CarmonaNahuelhual_2012.pdf

at SciVerse ScienceDirect

Applied Geography 32 (2012) 904e915

Contents lists available

Applied Geography

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

Combining land transitions and trajectories in assessing forest cover change

A. Carmonaa,b, L. Nahuelhuala,c,*a Instituto de Economía Agraria, Universidad Austral de Chile, Casilla #567, Valdivia, Chileb Escuela de Graduados, Facultad de Ciencias Agrarias, Universidad Austral de Chile, Chilec Fundación Centro de los Bosques Nativos, FORECOS, Chile

Keywords:Land transitionsLand trajectoriesLand use changeDeforestationTemperate rain forestsSouthern Chile

* Corresponding author. Instituto de Economía AgChile, Casilla #567, Valdivia, Chile. Tel./fax: þ56 63 2

E-mail address: [email protected] (L. Nahuelhu

0143-6228/$ e see front matter � 2011 Elsevier Ltd.doi:10.1016/j.apgeog.2011.09.006

a b s t r a c t

One restriction of landscape studies is that land use and cover change is often regarded as irreversible. Ahighly dynamic landscape in southern Chile was selected to show that forest cover change involvesa series of complex transitions and trajectories. Using Landsat images from 1976, 1985, 1999 and 2007 anin-depth analysis of the transition matrix was conducted to separate random and systematic transitionswhich were grouped into trajectories using a pixel-history approach. Main trajectories were linked tofragmentation indices and farming systems through cluster analysis. Of the 247 trajectories identified,old growth forest persistence comprised 22% of the landscape, whereas deforestation trajectoriescomprised 20.9% and were mostly composed of changes from old growth forest to shrubland (13.9%).Trajectories of forest degradation from old growth to secondary forest comprised 19.7% of the landscape.The periods 1976e1985 and 1999e2007 concentrated the most systematic deforestation and degrada-tion transitions. In turn, random transitions predominated between 1985 and 1999, probably in responseto economic factors that acted suddenly on the landscape during the 80’s, such as the woodchip exportand aquaculture booms. A close relationship between landscape fragmentation and the proportion ofsystematic transitions and farming systems was found; specifically, the highest entropy indices occurredin clusters which exhibited the lowest proportion of systematic transitions and the highest proportion(>70%) of peasant agricultural systems. Understanding the complexity of forest cover change trajectoriesis relevant for improving the prediction of possible landscape evolutions and establishing landscapemanagement priorities.

� 2011 Elsevier Ltd. All rights reserved.

Introduction

Recently emerged land change science aims to understand thepatterns and dynamics of land use and land cover (LULC) changeand its biophysical and human drivers (Rindfuss, Walsh, Turner,Fox, & Mishra, 2004). Yet, a typical limitation of land changestudies is that LULC change is regarded as a simple and irreversibleconversion from one cover type to another (Mertens & Lambin,2000). On the contrary, since human societies co-evolve withtheir environment, LULC change is predominantly non-linear andrelated to other social and biophysical changes through a multi-plicity of land transitions and trajectories (Raskin et al., 2002).

In the broadest sense, a land transition is a process of systemchange in which the structural nature of the land system is trans-formed (Martens & Rotmans, 2002, 135pp). More narrowly, the

raria, Universidad Austral de21235.al).

All rights reserved.

concept refers to any change in land use systems from one state toanother de.g. from a system dominated by marginal agriculture toa system with industrial tree plantations in response to newfinancial incentives.

Furthermore, transitions have been classified as random andsystematic (Braimoh, 2006; Pontius, Gilmore, Shusas, & McEachern,2004). Random transitions are defined as those influencedby unintentional or unique processes of change, characterizedby abrupt changes, sometimes followed by ecosystem recovery,depending on resilience and feedback mechanisms (Lambin, Geist,& Lepers, 2003; Tucker, Dregne, & Newcomb, 1991). Drivers ofrandom transitions are usually factors that act unexpectedly, such asmigration waves, land conflicts, or economic shocks (Barbier, 2000;Lambin et al., 2003). In turn, systematic transitions are driven bymore stable processes of change that evolve steadily or gradually.Drivers of systematic transitions are more permanent forces such aspopulation and market expansion, or changes in institutions thatcontrol the access to resources (Lambin et al., 2003).

Land transitions can be combined to create a high diversity ofchange trajectories (Verburg, van Berkel, van Doorn, van Eupen, &

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915 905

van den Heiligenberg, 2010), defined as temporal trends that reflectthe relationships among the factors that underpin the dynamicnature of coupled human-natural systems (Kasperson, Kasperson,& Turner, 1995; Liu et al., 2007). They can take many differentforms and depend on local circumstances (Geist & Lambin, 2002;Geist et al., 2006).

In this study we propose that the simultaneous consideration ofland transitions and trajectories allows for a better understandingof the complexity of LULC change and its long-term causes. In turn,this knowledge can contribute to improving predictions of possiblelandscape evolutions (Verburg et al., 2010) and establishingmanagement priorities for those landscape areas most vulnerableto change (Mena, 2008).

We selected a study area on Chiloé Island within the ValdivianTemperate Rainforest Ecoregion to assess the diverse compositionof transitions and trajectories involved in native forest change overa 31-year period (1976e2007). Unlike previous studies conductedelsewhere, focusing on land transitions and land trajectories(Braimoh, 2006; Mertens & Lambin, 2000; Pontius et al., 2004;Vågen, 2006), our analysis extends the time scale and discussesthe nature of the transitions within specific trajectories. Further-more, previous studies conducted in Chile have ignored complexsequences of land cover changes, as they have only assessed theconversions from one category to another (Díaz, Nahuelhual,Echeverría, & Marín, 2011; Wilson, Newton, Echeverria, Weston,& Burgman, 2005). The underlying assumption in these studies isthat the change (e.g. deforestation or forest re-growth) ispermanent.

Chiloé, in southern Chile, offers a scenario suited for studies thatunderpin our understanding of the complex interactions betweenLULC change and transformation of the rural landscape in devel-oping countries outside the tropics. Until the early 1970’s, the study

Fig. 1. Study area in Ancud Municipality

area remained largely isolated from the continent; later, in the80’s, this territory was strongly affected by globalization pressures.At present, a dichotomy exists between a development strategybased on the expansion of large-scale industrial aquaculture(salmon and mussel farming) and an endogenous developmentstrategy based on cultural heritage and local tourism (Díaz et al.,2011).

In 2008, Chiloé Island was proposed by FAO as one of the eightpilot sites of Global Importance Agricultural Heritage Systems(GIAHS) (FAO, 2008) for its outstanding land use system andlandscape, and its rich biological and cultural diversity. Traditionallandscapes like this are changing at an increasing rate and, conse-quently, important cultural heritage is being lost (Antrop, 2005),with LULC change being one of the main causes.

As far as we know, this research represents the longest time-span and most detailed spatio-temporal analysis of forest coverchange conducted in southern Chile. Furthermore, it is part of anintegrated research project aimed at understanding land transi-tions and trajectories as a way to improve our knowledge about therelations between coupled human-natural systems with potentialconsequences for rural livelihoods and sustainability (Marín,Nahuelhual, & Echeverría, 2011).

Overview of the study area

Themunicipality of Ancud (73� 150 and 74� 150 Wand 41� 500 and42� 150 S) is located in the northern portion of Chiloé Island (Fig. 1)in southern Chile and is part of the Valdivian Temperate RainforestEcoregion (Di Castri & Hajek, 1976). Chiloé Island is one of theVavílov centers of crop diversity origin such as, for example, potato(Solanum tuberosum), with around 200 documented varieties ofnative potatoes still managed today. Furthermore, the Island has

, on Chiloé Island, Southern Chile.

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915906

been classified as one of the 25 priority areas for ecosystemconservation in the world (FAO, 2008).

Ancud covers a territory of 172,400 ha of which less than 1% isclassified as urban. During the last decades Ancud has experiencedrapid population expansion, particularly during the 80’s with anannual growth rate that reached 2.4% between 1982 and 1992. Inturn, the rural population has decreased continuously from 41.9%(12,325 people) in 1982 to 37.1% (13.921 people) in 1992 and 31.7%(12,654 people) in 2002 (INE, 1982, 1992, 2002). At present, of thetotal population (39,946 people) 31.7% is rural (INE, 2002).

With an annual temperature of 9.9 �C, annual rainfall of3046.8 mm, and 74.3% of soils of forest aptitude, the study areapresents serious restrictions for agriculture. A recent studyconcluded that 94% of the farm properties in Ancud (2746 farms)correspond to peasant agricultural systems (subsistence and multi-functional small-scale farms), which manage reduced amounts ofland, pastures, and livestock. The remaining 6% consists of forestry-based systems (28 farms) and more specialized dairy farms (137farms) (Carmona, Nahuelhual, Echeverría, & Báez, 2010).

A large part of the municipality is covered by native forests(Carmona et al., 2010), of which 11,776 ha are publicly protected byChiloé National Park (Fig. 1). These temperate rain forests arecharacterized by their high degree of endemism, including relicts ofancient biotas largely lost or transformed by Pleistocene climatechange (Armesto, León-Lobos, & Kalin, 1996; Villagrán & Hinojosa,1997). A recent history (less than 200 years) of widespreadburning and logging to clear land for pastures and selective loggingof many forest patches (Torrejón, Cisternas, & Araneda, 2004;Willson & Armesto, 1996) has resulted in a rural landscape whereold growth forest stands are part of a mosaic of secondary forests,shrublands, exotic tree plantations and artificial grasslands.

Methods and data

The methodology consisted of three steps that are describedbelow.

LULC change assessment and construction of transition matrices

To assess LULC change we used Landsat scenes from 1976 (MSS),1985 (TM), and 1999 (ETMþ), which had been previously classifiedby Echeverría (2005), and a scene from 2007 (ETMþ) classified byCarmona et al. (2010). Image classification details and accuracyassessment measures for these four images can be found inEcheverría, Coomes, Hall, and Newton (2008), Carmona et al.(2010), and Díaz et al. (2011).

Based on the previous analysis, the following categories of landcover were identified: (i) agricultural land, including crops andpasture land (APL); (ii) shrubland (SH), corresponding to a landcover type where trees cover less than 10% and shrubs coverbetween 10% and 75% of the area; (iii) arboreous shrubland (ASH),which is an intermediate successional stage between shrublandand secondary forest; (iv) secondary forest (SF), resulting fromnatural or anthropogenic disturbances, with a more homogeneousstructure and age composition, characterized by a tree crown coverover 25% and under 75%; (v) old growth forest (OGF) characterizedby a tree crown cover over 75%, more heterogeneous in structuralcomposition, canopy cover and age (CONAF et al., 1999); vi) exotictree plantations (PL), composed almost exclusively of Eucalyptusspp. whichmeet a minimum area requirement of 0.5 ha, tree crowncover of at least 25% of the land area, and a total height of adulttrees above 2 m (FAO, 1996); and vii) other uses (OU), mainly urbanland. Since the research focused on forest transformation, forestcover was masked, while pixels without forest cover in any givenyear were excluded from the analysis. The spatial resolution used

was 30 � 30 m and all land cover maps were analyzed to assessLULC change.

The analysis was developed between consecutive satelliteimages using transition matrices, which is the most conventionalmethod for assessing land cover change. A filter of 0.25 ha wasapplied in the assessment of LULC change for two reasons: i) toavoid classification errors (accuracy of satellite images rangedbetween 88.8% and 90.1%); and ii) because 0.25 ha was consideredas the minimum feasible farming size on which a land holdermakes land use decisions; land cover changes in areas of less than0.25 ha may be attributable to other reasons (e.g. natural tree fall).Land use and cover change analysis was performed using LandChange Modeler ArcGIS (9.3) extension.

Transition matrices represent the area of landscape that suffersa transition from class i to class j between two consecutive images(Pontius et al., 2004). The proportion of the landscape that indicatesa class i at t1 is represented by Ciþ and is given by the followingequation:

Ciþ ¼Xni¼1

Cij isj (1)

where Cij indicates the proportion of the landscape that experi-enced a transition from class i to class j between t1 and t2. The maindiagonal element represented by Cjj indicates the proportion ofland classes that exhibits persistence of class j. In turn, theproportion of the landscape that indicates a class j at t2 is repre-sented by Cþj and is given by the equation:

Cþj ¼Xnj¼1

Cij isj (2)

The analyses that emerge from this matrix are net change andswap change. The latter indicates a change in the location ofa category, while the quantity remains the same. It is possible, forexample, that change occurs in such a manner that gains and lossesof a certain class are equivalent, and, consequently, the net changewould be zero, which might conceal the true dynamics of thelandscape. Hence, the concept of swap change allows us to avoidunderestimating the total change on the landscape. The amount ofswap was calculated as in equation (3) and total change for eachcategory was represented by net change (gains minus losses) andswap change.

Siþ ¼ 2�min�Cjþ þ Cjj; Cþj � Cjj

�isj (3)

Also annual rates of change were calculated using the formulaproposed by FAO (1996):

P ¼�

100t2 � t1

�� ln

�S2S1

�(4)

where P corresponds to the annual percent of change of a singleland cover; S1 represents the area of the specific land cover underanalysis in t1; and S2 represents the area of the specific land coverunder analysis in t2.

Identification of random and systematic transitions

An in-depth analysis of the classic transition matrix was usedto separate random and systematic transitions in each time interval(1976e1985; 1985e1999; and 1999e2007). We followed the threesteps proposed by Braimoh (2006). The first step computesthe expected loss using the formula proposed by Pontius et al.(2004) as:

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915 907

Liþ ¼ 2�min�Cjþ þ Cjj; Cþj � Cjj

�isj (5)

Eq. (5) assumes that the loss of each class and the proportion ofeach class in the second period are given a priori. The loss is thendistributed in each row across the other classes relative to theirproportions in t1.

The second step computes the differences between theobserved and expected proportions under a random process oflosses equivalent to those of gains. The higher the positive differ-ence between observed and expected proportions for the transitionbetween classes X and Y, the higher the inclination of class X tosystematically lose to class Y. In turn, the higher the negativedifference between the observed and expected, the higher theaversion of class X to systematically lose to class Y. In the statisticalsense, a land cover category is said to gain randomly from others ifthe gains are in proportion to the availability of those losing cate-gories. Similarly, the land cover category is said to lose randomly toothers if such losses are in proportion to the size of other gainingcategories. Any large positive or negative deviation from thoseproportions is referred to as a systematic landscape change(Braimoh, 2006). The third step, computes a ratio equal to thedifference between the observed and expected value divided by theexpected value, which represents how systematic a transition is inrespect to its expected value.

Both the magnitude (ha) of positive or negative deviations fromzero between observed and expected values and the aforemen-tioned ratio were used to identify the most systematic transitions(Pontius et al., 2004).

Temporal composition of land trajectories and cluster analysis

From a methodological standpoint, trajectories are the temporalsequences of land cover classes at the pixel level which aredescribed through a time series of classified satellite images (Mena,2008). Assessing the composition of LULC trajectories impliesexploring the landscape’s temporal dynamics by describingpatterns of change (e.g. cycles) represented within the satelliteimage time series (Mena, 2008).

The trajectories were identified through a pixel-historyapproach and incorporated into a cluster analysis with thepurpose of: i) identifying sectors that could be grouped according tothe proportion of specific trajectories and transitions that theycontained; and ii) characterizing these sectors based on relevantspatial features reflecting the impact of human activities. The spatialfeatures selected were landscape fragmentation, as measured byShanon’s Entropy index which has proved to be useful to assess thedegree of fragmentation (Pavanaguru & Latha, 2006), and farmmanagement practices, represented by a spatially-explicit typologyof Ancudmunicipality’s farming systems (Carmona et al., 2010). Therelationship between landscape composition and configuration canbe expected to change with increasing human activity for tworeasons. First, human activities introduce new land covers intolandscapes, thus affecting landscape composition. Second, humanstend to create patches in landscapes (e.g., crop fields, clear-cuts)that are scaled to human activities which constrains the scale ofpattern development and the resulting landscape configuration(Proulx & Fahrig, 2010).

Shanon’s Entropywas incorporated into cluster analysis throughentropy layers for each period of analysis which were created usingthe Land Change Modeler extension (ArcGis 9.3). Shanon’s Entropy(E) was calculated over the local neighborhood of each pixel usingthe formula:

E ¼Xp� lnðpÞ

lnðnÞ (6)

where p is the proportion of each class within the neigborhood,ln is the natural logaritm and n is the number of classes. Shanon’sEntropy indicates the level of disorder (increase in entropy) due tothe partitioning of the vegetation into spatially separated patches.In simple terms, patches consist of communities surrounded bya matrix with a dissimilar community structure or composition.According to equation (6), Shannon’s Entropy (E) ranges from 0 to 1,with 0 indicating a case where landcover is uniform within theneighborhood (minimum fragmentation) and 1 indicating themaximun diversity possible within the neigborhood (maximumfragmentation).

Due to the resolution of the Landsat images (0.09 ha), Shannon’sEntropy was calculated with a 7� 7 moving window. This selectionwas supported by image accuracy data and field data obtained byEcheverría’s (2005) study which reports patch areas between0.45 ha and 42,785 ha. This filter was expected to improve thequality of the result by eliminating spurious data.

Farming systems, as a level of organization, mediate the rela-tionship between farmers and the landscape by bringing togetherbiophysical, economic, social, cultural, and political factors thatconstrain or promote LUCC (Dixon, Gulliver, Gibbon, & Hall, 2001;Duvernoy, 2000; Nyssen, Getachew, & Nurhussen, 2009; Thenail& Baudry, 2004). To account for these factors, cluster analysisincluded a spatially-explicit typology of farming systems previouslyconstructed by Carmona et al. (2010) based on farm size, amount ofdairy livestock, months of milk production, size of sheep livestock,total animal carrying capacity, number of non-paid workers, area ofnative forest and shrubland, area of pastures, proportion of agri-cultural land (pastures and crops) with respect to total farm area,and people living on the farm. This typology included four farmtypes, each of them representing a “typical” farming system andcomprising 57% (type I subsistence farms), 37% (type II multifunc-tional farms), 1% (type III forest farms) and 5% (type IV specializeddairy farms) of the 2746 farms in the Ancud municipality. Property-level analysis provided us with insights into landscape change notavailable by landscape-level analysis.

Cluster analysis was carried out using a multivariate statisticalanalysis conducted in two steps: i) factor analysis through a spatialprincipal component analysis (PCA, IDRISSI extension); and ii)spatial cluster analysis (Cluster, IDRISSI extension). The clusterswere characterized based on the composition of the dominanttrajectories, the degree of landscape fragmentation, the presence ofdistinctive farming systems, and an index that defined the type oftransition for the entire period. This index was a sum of binarymaps that indicated for each period if the transitions were random(0) or systematic (1) resulting in an output layer (map) that rangedbetween 0 and 3.

To assess the cluster’s spatial distribution we applied ArcGis9.3’s Kernel density estimation tool. Kernel density estimationcalculates the 50% contour volume around point features thatbelong to a particular class, showing areas of concentration for eachclass.

Results

LULC change assessment: net change, swap and total change

The landscape area involved in forest dynamics between 1976and 2007 represented 75% of the entire rural area of the Ancudmunicipality. This is the relevant area of analysis in this study andall the results presented in this section are based on it. Given thecomplexity of LULC changes and the extent of the results derivedfrom the transition matrices, we focus on forest cover changes,particularly, on forest cover losses. Fig. 2 contains maps showingthe temporal changes in old growth forest (OGF) and secondary

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915908

forest (SF) cover over the 31-year study period, along witha graphical synthesis of the temporal changes of the main land usesand covers.

The dominant change in the 31-year period was the lossof OGF, which decreased from 79% of the landscape (100,125 ha)in 1976 to only 29% by 2007 (37,049 ha). Secondary forest (SF),

Fig. 2. Temporal changes in old growth forest (OGF) and sec

with an initial area of 13,071 ha, decreased from 10% of thelandscape in 1976 to 6% in 1985 (7115 ha) and 5% (5909 ha) in1999. However, between 1999 and 2007 this trend changed andSF increased from 5909 ha to 40,236 ha (32% of the landscape) asa result of the degradation of OGF, particularly during the thirdperiod.

ondary forest (SF) cover over the 31-year study period.

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915 909

The net reduction in OGF (sum of the net change in each period)throughout the study period was equivalent to 63,076 ha. Thisdecreasewas particularly important between 1999 and 2007where46,008 ha of OGF were lost at an annual rate of 10.1% compared tothe 11,272 ha lost in the first period at a rate of 1.3% and the 5795 halost in the second period at a rate of 0.5% (Table 1). By 2007OGF hadpersisted almost exclusively within Chiloé National Park (Fig. 2).

In contrast, SF exhibited a positive net change of 27,165 ha. Thelargest net change in SF was registered in the third period with anincrease of 34,327 ha and an annual increase rate of 24%. In fact, theoverall positive net change of SF was exclusively due to the gainswhich occurred between 1999 and 2007. During the first andsecond periods, net change in SF was negative reaching 5956 ha inthe first period and 1206 ha in the second, with annual loss rates of6.8% and 1.3%, respectively (Table 1).

During the first and second periods, total change in OGF waslargely represented by swap changes with 63% and 83%, respec-tively, while net change was 37% and 17%, respectively. In the thirdperiod, this situation reversed and net change became the mostrelevant (83%) as compared to swap (17%). Secondary forest fol-lowed the same trend, corroborating the high dynamism of forestcover in the first two periods (Table 1).

Of the OGF cover existing outside Chiloe National Park in 1976,only 12.6% persisted by 2007, whereas of the secondary forest coverin 1976 (12,194.6; 9.6% of the landscape), only 2.15% remained by2007.

Inter-category LULC change transitions

Tables 2, 3 and 4 show the transitions affecting forests in eachtime interval. For each land category, the first row shows theobserved transition value, as determined from satellite images. Thesecond row shows the expected value of losses if changes were tooccur randomly. The third row shows the difference between theobserved and expected values. The last row shows the ratiobetween this difference and the expected value.

Old growth forest dynamicsBetween 1976 and 1985 (Table 2), 62.9% of OGF persisted while

losses reached only 16.4%. These losses were related to processes offorest degradation causing the change from OGF to SF (4.2%) anddeforestation comprising the changes from OGF to ASH (6.2%), SH(2.7%), APL (agriculture and pasture land) (2.1%) and other uses(OU) (1.2%). During this period, the analysis of the transition matrixshows that when OGF lost cover, it was systematically replaced byASH and APL (deforestation), which is supported by the differencebetween the observed and expected values (4.0% and 1.7%,respectively). In turn, these percentages imply that the change ofOGF to ASH was 1.8 times higher than expected, while the change

Table 1Summary of main land categories, net changes and swap changes in hectares (regular fon(%) of each land category.

Period 1976e1985 1985e1999 1999e2007 1976e1985 1985e1999 1999e2

Land category Net change Change rate

OGF �11,272 �5,795 �46,008 �1.2% �0.5% �10.1%37% 17% 83%

SF �5,956 �1.206 34,327 �6.8% �1.3% 24%30% 14% 85%

ASH 11,726 6946 5874 16.3% 2.7% 2.9%67% 32% 19%

SH �2,030 �1,468 4040 �3.1% �1.9% 7.5%15% 15% 40%

APL 5.411 �341 3523 25.1% �0.4% 6.0%89% 4% 37%

fromOGF to APLwas 4.3 times higher than expected from a randomprocess.

The change of OGF to SF (OGF degradation) had the secondmostimportant observed value; however, the amount of change was 0.5times lower than the expected value, showing that when OGF lostcover it was not systematically replaced by SF (Table 2).

During the second period (Table 3), 54.3% of the area covered byOGF in 1985 persisted until 1999, while losses represented 16.2%.The largest losses of OGF related to the change toward ASH (8.9%)and SF (2.5%). In this period, the observed values of losses in OGFwere remarkably similar to the expected values (differences for allvalues were near 0), except for the change from OGF to APL whichwas �1.4%, implying that this change occurred 0.5 times less thanexpected. This suggests the presence of factors which slowed therate of forest clearing for agricultural expansion during this period.

Between 1999 and 2007 (Table 4), OGF persistence was 25.6%while losses reached 40.2% and gains 3.8%. As in previous periods,SF and ASH had the greatest observed losses. The differencesbetween observed and expected values for all transitions involvingOGF were negative, indicating random processes of loss, with theexception of the changes from OGF to SF and fromOGF to EP (exotictree plantations) which were highly systematic. Secondary foresthad a large positive difference of 19.5% which indicates that thechange from OGF to SF was 3.5 times higher than expected.Although the area of EP was still small compared to other landcovers, the transition of OGF to this category was highly systematic,occurring 5 times more than expected from a random process.Furthermore, this change indicates that the expansion of planta-tions was a direct cause of deforestation.

Secondary forest dynamicsBetween 1976 and 1985 (Table 2), SF losses reached 10.2%, while

persistence was only 0.2%. The transition analysis revealed thatwhen SF cover was lost, it was systematically replaced by ASH(deforestation) with a large positive difference of 4.5% betweenexpected and observed values, indicating that this changehappened 14.4 times more than expected from a random process.

Other values were near zero due to the extent of the areasaffected rather than to the small difference between observed andexpected values. Thus, for example, the change from SF to APL(deforestation) was 10.2 times more than expected but involvedonly 0.6% of the landscape.

During the second period (Table 3), all losses in SF coverexhibited small and negative differences between observed andexpected values which suggest that SF transitions were predomi-nantly random during this period.

Between 1999 and 2007 (Table 4) losses of SF to all other landcovers did not exhibit large differences between observed and ex-pected values. As in the case of OGF, while the area involved in the

t numbers; first line) and percentages (bold numbers; second line), and change rates

007 1976e1985 1985e1999 1999e2007 1976e1985 1985e1999 1999e2007

Swap Total Change

18,885 29,162 9546 30,157 34,957 55,55463% 83% 17% 100% 100% 100%

13,748 7482 6276 19,704 8688 40,60370% 86% 15% 100% 100% 100%5697 15,046 24,423 17,423 21,992 30,29733% 68% 81% 100% 100% 100%

11,378 8490 6072 13,408 9958 10,11285% 85% 60% 100% 100% 100%653 8402 5875 6064 8061 939811% 96% 63% 100% 100% 100%

Table 2Forest cover transition matrix 1976e1985. First row in bold, shows the observed transitions between categories; Second row in italics, shows the expected value for a randomtransition; Third row shows the difference between observed and expected value; Fourth row shows the ratio of how systematic a transition is in respect to the expected value.

OGF SF ASH SH APL OU EP Total 1976 Loss

OGF 62.9% 4.2% 6.2% 2.7% 2.1% 1.2% 0.0% 79.4% 16.4%62.9% 8.2% 2.2% 5.3% 0.4% 0.30% 0.0% 79.4% 16.4%0.0% �4.0% 4.0% �2.6% 1.7% 0.1% 0.0% 0.0% 0.0%0.0 �0.5 1.8 �0.5 4.3 3.3 0.0

SF 2.9% 0.2% 4.8% 1.4% 0.6% 0.5% 0.0% 10.4% 10.2%9.0% 0.2% 0.3% 0.7% 0.1% 0.0% 0.0% 10.4% 10.2%

�6.1% 0.0% 4.5% 0.6% 0.6% 0.4% 0.0% 0.00% 0.00%�0.7 0.0 14.4 0.8 10.2 10.2 0.0

Total 1985 70.4% 5.6% 12.1% 5.1% 4.8% 2.0% 0.0% 100.00% 35.4%79.4% 9.4% 3.2% 6.8% 0.7% 0.5% 0.0%�8.95% �3.8% 8.9% �1.7% 4.1% 1.6% 0.0%

Gains 7.5% 5.5% 11.6% 4.5% 4.6% 1.9% 0.0% 35.4%16.4% 9.2% 2.7% 6.3% 0.5% 0.4% 0.0%�9.0% �3.8% 8.9% �1.7% 4.1% 1.6% 0.0%

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915910

change from SF to EP (deforestation) was small, the changehappened 24.3 times more than expected, showing that theexpansion of plantations in the landscape occurred in a highlysystematic way driven by stable and permanent forces.

Temporal composition of main trajectories

A total of 247 different trajectories were identified comprisingareas that ranged from 27,714 hae0.5 ha. In this section we presentthe most relevant trajectories (35) in terms of the area involved,which comprised 75% of the landscape (Table 5). Overall, the resultsshow that the landscape has been dominated by trajectories ofdeforestation (OGF to SH) and forest degradation (OGF to SF).

The single most important trajectory, comprising 22% of thelandscape, was the persistence of OGF (T1). The second mostimportant group of trajectories (T2 to T17), representing 20.9% ofthe landscape, was comprised of early, recent and very recentdeforestation, depending if the transition occurred in the first,second or third period, respectively. Among these, early defores-tation affecting both OGF and SF represented 4.9% of the landscape,whereas recent and very recent deforestation accounted for 7.9%and 8%, respectively. It is important to notice that deforestationis represented by the loss of OGF and SF to different land coversand uses. Nonetheless, in most cases deforestation correspondedto the change from OGF to ASH (13.9%). Another important findingis the evidence of recent and very recent OGF clearing (T13 and T17)for the expansion of agriculture.

Table 3Forest cover transition matrix 1985e1999. First row in bold, shows the observed transitiotransition; Third row shows the difference between observed and expected value; Fourth

OGF SF ASH SH

OGF 54.2% 2.5% 8.9% 2.2%54.2% 3.1% 6.6% 2.8%0.00% �0.6% 2.3% �0.6%0.0 �0.2 0.4 �0.2

SF 3.4% 1.7% 0.2% 0.1%2.7% 1.7% 0.7% 0.2%0.7% 0.00% �0.5% �0.1%0.3 0.0 �0.7 �0.7

Total 1999 65.8% 4.7% 17.6% 3.9%68.9% 5.8% 14.6% 3.7%�3.1% �1.1% 2.9% �0.2%

Gains 11.6% 3.0% 11.5% 3.4%42.7% �4.0% �8.5% �3.2%

The third most important group of trajectories, comprising19.7% of the landscape, was represented by the degradation ofOGF in all periods (T24 to T26). The transition of OGF to SF ismostly caused by selective logging, where the best individuals areextracted, thus compromising the quality of the forest. It isimportant to note that the very recent change from OGF to SF (at anannual rate of 6.5%) after a long period of OGF persistence (T26)accounted for 18.5% of all forest degradation. This transition wasalso highly systematic which suggests that new, but stable, forcesare driving the degradation of large areas of OGF that until recentlyhad persisted without notorious human intervention.

Finally, the fourth group of trajectories comprising a smallpercentage of the landscape, was represented by a range of cyclicaltrajectories including cyclical deforestation (T18), deforestationreversals (T19 to T22), degradation reversals (T27 to T31), and re-growth reversal trajectories (T33 to T35).

Configuration of land trajectories and cluster analysis

Based on the results of a factor analysis, we obtained spatialcomponents for further scrutiny. These components were con-structed following the criteria of an eigenvalue greater than 0.5,and from the four variables mentioned in section 3.3 represented asinformation layers (trajectory composition, index of systematic andrandom transitions, degree of fragmentation expressed as entropyvalues, and farm types). These spatially explicit variables werecontained in three components that explained 81.9% of the total

ns between categories; Second row in italics, shows the expected value for a randomrow shows the ratio of how systematic a transition is in respect to the expected value.

APL OU EP Total 1985 Loss

1.2% 1.4% 0.0% 70.4% 16.2%2.6% 1.11% 0.0% 70.4% 16.2%

�1.4% 0.25% 0.0% 0.0% 0.0%�0.5 0.2 0.0

0.1% 0.1% 0.0% 5.6% 3.9%0.2% 0.14% 0.0% 5.6% 3.9%

�0.1% �0.05% 0.0% 0.0% 0.0%�0.5 �0.3 0.0

4.5% 3.5% 0.0% 100.00% 35.7%4.6% 2.04% 0.0%

�0.1% 1.47% 0.0%

3.3% 3.1% 0.0% 35.7% 0.00%�3.4% �1.57% 0.0%

Table 4Forest cover transition matrix 1999e2077. First row in bold, shows the observed transitions between categories; Second row in italics, shows the expected value for a randomtransition; Third row shows the difference between observed and expected value; Fourth row shows the ratio of how systematic a transition is in respect to the expected value.

OGF SF ASH SH APL OU EP Total 1999 Loss

OGF 25.6% 25.0% 10.1% 1.5% 2.3% 0.7% 0.1% 65.8% 40.2%25.6% 5.5% 0.7% 4.6% 5.3% 4.1% 0.0% 65.8% 40.2%0.0% 19.5% �10.0% �3.1% �3.0% �3.5% 0.0% 0.0% 0.0%0.0 3.5 �0.5 �0.7 �0.6 �0.8 5.0

SF 1.9% 2.2% 0.4% 0.1% 0.1% 0.0% 0.0% 4.7% 2.6%1.7% 2.2% 0.5% 0.1% 0.1% 0.1% 0.0% 4.7% 2.49%0.2% 0.0% �0.1% �0.1% 0.0% �0.1% 0.0% 0.0% 0.0%0.1 0.0 �0.2 �0.5 0.0 �0.6 24.3

Total 2007 29.3% 31.9% 22.2% 7.1% 7.3% 1.6% 0.5% 100.00% 60.3%40.5% 8.7% 30.5% 1.7% 8.4% 4.8% 0.3%

�11.0% 23.2% �8.3% 0.9% �1.1% �3.2% 0.2%

Gains 3.8% 29.7% 14.3% 5.6% 5.1% 1.3% 0.5% 0.0% 60.3%14.9% 6.5% 22.6% 0.2% 6.2% 4.81% 0.33%

�11.1% 23.2% �8.3% 0.9% �1.1% �3.49% 0.16%

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915 911

variation, each of themwith an own value greater than 0.75. Clusteranalysis indicated the existence of five clusters whose character-istics are summarized in Table 6. Fig. 3 shows the spatial distribu-tion of each cluster and the main polygons of cluster concentration.Fig. 4 shows the entropy layer for each period and correspondingentropy values for each cluster as a measure of landscapefragmentation.

Cluster 1 comprised the largest area with 49,668 ha (33.1% oftotal). This cluster’s major concentration zones were located in thenorthern part of the study area, near the city of Ancud and

Table 5Temporal composition of main forest change trajectories between 1975 and 2007.

Change trajectory Cove

T1 OGF OGF OGF OGF 27,7

T2 SF ASH ASH ASH 18,3T3 OGF ASH ASH ASH 1667T4 SF ASH ASH SH 980.T5 OGF SH APL APL 632.T6 OGF ASH ASH SH 571.T7 OGF APL APL APL 557.T8 OGF OGF ASH ASH 4711T9 OGF OGF ASH SH 1532T10 OGF OGF SH SH 1203T11 OGF OGF ASH APL 752.T12 OGF OGF SH ASH 740.T13 OGF OGF APL APL 551.T14 OGF OGF SH APL 496.T15 OGF OGF OGF ASH 8313T16 OGF OGF OGF SH 1111T17 OGF OGF OGF APL 730.T18 OGF ASH OGF ASH 660.T19 OGF ASH ASH SF 830T20 SF ASH ASH SF 549.T21 OGF OGF ASH SF 2290T22 OGF OGF ASH OGF 777.T23 OGF ASH OGF SF 1107T24 OGF SF SF SF 685.T25 OGF OGF SF SF 890.T26 OGF OGF OGF SF 23,2T27 OGF SF OGF SF 1485T28 OGF SF OGF OGF 1334T29 OGF SF SF OGF 661.T30 OGF OGF SF OGF 1202T31 SF OGF OGF SF 1049T32 SH OGF OGF OGF 772.T33 SH OGF OGF ASH 824.T34 SH OGF OGF SF 1162T35 ASH OGF OGF SF 532.

surrounding the coastline (Fig. 3). This cluster exhibited a lowpercentage of persistence (3.1%) while the predominant trajectorieswere the recent (1985e1999) and very recent (1999e2007) defor-estation from OGF to ASH, comprising 7.4% and 4.6% of the cluster’sarea, respectively. This cluster was dominated by random transi-tions which occurred mainly in the intermediate period and cor-responded to the change from OGF to SH and ASH, whereassystematic transitions accounted only for 4.7% of the cluster’s area.Main deforestation trajectories were linked to subsistence andmultifunctional farming systems, which together accounted for 71%

rage (ha) % Trajectory name

14.1 22.0 Persistence

7.4 1.5 Early deforestation.3 1.3 Early deforestation6 0.8 Early deforestation3 0.5 Early deforestation1 0.5 Early deforestation6 0.4 Early deforestation.8 3.7 Recent deforestation.9 1.2 Recent deforestation. 1.0 Recent deforestation5 0.6 Recent deforestation2 0.6 Recent deforestation5 0.4 Recent deforestation1 0.4 Recent deforestation.4 6.6 Very recent deforestation.1 0.9 Very recent deforestation3 0.6 Very recent deforestation9 0.5 Cyclical deforestation

0.7 Early deforestation-Regrowth5 0.4 Early deforestation-Regrowth.1 1.8 Recent deforestation-Regrowth2 0.6 Recent deforestation-Regrowth.6 0.9 Early deforestation-Regrowth-Degradation8 0.5 Early OGF degradation3 0.7 Recent OGF degradation90 18.5 Very recent OGF degradation.2 1.2 Cyclical OGF degradation.8 1.1 Early degradation-Regeneration9 0.5 Early degradation-Regeneration.5 1.0 Recent degradation-Regeneration.6 0.8 Regeneration-Degradation4 0.6 Regrowth-Persistence4 0.7 Regrowth-Deforestation.8 0.9 Regrowth-Degradation5 0.4 Regrowth-Degradation

Table 6Characteristics of each cluster identified in the study area according to main trajectories, transitions and ecological and socioeconomic attributes.

Attributes Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5

Total area (ha) 49,669 40,306 27,378 24,809 7762Land cover persistence (%) 3.1% 51.2% 19.7% 0% 0%Main trajectory

(proportion of cluster area)OGF-OGF-ASH-ASH (7.4%) OGF-OGF-OGF-SF (13%) OGF-OGF-OGF-SF (35%) SH-OGF-OGF-SF (0.1%) OGF-OGF-OGF-SF (94.7%)

Second most importanttrajectory (proportion ofcluster area)

OGF-OGF-OGF-ASH (4.6%) OGF-OGF-OGF-ASH 7.6% OGF-OGF-OGF-ASH 10.9% OGFeSHeOGF-ASH 0% SF-ASH-ASH-EP 1.4%

Proportion of mostsystematic transitions

4.7% 13.9% 44.8% 0% 97.5%

Proportion under subsistencefarming systems (%)

36.6% 7.9% 53.5% 38.1% 25.1%

Proportion undermultifunctional farmingsystems (%)

34.4% 10.5% 29.9% 41.4% 14.2%

Proportion underforest-oriented systems (%)

4.8% 11.7% 5.0% 1.8% 37%

Proportion under specializeddairy systems (%)

21.3% 16.2% 10.3% 16.9% 22.7%

Proportion under ChiloéNational Park

3% 56.6% 1.3% 1.7% 1%

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915912

of the farm types present in the cluster. It can be observed that themean entropy value for the first period was 0.56 with a standarddeviation (s.d.) of 0.17; in the second period the value decreased to0.53 (s.d. 0.14), whereas in the third period it increased againreaching its highest value with a mean entropy of 0.63 (s.d. 0.14).

Cluster 2 comprised an area of 40,306 ha (26.9% of total) and itwas concentrated in Chiloé National Park (Fig. 3), which largelyexplains the high persistence of OGF (51.2%). Outside the NationalPark this cluster was highly concentrated in two polygons that canbe identified clearly in the landscape (Fig. 3). In these areas, the

Fig. 3. Spatial distribution of clusters and m

cluster was characterized by the predominance of specialized dairyfarms (16.2%) and forest-oriented systems (11.7%), where the maintrajectories were very recent OGF degradation to SF (13.6%) anddeforestation from OGF to ASH (7.9%). Most of these farms heldlarge areas of native forest ranging from 802 ha to 4000 ha andexhibited some degree of forest persistence. The highest systematictransitions within this cluster accounted for 13.9% of the cluster’sarea. Entropy indices were lower than those of cluster 1, but withincreasing values over time (Fig. 4). For the first period the entropyvalue was 0.19 (s.d. 0.12) indicating a low level of fragmentation. In

ain polygons of cluster concentration.

Fig. 4. Entropy layer for each period and corresponding entropy values for each cluster as a measure of landscape fragmentation.

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915 913

the second and third periods the indices increased to 0.21 (s.d. 0.12)and 0.32 (s.d. 0.14), respectively.

Cluster 3 comprised 27,378 ha (18.3% of the total) and it wasconcentrated around Chiloé National Park, comprising zones withthe largest patches of OGF (Figs. 2 and 3). Like previous clusters, itsmain trajectories were very recent OGF degradation to SF (35%) andvery recent deforestation (10.9%), while persistence represented19.7% of the area. Systematic transitions predominated, accountingfor 44.8% of the cluster’s area. Like cluster 1, cluster 3 was charac-terized by the predominance of subsistence and multifunctionalfarming systems (53.5% and 29.9%, respectively) and it was thethird most important in terms of fragmentation level. Entropyincreased over time from a value of 0.3 (s.d. 0.17) in the first period,to 0.33 (s.d. 0.15) and 0.5 (s.d. 0.13) in the second and third period,respectively.

Cluster 4 comprised 24,809 ha (16.6% of the landscape) and wascharacterized by a complete absence of forest persistence andsystematic transitions. It was distributed widely across the studyarea, with its highest concentration at the center of themunicipalityand near main roads. The dominant trajectory was cyclical,combiningOGFre-growth fromSHandvery recentOGFdegradationto SF. The second most important trajectory was cyclical defores-tation fromOGF to SHandASH. This clusterwas characterized by thedominance of subsistence (38.1%) and multifunctional (41.4%)farming systems and reached the second highest level of entropy,but, unlike other clusters, the level of fragmentation remainedmoreconstant over time ranging between 0.4 (s.d 0.13) and 0.5 (s.d. 0.14).

Cluster 5 comprised the smallest area with 7762 ha (5.2% of thelandscape). Its distribution was similar to cluster 1, but with

a higher concentration around the city of Ancud (Fig. 3). While themain trajectory coincided with clusters 2 and 3 (late degradation ofOGF to SF) a remarkable difference is the high percentage ofsystematic transitions (97.5%). Specifically, the most systematictransitions included LULC changes that ended in exotic plantations.Like cluster 1, this cluster was characterized by the absence of forestpersistence. Simultaneously, the predominant farming systemswere forest-oriented farms (37%) and commercial agriculturalsystems (22.7%). This cluster presented the lowest level of entropyin all periods, ranging from a mean of 0.1 (s.d. 0.09) in the firstperiod and 0.3 (s.d. 0.1) in the third period.

Discussion and conclusions

This work provides an empirical assessment of LULC changedynamics in southern Chile. The results show that forest coverchange involves a series of complex trajectories, some of which arecyclical and reversible, while others are more linear and perma-nent. These diverse trajectories are consistent with a highlydynamic landscape dominated by forms of small-holder land usethat reflect heterogeneous livelihood strategies.

The in-depth analysis of the transition matrices allowed us toseparate systematic from random transitions, which revealedunexpected dynamics. Usually, in rural landscapes dominated bypeasant farming systems, forest cover loss is attributed to shiftingcultivation. Our results, however, show that native forests havebeen systematically replaced by a range of other covers and landuses over time, and that agricultural expansion is just one of thedirect causes of forest decline (Marín et al., 2011).

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915914

During the first and second periods, the main trajectories weredominated by deforestation transitions that led to the decline of oldgrowth forests and the increase of arboreous shrubland as a resultof logging practices. A remarkable finding is, however, that thetransition from old growth forest to arboreous shrubland changedfrom highly systematic in the first period to highly random in thesecond, similar to the majority of the transitions affecting nativeforest cover between 1985 and 1999. This finding suggests that thesame type of transition (deforestation in this case) can be caused byeither permanent or sudden forces that take place in the landscape.In the study area, the period of random changes (and coincidentallyof a large amount of swap change) coincides with the beginning ofthe globalization process, characterized by trade liberalizationpolicies and structural adjustment reforms which opened up theeconomy to international trade, favored international investments,and reduced the role of the state in favor of market mechanisms todrive development (Hecht & Saatchi, 2007). The arrival of salmonand mussel farming and the transnational processing industriesshows us how the globalization process manifested itself in thestudy area (Ramírez et al., 2009). During the mid-80 and 90’s, ruralmigration rates and urban population increased, thus expandingthe demand for firewood, the main product extracted from nativeforests in southern Chile. Added to this increased logging, the“woodchips exporting boom” (early 80’s tomid-90’s), led to abruptdeforestation, as indicated by the direct change from old growthand secondary forest to shrublands through clearcutting.

In the last period (1999e2007), most forest cover transitionsbecame systematic again, driven by new forces that led to differentcycles of old growth forest decline. The most systematic transitionand relevant in terms of magnitude, was the change of old growthto secondary forest at an average annual rate of 5.9%. This veryrecent forest degradation relates mostly to peasant agriculturalsystems and can be associated to an increasing firewood demandfrom an expanding population in urban areas outside Ancud(Carmona et al., 2010; Marín et al., 2011).

Another highly systematic transition during this period(although less important in terms of magnitude) was the changefrom forest and shrubland to exotic tree plantations, (almostentirely Eucalyptus spp) which increased from 8.7 ha in 1999 to615.5 ha in 2007. Of this, more than half directly replaced nativevegetation (old growth forest, secondary forest, and arboreousshrubland) and the remaining corresponded to afforestation onmarginal agricultural lands and flooded areas. The expansion ofplantation-based forestry has been pointed to as one of the threemajor continuing trends leading to the loss of remnant old growthforests in southern Chile, along with selective and stand-scalelogging and extraction of timber, firewood and woodchips fromnative trees and forest fires, predominantly as a result of humanactivities, such as land clearing and opening of forests for timberextraction (Armesto et al., 2009).

Compared to trajectories of forest loss, trajectories of forestrecovery from abandoned agricultural lands reached only 5.3% inthe area under analysis. Spontaneous abandonment of agriculturalproduction by farmers inmarginalized land has been reported to beone of the main factors behind deforestation reversal and conse-quent forest re-growth in Latin America (Grau et al., 2003, Grau,Perez, Martinuzzi, Encarnación, and Aide, 2008; Hecht, Kandel,Gomes, Cuellar, & Rosa, 2006; Farley, 2007; Izquierdo & Grau,2009). However, in most developing countries, and Chile inparticular, deforestation continues to be the dominant pattern offorest cover change (Díaz et al., 2011; García-Barrios et al., 2009;Marín et al., 2011).

Deforestation and degradation trajectories coincide withincreased forest fragmentation. In 31 years, the landscape outsideChiloé National Park has changed from one dominated by extensive

and continuous areas of old growth forest and agricultural andpasture lands to a landscape dominated by a mosaic of shrubland,arboreous shrubland and secondary forests, with small and fewpatches of remnant old growth forests.

Although our analysis cannot establish causality, we founda close relationship between fragmentation and the proportion ofsystematic transitions and farming systems; specifically, the high-est entropy indices occurred in clusters 1 and 4 which exhibited thelowest proportion of systematic transitions and the highestproportion (>70%) of peasant agricultural systems (types I and II).This suggests that forces that act suddenly in the landscape can leadto higher levels of landscape fragmentation. In general terms,a core-periphery fragmentation dynamic was observed around themain roads confirming the connection between land use andlocation of the farm property (e.g. access of labor and trans-portation costs to markets).

Overall, the results reflect the conflicting interactions betweenphysical and human systems in the study area. In this respect, a keyquestion to address is how to generate the incentives that moveindividuals from conflicting relations with their natural system,toward more sustainable landscape transitions and trajectorieswithout the regulatory presence of the government (e.g. a ban onlogging). Worldwide, land is private property and its usufruct is animportant right for the landowner, which implies its free use andalso determines its value. The forest dynamics described in thisstudy involve 2746 landowners managing between 0.5 and over4000 ha who respond to systematic economic forces such as fire-wood and industrial timber demand. If these landowners continueto degrade their forest resources at the rates observed between1999 and 2007, by 2020 few and small patches of old growth forestcan be expected to remain (Marín et al., 2011).

Simultaneously, this landscape is considered as a commonheritage and as such transgresses property boundaries. In fact, theGIAHS project is expected to help design policies which recognizeand conserve fundamental resources, an effort in which ruralcommunities play an active role and are recognized as the maincustodians of this global heritage (Koohafkan, 2009). Nonetheless,the predominant trends observed reveal that the institutionalfactors for forest management and conservation have not yetprovided a sustainable answer to forest loss, nor have theypromoted forest recovery among private landowners.

References

Antrop, M. (2005). Why landscapes of the past are important for the future.Landscape and Urban Planning, 70, 21e34.

Armesto, J., León-Lobos, P., & Kalin, M. (1996). Los bosques templados del sur deChile y Argentina: una isla biogeográfica. In J. Armesto, C. Villagrán, & M. Kalin(Eds.), Ecología de los Bosques Nativos de Chile (pp. 23e28). Santiago, Chile:Editorial Universitaria.

Armesto, J., Smith-Ramírez, C., Carmona, M., Celis-Diez, J. L., Díaz, I., Gaxiola, A.,et al. (2009). In C. Wirth, et al. (Eds.), Old growth forests, ecological studies, Vol.207 (pp. 367). doi:10.1007/978-3-540-92706-8_16.

Barbier, E. B. (2000). The economic linkages between rural poverty and landdegradation: some evidence from Africa. Agriculture, Ecosystems and Environ-ment, 82, 355e370.

Braimoh, A. K. (2006). Random and systematic land-cover transitions in northernGhana. Agriculture, Ecosystems and Environment, 113, 254e263.

Carmona, A., Nahuelhual, L., Echeverría, C., & Báez, A. (2010). Linking farmingsystems to landscape change: an empirical and spatially explicit study insouthern Chile. Agriculture, Ecosystems and Environment, 139, 40e50.

Corporación Nacional Forestal: CONAF. (1999). Comisión Nacional del MedioAmbiente: CONAMA, Banco Internacional de Reconstrucción y Fomento: BIRF,Universidad Austral de Chile, Pontificia Universidad Católica de Chile y Uni-versidad Católica de Temuco. In Catastro y evaluación de los recursos vegeta-cionales nativos de Chile. Informe nacional con variables ambientales (Cadastreand Evaluation of Native Vegetation Resources of Chile. National Report withEnvironmental Variables), Authors, Santiago, Chile (in Spanish).

Díaz, G. I., Nahuelhual, L., Echeverría, C., & Marín, S. (2011). Drivers of land aban-donment in Southern Chile and implications for landscape planning. Landscapeand Urban Planning, 99, 207e217.

A. Carmona, L. Nahuelhual / Applied Geography 32 (2012) 904e915 915

Di Castri, F., & Hajek, E. (1976). Bioclimatología de Chile (1a Edición).. Santiago: Chile.127 pp.

Dixon, J., Gulliver, A., Gibbon, D., & Hall, M. (2001). Farming systems and poverty:Improving farmers’ livelihoods in a changing world. FAO and World Bank.

Duvernoy, I. (2000). Use of a land cover model to identify farm types in the Mis-iones agrarian frontier (Argentina). Agricultural Systems, 64(3), 137e149.

Echeverría, C. (2005). Fragmentation of temperate rain forest in Chile: patterns, causesand impact. Dissertation University of Cambridge for the degree of Doctor ofPhilosophy. (Fitzwilliam College Cambridge).

Echeverría, C., Coomes, D., Hall, M., & Newton, A. C. (2008). Spatially explicit modelsto analyze forest loss and fragmentation between 1976 and 2020 in SouthernChile. Ecological Modelling, 212, 439e444.

Farley, K. A. (2007). Grasslands to tree plantations: forest transition in the Andes ofEcuador. Annals of the Association of American Geographers, 97(4), 755e771.

Food and Agricultural Organization: FAO. (1996). Forest resources assessment. 1990.Survey of tropical forest cover and study of change processes. Rome Italy: FAOForestry. Paper 130.

Food and Agricultural Organization: FAO. (2008). Terminal report, conservation andadaptive management of Globally Important Agricultural Heritage Systems(GIAHS). Rome, Italy.

García-Barrios, L., Galván-Miyoshi, Y. M., Valdivieso-Pérez, I. A., Masera, O. R.,Bocco, G., & Vandermeer, J. (2009). Neotropical forest conservation, agriculturalintensification, and rural out-migration: the Mexican experience. Bioscience,59(10), 863e873.

Geist, H. J., & Lambin, E. F. (2002). Proximate causes and underlying driving forces oftropical deforestation. Bioscience, 52, 143e150.

Geist, H., Mc Connell, W., Lambin, E. F., Moran, E., Alves, D., & Rudel, T. (2006).Causes and trajectories of land-use/cover change. In E. F. Lambin, & H. Geist(Eds.), Land-use and landcover change: Local processes and global impacts (pp.41e70). Heidelberg: Springer.

Grau, H. R., Aide, T. M., Zimmerman, J. K., Thomlinson, J. R., Helmer, E., & Zou, X.(2003). The ecological consequences of socioeconomic and land use change inpostagriculture Puerto Rico. Bioscience, 12, 1159e1168.

Grau, H. R., Perez, X., Martinuzzi, S., Encarnación, X., & Aide, T. M. (2008). Cambiossocioeconómicos y regeneración del bosque en República Dominicana. [Socio-economic changes and forest regrowth in República Dominicana]. InM. Gonzales-Espinoza, J. M. Rey-Benayas, & N. Ramirez-Marcial (Eds.), Restau-ración de Bosques en América Latina, forest Restoration in Latin America (Thirded.). (pp. 211e227) México, DF: Fundación Internacional para la Restauración deEcosistemas (FIRE) y Editorial Mundi Prensa México, (in Spanish).

Hecht, S. B., Kandel, S., Gomes, I., Cuellar, N., & Rosa, H. (2006). Globalization, forestresurgence and environmental politics in El Salvador.World Development, 34(2),308e323.

Hecht, S. B., & Saatchi, S. S. (2007). Globalization and forest resurgence: changes inforest cover in El Salvador. BioScience, 57(8), 663e672.

INE (Instituto Nacional de Estadísticas). (1982). Censo de Población y Vivienda 1982.http://www.ine.cl/canales/usuarios/cedoc_online/censos/pdf/censo_1982_volumen_I.pdf.

INE (Instituto Nacional de Estadísticas). (1992). Censo de Población y Vivienda 1992.http://www.ine.cl/canales/usuarios/cedoc_online/censos/pdf/censo_1992.pdf.

INE (Instituto Nacional de Estadísticas). (2002). Censo de Población y Vivienda 2002.http://www.ine.cl/canales/usuarios/cedoc_online/censos/pdf/censo_2002_volumen_I.pdf.

Izquierdo, A., & Grau, R. (2009). Agriculture adjustment, land-use transition andprotected areas in Northwestern Argentina. Journal of Environmental Manage-ment, 90, 858e865.

Kasperson, J. X., Kasperson, R. E., & Turner, B. L., II (1995). Regions at risk: Compar-isons of threatened environments. United Nations. Tokyo: Univ. Press.

Koohafkan, P. (2009). Conservation and adaptive management of globally importantagricultural heritage system (GIAHS). Resources Science, 31(1), 4e9.

Lambin, E. F., Geist, H., & Lepers, E. (2003). Dynamics of land use and cover changein tropical regions. Annual Review of Environment and Resources, 28, 205e241.

Liu, Z., Wang, Y., Gallimore, R., Gasse, F., Johnson, T., de Menocal, P., et al. (2007).Simulating the transient evolution and abrupt change of Northern Africa atmo-sphere ocean terrestrial ecosystem in the holocene.

Marín, S. L., Nahuelhual, L., & Echeverría, C. (2011). Projecting landscape changes insouthern Chile: simulation of human and natural processes driving landtransformation. Ecological Modelling, . doi:10.1016/j.ecolmodel.2011.04.026.

Martens, P., & Rotmans, J. (2002). Transitions in a globalising World. Lisse, Neth:Swets & Zeitlinger.

Mena, C. F. (2008). Trajectories of land use and land cover in the northern Ecua-dorian Amazon: temporal composition, spatial configuration, and probability ofchange. Photogrammetric Engineering and Remote Sensing, 74(6), 737e752.

Mertens, B., & Lambin, E. F. (2000). Land-cover change trajectories in southernCameroon. Annals of the Association of American Geographers, 90(3), 467e494.

Nyssen, J., Getachew, S., & Nurhussen, T. (2009). An upland farming system undertransformation: proximate causes of land use change in BelaeWelleh catchment(Wag, Northern Ethiopian Highlands). Soil Tillage Research, 103(2), 231e238.

Pavanaguru, R., & Latha, A. (2006). Entropy as an indicator of fragmented landscape.Current Science, 91(3), 10.

Pontius, R., Jr., Gilmore, R., Shusas, E., & McEachern, M. (2004). Detecting importantcategorical land changes while accounting for persistence. Agriculture, Ecosys-tems and Environment, 101(2e3), 251e268.

Proulx, R., & Fahrig, L. (2010). Detecting human-driven deviations fromtrajectories in landscape composition and configuration. Landscape Ecology, 25,1479e1487.

Ramírez, E., Modrego, F., Macé, J.C., Yáñez, R. (2009). Dinámicas territoriales enChiloé central: La fuerza de las coaliciones extra territoriales (Territorialdynamics in central Chiloé: the strength of extra-territorial coalitions). Doc-umento de Trabajo No 54. Programa Dinámicas Territoriales Rurales, Rimisp.Santiago, Chile (in Spanish).

Raskin, P., Banuri, T., GallopÃn, G., Gutman, P., Hammond, A., Kates, R. W., et al.(2002). Great transition: the promise and lure of the times ahead (e-book)[First published in the Encyclopedia of Earth November 26, 2008; Last revisedDate November 26, 2008; Retrieved June 23, 2011]. In C. J. Cleveland (Ed.),Encyclopedia of Earth. Washington: D.C.: Environmental Information Coalition,National Council for Science and the Environment. http://www.eoearth.org/article/Great_Transition:_The_Promise_and_Lure_of_the_Times_Ahead_%28e-book%29.

Rindfuss, R. R., Walsh, S. J., Turner, B. L., Fox, J., & Mishra, V. (2004). Developinga science of land change: challenges and methodological issues. In. Proceedingsof the National Academy of Sciences of the United States of America, 101 (pp.13976e13981).

Thenail, C., & Baudry, J. (2004). Variation of farm spatial land use pattern accordingto the structure of the hedgerow network (bocage) landscape: a case study innortheast Brittany. Agriculture, Ecosystems and Environment, 101(1), 53e72.

Torrejón, F., Cisternas, M., & Araneda, A. (2004). Efectos ambientales de la colo-nización española desde el río Maullín al archipiélago de Chiloé, sur de Chile.Revista Chilena de Historia Natural, 77, 661e677.

Tucker, C. J., Dregne, H. E., & Newcomb, W. W. (1991). Expansion and contraction ofthe Sahara Desert from 1980 to 1990. Science, 253, 299.

Vågen, T. G. (2006). Remote sensing of complex land use change trajectoriesda casestudy from the highlands of Madagascar. Agriculture, Ecosystems and Environ-ment, 115, 219e228.

Verburg, P. H., van Berkel, D. B., van Doorn, A. M., van Eupen, M., & van denHeiligenberg, H. A. R. M. (2010). Trajectories of land use change in Europe:a model-based exploration of rural futures. Landscape Ecolology, 24(2), 217e232.

Villagrán, C., & Hinojosa, L. F. (1997). Historia de los bosques del sur de Sudamérica,II: análisis fitogeográfico. Revista Chilena de Historia Natural, 70, 241e267.

Willson, M. F., & Armesto, J. J. (1996). The natural history of Chiloé: on Darwin’s trail.Revista Chilena de Historia Natural, 69, 149e161.

Wilson, K., Newton, A., Echeverria, C., Weston, C., & Burgman, M. (2005).A vulnerability analysis of the temperate forests of south central Chile. BiologicalConservation, 122, 9e21.