).

Dependent variable Fixed variable df AIC DAIC Likelihood ratio p Pattern Soil moisture Patch size 6 4419.1 6.3 10.304 0.0058 Zone 7 4427.1 4.8 42.569 ,0.0001 Mixed Patch size 3 Zone 15 4461.4 34.3 25.472 0.0013 Mixed-Sym Plant height Patch size 6 9163.8 2 2.051 0.3587 Zone 7 9107.1 54.7 60.724 ,0.0001 Sym Patch size 3 Zone 15 9115.6 8.5 7.516 0.4821 Leaf litter depth Patch size 6 673.1 1.6 2.360 0.3073 Zone 7 644.8 1.6 32.690 ,0.0001 Sym Patch size 3 Zone 15 643.9 26.7 18.929 0.0309 Sym Understory light Patch size 6 623.0 2.6 1.402 0.4962 Zone 7 602.2 18.3 24.205 ,0.0001 Mixed Patch size 3 Zone 15 595.1 7.1 23.202 0.0031 Mixed-Sym Soil carbon Patch size 6 1000.9 1.9 2.117 0.3470 7.18 13.110 Zone 7 991.9 7.18 13.110 0.0044 Mixed Soil nitrogen Patch size 6 140.6 1.9 2.161 0.3394 Zone 7 133.4 5.3 11.302 0.0102 Mixed C:N Patch size 6 663.6 3.3 0.687 0.7093 Zone 7 661.4 1.1 4.812 0.1861 Community composition Patch size 6 1194.9 2.1 1.911 0.3846 Zone 7 1246.7 49.7 55.796 ,0.0001 Mixed Patch size 3 Zone 15 1250.1 3.4 19.363 0.0130 Mixed-Sym Notes: In all models patch identity and location within patch were considered random effects. Single fixed effect models are compared to the null model, interaction models are compared to the relevant significant single fixed effect model. Data available was insufficient to fit full interaction models for soil carbon and soil nitrogen. Soil moisture, plant height, leaf litter and light availability data were square-root transformed for the analysis to conform to assumptions of normality. P values , 0.05 are shown in boldface. The patterns of resource distribution across zones for each significant model are classified as symmetrical (‘‘Sym’’, as in Fig. 1B), asymmetrical (‘‘Asym’’, Fig. 1D) or mixed (‘‘Mixed’’, Fig. 1F).

All statistical analyses were performed using Vegetation community composition was ana- the open-source statistical software program R (R lyzed using Principal Coordinates Analysis Development Core Team 2012). The distributions (PCO). We computed floristic similarity between of above and below-ground variables were locations using a Sorensen dissimilarity and evaluated by linear mixed effects models maxi- computed the two first axes of the PCO projection mising log-likelihood using the function lme in R using R package labdsv (Roberts 2012). The first package nlme (Pinheiro et al. 2013). Soil moisture, axis of the PCO provided a single variable plant height, leaf litter and light availability data descriptor of the community assemblage of each were square-root transformed for the analysis to transect point. Linear mixed models using the conform to assumptions of normality and heter- PCO 1st axis as the dependent variable, patch size oscedasticity. Patch identity and zone were used and/or zone as fixed effects and patch identity and as random effects and patch size, zone and the zone as random effects were created and tested as interaction of patch size and zone applied as above to determine the distribution of vegetation fixed effects. Likelihood ratio tests were used to across patch zones and patch sizes. Furthermore, determine the best model for each. Single fixed the proportional distribution of individual species effect models are compared to the null model, across patch zones was evaluated. interaction models are compared to the relevant significant single fixed effect model. Data avail- RESULTS able was insufficient to fit full interaction models for soil carbon and soil nitrogen. To determine Spatial patterns of abiotic variables the pattern underlying significant fixed effects All but one (C:N ratio) of the above and below- we conducted Tukey multiple comparisons ap- ground abiotic variables measured varied signif- plied to the fullest significant LME using R icantly with zone within the landscape (Table 1). package multcomp (Hothorn et al. 2008). Furthermore there were interactions between

v www.esajournals.org 6 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

Fig. 3. Boxplots of distributions of (A) plant height, (B) leaf litter depth, (C) total soil carbon, and (D) total soil nitrogen with location within patches (windward edge, core, leeward edge and surrounding matrix). Thick lines represent the median, boxes represent the interquartile range, whiskers represent maxima and minima within 1.5 times the interquartile range and open circles show outliers. Letters indicate significantly different groups (p , 0.05) as determined by Tukey HSD multiple comparisons applied to an LME model with zone as a fixed effect (see Methods and Table 2). zone and patch size for soil moisture, leaf litter Large patches showed the most symmetrical depth and understory light. Plant height was within-patch distributions (Fig. 4). Contrary to symmetrically distributed around the core (Fig. predictions soil C and N were were marginally 3A), which is partly driven by the height-based asymetrical (z ¼2.550, P , 0.05217 and z ¼ definition of the zonation. Leaf litter depth did 2.408, P , 0.07464, respectively) with the not differ within patches but was significantly greatest values found at the leeward edge and greater than in the surrounding matrix (z ¼5, P core (Fig. 3C, D). , 0.0001; Fig. 3B), the only interaction with patch size being the significance of the difference Plant community composition between the patch and the surroundings. No The PCO first axis was able to represent 38.5% variables were found to be completely asymmet- of the variance in plant community composition. rically distributed (Windward Edge . Core . The woody plant community showed a signifi- Leeward Edge . Matrix; Fig. 1D), however soil cant response to patch zone (Table 1). Although moisture, light, soil C and soil N all showed patches always differed from the surrounding mixed symmetry (Windward Edge ¼ Core . matrix, the within-patch distributions varied Leeward Edge . Matrix; Fig. 1F, Fig. 3C, D, Fig. with patch size, from asymmetrical in small 4A, B). Windward edges and cores were wettest patches to symmetrical around the core in and most shaded in the small patches, but not in medium and large patches (Fig. 4C). When the medium and large patch (Fig. 4A). Small individual species are considered the patterns patches showed asymmetrical distributions, with are even more clearly pronounced. Species with a the degree of symmetry decreasing with size. strong temperate wet forest affinities (Villagra´n

v www.esajournals.org 7 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

Fig. 4. Boxplots of distributions (A) of soil moisture, (B) understory light availability, and (C) woody plant community composition with patch size (small, medium, large) and location within patches (windward edge, core, leeward edge and surrounding matrix). Thick lines represent the median, boxes represent the interquartile range, whiskers represent maxima and minima within 1.5 times the interquartile range and open circles show outliers. Letters indicate significantly different groups (p , 0.05) as determined by Tukey HSD multiple comparisons applied to the LME model of the Patch Size 3 Patch Zone interaction (see Methods and Table 2). et al. 2004) were predominantly found inside tion either in symmetry or in the form of patches (Table 2). In small patches they showed asymmetry. asymmetrical preferences for the windward edge Differences in tree survival and foliage reten- and core with a reduced presence at the leeward tion are also a likely explanation for the striking edge. In large patches the distribution was more asymmetry in understory light availability (Sal- frequently symmetrical, centering around the gado Negret et al. 2013). Although we predicted core of the patch for the trees Aextoxicon that the vertical competition for light within punctatum, Drimys winteri and Raphithamnus patches would lead to a symmetrical distribution spinosus, but not for sclerophyllous trees Azara (Fig. 1A, B), understory light availability actually microphylla and correifolia and appears to show an inverse response to soil water woody vine Griselinia scandens. content (Fig. 4). We observed considerably denser living vegetation at windward than DISCUSSION leeward edges, and high drought-induced mor- Above-ground variables tality and leaf loss probably allow for far greater The fog-fed forest patches were poorly de- light penetration. The increased light penetration scribed by the traditional symmetrical model, would then create a positive feedback by and showed strong directionality in several increasing evaporation from the soil surface. ecosystem properties. Many environmental var- The reduced insolation on the windward edge iables showed horizontally asymmetrical distri- should also translate into lowered soil tempera- butions (Table 1). Although some of these tures and reduced evaporation rates from the distributions matched those predicted by our soil, which may translate into reduced drought conceptual model, others differed from predic- and greater canopy density.

v www.esajournals.org 8 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

Table 2. Species and distribution of woody vascular plants (and the comparably sized bromeliad Puya) large found in forest patch transects.

Small patches (%) Large patches (%) Species Family WE C LE M WE C LE M Aextoxicon punctatum Aextoxicaceae 32 49 3 15 6 76 17 1 Ageratina glechonophylla Asteraceae 50 0 17 33 11 0 0 89 Azara microphylla Salicaceae 0 89 0 11 36 19 17 28 Baccharis linearis Asteraceae 0 0 0 100 0 0 0 0 Baccharis vernalis Asteraceae 5 3 5 87 0 0 9 91 Bahia ambrosoides Asteraceae 0 0 0 100 0 0 0 0 Berberis actinacantha Berberidaceae 20 0 0 80 0 0 0 100 Calceolaria integrifolia Calceolariaceae 0 0 0 0 0 0 0 100 Colletia spinosa Rhamnaceae 0 0 0 100 0 0 0 100 Colliguaja odorifera Euphorbiaceae 0 0 0 0 100 0 0 0 Drimys winteri Winteraceae 0 0 0 0 5 92 3 0 Echinopsis chilensis Cactaceae 0 0 0 100 0 0 0 0 Erigeron luxurians Asteraceae 0 0 2 98 0 0 0 100 Eupatorium salvia Asteraceae 0 0 8 92 6 4 2 88 Fuchsia lysioides Onagraceae 0 0 0 100 0 0 0 0 Griselinia scandens Griselinaceae 26 32 16 26 10 28 47 16 Haplopappus foliosus Asteraceae 0 0 0 100 0 0 0 0 Kageneckia oblonga Rosaceae 0 0 0 100 0 0 0 100 Myrceugenia correifolia 22 37 18 22 9 31 38 23 Puya chilensis Bromeliaceae 0 0 0 100 0 0 0 0 Raphithamnus spinosus Verbenaceae 8 58 8 25 8 69 18 5 Ribes punctatum Grossulariaceae 0 7 0 93 0 0 29 71 Senecio planiflorus Asteraceae 9 0 0 91 0 0 0 100 Notes: Organized by distribution with transects (WE, windward edge; C, core; LE, leeward edge; M, matrix) in small and medium-large patches. Species highlighted in bold were identified by Villagra´n et al. (2004) to exibit a disjunct distribution between Fray Jorge and wet temperate forest.

Below-ground variables Fundamental differences between patch types Soil characteristics were distinctly asymmetrical Small and large patches differed considerably along a windward to leeward axis, as predicted, in their spatial structure, both above- and below- however the details of the distributions differed ground. Barbosa et al. (2010) characterized the markedly from our hypotheses. Soil moisture, microclimatic and structural characteristics of which is strongly influenced by fog water inputs, forest patches (including a subset of those was expected to be greatest at the windward edge sampled in the present study) representative of and decrease across the patch due to the different sizes. One of the traits reported but not progressive ‘filtering’ out of fog-droplets from commented on is that small patches occur on flat the air by trees, as described in simpler fog- ground whereas most medium and large patches influenced banded vegetation (Borthagaray et al. are found on steep slopes (30–458). Windflow 2010). However, although soil water content was over flat areas will be strongly affected by the high at the windward edge, it was comparable or boundary layer created by a forest edge, and greater in the patch core (Fig. 4A). Trees were forest patches will leave a long wakes in which significantly taller in the patch core, which may little to no fog water is available, until airflow allow them to access fog water unavailable to (and fogwater) are replenished downstream (Oke shorter trees, thereby partially escaping the 1987). These ‘fog-shadows’ (del Val et al. 2006) interference effects of upwind competitors. Soil are likely to be far less pronounced or potentially carbon and nitrogen increased greatly from the absent on steep slopes (Borthagaray et al. 2010), windward edge to the core, before decreasing reducing or eliminating the competition for fog- again more gradually leewards. This suggests that water between trees. This difference in topogra- the availability of soil nutrients is not a simple phy may explain reduced asymmetry in large function of moisture and litter inputs, and may and medium patches (Fig. 4). In the large and instead be indicative of more complex ecosystem medium patches topography may still play some dynamics, as discussed below. role: the leeward edge is always associated with

v www.esajournals.org 9 May 2013 v Volume 4(5) v Article 65 STANTON ET AL. the flattening out of the terrain at a ridge crest, inputs (light and rainfall, Fig. 1A, B) and hori- whereas the small patches are topographically zontal inputs (fog water and nutrients, Fig. homogeneous and flat throughout. 1C, D). Principal components ordination clearly It is also important to clarify that several of the distinguished between forest and matorral plant small patches sampled in this study (but not in communities (Fig. 4C). Contrary to our predic- Barbosa et al. 2010) do occur on steep slopes. tions, understory light availability was strongly However, they are located such that the slope horizontally asymmetrical in all patches, and does not interact with wind direction (see Fig. 2), itself possibly driven by positive feedbacks with and therefore there is little to no sloping along asymmetric soil water availability (Fig. 4A, B). As the actual windpath. This observation supports such, plant community composition was also our interpretation the asymmetries are due to strikingly asymmetrical across patches, especially directional fog inputs rather than by the differing in small patches. Larger patches were symmetri- solar radiation that can be created by sloping cal in nature, with some more arid adapted terrain (e.g., Tian et al. 2001, Allen et al. 2006). shrubs (Myceugenia correifolia, Kageneckia oblonga- Variations on incoming solar radiation associated ta) present at both edges (Table 2). The differen- to slope steepness and orientation may favor tial microclimatic conditions across these patches moist retention and most likely plays an impor- may also lead to ecophysiological differences tant role in ecosystem dynamics, however, in the between individuals in those species that span present study it is unlikely to be the primary the patches (Salgado Negret et al. 2013). factor in the formation of patch asymmetries. If water availability is indeed a primary driver Forest patches as self-organizing ecosystems of spatial distributions of other ecosystem prop- The spatial asymmetries of soil carbon and erties, then it is perhaps unsurprising that small nitrogen content (Fig. 3C, D), while differing patches, in which competition for fog water will from those predicted, are in line with a dynamic be strongly asymmetrical, show far more marked view of forest patches. Del Val et al. (2006) have differences between windward edge and core proposed, based on the strong asymmetry in than do the larger patches, in which windward recruitment and mortality between edges, that and core trees likely have access to comparable forest patches at Fray Jorge may be progressively water inputs. This fundamental biophysical moving windward across the landscape. Under difference leads to a reinterpretation of Barbosa such a scenario, windward edges would be the et al.’s (2010) finding that small patch microcli- youngest, and considering that matorral soils are mates were more strongly impacted by edge very carbon and nitrogen-poor, the greater effects. Flat-ground (small) patches will have carbon and nitrogen content in core and leeward greater depletion of fogwater by the windward soils (Fig. 3C, D) may in fact reflect the greater edge, such that the patch interior and leeward accumulation of organic matter and nutrients. edges will be dryer than in larger patches. This The transition from matorral to forest soils and effect will amplify the edge effect (in the usual communities across very small spatial scales (at sense of the term) of the mostly dead leeward times ,5 m) may therefore reflect the build-up of edge allowing for increased insolation of the water and nutrient cycling facilitated by fog patch interior. collection. Such a self-organization of forest patches will leave a trail of modified above- Plant community as an integration and below-ground ecosystem attributes in its of co-limiting factors leeward wake. The presence of such a ‘foot-print’ communities are often struc- of forests past can indeed be identified, and will tured by competition for numerous potentially be the subject of a forthcoming paper (Stanton et limiting resources, such as light, water and al., unpublished manuscript). nutrients. Having predicted that these different Several forest associated species, such as resources would have different spatial distribu- Aextoxicon punctatum, Griselinia scandens and tions driven by directionality of resource deliv- Myrceugenia correifolia were occasionally found ery, we hypothesized that plant communities outside of the forest patches (Table 2). These would reflect overlapping effects of vertical individuals, when not just windward of the

v www.esajournals.org 10 May 2013 v Volume 4(5) v Article 65 STANTON ET AL. forest edge, formed small ‘mini-patches’ that may extend special thanks to Patricio Valenzuela, Marı´a be incipient forest patches. The long-term persis- Fernanda Perez and the CONAF staff at Fray Jorge for tence of windward migrating forest patches support in the field, Aurora Gaxiola, Pablo Marquet, requires the regeneration of patches downwind. Adam Wolf, Carla Staver and members of the Armesto and Perez labs for their support and discussion of The mechanisms for formation of these patches ideas as well as Madhur Anand and two anonymous are unknown, and may be associated with reviewers for suggestions that have greatly improved exceptional weather events, such as large El the manuscript. Nin˜o-Southern Oscillation (ENSO) events, as is the case for tree recruitment in other semi-arid LITERATURE CITED locations (Holmgren et al. 2006). It is well understood that species will assort Allen, R. G., R. Trezza, and M. Tasumi. 2006. along environmental gradients such as those Analytical integrated functions for daily solar found across forest patches. However, general radiation on slopes. Agricultural and Forest Mete- models of how these gradients themselves form orology 139:55–73. are more often overlooked or implicitly assumed. Barbosa, O., P. A. Marquet, L. D. Bacigalupe, D. A. Christie, E. Del-Val, A. G. Gutierrez, C. G. Jones, We have shown that the direction of delivery of K. C. Weathers, and J. J. Armesto. 2010. Interactions limiting resources drives the spatial asymmetries among patch area, forest structure and water fluxes in forest structure. Symmetrical forest patches in a fog-inundated forest ecosystem in semi-arid consisting of a core and periphery are but a Chile. Functional Ecology 24:909–917. special (albeit widespread) case of forest patch Borthagaray, A. I., M. A. Fuentes, and P. A. Marquet. structure, in which the primary directional 2010. Vegetation pattern formation in a fog- limiting resources (water and light) are delivered dependent ecosystem. Journal of Theoretical Biol- vertically. Not all natural ecosystems incorporat- ogy 265:18–26. Chen, J., J. Franklin, and J. Lowe. 1996. Comparison of ing a mosaic of small forest patches may show abiotic and structurally defined patch patterns in a the same directionality. For example, Silva and hypothetical forest landscape. Conservation Biolo- Anand (2011) studied Araucaria forest patches gy 10:854–862. that exhibited asymmetries, but without the del Val, E., J. Armesto, O. Barbosa, D. Christie, A. strong directionality that we have documented Gutie´rrez, C. Jones, P. Marquet, and K. Weathers. in Fray Jorge. In such cases feedbacks from the 2006. Rain forest islands in the Chilean semiarid surrounding matrix (e.g., fire, competition with region: fog-dependency, ecosystem persistence and shrubs or grasses, different water and nutrient tree regeneration. Ecosystems 9:598–608. availability, soil microbial communities) may act Eppinga, M., M. Rietkerk, W. Borren, E. Lapshina, W. Bleuten, and M. Wassen. 2008. Regular surface to stabilize patches. In yet other ecosystems the patterning of peatlands: confronting theory with driver of asymmetries may be seed rain, nutrient field data. Ecosystems 11:520–536. deposition (Weathers 1999, Ewing et al. 2009), Eppinga, M. B., P. C. De Ruiter, M. J. Wassen, and M. runoff (Klausmeier 1999, Van De Koppel and Rietkerk. 2009. Nutrients and hydrology indicate Rietkerk 2004, Saco et al. 2007, Ke´fi et al. 2010) or the driving mechanisms of peatland surface pat- frost damage (Watt 1947, Sprugel and Bormann terning. American Naturalist 173:803–818. 1981, Sato¯ and Iwasa 1993). The conceptual Ewing, H. A., K. C. Weathers, P. H. Templer, T. E. framework and empirical confirmation presented Dawson, M. K. Firestone, A. M. Elliott, and V. K. S. Boukili. 2009. Fog water and ecosystem function: here are a step towards a more inclusive heterogeneity in a California redwood forest. understanding of forest patches and their inter- Ecosystems 12:417–433. nal and external dynamics. Gutie´rrez, A., O. Barbosa, D. Christie, E. Del-Val, H. Ewing, C. Jones, P. Marquet, K. Weathers, and J. ACKNOWLEDGMENTS Armesto. 2008. Regeneration patterns and persis- tence of the fog-dependent Fray Jorge forest in This research was funded by NSF DDIG award semiarid Chile during the past two centuries. # 0909984 to L. H. and D. S.; Princeton Latin American Global Change Biology 14:161–176. Studies Travel Grants and a Princeton President’s Holmgren, M., B. Lo´pez, J. Gutie´rrez, and F. Squeo. Award to D. S. and CONICYT fellowship 24110074 2006. Herbivory and plant growth rate determine to B.S-N. Research in Chile was conducted under the success of El Nin˜o Southern Oscillation-driven CONAF research permit 06/08. We would like to tree establishment in semiarid South America.

v www.esajournals.org 11 May 2013 v Volume 4(5) v Article 65 STANTON ET AL.

Global Change Biology 12:2263–2271. Salgado Negret, B., F. Pe´rez, L. Markesteijn, M. J. Horn, H. S. 1971. The adaptive geometry of trees. Castillo, and J. J. Armesto. 2013. Diverging Princeton University Press, Princeton, New Jersey, drought-tolerance strategies explain tree species USA. distribution along a fog-dependent moisture gra- Hothorn, T., F. Bretz, and P. Westfall. 2008. Simulta- dient in a temperate rain forest. Oecologia doi: 10. neous inference in general parametric models. 1007/s00442-013-2650-7 Biometrical Journal 50:346–363. Sato¯, K., and Y. Iwasa. 1993. Modeling of wave Hylander, K. 2005. Aspect modifies the magnitude of regeneration in subalpine Abies forests: population edge effects on bryophyte growth in boreal forests. dynamics with spatial structure. Ecology 74:1538– Journal of Applied Ecology 42:518–525. 1550. Ke´fi, S., M. B. Eppinga, P. C. Ruiter, and M. Rietkerk. Silva, L. C. R., and M. Anand. 2011. Mechanisms of 2010. Bistability and regular spatial patterns in arid Araucaria (Atlantic) forest expansion into southern ecosystems. Theoretical Ecology 3:257–269. Brazilian grasslands. Ecosystems 14:1354–1371. Klausmeier, C. 1999. Regular and irregular patterns in Sprugel, D., and F. Bormann. 1981. Natural distur- semiarid vegetation. Science 284:1826–1828. bance and the steady state in high-altitude balsam Manor, A., and N. Shnerb. 2008. Facilitation, compe- fir forests. Science 211:390–393. tition, and vegetation patchiness: From scale free Squeo, F., A. Gutie´rrez, and I. Herna´ndez, editors. distribution to patterns. Journal of Theoretical 2004. Historia Natural del Parque Nacional Bosque Biology 253:838–842. Fray Jorge. Ediciones Universidad de La Serena, Matlack, G. R. 1993. Microenvironment variation Chile. within and among forest edge sites in the eastern Tian, Y. Q., J. Davies-Colley, P. Gong, and B. W. United-States. Biological Conservation 66:185–194. Thorrold. 2001. Estimating solar radiation on Matlack, G. R. 1994. Vegetation dynamics of the forest slopes of arbitrary aspect. Agricultural and Forest edge: trends in space and successional time. Meteorology 109:67–74. Journal of Ecology 82:113–123. Van De Koppel, J., and M. Rietkerk. 2004. Spatial Murcia, C. 1995. Edge effects in fragmented forests: interactions and resilience in arid ecosystems. implications for conservation. Trends in Ecology American Naturalist 163:113–121. and Evolution 10:58–62. Oke, T. R. 1987. Boundary layer climates. Routledge, Villagra´n, C., J. Armesto, F. Hinojosa, J. Cuvertino, F. L. Oxon, UK. Perez, and C. Medina. 2004. Historia Natural del Pinheiro, J., D. Bates, S. DebRoy, and D. Sarkar, and R Parque Nacional Bosque Fray Jorge, Chapter El Core Team. 2013. nlme: Linear and nonlinear enigma´tico origen del bosque relicto de Fray Jorge. mixed effects models. http://CRAN.R-project.org/ Ediciones Universidad de La Serena, La Serena, package=nlme Chile. R Development Core Team. 2012. R: A language and Wales, B. A. 1972. Vegetation analysis of north and environment for statistical computing. R Founda- south edges in a mature oak-hickory forest. tion for Statistical Computing, Vienna, Austria. Ecological Monographs 42:451–471. Rietkerk, M., and J. Van De Koppel. 2008. Regular Watt, A. 1947. Pattern and process in the plant pattern formation in real ecosystems. Trends in community. Journal of Ecology 35:1–22. Ecology and Evolution 23:169–175. Weathers, K. 1999. The importance of cloud and fog in Roberts, D. W. 2012. labdsv: Ordination and multivar- the maintenance of ecosystems. Trends in Ecology iate analysis for ecology. http://CRAN.R-project. and Evolution 14:214–215. org/package¼labdsv Wetzel, P. R., A. G. Valk, S. Newman, C. A. Coronado, Saco, P., G. Willgoose, and G. Hancock. 2007. Eco- T. G. Troxler-Gann, D. L. Childers, W. H. Orem, geomorphology of banded vegetation patterns in and F. H. Sklar. 2008. Heterogeneity of phosphorus arid and semi-arid regions. Hydrology and Earth distribution in a patterned landscape, the Florida System Sciences 11:1717–1730. Everglades. Plant Ecology 200:83–90.

v www.esajournals.org 12 May 2013 v Volume 4(5) v Article 65