Printer Friendly

Spatial patterns and edge effects on soil organic matter and nutrients in a forest fragment of southern Brazil.


Forest edges are an increasingly important feature of terrestrial landscapes (Wade et al. 2003). They occur naturally in forest-grassland transitions due to climatic or soil limitations to tree growth. However, most of the scientific interest in forest edges is due to increasing forest fragmentation, mostly for human-induced reasons (Wade et al. 2003). Yet, the long-term effects of such landscape changes are not fully known. In Brazil, more attention has been given to the recent forest fragmentation due to deforestation and logging in the Amazon (Laurance et al. 2000; Vasconcelos and Luizao 2004; Pellegrini et al. 2014). At the same time, the Atlantic forest, an ecosystem with large rates of endemism, has been shown to have more than 20% of its area in small fragments and almost 50% located within less than 100 m of a forest edge (Ribeiro et al. 2009).

Edge effects may be defined as changes in the ecosystem that occur at the boundaries of two habitats (Levin et al. 2009). In forest ecology these effects are commonly related to forests adjacent to human-made grasslands or agriculture, but may also extend to a managed or unmanaged forest matrix (Baker et al. 2016). The main effect on these edges is due to a microclimatic gradient, with moisture and temperature variation perpendicular to the edge (Berges et al. 2013). This causes differences in the production of litter and its nutrient composition (Vasconcelos and Luizao 2004) and, consequently, in its decomposition and soil fauna (Riutta et al. 2012). Forest edges may also be prone to greater rates of nutrient and pollutant deposition via throughfall (Wuyts et al. 2008). Furthermore, these changes alter tree forest composition (Harper et al. 2005) and fauna (Vetter et al. 2013).

Soil fertility is an important driver of forest productivity and the aboveground forest biomass stock (Fernandez-Martinez et al. 2014). Therefore, the nutrient pool variation from forest edge to forest interior might have large consequences on forest carbon sequestration capability in a fragmented landscape (Laurance et al. 2000). Studies have generally focused on soil organic matter or other biological indicators of soil fertility (Berges et al. 2013), but very few have assessed the availability of a wide array of mineral nutrients (Johnson and Wedin 1997), even though phosphorus (P) (Cleveland et al. 2011) or potassium (K.) (Wright et al. 2011) might be the most limiting nutrients for forest growth and accumulation of aboveground forest biomass.

Edge effects are subject to the influences of a large number of variables which are dynamic in space and time (Ewers and Didham 2006a). The changes in a forest edge must be viewed and analysed as a continuous gradient from the matrix of surrounding habitat to the forest interior instead of as an abrupt change (Ewers and Didham 2006b). The depth of edge influence (Harper et al. 2005) can be viewed as the distance from the edge that is influenced by the edge effects which, in turn, causes significant alterations in the ecosystem. Theoretical models expect edge effects to produce three types of responses: neutral, positive or negative (Ries et al. 2004). Positive and negative responses arc expected to exhibit sigmoid-shaped trends for a given variable in analysis of the depth of edge influence, but other shapes of trends may also be observed (Ewers and Didham 2006b).

Edge effects are a spatial phenomenon per se (Ewers and Didham 2006a); however, there are other underlying spatial effects on edge ecology. Tree species composition is also subject to spatial variation and complicates unbiased quantification of edge effects on the plant community (Ewers and Didham 2006a). Soil nutrients are also subject to spatial variation (Yuan et al. 2013) and may present a large variation in short distances due to diverging pedogenesis processes (Laliberte et al. 2013). Failing to account for this underlying correlation in soil processes may lead to biased estimation (Isaak et al. 2014) of edge effects on soil variables. There are several methods to detect and model spatial processes (Dormann et al. 2007). Generalised Additive Model (GAM) mean fitted surfaces are a powerful exploratory and predicting tool that accounts for trends in space and can produce results as good as the more traditional methods (Drexler and Ainsworth 2013).

The aims of this study were to quantify the change in availability of topsoil nutrients in a continuous gradient from grassland to forest interior and to identify the depth of edge influence in both habitats. We also aimed to evaluate the underlying spatial influence on these soil characteristics due to topography and pedogenesis, and to determine if the edge effect or the spatial variation on soil properties were dominant in determining availability of nutrients in a forest fragment of the Atlantic forest in southern Brazil.

Materials and methods

Site description

The study was carried out in Santa Maria county, Rio Grande do Sul State, Brazil. The climate is classified as Cfa (humid subtropical) (Alvares et al. 2013) with a mean annual temperature of around 19[degrees]C, mean temperatures exceeding 30[degrees]C in summer and some occurrence of temperatures below freezing during winter. Mean annual rainfall is 1600 mm with no dry season, although there are water deficits for as long as 7 months during summer and common random droughts (Heldwein et al. 2009). The region is included in the Parana sedimentary basin composed of sedimentary and igneous rocks and was formed through fluvial erosions that generated rolling hills and slopes (Sartori 2009). Soils are very heterogeneous with dominance of Ultisols on top of the hills and Alfisols and Entisols near water streams (Dalmolin and Pedron 2009).

The forest fragment under study is approximately 105- 150 m above sea level, with a clear Ultisol-Entisol (USDA 1999) east-west gradient. Ultisols (Kandiudults) are formed when precipitation exceeds potential evapotranspiration and water storage capacity during some periods of the year, occurring generally under forest cover in tropical areas. Ultisols commonly feature an E (eluvial) horizon above an illuvial horizon with clay precipitation and have relative low base status. Entisols (Psammaquents) are hydro-morphic soils formed in permanent or periodic waterlogged environments. This soils feature no abrupt horizon transition, with characteristic gleyic colour pattern in the lower horizons. Due to anaerobic conditions the soil ferric oxide is reduce to ferrous oxide, diminishing the soil cation exchange capacity.

The forested area is around 13.3 ha surrounded by grassland, with a fragment-shape index (Cochrane and Laurance 2002) of 3.8 in a scale for which an index equal to 1 indicates perfect circular fragments and 8 or higher indicates very irregular fragments. No visual changes in shape or size of the fragment could be observed in the Landsat satellite imagery from 1975 and no logging has been reported since the 1960s, therefore, forest structure and edges might be considered to be stabilised. The vegetation is classified as a subtropical moist deciduous forest with the canopy dominated by Luehea divaricata Mart. (Malvaceae) and Patagonula americana L. (Boraginaceae). Both forest and grassland are subject to light grazing by cattle.

Data collection

A regular 80 m x 80 m sample grid compromising 30 points was chosen to provide full coverage of the study area. However, this approach might limit estimation of close range variation in soil properties (Diggle and Ribeiro 2007). To address this problem we added 16 infill sampling points, resulting in a 46-point sampling grid (Fig. 1a). The distances from the forest edge were calculated based on planar coordinates of points and a forest polygon observed from Google Earth data. Sampling points outside the forest were given negative values to the estimated distances from the edge in meters. The sampling intensity of the resulting distances from edge had a clear unimodal shape (Fig. 1 b). This sampling scheme provides better estimates for soil parameter changes at a position near the forest edge, where the most marked differences should occur.

During the winter of 2015, the sampling points were located in the field using a GPS. At each sample location, the litter layer was removed and three subsamples (0.5-m sample point radius) to a depth of 20 cm using a 10-cm diameter cylinder were obtained. The subsamples were mixed in sterile plastic bags to ensure the sample was representative. All remaining twigs or roots in the sample were carefully removed. Nutrient availability was determined using a standard protocol for soil analysis in southern Brazil. Samples were dried, P (mg/L) was determined by the colourimetric method and K. (mg/L) by photometry, both using Mehlich I extractant solution. Calcium (Ca, [cmol.sub.c]/L) and magnesium (Mg, [cmol.sub.c]/L) were determined by spectrophotometric method using potassium chloride extractant. Soil pH was determined by a pH meter in 1: 1 water solution. Soil organic matter (SOM) (%) was measured by humid combustion colourimetry using a strong oxidising agent (sodium dichromate) in the presence of sulfuric acid; soil nitrogen (N) availability may be indirectly obtained from the SOM proportion.

Data analysis

GAMs are extensions of Generalised Linear Models in which the total variance is explained through a link function and a sum of parametric covariates or local smoothers (Hastie and Tibshirani 1986). The main reason that GAMs were used was their flexibility in identifying non-linear trends that are common in edge effects and ecology in general (Ries et al. 2004; Ciannelli et al. 2008). Spatial effects modelling was accomplished by fixed-effects two-dimensional thin plate regression splines (Wood 2003). Tensor products were set to explain the edge distance effect on soil variables as they produce relatively low rank and more interpretable smooths (Wood 2006a). The number of degrees of freedom for each smooth was estimated by generalised cross-validation --a leave-one-out method to estimate predictive square error (Wood 20066). To guarantee that the model was identifiable, each function estimate was constrained to a zero average (centred variables) over the entire dataset (Ciannelli et al. 2008). GAM assumes only additivity among covariate events, although this is a limitation in ecological studies (Ciannelli et al. 2008) where covariates are never really orthogonal. However, we preferred this limitation to dividing the variables into classes and analysing the class interactions. All estimates were obtained using the mgcv package (Wood 20066) of the R statistical environment (R Core Team 2016).

We tested three models assuming normal distribution for each of the soil variables. First we fitted a full model in the following way (Eqn 1):

Soil variable = c + [f.sub.1] (Latitude, Longitude) + [f.sub.2](Distance from edge) + [epsilon] (1)

where c is the constant representing the mean of the dependent variable, [f.sub.1] is a thin plate regression spline, [f.sub.2] is a tensor product and [epsilon] is a random error term. In order to test the influence of both spatial variables in the model, we fitted individual models exclusively with the spatial mean trend surface (Eqn 2) and the edge effect distance (Eqn 3):

Soil variable = c + [f.sub.1] (Latitude, Longitude) + [epsilon] (2)

Soil variable = c + [f.sub.1] (Distance from edge) + [epsilon] (3)

The models were compared through plots of the estimated relationships, and by comparing their shapes with each other and the possible theoretical shapes (Ries et al. 2004; Ewers and Didham 2006b). We also compared the models' overall fit using Akaike Information Criteria (AIC), where the model with the lowest AIC is best supported by the data. We used a likelihood ratio test to determine whether the inclusion or exclusion of one of the variables provided a model with better fit (Bolker et al. 2009; Brewer et al. 2016). However, as the aim was to compare how models performed in the presence or absence of the underlying spatial trend and edge effects, we did not use these statistics as a means to build the 'best model' and they served only as tools to evaluate the ecological process (Johnson and Omland 2004). These tools were used along with visual analysis of plots of the fitted trends and surfaces as means to evaluate model fit and overall realism.


The east-west changes in availability of nutrients were clear in the mean fitted surface models (Fig. 2), as expected due to the Ultisol-Entisol gradual transition, except for P, which showed a concentric mean fitted surface in the spatial trend-only model. This indicated no realism in this last model, as P availability was greatly affected by edge distance; therefore this model should be rejected as an explanation for spatial P variation. Lower pH, SOM and cation (K, Ca and Mg) values were observed in soil with higher water influence (Entisol), although there was a higher availability of P in such sites according to the full model. There were considerable changes in the modelled spatial mean surfaces comparing full models with the spatial trend-only models. The spatial patterns for SOM, pH and P changed drastically with the inclusion of the edge distance variable (full model) but estimated trend surfaces for cations were little affected by this variable.

There was a clear edge effect on soil nutrient availability (Fig. 3) as sites tend to be increasingly more fertile under the forest than under grassland. With increased distance from the forest edge into the forest, the higher were the observed nutrient availability and pH. The shape of edge distance and nutrient availability relationships were unchanged for P (sigmoid) and Mg (linear) between the full and edge effects-only models. However, for SOM, K. and Ca there was a linear relationship in the edge effects-only model and a sigmoid curve in the full model. The clearest change in the edge distance relationship was for soil pH, where the edge effects-only model predicted a negative quadratic curve, which is not biologically plausible. This relationship of pH with edge distance was linear in the full model. The sigmoidal relationships in the data indicate that the depth of edge influence was ~50 m, and that there were effects on the availability of nutrients in the grassland soil near the forest fragment.

The best model fits using AIC were found for each of the three tested models (Table 1), but when the best model was obtained by the mean estimated surface or by the edge effects models, they were never significantly better than the full model. This indicates that underlying spatial trends were more important in determining availability of nutrients than vegetation type. Furthermore, this shows the importance of evaluating ecological processes not only through the 'best model' fit, which might generate misleading conclusions due to overfitting. We believe that the shapes of the relationships in space were better represented in the full models than the other options, and that the higher A1C values observed in some instances were products of the higher number of model parameters, which AIC does penalise for in such instances.


The 'best model' approach might give misleading information about data in ecology and other sciences (Bates et al. 2015), even if it represents a development in comparison with pure hypothesis testing (Johnson and Omland 2004). Although a parsimonious estimation is important to yield an easily interpretable inference, selections among models based solely on statistics have disadvantages (Bolker et al. 2009). These were clear in our analysis that included two spatial effects--when one was ignored in building a model the results were drastically changed. Therefore, full models, including underlying spatial trends, in transversal studies should be preferred over simpler models (Whittingham et al. 2006). In quantifying the continuous edge effects, we found that the theoretical shapes of the relationships, with overall realism, were mostly uncovered when the spatial variation in soil heterogeneity was taken into account.

The east-west spatial pattern in the study area is related to the local pedogenesis as there is a topographic trend in this direction. As expected soil P had higher availability in lower areas of the region under study, due to the high mobility of this nutrient in soil. However, the other nutrients had lower availability in these lower areas, probably due to the periodic waterlogged environment in which the soil has lower cation exchange capacity, causing cations to be easily leached. Nutrient variation is a result of the movement of water and soil on hillslopes (Yoo et al. 2009), but the shape of the hills might influence the patterns of nutrient availability (Chadwick and Asner 2016). Our results showed opposite trends for P, K, Ca and Mg to those observed in areas with

more marked erosion patterns (Chadwick and Asner 2016), where plateaus and then 'ravines' are formed by water flow. Our SOM patterns resembled those found in more gentle slopes (Yoo et al. 2009). This topographic control of availability of nutrients was not the aim of this study, and is accounted for as an underlying spatial effect. Furthermore, our study area had heterogeneous relief, with gentle relief being dominant except for the south-western part, where 'ravine' relief was found. We believe that studies related to topographic control over availability of nutrients would profit from considering underlying spatial effects in their models as a method to reduce unexplained errors, as there was a strong underlying spatial effect even at the relatively small spatial scale used in our study.

In addition to spatial effects, there is a strong relationship of vegetation structure and availability of soil nutrients (Pellegrini et al. 2014). One mechanism behind this trend is the higher net primary production of N-fixing trees compared with the herbaceous counterparts. The resulting lower C : N ratio of the forest litter, compared with the nutrient cycling in grasslands, would provide higher SOM in forests and therefore a higher effective cation exchange capacity, allowing for a higher overall nutrient availability (Pellegrini et al. 2014). Our results showed this trend, with higher Ca, P and SOM contents and soil pH in the forest soil compared with grassland soil. The larger biomass of forest ecosystems compared with grasslands depends on a larger nutrient cycling pool in such conditions, which would explain the abundant availability of soil nutrients because the presence of Fabaceae trees in the fragment under study was very low.

This higher availability of nutrients seemed to increase from the forest edge up to 50 m inside the forest and then P, Ca and SOM reached stable concentrations in the soil. This is the depth of edge influence commonly reported in edge effects studies (Ewers and Didham 2008). However, our pH model showed a linear reduction of soil acidity from grassland to forest interior. This might be due to our sampling being limited to 100 m from the forest edge, as there is evidence that there are increasing pH values in forests at greater distances from the edge (Berges et al. 2013). As well as higher availability of nutrients inside forest borders, there was better soil fertility in grasslands closer to forest edges. Samples of grassland soil taken at distances less than 25 m from the forest edge had slightly higher concentrations of P, Ca and SOM than those farther from the edge. Thus, edge effects occurred on both sides of the forest-grassland landscape matrix.

However, there was no evident change in soil availability of Mg and K between forest and grassland samples. These nutrients lack a gaseous phase and in soil are mostly rock-derived. This explains why trend surface-only models had a better fit than the full model in our analysis for these nutrients, suggesting that variation in their availability was controlled by soil parent material and little influenced by vegetation dynamics. On the other side of this spectrum was the soil concentration of P, the availability of which is strongly controlled by vegetation structure. However, determining if soil nutrients control plant biomass and diversity (John et al. 2007), or if composition of vegetation and species control availability of soil nutrients (Binkley and Giardina 1998) is difficult, mainly because all these variables have synergistic correlations (Vitousek et al. 2010). Therefore, it may be impossible to determine cause-effect relationships in soil-plant interactions by empirical methods, and we may only quantify the effects of these relationships.


The spatial patterns of availability of soil nutrients are affected by a large number of variables and play equally large roles in ecosystem functioning. Unveiling the magnitude of the relationships between nutrients and environment requires testing these relationships both individually and collectively using an adequate sampling and statistical scheme. There are several tools for analysis of spatial trends, but spatial statistics is not limited to the analysis to a bivariate coordinate space, and simple distance to environmental features, such as the forest edge, may be much more meaningful in evaluating ecological relationships--as was the case for P in this study. These soil nutrient, vegetation and parent material patterns may have a large influence on global carbon quantification as forests become fragmented across landscapes. How much forest productivity and total biomass are affected by edge effects and soil nutrient patterns may have to be defined for each ecosystem.

Received 18 July 2016, accepted 1 February 2017, published online 28 February 2017


Alvares CA, Stape JL, Sentelhas PC, de Moraes Gongalves JL, Sparovek G (2013) Koppen's climate classification map for Brazil. Meteorologische Zeitschrift 22, 711-728. doi:10.1127/0941-2948/2013/0507

Baker TP, Jordan GJ, Baker SC (2016) Microclimatic edge effects in a recently harvested forest: do remnant forest patches create the same impact as large forest areas? Forest Ecology and Management 365, 128-136. doi: 10.1016/j.foreco.2016.01.022

Bates DM, Kliegl R, Vasishth S, Baayen H (2015) Parsimonious mixed models. arXiv preprint arXiv: 150604967. Available at [Accessed 14 January 2016]

Berges L, Pellissier V, Avon C, Vertieyen K, Dupouey JL (2013) Unexpected long-range edge-to-forest interior environmental gradients. Landscape Ecology 28, 439-453. doi: 10.1007/s10980-012-9841-1

Binkley D, Giardina C (1998) Why do tree species affect soils? The warp and woof of tree-soil interactions. Biogeochemistry 42, 89-106. doi:10. 1023/A: 1005948126251

Bolker BM, Brooks ME, Clark CJ, Geange SW, Poulsen JR, Stevens MHH, White J-SS (2009) Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution 24, 127-135. doi: 10.1016/j.tree.2008.10.008

Brewer MJ, Butler A, Cooksley SL (2016) The relative performance of AIC, AICc and BIC in the presence of unobserved heterogeneity. Methods in Ecology and Evolution 7, 679-692. doi:10.1111/204121OX. 12541

Chadwick KD, Asner GP (2016) Tropical soil nutrient distributions determined by biotic and hillslope processes. Biogeochemistry 127, 273-289. doi: 10.1007/s10533-015-0179-z

Ciannelli L, Fauchald P, Chan KS, Agostini VN, Dingsor GE (2008) Spatial fisheries ecology: recent progress and future prospects. Journal of Marine Systems 71, 223-236. doi:10.1016/j.jmarsys.2007.02.031

Cleveland CC, Townsend AR, Taylor P, Alvarez-Clare S, Bustamante MMC, Chuyong G, Dobrowski SZ, Grierson P, Harms KE, Houlton BZ, Marklein A, Parton W, Porder S, Reed SC, Sierra CA, Silver WL, Tanner EVJ, Wieder WR (2011) Relationships among net primary productivity, nutrients and climate in tropical rain forest: a pantropical analysis. Ecology Letters 14, 939-947. doi: 10.111 l/j.14610248.2011.01658.x

Cochrane MA, Laurance WF (2002) Fire as a large-scale edge effect in Amazonian forests. Journal of Tropical Ecology 18, 311-325. doi:10.1017/S0266467402002237

Dalmolin RSD, Pedron F de A (2009) Solos do municipio de Santa Maria. Ciencia & Amhiente 38, 59-78.

Diggle P, Ribeiro PJ (2007) 'Model-based geostatistics.' (Springer: New York, NY)

Dormann CF, McPherson JM, Araujo MB, Bivand R, Bolliger J, Carl G, Davies RG, Hirzel A, Jetz W. Kissling WD, Kuhn I, Ohlemuller R, Peres-Neto PR, Reineking B, Schroder B, Schurr FM, Wilson R (2007) Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography 30. 609-628. doi: 10.1111/j.2007.0906-7590.05171.x

Drexler M, Ainsworth CH (2013) Generalized additive models used to predict species abundance in the Gulf of Mexico: an ecosystem modeling tool. PLoS One 8, e64485. doi:10.1371/journal.pone.0064458

Ewers R, Didham R (2006a) Confounding factors in the detection of species responses to habitat fragmentation. Biological Reviews of the Cambridge Philosophical Society 81, 117-142. doi: 10.1017/S1464793 105006949

Ewers R, Didham R (2006b) Continuous response functions for quantifying the strength of edge effects. Journal of Applied Ecology 43, 527-536. doi: 10.1111/j.1365-2664.2006.01151.x

Ewers RM, Didham RK (2008) Pervasive impact of large-scale edge effects on a beetle community. Proceedings of the National Academy of Sciences of the United States of America 105, 5426-5429. doi: 10.1073/ pnas.0800460105

Fernandez-Martinez M, Vicca S, Janssens IA, Sardans J, Luyssaert S, Campioli M, Chapin FS Ciais P, Malhi Y, Obersteiner M, Papale D, Piao SL, Reichstein M, Roda F, Penuclas J (2014) Nutrient availability as the key regulator of global forest carbon balance. Nature Climate Change 4, 471-476. doi:10.1038/ncIimate2177

Harper KA, Macdonald SE, Burton PJ, Chen J, Brosofske KD, Saunders SC, Euskirchen ES, Roberts D, Jaiteh MS, Esseen P-A (2005) Edge influence on forest structure and composition in fragmented landscapes. Conservation Biology 19, 768-782. doi:10.1111/j.1523-1739.2005.00 045.x

Hastie T.I, Tibshirani R (1986) Generalized additive models. Statistical Science 1, 297-318. doi: 10.1214/ss/1177013604

Heldwein AB, Buriol GA, Streck NA (2009) O clima de Santa Maria. Ciencia & Amhiente 38, 43-58.

Isaak DJ, Peterson EE, Ver Hoef JM, Wenger SJ, Falke JA, Torgersen CE, Sowder C, Steel EA, Fortin M-J, Jordan CE, Ruesch AS, Som N, Monestiez P (2014) Applications of spatial statistical network models to stream data. WIREs. Water 1, 277-294. doi: 10.1002/wat2.1023

John R, Dalling JW, Harms KE, Yavitt JB, Stallard RF, Mirabello M, Hubbell SP, Valencia R, Navarrete H, Vallejo M, Foster RB (2007) Soil nutrients influence spatial distributions of tropical tree species. Proceedings of the National Academy of Sciences of the United States of America 104, 864-869. doi:10.1073/pnas.0604666104

Johnson J, Omland K (2004) Model selection in ecology and evolution. Trends in Ecology & Evolution 19, 101-108. doi: 10.1016/j.tree.2003.10. 013

Johnson NC, Wedin DA (1997) Soil carbon, nutrients, and mycorrhizae during conversion of dry tropical forest to grassland. Ecological Applications 7, 171-182. doi: 10.1890/1051-0761 (1997)007[0171: SC NAMD]2,0.CO;2

Laliberte E, Grace JB, Huston MA, Lambers H, Teste FP, Turner BL, Wardle DA (2013) How does pedogenesis drive plant diversity? Trends in Ecology & Evolution 28, 331-340. doi:10.1016/j.tree.2013.02.008

Laurance WF, Delamonica P, Laurance SG, Vasconcelos HL, Lovejoy TE (2000) Rainforest fragmentation kills big trees. Nature 404, 836. doi: 10.1038/35009032

Levin SA, Carpenter SR, Godfray HCJ, Kinzig AP, Loreau M, Losos JB, Walker B, Wilcove DS (2009) 'The Princeton guide to ecology.' (Princeton University Press: Princeton, NJ)

Pellegrini AFA, Hoffmann WA, Franco AC (2014) Carbon accumulation and nitrogen pool recovery during transitions from savanna to forest in central Brazil. Ecology 95, 342-352. doi: 10.1890/13-0290.1

R Core Team (2016) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Available at

Ribeiro MC, Metzger JP, Martensen AC, Ponzoni FJ, Hirota MM (2009) The Brazilian Atlantic forest: how much is left, and how is the remaining forest distributed? Implications for conservation. Biological Conservation 142, 1141-1153. doi:10.1016/j.biocon.2009.02.021

Ries L, Fletcher RJ, Battin J, Sisk TD (2004) Ecological responses to habitat edges: mechanisms, models, and variability explained. Annual Review of Ecology Evolution and Systematics 35, 491-522. doi: 10.1146/annurev. ecolsys.35.112202.130148

Riutta T, Slade EM, Bebber DP, Taylor ME, Malhi Y, Riordan P, Macdonald DW, Morecroft MD (2012) Experimental evidence for the interacting effects of forest edge, moisture and soil macrofauna on leaf litter decomposition. Soil Biology & Biochemistry 49, 124-131. doi: 10.1016/j.soilbio.2012.02.028

Sartori PLP (2009) Geologia e geomorfologia de Santa Maria. Ciencia & Ambiente 38. 19-42.

USDA (1999) 'Soil taxonomy: a basic system of soil classification for making and interpreting soil surveys.' (USDA: Washington, DC)

Vasconcelos JF, Luizao HL (2004) Litter production and litter nutrient concentrations in a fragmented Amazonian landscape. Ecological Applications 14, 884-892. doi: 10.1890/03-5093

Vetter D, Rucker G, Storch I (2013) A meta-analysis of tropical forest edge effects on bird nest predation risk: edge effects in avian nest predation. Biological Conservation 159, 382 395. doi:10.1016/j.biocon.2012.12.023

Vitousek PM, Porder S, Houlton BZ, Chadwich OA (2010) Terrestrial phosphorus limitation: mechanisms, implications, and nitrogen-phosphorus interactions. Ecological Applications 20, 5-15. doi: 10.18 90/08-0127.1

Wade TG, Riitters KH, Wickham JD, Jones KB (2003) Distribution and causes of global forest fragmentation. Conservation Ecology 7, 7. doi:10.5751/ES-00530-070207

Whittingham MJ, Stephens PA, Bradbury RB, Freckleton RP (2006) Why do we still use stepwise modelling in ecology and behaviour? Journal of Animal Ecology 75, 1182-1189. doi: 10.1111/j.1365-2656.2006.01141.x

Wood SN (2003) Thin-plate regression splines. Journal of the Royal Statistical Society. Series A. (Statistics in Society) 65, 95-114. doi: 10.11 11/1467-9868.00374

Wood SN (2006a) Low-rank scale-invariant tensor product smooths for generalized additive mixed models. Biometrics 62, 1025-1036.

Wood S (2006b) 'Generalized additive models: an introduction with R.' (CRC Press: Boca Raton, FL)

Wright SJ, Yavitt JB, Wurzburger N, Turner BL, Tanner EVJ, Sayer FJ, Santiago LS, Kaspari M, Hedin LO, Harms K.E, Garcia MN, Corre MD (2011) Potassium, phosphorus, or nitrogen limit root allocation, tree growth, or litter production in a lowland tropical forest. Ecology 92, 1616-1625. doi: 10.1890/10-1558.1

Wuyts K, De Schrijver A, Staelens J, Gielis L, Vandenbruwane J, Verheyen K (2008) Comparison of forest edge effects on throughfall deposition in different forest types. Environmental Pollution 156, 854-861. doi:10. 1016/j.envpol.2008.05.018

Yoo K, Mudd SM, Sanderman J, Amundson R, Blum A (2009) Spatial patterns and controls of soil chemical weathering rates along a transient hillslope. Earth and Planetary Science Letters 288, 184-193. doi: 10.1016/j.epsl.2009.09.021

Yuan Z, Gazol A, Lin F, Ye J, Shi S, Wang X, Wang M, Hao Z (2013) Soil organic carbon in an old-growth temperate forest: Spatial pattern, determinants and bias in its quantification. Geoderma 195-196, 48-55. doi: 10.1016/j.geoderma.2012.11.008

Thomas Schroder (A,B) and Frederico D. Fleig (A)

(A) Universidade Federal de Santa Maria, Avenida Roraima, 1000, Santa Maria, RS 97105900, Brazil.

(B) Corresponding author. Email:

Caption: Fig. 1. (a) Localisation, shape and sample points of the forest fragment under study, (b) histogram of distance from forest edge of soil samples and (c) mean estimated elevation above sea level (m).

Caption: Fig. 2. Mean trend surfaces of the soil variables estimated by Generalised Additive Models under the full model, in which spatial trend and distance from forest edge are independent variables, or by trend surface model where only the spatial trends are considered. Measuring units: pH ([H.sub.2]O), SOM (%), P and K. (mg/L), and Ca and Mg ([cmol.sub.c]/L).

Caption: Fig. 3. Estimated influence of distance from forest edge on soil variables under the full model, in which spatial trend and distance from forest edge are independent variables, or by the edge effect model where only the distance from forest edge is considered. Shaded areas are confidence intervals including uncertainty about the overall mean. Modelled variables are centred.
Table 1. Akaike Information Criteria of Generalised Additive
Models of different soil variables of an Atlantic forest
fragment in southern Brazil

Bold values indicate the 'best' model. Superscripts are based
on likelihood ratio test for pairwise fit comparison with the
full model: single asterisks indicate 0.05 significance level,
double asterisks indicate 0.01 significance level and ns
indicates non-significance

Soil variable        Full       Trend surface       Edge effects
                  model (A)       model (B)          model (C)

PH                 53.2           52.1# (ns) ***     55.6 *
SOM                95.7# ***       98.9 *           101.1 **
P                 241.4           241.1 (ns)        238.9# ns ***
K                 491.3          490.3# (ns) ***    499.9 **
Ca                786.8# ***     792.2 **           809.1 **
Mg                697.2          695.5# (ns) ***    701.9 *

(A) Full models include edge effects and spatial trend.

(B) Trend surface models take into account only the spatial

(C) Edge effects are based on distance from forest edge.

Note: Bold values that indicate the 'best' model are indicated with #.
COPYRIGHT 2017 CSIRO Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2017 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Schroder, Thomas; Fleig, Frederico D.
Publication:Soil Research
Article Type:Report
Geographic Code:3BRAZ
Date:Oct 1, 2017
Previous Article:Quantifying above and belowground biomass carbon inputs for sugar-cane production in Brazil.
Next Article:Pronounced surface stratification of soil phosphorus, potassium and sulfur under pastures upstream of a eutrophic wetland and estuarine system.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters