Cork oak woodlands patchiness: A signature of ...

Applied Geography 54 (2014) 18e26

Contents lists available at ScienceDirect

Applied Geography journal homepage: www.elsevier.com/locate/apgeog

Cork oak woodlands patchiness: A signature of imminent deforestation? Augusta Costa a, b, *, Manuel Madeira c, Tobias Plieninger d a Universidade Nova de Lisboa, Faculdade de Ci^ encias e Tecnologia, Center for Environmental and Sustainability Research (CENSE), Campus de Caparica, 2829-516 Caparica, Portugal b ~o Agra ria e Veterina ria, I.P., Quinta do Marqu^ Instituto Nacional de Investigaça es, Av. da República, 2780-159 Oeiras, Portugal c Universidade de Lisboa, Instituto Superior de Agronomia, Centro de Estudos Florestais, Tapada da Ajuda, 1349-017 Lisboa, Portugal d University of Copenhagen, Department of Geosciences and Natural Resource Management, Rolighedsvej 23, 1958 Frederiksberg C, Denmark

a b s t r a c t Keywords: Quercus suber L. Mediterranean landscape Forest degradation Fractional canopy cover Patch sizeefrequency distribution Power-law function

The cork oak (Quercus suber L.) woodlands of the agroforestry landscapes of Southwestern Iberia are undergoing drastic change due to severe natural and anthropogenic disturbances. These may eventually result in woodland loss or deforestation, the final step of an ongoing process of woodland degradation. Monitoring changes in the spatial patterns of woodlands e especially fractional canopy cover of woodlands and/or their patchiness in the landscape mosaic e potentially enables forecasting of loss and responding to it at an early stage. We examine the degradation process in two cork oak woodlands, resulting from distinct disturbances, wildfire and oak mortality, aimed at evaluating the changes, trends and deviations of the spatial attributes of these woodlands as they move from an initial (less disturbed ecosystem) to a final state (more disturbed ecosystem). While undergoing disturbances, both woodlands exhibited similar trends of decreasing fractional canopy cover and decreasing number of larger patches. Patchiness rather than fractional canopy cover seems, however, to be potentially more useful as a signature of imminent oak woodlands deforestation, given that its contrast before and after disturbance was much higher. The structural dynamics of oak woodlands is a crucial but neglected issue that needs greater attention from policy forums working toward their conservation and restoration as well as from stakeholders and society as a whole. © 2014 Elsevier Ltd. All rights reserved.

Introduction Worldwide deforestation is driven by mixed natural and anthropogenic disturbances (Lambin et al., 2001). Over time, these disturbances re-shape the spatial configuration of forests (i.e., physical organization, complexity and mosaicing) and affect biotic diversity (Sala et al., 2000), climate change and vegetative carbon release to the atmosphere (Bonan, 2008; Dale et al., 2001). Typically, in most studies on deforestation, the temporal dynamics of woodlands are assessed by applying a simple forest/nonforest classification to woodland landscape snapshots, though they often exhibit drastic variations in patch composition and structure (Grossmann & Mladenoff, 2007). However, deforestation, which * Corresponding author. Instituto Nacional de Investigaç~ ao Agr aria e Veterin aria, ^s, Av. da República, 2780-159 Oeiras, Portugal. Tel.: þ351 21 I.P., Quinta do Marque 4463740; fax: þ351 21 4463701. E-mail addresses: [email protected], [email protected] (A. Costa), [email protected] (M. Madeira), [email protected] (T. Plieninger). http://dx.doi.org/10.1016/j.apgeog.2014.07.006 0143-6228/© 2014 Elsevier Ltd. All rights reserved.

denotes the definitive loss of trees, can also be interpreted as a continuous process of forest degradation, defined as tree thinning and damage to the forest canopy, which eventually end in forest loss. Yet this is a process that can be reversed, if addressed in a timely manner, as woodlands are expected to regenerate (Hosonuma et al., 2012). The forest degradation process usually results in changes in forest spatial pattern attributes, especially woodland composition and structure, which do not immediately affect overall forest area and, consequently, remain undetected in simple land cover change assessments. Monitoring forest degradation processes can potentially help in detecting of changes and forecasting forest loss, which may enable adequate policy or management responses at an early stage (Wang, Qi, & Cochrane, 2005). Evergreen oak woodlands are widely distributed within the Mediterranean climate-type regions (e.g. in California, Chile, Australia) and are among the most representative ecosystems in Southwestern Iberia, where they occupy about 4 million ha (Olea & San Miguel-Ayanz, 2006). Here, extensive woodlands comprised of

A. Costa et al. / Applied Geography 54 (2014) 18e26

Quercus suber L. (cork oak) and Quercus ilex ssp. rotundifolia Lam. (holm oak) are called montados in Portugal and dehesas in Spain. They are mainly found in oligotrophic soils. The prevailing land-use system in these areas is distinguished by a systematic combination of pastoral, agricultural, and forestry activities, with livestock raising being the dominant one that determines all other land uses. Uses are well adapted to the unpredictability of the Mediterranean climate. Oak stands are regularly cleared and thinned to densities of 10e50 trees ha1 to enhance herb growth as well as to ensure maximum yields of cork and acorns. Traditionally, rotational plowing has been a common management strategy for grain cultivation and for the control of shrub encroachment. These ecosystems have a long history of deforestation within intensive human-managed landscapes and have been shaped through time by natural disturbances and human activity (Costa, Madeira, Santos, & Plieninger, 2014). Nevertheless, it is only recently that several authors assessed the importance of disturbances in shaping woodland configuration over time in such landscapes and their consequences on ecosystem functions (Costa, Madeira, Santos, & Oliveira, 2011; Costa et al., 2014; Pinto-Correia & Mascarenhas, 1999; Plieninger, 2006; Plieninger & Schaar, 2008; ~o n, 2008), revealing many similarities to Urbieta, Zavala, & Maran other scattered-tree savannah-type ecosystems in the world (Manning, Gibbons, & Lindenmayer, 2009). Despite their remarkable resilience, novel disturbances, sometimes acting recurrently, may surpass their ability to control important ecosystem processes, such as degradation, which eventually result in definitive change and woodland loss. The loss of evergreen oak woodlands has been recently attributed to consistent management extensification (e.g. shrub encroachment) or intensification (e.g. livestock grazing) trends, under stressful climate conditions and “strong” soil degradation cio, Holmgren, Jansen, & Schrotter, 2007; Brooker, Maestre, & (Aca Callaway, 2008; Costa, Madeira, Santos, Plieninger, & Seixas, 2014; Costa et al., 2011; Maestre, Valladares, & Reynolds, 2005; Pausas, Ribeiro, Dias, Pons, & Beseler, 2006; Plieninger, Rolo, & Moreno, 2010; Pons & Pausas, 2006; Torre & Díaz, 2004). Woodland loss also occurs abruptly, under disturbances such as oak mortality (Brasier, 1996; Costa, Pereira, & Madeira, 2009, 2010; Shifley, Fan, Kabrick, & Jensen, 2006; Wargo, 1996) or wildfire occurrence cio, Holmgren, Rego, Moreira, & Mohren, 2009; Catry, Moreira, (Aca Cardillo, & Pausas, 2012; Costa, Madeira, & Santos, 2014). In Southwestern Iberia, evergreen oak woodland loss seems hard to reverse and corresponds to a continuous process of woodland degradation. Based on observations from other sorts of vegetation in semi-arid regions, such degradation should become visible through a transition of both fractional canopy cover (Cano, Navarro, & Ferrer, 2003; Carreiras, Pereira, & Pereira, 2006; Costa, Madeira, & Oliveira, 2008, Costa et al., 2010; Plieninger & Schaar, 2008) and patchiness fi et al., 2007; Maestre & (patch sizeefrequency distribution) (Ke , 2007). Although canopy-cover persistence is Escudero, 2009; Sole already an important feature of woodland conservation, woodland patchiness should also be seen as far from being negligible for understanding factors affecting woodlands persistence. In fact, the concept of patchiness has been used in arid Mediterranean ecosystems, where fitted power functions (Xiao, White, Hooten, & Durham, 2011) describing the patch sizeefrequency relationship fi et al., have been tested to predict the onset of desertification (Ke , 2007). 2007; Sole Given the socio-economic and ecological importance of Mediterranean evergreen oak woodlands, for timely prevention of deforestation it is important to understand how woodland degradation processes occur in response to disturbances. Monitoring changes in woodland fractional canopy cover and patch sizeefrequency distribution can be potentially useful tools for providing

19

such knowledge in a cost-effective manner. However, so far no studies have compared the spatial pattern changes of oak woodlands to examine trends and deviations for initial and final ecosystem states, before and after disturbances. The present study is therefore exploratory in nature and aims to fill this gap by analyzing the fractional canopy cover and patchiness trends of cork oak woodlands, considering distinct disturbances (wildfires and cork oak mortality, see Table 1) that may eventually result in deforestation (Costa, Madeira, Santos, Plieninger, & Seixas, 2014). We investigated the degree to which these oak woodland attributes have changed under disturbance effects by characterizing an initial state before disturbance (a less disturbed ecosystem) and a final state after disturbance (a more disturbed ecosystem) and then quantifying the deviations indicative of intrinsic ecological organization dynamics. The study was carried out under the following three hypotheses: Firstly, we assumed that, as a result of disturbances, woodland fractional canopy cover would increase, indicating an increase of canopy gaps. Secondly, as another result of disturbances, transition in patch sizeefrequency distribution would occur, from a powerlaw function to a truncated-power-law function, indicating decrease of larger patches. Thirdly, based on the theory predicting that sizeefrequency distribution of spatial pattern of forests will follow a power-law distribution only if the ecosystem is not disturbed (Foster & Reiners, 1986; Marco, Montemurro, & Cannas, 2011), we assumed that, when a deviation is observed in these dynamics, the ecosystem is not at equilibrium. Therefore, based on these assumptions, we hypothesized that patchiness, rather than canopy cover, would turn out to be a more accurate signature of imminent transition toward deforestation in cork oak woodlands. Our approach is aimed at contributing toward obtaining improved insights into the changing spatial patterns of cork woodlands. Knowledge on fractional canopy cover and patchiness is sorely needed to address landscape conservation and restoration issues (Fahir, 2003; Hill & Curran, 2001; Long, Nelson, & Wulder, 2010; Remmel & Csillag, 2003). Material and methods Study area The study was conducted in two cork oak woodlands in Southern Portugal, located in the littoral region, one of the most Table 1 Characteristics of the two study areas, Ulme (UL) and S. Bartolomeu da Serra (SB). Study areas

Ulme (UL)

S. Bartolomeu da Serra (SB)

Location Dominant oak species Dominant woodland ecosystem Oak woodland initial area (Total area) (ha) Mean annual temperature ( C) Annual rainfall (mm) Average air humidity (%) Aridity index (R/PET) Lithology

39 230 Ne08 300 W Q. suber Dense forest

38 040 Ne08 400 W Q. suber Dense forest

4,489 (12,197)

4,929 (6,224)

15.9

16.0

637 75e80

629 80e85

0.50e0.65 Sedimentary formations 20e190 Flat and steeply undulating Wildfire occurrence 1995e2007

0.50e0.65 Schist formations

Altitude (m) a.s.l. Slope Main disturbance Addressed disturbance period

80e260 Steeply undulating (heterogeneous) Oak mortality 1995e2007

20


iconic areas for such woodlands in Iberia. The climate is of a Mediterranean type, and rain mostly occurs from late autumn to early spring, with great annual irregularities. The S. Bartolomeu da Serra (SB) and Ulme (UL) municipalities, which form our study areas, are situated close to the ocean. Accordingly, there is some Atlantic influence. Values for rainfall and temperature are similar in both areas (Table 1). At UL, the landscape is dominated by Pliocenic and Miocenic formations, which are densely incised by water lines. The land is flat in valley bottoms and on top of the Pliocenic terraces. Predominant soils in valleys are Gleysols and Fluvisols (sensu WRB, 2006); soils developed on flat tops are mostly Stagnic Luvisols, while in undulating steep areas Regosols and Cambisols are predominant. The landscape is gently undulating at SB, and geological formations correspond to Carbonic schists (grauwakes, siltstones, carbonaceous schists) with vertical stratification. Soils in SB are mostly Haplic Leptosols, combined to a small degree with Haplic and Epileptic Luvisols (Table 1). In the two study areas, oak woodland structure and composition were analyzed for the years 1995 and 2007. The common element shared by each of the areas is that the economic value of montado land relies on relatively large farm sizes, low-intensity farming systems and cork production (Moreno & Pulido, 2009). The decision to study the 1995e2007 period was driven by photo availability and because of the substantial disturbances that had impacted the study areas in that period. The oak woodlands at UL and SB do differ; however, with respect to the density of oak forest, oak mortality and wildfire incidence (see Table 1). At UL, recent oak woodland changes were mostly driven by a large wildfire in 2003 (Costa, Madeira, & Santos, 2014), whereas at SB changes were driven by elevated oak mortality, which has been reported for Southwestern Portugal since the early 1990s (Costa et al., 2010). It has been described as a widespread and complex phenomenon, similar to oak mortality that has occurred elsewhere in Europe (Oszako, 2000) and North America (Oak, Tainter, Williams, & Starkey, 1996).

Data analysis Data acquisition and processing In 1995 and 2007, national aerial photography plans (commissioned by the Portuguese Geographic Institute) captured infrared color, ortho-rectified, digital aerial photographs, with spatial resolutions of 1 m and 0.5 m, respectively. For built categorical maps, the classification scheme used in this study comprised a total of nine land cover classes, with seven vegetation cover classes and two non-vegetation classes (water and urban areas). Vegetation cover classes included open farmland (arable and permanent pastures), shrubland and five forest classes: a) evergreen oak woodlands, b) eucalypt and c) pine stone plantations, d) other broad-leaved forests and e) other coniferous forests. First, we digitized a categorical land cover map of the study areas from 2007 on the basis of visual photo interpretation (i.e., delineation of patches to a minimum size of 1 ha on the basis of within-patch uniformity of structure and composition attributes) of the 2007 orthophotographs. Next, we followed an adapted regressive photo interpretation method (see Andrieu, Ladet, Heintz, & Deconchat, 2011) by using a duplicate of the 2007 layer for photo interpretation of the photographs from 1995. We consistently used the same attributes, standards, equipment and photo interpreters for the two time layers. Similar methods have been used by Plieninger, Schleyer, Mantel, and Hostert (2012) and Skalos and Engstova (2010), among others.

Four weeks of ground verification were performed between June and November 2008 to assess the accuracy of the photo interpretation, meaning here the accuracy with which vegetation classes had been mapped. A randomly selected sample was built with 15 patches per vegetation-type present in the study area, which included open farmland, shrubland and the forest classes of oak woodlands and eucalypt and pine stone forests (about 60 for UL and 45 for SB). The accuracy of each specific vegetation class was calculated based on a simple cross-tabulation matrix (confusion matrix), where mapped data at patch level (photo-interpreted data) was compared against ground data (Foody, 2002). The elements on the diagonal of this matrix contained the percentage of the total number of cases where the photo-interpreted and ground data agreed. All patches interpreted initially as forest plantations were only completely classified after the ground survey (and for that reason were initially excluded from the confusion matrix), such that, for each patch generally labeled forest plantation, the final mapped label contained indicated a specific type (e.g. oak woodland, eucalypt or stone pine plantation). For oak woodland, agreement between photo-interpreted and ground data was 88.8% in UL and 86.4% in SB. The pattern of misclassification of oak woodlands was likely the result of tree mortality, as most errors seem to have been related to (old or less dense) oak woodlands in the mapping that had disappeared and, to a lesser extent, there were errors of omission. In particular, some forest areas (new plantations) were missing from the maps but were found to exist on the ground. Fractional canopy-cover maps Based on woodland composition maps (Costa, Madeira, Santos, Plieninger, & Seixas, 2014), for each study area we built fractional canopy-cover (FC) maps, one before (1995) and another after (2007) the occurrence of a disturbance. Fractional canopy cover is defined as the percentage of a horizontal vegetated area occupied by the vertical projection of tree crowns (Gonsamo, D'odorico, & Pellikka, 2013). The oak woodlands FC maps discriminated the total oak woodlands area into four classes: three classes of oak forest where oaks were the dominant species, with a density of over 20 trees ha1 (dead trees excluded) e dense oak forest (FC > 0.50), typical oak forest (0.25 < FC < 0.50), and open oak forest (FC < 0.25) e and one class of scattered trees, with oak density between 5 and 20 trees ha1 (dead trees excluded) and with an understory of natural grasslands for grazing or occasional cereal crops (oak farmland) and encroaching shrub species (oak shrubland). Using data from two different years, 1995 and 2007, on canopycover attributes enabled the monitoring of forest degradation based on three transition indicators: 1) clearance, defined as the rate of canopy-cover decrease or opening; 2) expansion, defined as the rate of canopy-cover increase or closure; and 3) persistence, defined as the rate at which oak woodland fractional canopy cover remained the same, meaning no change in canopy cover. These indicators were estimated based on oak woodland transition matrices (Luque, Lathrop, & Bognar, 1994; Teixido, ~ o, & Gutie rrez, 2010). The matrices, [T]44, had Quintanilla, Carren a total of four established oak woodland FC classes in the oak woodland fractional canopy-cover maps: scattered trees (in oak farmland or in oak shrubland) constitute the least dense class, followed by open, typical and dense oak forest, with the last being the most dense class. For each element of the [T]44 matrices, aif represented the polygonal area (in ha) of an FC class that changed from i (defined as the initial class, with i ¼ 1,…,4), ordered by the woodland's FC degree, i þ 1 > i, undergoing transition to class f (defined as the final class, with f ¼ 1,…,4), also ordered by the woodland's FC degree, f þ 1 > f. Based on these matrices, [T]44, proportion matrices, [P]44, were then generated, where each element, aif, was converted into proportions (annual percentage),


pif, of the row total value, Ai, within a determined period of years, n, before and after the disturbance occurrence, as defined by the following formula:

pif ¼

aif Ai

100

; with Ai ¼

n

4 X

aif

(1)

f ¼1

The transition indicators for each oak woodland class were then computed, according to the following algorithms: Clearance of a class i (Ci), defined as the sum of the row elements on the left side of diagonal element for each class i, P Ci ¼ i1 f ¼1 pif ; i > 1; Expansion of a class i (Ei), defined as the sum of the row elements on the right side of the diagonal element for each class i, P Ei ¼ 4f ¼iþ1 pif ; i < 4; and Persistence of a class i (Pi), is the diagonal element of the [T]44 matrices, considered the complement of clearance and expansion, defined for each class i, Pi ¼ 100 (Ci þ Ei). Rates of oak woodland outflow were calculated by overlaying the categorical maps from 1995 and 2007, discriminated into the categories of abandonment, intensification and change. Oak woodland outflow was defined as the rate of oak woodland transition to i) another forest vegetation class, as change; ii) the open farmland vegetation class, as intensification and; iii) to the shrubland vegetation class, as abandonment. Oak woodland outflow (formula (2)) was obtained from information from the composition maps on oak woodland landscapes, considering the previous oak woodland classes plus three vegetation cover classes: open farmland (arable and permanent pastures), shrubland and other forests (eucalypt or stone pine plantations) (Costa et al., 2011). The land cover transition matrices, [T]77, and correspondent proportion matrices, [P]77, (with 7 corresponding to four FC oak woodlands classes e scattered oak farmland and oak shrubland and open, typical and dense oak forest e plus the three aforementioned vegetation cover classes of shrubland, open farmland and other forest), with each of the elements aif and pif representing, respectively, the area and the correspondent proportion (annual percentage) undergoing transitions

pif ¼

aif Ai

100 n

; with Ai ¼

7 X

aif

(2)

f ¼1

Change in land cover was defined as the transition from each oak woodland class to a different kind of forest; intensification of land cover as the transition from each oak woodland class to open farmland; and abandonment of land cover as the transition from each oak woodland class to shrubland. Patch sizeefrequency distribution In each of the study oak woodlands, before and after their disturbances occurred, we analyzed the number (N) and size (S) of the oak woodland patches and plotted their frequency distribution using a linear binning approach (White, Enquist, & Green, 2008) (see Fig. 2). A constant bin-width of 103 m2 was chosen based on the observations range and on a trade-off between frequency distribution resolution and accuracy in the studied oak woodlands, given the FC resolution maps. These oak woodland patch sizeefrequency distributions were then fitted to two different models for fi abundance distribution of oak woodlands: 1) a power law (Ke et al., 2007; White et al., 2008)

NðSÞ ¼ cSl

(3)

21

where c and l are constants and l is the estimated scaling exponent, typically negative (l < 0), and 2) a truncated-power law

NðSÞ ¼ cSl eSx S

(4)

where Sx was the observed patch sizes (in 1000 m2) above which fi et al., 2007). the N(S) decreased faster than in the power law (Ke The goal was to first focus on the frequency distribution of oak woodlands before disturbance. Given the long-term history of human disturbances induced in each type of woodland prior to the occurrence of the recent disturbance effects under study, which we call here the initial stage, deviations from a typical power-law species abundance distribution (with l z 1) were already expected. The truncated-power-law model is better suited, therefore, at this initial stage to describe the oak woodland patch sizeefrequency distribution. In addition, differing local characteristics and patterns between the two studied woodland ecosystems (Costa et al., 2011) due to the recent and contrasting disturbance effects should determine how the patch sizeefrequency distribution changes and which disturbance occurrences are likely to lead to high deviations between the scaling exponents at what we are calling a final stage, after disturbance. In the studied woodlands, the fit of the models e both before and after the disturbances occurred e was compared, based on the coefficient of determination (R2) and the residual sum of squares (SSR). This method was developed to allow great potential for application when using remote-sensing data for monitoring woodland composition, based on the assumption that, under specific disturbance effects, the patch sizeefrequency distribution of oak woodlands will be altered. The greater the intensity of deviations in the patch sizeefrequency distribution (e.g. the more different exponents of fitted truncated-power laws), the higher is the degree to which the ecosystem will be disturbed in its structural components. Results Fractional canopy cover The FC maps showed distinct initial and final states (1995 and 2007, respectively) in terms of fractional canopy cover of the two cork oak woodlands investigated (Fig. 1), with the opening of canopy cover being most noticeable. Dense oak forest exhibited a higher fractional canopy cover contrast between 1995 and 2007, though this can also be observed in all other oak woodland classes, including to some extent the scattered trees class. Between 1995 and 2007, the dense oak forest showed annual losses of 3.9% yr1 and 3.7% yr1 at UL and SB, respectively, corresponding to losses of 86 and 129 ha yr1 (Table 2). The scattered trees (in oak farmland and shrubland) showed annual gain rates of 27.3% yr1 in SB and 11.8% yr1 in UL, corresponding to gains of 37.0 and 30.8 ha yr1, respectively. The oak woodland degradation revealed by our comparison resulted largely in canopy opening or clearance (Table 2). Clearance values were, in most cases, twice expansion values, with rates higher than 2% yr1, with the highest values of 3.7 and 3.4% yr1 for SB's and UL's cork oak forests, respectively, which corresponded to canopy-cover-opening rates of 23.3 and 20.5 ha yr1. Expansion rates of these cork oak woodlands did not reach 2% yr1, with 1.3 and 1.8% yr1 for UL and SB, respectively, corresponding to 7.8 and 9.1 ha yr1. Despite large similarities, when comparing both cork oak woodlands, SB seems to be more affected in terms of fractional canopy cover than UL. At SB, suffering from cork oak mortality, the annual clearance area was eight times the expansion area against five times at UL, due to wildfire occurrence (Table 2).

22


Fig. 1. Location of the study areas in Southern Portugal, Iberia. Cork oak woodland composition at UL (first row) and SB (second row). From left to right: before (1995) and after (2007) disturbance: oak farmland and oak shrubland (scattered-tree area) in black; oak forest fractional canopy-cover classes are in intermediate grey tones, from moderately light grey tones (open oak forest) to darker grey tones (dense oak forest).

The oak woodland outflow dynamics also contributed to the degree of canopy-cover opening, specifically through the emergence of canopy gaps. Outflow dynamics were marked by abandonment (Table 3). At UL, due to the wildfire, abandonment was the highest and ranged between 2.1 and 3.2% yr1, resulting in a loss of 46.6 ha yr1 of dense oak forest and 31.9 ha yr1 of open oak forest to shrubland areas, respectively. At SB, as a result of oak mortality, the outflow dynamics exhibited a relatively lower oak forest

abandonment rate, ranging between 0.4 and 1.3% yr1 or 14.0 ha yr1 of dense oak forest and 6.7 ha yr1 of open oak forest. The oak woodland canopy intensification, on the other hand, was most noticeable at SB, ranging from 0.4 to 1.8% yr1 (i.e., 14.9 ha yr1 of dense oak forest and 9.3 ha yr1 of open oak forest), while in UL the intensification ranged between only 0.1 and 0.2% yr1, meaning an outflow of 1.4 ha yr1 of dense oak forest and 1.8 ha yr1 of open oak forest to open farmland (Table 3).

Fig. 2. Cork oak woodland composition before (1995) and after (2007) disturbance occurrence. From left to right: UL and SB. In brackets, scattered-tree area and oak farmland are marked in white, with oak shrubland in lighter grey.

A. Costa et al. / Applied Geography 54 (2014) 18e26 Table 2 Clearance, persistence and expansion of fractional canopy cover of woodland classes in % yr1 (in ha yr1, in brackets). Transitions based on the proportion matrix [P]44. Oak woodland class UL Scattered trees Oak forest Open Typical Dense SB Scattered trees Oak forest Open Typical Dense

Clearance

Persistence

Expansion

e

97.5 (685.4)

2.5 (17.6)

2.4 (24.3) 3.4 (20.5) 2.9 (63.0)

96.4 (975.6) 95.3 (573.7) 97.1 (2,109.0)

1.2 (12.1) 1.3 (7.8) e

e

98.1 (261.9)

1.9 (5.1)

2.2 (11.1) 3.7 (23.3) 3.2 (112.9)

96.0 (484.8) 94.9 (596.9) 96.8 (3,415.1)

1.8 (9.1) 1.4 (8.8) e

Table 3 Change, intensification and abandonment of oak woodland classes in % yr1 (in ha yr1 in brackets). Transitions were based on the proportion matrix [P]77. Oak woodland class UL Scattered trees Oak forest Open Typical Dense SB Scattered trees Oak forest Open Typical Dense

Change

Intensification

Abandonment

0.5 (3.7)

0.5 (3.4)

2.6 (18.2)

0.1 (1.4) 0.1 (0.6) 0.2 (4.6)

0.2 (1.8) 0.2 (1.4) 0.1 (1.7)

3.2 (31.9) 2.2 (13.3) 2.1 (46.6)

0.0 (0.0)

2.4 (6.3)

2.3 (6.3)

0.0 (0.0) 0.0 (0.0) 0.0 (0.3)

1.8 (9.3) 0.6 (4.0) 0.4 (14.9)

1.3 (6.7) 1.0 (6.3) 0.4 (14.0)

23

concomitantly with an increase of smaller patches, which resulted in an increase in the negative slope of the fitted linear regressions to log-transformed values of number (N) and size of patches (S) (Fig. 3). At SB, the deviations in the oak woodland patch sizeefrequency distribution before and after disturbance was most noticeable (Dl ¼ 0.49, with l1995 of 0.88 and l2007 of 1.37) when compared to the deviations found in the woodlands at UL (Dl ¼ 0.18, with l1995 of 0.1.06 and l2007 of 1.24). These highest deviations in SB were associated with its higher frequency of reduction of larger patches (larger than 240 104 m2) and higher frequency of increase of small patches (smaller than 3 104 m2) after the disturbance in 2007, which had not occurred at UL (Fig. 3). The values of the power-law scaling exponent, g, fitted to cork oak woodlands patch number and size distributions were higher in SB, with g z 1 (Table 4). However, patch sizeefrequency distribution in both cork oak woodlands was poorly adjusted by the power-law distribution when compared to truncated-power-law distribution. Moreover, the highest R2 value, of 0.94, was found at SB. At UL, the patch sizeefrequency distribution was poorly adjusted to a power-law function (R2 < 0.70) during both study years and a truncated-power law gave a better fit (Table 4). This indicates a severe deficiency of large patches in cork oak woodlands, already noticed in the initial state before wildfire occurrence. The estimated scaling exponents for (best fitted) truncatedpower-law models were always smaller than the ones estimated for the power laws. After the disturbances, the degradation process cooccurred with the absence of larger patches (which resulted in decreased truncation of the woodlands patch sizeefrequency distribution). The highest difference between scaling exponents was found for UL, 0.39, while at SB it was only 0.21. Discussion

The two kinds of disturbances seem to have had distinct effects on the fractional canopy cover of the cork oak woodlands studied. Due to wildfire, for example, the woodlands at UL exhibited three times greater forest gap areas than SB (92 ha yr1 in UL against 27 ha yr1 in SB). Meanwhile, at SB canopy opening due to cork oak mortality was only slightly higher than at UL (147 ha yr1 against 108 ha yr1) (Tables 2 and 3).

Patch sizeefrequency distribution The patchiness of the cork oak woodlands exhibited a wide range of sizes for 1995 and 2007, despite the decreasing trend in the patch area range: on average, at SB the largest patch size decreased from 1,727 to 217 ha, whereas at UL they went from 250 to 144 ha. This trend corresponded to a decrease of larger patches

Fractional canopy cover of cork oak woodlands Our analysis revealed that, despite having undergone quite different disturbances (wildfire, oak mortality), the oak woodlands under study shared similar clearance trends. Canopy-cover dynamics were affected negatively in both cases, which confirmed our first hypothesis, which expected a clearance of canopy cover and an increase of canopy gaps as a result of disturbances. Our results also showed that the annual rate of clearance in these cork oak woodlands (maximum of 3.9 and 3.7% yr1 for UL and SB, respectively) was lower than that reported by Cano et al. (2003) for a cork oak diz (4.4% yr1), in the 1977e2000 period. dehesa in Ca The rates of clearance found in this study are clearly above those for the expansion of oak canopy cover, which suggests that the two cork oak woodlands are becoming unstable and likely to change.

Fig. 3. Patch sizeefrequency distribution of oak woodland area. From left to right: UL and SB. Black squares are before disturbance (1995) and white squares after disturbance (2007). Fitted linear regressions: solid line means before disturbance (1995) and dotted line means after disturbance (2007).

24


Table 4 Power-law and truncated-power-law functions fitted to the patch sizeefrequency distribution of cork oak woodland area, before (1995) and after (2007) disturbances in UL and SB. Oak woodland UL Before (1995) After (2007) SB Before (1995) After (2007)

gPL

R2PL

0.85 0.87

0.70 0.68

5552.4 5781.2

1.03 1.22

0.75 0.92

10282.3 22837.2

SSRPL

Pvalue

gTPL

R2TPL

849.5 775.7