Landscape Ecology 14: 1–15, 1999. © 1999 Kluwer Academic Publishers. Printed in the Netherlands.

1

Fire-induced changes in northern Patagonian landscapes Thomas Kitzberger1 and Thomas T. Veblen2 ´ Universidad Nacional del Comahue, E.P. Universidad, 8400 Bariloche, Argentina; de Ecologia, of Geography, University of Colorado, Campus Box 260, Boulder CO 80309, USA

1 Departamento 2 Department

Received 20 June 1997; Revised 15 December 1997; Accepted 7 February 1998

Key words: Austrocedrus, change detection, disturbance, fire exclusion, landscape pattern, Nothofagus, Patagonia

Abstract In northern Patagonia, Argentina we quantify changes in fire frequency along a gradient from mesic Nothofagus dombeyi forest to xeric woodlands of Austrocedrus chilensis at the steppe ecotone, and we examine patterns of vegetation change coincident with the changes in fire regimes across a range of spatial scales. At a regional scale changes in land cover types are documented by comparing 1:250 000 scale cover type maps from 1913 and 1985. Changes in landscape structure are analyzed by comparing vegetation patterns on 1:24 000 scale aerial photographs taken in 1940 and 1970. Fire frequency peaked in the late nineteenth-century due to widespread burning and clearing of forests by European settlers late in the century. Subsequently, fire frequency declined dramatically about 1910 due to the cessation of intentional fires and has remained low due to increasingly effective fire exclusion. At a regional scale there has been a dramatic increase during the twentieth century in the proportion of forest cover relative to areas mapped as recent burns or shrublands in 1913. Remnant forest patches that survived the widespread late-nineteenth century burning have coalesced to form more continuous forest covers, and formerly continuous areas of shrublands have become dissected by forest. Under reduced fire frequency there has been a shift in dominance from short-lived resprouting species (mostly shrubs) towards longer-lived species and obligate seed-dispersers such as Austrocedrus chilensis and Nothofagus dombeyi. Due to limited seed dispersal of these tree species, the spatial configuration of remnant forest patches plays a key role in subsequent changes in landscape pattern.

Introduction Anthropogenic changes in fire regimes, in particular suppression of formerly frequent fires, can result in major landscape-scale changes in natural vegetation (Baker 1989, 1992; Agee 1994; Covington and Moore 1994; Turner and Romme 1994). As a consequence of fire exclusion, entire landscapes may change in terms of cover types and landscape configuration as well as patch structure and composition. Common patterns associated with anthropogenic changes in fire regimes include increases in mean sizes and ages of patches, greater patch connectivity, and increased dominance by long-lived obligate seed-dispersed species at the expense of short-lived resprouting species. To understand such changes, knowledge of stand-level processes must be combined with spatial analyses of

vegetation patterns at landscape scales. In this study, we examine changes in vegetation patterns associated with fire exclusion in northern Patagonia, Argentina across a range of spatial scales from local changes in patch boundaries (e.g., a few m2 ) to regional changes in broad cover types. Fire, both natural and anthropogenic, has been shown to have a major influence on vegetation patterns in northern Patagonia (c. 37 ◦ to 43◦ S) in the transitional area from warm temperate forests to the Patagonian steppe (Veblen and Lorenz 1987; Veblen et al. 1992a; Burns 1993). Changes in fire regimes are believed to be major determinants of cover types. Both analyses of tree population age structures and comparison of modern and historical (1880 to 1920) landscape photographs document a dramatic expansion of the conifer Austrocedrus chilensis into areas of former

2 steppe throughout northern Patagonia. This tree invasion coincides with a marked decline in fire frequency about 1900 due both to the elimination of intentional fires set by Native American hunters and increasingly effective fire suppression during this century (Veblen and Lorenz 1988; Veblen and Markgraf 1988; Veblen et al. 1992a; Kitzberger et al. 1997). To the west, in more mesic areas in the Andean foothills, mosaics of even-aged Nothofagus and Austrocedrus forests reflect massive tree establishment following extensive late 19th and early 20th century stand-replacing fires set by white settlers in failed attempts to convert forest to pasture (Veblen and Lorenz 1987, 1988). Previous studies of vegetation changes in northern Patagonia have focused on stand-level changes based on dendroecological techniques and also have used matched historical and modern landscape photographs to qualitatively document changes in landscape patterns that coincide with marked changes in fire regimes. In this study, we quantify changes in land cover at a regional scale based on comparison of 1:250 000 vegetation maps from 1913 and 1985, and the changes in patch composition and landscape structure based on comparison of aerial photographs at the scale of 1:24 000. Changes in vegetation patterns are related to dispersal and life history traits of the dominant species in the context of the region’s recent fire history.

Study area Topographically, from west to east, the study area includes the Andean cordillera, the lower foothills intersected by glacial lakes and valleys, and the Patagonian plains at c. 700 m. Throughout the region, soils are derived from ash deposited by relatively recent volcanic activity. Due to the rainshadow effect of the Andes on the westerlies, mean annual precipitation declines from c. 3000 mm at the continental divide to less than 500 mm only 80 km to the east in the steppe. A dramatic west-to-east vegetation gradient parallels the precipitation gradient. At c. 40◦ S western montane rain forests (c. 800 to 1100 m elevation) are dominated by tall evergreen Nothofagus dombeyi and in the wettest areas they also include shade-tolerant trees such as Laurelia philippiana, Saxegothaea conspicua and Dasyphyllum diacanthoides. South of 41◦ the long-lived conifer Fitzroya cupressoides co-occurs with N. dombeyi in the wettest areas (Veblen et al. 1996). Eastwards, as precipitation declines, there is an extensive zone of pure N. dombeyi with understories

of the 3 to 6 m-tall bamboo Chusquea culeou. With increasing aridity, the conifer Austrocedrus chilensis forms mixed stands with N. dombeyi and then pure stands; understories become less dense as Chusquea is replaced by shrubs and small trees such as Aristotelia chilensis and Lomatia hirsuta. Finally, near the ecotone with the steppe Austrocedrus stands form open woodlands with bunch grasses and low shrubs such as Discaria articulata and Mulinum spinosum. Midslopes and valley bottoms from 900 to 1200 m at intermediate positions along the precipitation gradient are dominated by shrublands of xerophyllous small tree species such as Lomatia hirsuta, Schinus patagonicus, Embothrium coccineum, Maytenus boaria, the deciduous Nothofagus antarctica and the bamboo Chusquea culeou. Subalpine forests of the deciduous Nothofagus pumilio occur at elevations of 1100 to 1200 m, forming the upper border of midslope shrublands and the montane rainforests. North of c. 40◦ 300 and 40◦ 200 S, respectively, deciduous Nothofagus spp. and the coniferous Araucaria araucana also form part of the forest vegetation. N. obliqua and N. alpina typically occur at mid-elevations below subalpine N. pumilio forests and above the more xeric habitat of Austrocedrus. Araucaria occurs along a gradient from xeric sites in the steppe and shrublands of N. antarctica to more mesic forests with N. dombeyi or N. pumilio. Methods Fire frequency The four areas sampled (Table 1) range from mixed Austrocedrus-N. dombeyi forests (ES and MT), pure mesic N. dombeyi forests (WT), to wet forests of N. dombeyi with Fitzroya cupressoides (LR). Mean annual precipitation ranges from c. 1200 to 2200 mm (Barros et al. 1984). In area, they range in size from 648 ha to 1701 ha and in elevation from 800 to 950 m a.s.l. In terms of disturbance by logging and grazing, areas ES and MT are highly disturbed, area WT is moderately disturbed, and area LR is relatively free from logging and livestock. For each of the four sample areas, stand-origin maps (sensu Heinselman 1973) were constructed by delineating tentative boundaries of even-aged, postfire stands as interpreted from variations in canopy densities on 1:24 000 aerial photographs. Based on field sampling (described below), these tentative boundaries were redrawn as necessary. All

3 Table 1. Characteristics of time-since-fire sampling areas. Mean annual precipitation is estimated from Barros et al. (1983). Sites are listed from dry to wetter. Area code

Location

Vegetation type

Mean annual precipitation (mm)

Topographic position

ES

East L. Steffen

1200

W-E running valley

MT

Mid L .Traful

1400

W-E running N-facing slopes

1701

WT LR

West L. Traful Lago Roca

A. chilensis forest, Nothofagus antarctica shrubland and N. dombeyi forest A. chilensis-N. dombeyi and N. dombeyi forest N. dombeyi forest N. dombeyi forest with Fitzroya cupressoides

2000 2200

W-E running N-facing slopes W-E running lake basin

1251 648

mapped N. dombeyi-, N. dombeyi-Austrocedrus and Austrocedrus-dominated post-fire stands below 950 m elevation were visited, and randomly sampled at one to four sites depending on their size. These stands accounted for 50.9% of the combined surface area of contiguous sample areas MT and WT, 40.1% of area ES, and 57.2% of area LR. After 150 to 250 years of stand development, depending on site quality, stands could no longer be identified as representing first cohorts of post-fire stand development. Such stands typically had entered into a fine-scale gap phase of reproduction and were designated as old-growth. The areas not field sampled and not included in the stand-origin maps included old-growth N. dombeyi stands, subalpine Nothofagus pumilio forests, midslope Nothofagus antarctica shrublands, bogs and grasslands, and high elevation barren lands. At each sample site within a mapped unit, one to three partial cross sections were cut from live fire-scarred trees and fire dates were determined by cross-dating the rings containing fire scars (McBride 1983). Five to eight of the largest trees from the postfire cohort were cored to the pith at 20 cm above the ground for age determination. Ages at coring height were matched with the closest precisely dated fire scar which usually was 5–15 years younger because of the time required for seedlings to grow to a coring height of 20 cm. Thus, approximate maximum ages of postfire cohorts were used to confirm the stand boundaries interpreted from aerial photographs that were linked to precise fire dates from fire scars. Where no fire scars occurred in a sample unit, the fire ages estimated from cohort ages (i.e., age at coring height plus 10 years for the time to reach coring height; Veblen and Lorenz 1987) are only accurate to ± 10 years. Releases and

Area (ha) 889

suppressions on remnant trees were also used to confirm dates and stand boundaries of post-fire cohorts (Lorimer 1986). Stand boundaries and origin dates were transferred to topographic maps and digitized into a Geographic Information System (IDRISI, Eastman 1990) which was used to determine total areas burned at different dates. A graphical procedure was used to describe timesince-fire (TSF) distributions (Van Wagner 1978; Johnson and Van Wagner 1985). Starting from the date of the oldest post-fire stand, cumulative percentage area burned was plotted on semi-log paper. On the semi-log plots, homogenous fire frequency periods are represented as straight lines. Sharp changes in slopes of the lines indicate changes in fire frequency. Broad-scale land-cover changes Two large areas (113 392 ha and 87 564 ha) in south and central Nahuel Huapi National Park (SNH and CNH) and two smaller areas (22 030 ha and 29 021 ha) in south and central Lanín National Park (SL and CL) were selected for assessment of land-cover changes that occurred between 1913 and 1985 as depicted in two vegetation maps (Figure 1). The vegetation maps used were a 1:250 000 resource survey map compiled in 1913 (Willis 1914) and 1: 250 000 National Park Service floristic-structural cover type maps for Lanín and Nahuel Huapi National Parks derived from aerial photographs and satellite images from the 1980s (Mermoz and Martín 1986). Land cover on the 1913 map was classified for land-use purposes and, therefore, lacked detail on floristic composition. The six main land-cover types mapped in 1913 were: virgin forests, brushlands (that the author believed had developed after fires), re-

4 ground resolutions of 200 × 200 m and 70 × 70 m for the two larger and two smaller areas, respectively. Because of minor distortions in the historical map, registration accuracy between historical and modern maps was not satisfactory. To improve point-to-point correspondence, 15 to 20 landmarks that served as control points in each map pair, were registered to the modern map grid using a quadratic fit and nearest neighbor interpolation for the resampling of the landcover class to the output image. This co-registration resulted in root mean square (R.M.S.) errors between images that ranged from 127 to 185 m. Given these high R.M.S. errors, in our interpretation of the results we focus only on major changes in cover types. Percentage changes in areas mapped as forest, shrubland, burn, and grassland were calculated for both dates based on the area surveyed in 1913 (i.e., nonsurveyed areas in 1913 were masked out of the modern map). Finally, resampled 1913 images and modern map images were cross-classified on a pixel-by-pixel basis for the analysis of land-cover changes. The statistical significance of association among transitions was assessed by contingency analysis (Legendre and Legendre 1983). Figure 1. Maps of northern Patagonia showing areas included in the analysis of 1913 to 1985 land-cover change (cross-hatched areas) and areas sampled for fire history (WT, MT, LR and ES); see Table 1 for area codes). The solid box inside area WT is the site of the comparison of 1940 and 1970 aerial photographs.

cently burned areas, grasslands, agricultural lands and alpine or non-surveyed areas (Willis 1914). Preliminary ground checking indicated that most of the areas mapped by Willis as ‘virgin forests’ correspond to old-growth dominated by Nothofagus dombeyi or Austrocedrus chilensis and N. dombeyi. However, stand structure data (Veblen and Lorenz 1987) also indicate that some of the ‘virgin forests’ were actually old post-fire stands at the time of mapping. The brushland cover-type includes N. antarctica shrublands as well as shrublands dominated by other small trees and shrubs. Steppe communities are included in the grassland cover type, and agricultural areas correspond to sites of former forest, shrublands, bogs or steppe that were converted to pastures or crop land. ‘Alpine/nonsurveyed’ included mainly inaccessible sites of alpine vegetation, barren areas, or N. pumilio shrublands. The mapped polygons of the four areas in the modern and historical maps were digitized on a Calcomp 9600 tablet. Vector files were rasterized to pixel

Changes in the landscape mosaic Changes in landscape mosaics of forest, shrubland and grassland patches were analyzed by comparing 1:24 000 aerial photographs taken in 1940 and 1970 of a submesic area near the southwestern coast of Lake Traful in northern Nahuel Huapi National Park (Figure 1). The 1940 and 1970 air photos were taken at approximately the same time of day during late spring to early summer, at the same altitude, and had principal points that were only 1.8 km distant from each other on the ground (i.e., were taken from a similar airplane position). This implied that distortions inherent in the non-rectified aerial photographs and shading patterns in both images were nearly the same. Aerial photographs were scanned at a resolution of 300 dots per inch corresponding to a ground pixel resolution of 2 × 2 m. Once displayed on screen, two study areas: one coastal area covering 21.33 ha characterized by abundant forest cover (area LTC), and a contiguous inland area of 45.67 ha with a much less continuous forest cover (area LTI) were selected from the 1940 and 1970 images. To allow precise co-registration, the 1940 images were resampled on the corresponding 1970 images as follows. For each 1940–1970 pair the RESAMPLE module of the IDRISI GIS pack-

5 age was run based on 15 common ground points, fitting a quadratic equation and resampling the 1940 pixels on the 1970 grid system with bilinear interpolation (Eastman 1992). This procedure resulted in root mean square (R.M.S.) errors between 1940 and 1970 images of 1.31 m and 1.37 m for LTC and LTI, respectively. The possibility of co-registration errors of this magnitude is considered in the interpretation of the results. Three main vegetation cover types were classified a priori from the images: closed forests, shrublands and grasslands. For each cover class, training areas of known vegetation were digitized on both the 1940 and 1970 images. The distributions of digital numbers were fit to normal distributions and cutoff values for each cover class were determined based on 95% critical values. This calibration was performed separately for the 1940 and 1970 photographs to account for different tones in the photographs. Thematic accuracy of this classification was assessed by field checking 45 random points in the coastal area. Each point on the ground was classified as forest, shrubland or grassland based on percent cover of arboreal, shrub and herbaceous species and then compared in a confusion matrix with the classification from the aerial photographs. These ground observations indicated that classification of sites into one of the three cover types on the 1970 aerial photographs was accurate in 84.4% of the cases (Overall Index of Accuracy, Story and Congalton 1986). Structural changes in the landscape were evaluated by comparing a series of metrics that quantify landscape composition and pattern (McGarrigal and Marks 1993). Changes in cover types were analyzed by comparing the Percentage Area (PC) and Patch Density (PD) in 1940 and 1970. Changes in patch size distribution were assessed by computing the Largest Patch Index (LPI) (i.e., the percent area accounted for by the largest patch), and by comparing rankordered distributions of patch sizes in 1940 and 1970. Changes in the degree of dissection of the landscape were quantified with Edge Density (ED) (i.e., the total length of edges of a given cover class per hectare). The Area Weighted Mean Shape Index (WSI) and the Area Weighted Mean Patch Fractal Dimension (WFD) were used to assess changes in the complexity of patch shapes. WSI equals 1 when all patches are square and increases when patches become more irregular. WFD values range from 1 to 2 for patch configurations varying from simple patch shapes (e.g., circular) to highly convoluted patch perimeters, respectively.

Proximity of the same patch types was quantified with the Mean Proximity Index (MPI) using a 20-m search radius. MPI equals 0 when no patch has a neighbor of the same class within 20 m and increases as patches become less isolated from patches of the same class (McGarrigal and Marks 1993). To examine pixel-to-pixel changes in cover class, we cross-tabulated (CROSSTAB routine in IDRISI) the classified 1940 and 1970 images in both areas. This yielded: (1) a 3 × 3 cross-classification table that indicates class-to-class transition areas, and (2) a spatially referenced cross-classified image that contained the location of the transition areas. Forested areas in 1940 were assumed to be the seed sources from which establishment of seed-dispersed arboreal species occurred. Thus, an image was generated that represented the distance between each cell and the nearest 1940 forest patch (DISTANCE routine in IDRISI). This image was then cross-classified at 2.5 m intervals with the 1940–1970 transition image to obtain the distribution of areas of transition classes as a function of increasing distance from assumed seed sources. Statistical significance of associations among certain transitions and distance classes was assessed through contingency analysis. Results Regional fire frequency changes Time-since-fire (TSF) distributions for individual sample areas ES, MT, WT and LR showed similar trends, and thus stand age data of these areas were combined to represent temporal trends in regional fire frequency in the Nothofagus- and Austrocedrusdominated forests. This composite TSF distribution shows a temporally mixed regime with a sharp increase in fire frequency beginning in the late 19th century, followed by reduced fire frequency during the 20th century (Figure 2a). Differences in the slopes of regression lines suggest that fire frequency declined more 7-fold from approximately 1880–1910 to 1910–1980 (Figure 2b). The greater fire frequency of 1880–1910 coincides with massive forest clearing and burning by white settlers (Willis 1914), and the subsequent decline in fire frequency reflects both reduced intentional burning as well as increasingly effective fire exclusion after the creation of national parks in the region (Veblen et al. 1992a; Kitzberger et al. 1997). The vegetation changes analyzed below represent responses to the long-term effects of

6

Figure 2. Time-since-fire distributions for Nothofagus dombeyiand Austrocedrus-dominated forests (pooled sample areas ES, MT, WT, and LR) for the periods 1677–1972 (a) and 1882–1972 (b). Deviations of the distributions from straight lines on semi-log paper indicate temporally mixed fire frequency distributions (i.e., changes in fire frequencies at inflection points) as illustrated by the regression lines for the pre- and post-1910 periods in (b).

– – – – –

SL

100 100 100 100 100 100 100 100

– 19.9 66.7∗ 0.0 13.4∗ – 43.6∗ 35.1∗ 0.3 21.0 0.0 82.4∗ 0.0 16.6∗ 1.0 – – – – –

100 100 100 100 100

– – – – – 26.7 67.8∗ 5.5 0.0 0.0 1.3 59.3∗ 17.2 9.1 13.1 – 54.3∗ 41.1 4.6 0.0 – 98.4∗ 0.0 0.0 1.6 32.3∗ 7.4 55.5∗ 2.2 2.6 5.3∗ 0.0 80.3∗ 0.0 14.4 – 29.2∗ 69.5∗ 1.0 0.3

100

CL∗∗ SNH CNH SL CL SNH CNH SL

100

SNH

– 28.2∗ 64.0∗ 1.2 6.6

100

CL SL CNH 1913 state Recent burns Brushlands Virgin forests Agricultural lands Others/non surveyed

100 Total

37.3∗ 28.2∗ 14.6 8.9∗ 11.1 0.1 62.4∗ 3.4 27.6∗ 6.5 – 65.0∗ 15.6 19.4∗ 0.0 – 78.4∗ 13.1 7.9∗ 0.6 11.5 13.0 32.9∗ 0.0 42.6∗ 0.8 5.1 45.0∗ 0.0 49.1∗

N.pumilio forests CL

fire exclusion following the earlier period of frequent burning.

∗ P < 0.01. ∗∗ both N.pumilio and N. obliqua forests.

CNH

SNH

N. antarctica woodlands and shrublands CL SL CNH SNH A. chilensis woodlands A. chilensis forests N. dombeyi forests Present vegetation

Regional land-cover changes Relative proportions of land-cover types have changed markedly in northern Patagonia over the period included in our comparison of vegetation maps from 1913 to 1985. There has been a substantial increase in forest area that is consistent for all four areas examined (Figure 3). In the two northern areas (CL and SL), which were settled and burned 20 to 30 years earlier than the southern areas, this increase in forested

Table 2. Present (1985–1986) woody vegetation types according to their states in 1913 (%). Areas are arranged from north to south. CL = central Lan´in N.P., SL = south Lan´in N.P., CNH = central Nahuel Huapi N.P. and SNH = south Nahuel Huapi.

7 area clearly has been at the expense of post-fire shrublands that have succeeded to forest. In the two southern areas fires occurred more recently including immediately prior to and during the 1913 mapping by Willis. Other shrublands also were created by burning after the date of the map, so that in these areas there has been a less dramatic shift from shrubland to forest. Also, in the south many burned sites were mapped as ‘recent burns’ whereas in the northern areas recent burns were scarce at the time of the mapping because they had already succeeded to shrublands explaining the scarcity of recent burns in areas CL and SL in 1913 (Figure 3). Agricultural and pasture lands also declined from 1913 to 1985 over the entire region. Percentages in Table 2 indicate the 1913 state (i.e., cover type) of present vegetation types and the statistical significance of transitions from one type to another. Given that some differences between the maps may be attributable to co-registration errors, we emphasize only the major trends. In the four study areas, from 33 to 80% of present N. dombeyi and N. pumilio forests were mapped as virgin forests in 1913 which indicates that to some degree these forest types remained unchanged. However, large areas of present N. dombeyi forest also originated after burning. In the south, where burning was more recent, there are significant transitions from recent burn to N. dombeyi forest, and in the north from brushland to N. dombeyi forest (Table 2). Austrocedrus forests show a more uniform tendency of developing from pre-existing shrublands. In the one area (CNH) that includes Austrocedrus woodlands, there is also a significant transition from agricultural land (probably pasture) to woodland. Nothofagus obliqua forests, included only in area CL, also tend to have developed from brushlands. Strikingly, only low percentages of modern N. pumilio forests developed from recent burns or brushlands. Finally, in the four study areas 28 to 78% of modern N. antarctica shrublands and woodlands were also brushlands in 1913 which suggests that at many sites these are stable communities and/or, if they are seral to forest, repeated fires have prevented succession to forest. Structural changes in the landscape Major changes in patch size, shape and spatial arrangement have resulted in substantially different landscapes over the 30 years spanned by the 1940 and 1970 aerial photographs (Table 3). Due to inaccuracies in the co-registration of the photographs, only the major trends are interpreted. Both LTC and LTI

changed from landscapes dominated by shrublands to landscapes dominated by forests. Between 1940 and 1970, in the coastal (LTC) and inland (LTI) areas the percent area covered by closed forest (PC) increased 2.5 to 4.3 fold, respectively. In contrast, shrubland areas declined by factors of 2.6 and 1.5, respectively. Grasslands and open areas show 1.5 and 2.9-fold reductions in area, respectively. The largest continuous patch (LPI) of forest increased from 5.4% to 68.1% in LTC, and from 3.1% to 21.0% in LTI (Table 3). The landscape at LTC changed from one of dense (PD=1812 patches/ha), highly dissected small forest patches (LPI=5.4%) immersed in a shrubland matrix (LPI=64.7%) in 1940 to one of dense shrubland patches (PD=2136 patches/ha) surrounded by almost continuous forests (LPI=68.1%) in 1970. Grassland areas in LTC remained disjunct and dissected over the 1940 to 1970 period. In the more xeric LTI site, shrublands remain as the background-forming class (i.e., shrubland LPI is highest among cover classes), but forest patches increased considerably in size as indicated by the change in LPI from 3.1 to 21%. Grasslands in LTI were relatively important in 1940 but became highly dissected by 1970 as indicated by low LPI values and high patch density (Table 3). Hierarchical changes in matrix-forming patches are evident when patch-size distributions are arranged from large to small (Figure 4). In the LTC area, ranked patch size distributions of forest and shrubland reversed their patterns from 1940 to 1970. Single c. 200 000 m2 patches of shrubland and forest characterize the 1940 and 1970 landscapes, respectively. All other patches were at least 200 times smaller (< 1000 m2 ) in both years. In LTI, changes between 1940 to 1970 are much less pronounced, and shrublands continue to be the background matrix (Figure 4). The 1940 structure is characterized by a large difference between the largest shrubland patch (c. 300 000 m2 ) and the next largest patches (< 600 m2 ). In 1970, this difference is somewhat reduced (c. 180 000 m2 vs. < 6000 m2 ; Figure 4). Similar, but less pronounced trends, characterize the change in size sorting of grassland patches (from more hierarchical to less hierarchical size distributions). For both LTC and LTI, the dominant transition during the 1940 to 1970 interval is the shift from shrublands to closed forests: 43.4 and 36.4% of the total areas changed from shrubland to closed forest in LTC and LTI, respectively (Table 4). This suggests that either some advance tree regeneration was already present in the shrublands in 1940 or trees

8

Figure 3. Percentages of areas in different land cover proportions in 1913 (open bars) and in 1985 (solid bars) for selected areas of Lan´in and Nahuel Huapi National Parks. Area codes are: CL = central Lan´in N.P., SL = south Lan´in N.P, CNH = central Nahuel Huapi N.P. and SNH = south Nahuel Huapi N.P.

subsequently established beneath the shrub canopy or both. In contrast, grasslands appear to have a less dynamic behavior as only c. 4% of the grasslands in each area changed to forest and 3.7 to 15.8% of the grassland changed to shrublands (Table 4). Of the area that was grassland in 1940, 39% and 70% remained in the grassland state in 1970 in LTC and LTI, respectively (not shown in Table 4). Finally, shifts towards physiognomically lower cover types (i.e., forest to grassland and shrubland to grassland transitions) were minor, and probably reflect limited tree cutting followed by grazing. Measures of patch-shapes and proximity indices are consistent with the dissection of the landscape (Table 3). Forest patches change from simple nearcircular shapes in 1940 (WFD = 1.28 in both LTC and LTI) characteristic of island patches to more convoluted patches in 1970 (WFD = 1.48 and 1.47 in

LTC and LTI, respectively) as a result of the expansion and fusion of formerly disjunct patches. In contrast, shrubland patches maintain approximately the same shape complexity (WFD) suggesting that patch dissection is not simply the reverse spatial process of patch growth. In other words, patch expansion is generally a circular process except when patches fuse to create more irregular shapes. In contrast, when matrix patches are dissected by expanding patches the resulting smaller patches do not adopt simple circular shapes but retain most of their initial shape complexity. As a consequence of the process of patch expansion and fusion and matrix dissection, mean proximity increases dramatically among forest patches and conversely decreases among shrubland patches (Table 3). The net result of patch changes from 1940 to 1970 has been the differentiation of the LTC and LTI land-

9 Table 3. Change in landscape metrics from 1940 to 1970 for forest, shrubland and grassland patches in the coastal (LTC) and inland (LTI) study areas at the west end of Lake Traful. Landscape index

Landscape composition Class area (%) (PC)

Patch density and size distribution Patch density (patches/ha) (PD) Largest patch index (%) (LPI)

Edge and shape metrics Edge density (m/ha) (ED) Area weighted mean shape index (WSI) Area weighted mean patch fractal dimension (WFD) Landscape pattern Mean proximity index (MPI)

Cover class

LTC 1940

1970

LTI 1940

1970

forest shrubland grassland

26.6 66.1 7.3

69.8 25.2 5.0

11.6 60.1 28.3

50.5 39.7 9.8

forest shrubland grassland forest shrubland grassland

1812 378 978 5.4 64.7 1.3

434 2136 1245 68.1 5.5 0.3

1402 892 904 3.1 57.4 9.2

1292 1581 1426 21.0 30.5 0.8

forest shrubland grassland forest shrubland grassland forest shrubland grassland

993 1387 394 3.57 22.18 2.66 1.28 1.51 1.27

1300 1631 337 20.07 9.00 2.01 1.48 1.47 1.24

553 1502 948 4.18 30.16 8.53 1.28 1.53 1.41

1448 2028 583 16.23 36.94 3.12 1.47 1.57 1.30

forest shrubland grassland

404 6387 90

8350 622 45

210 9081 1094

3932 3973 106

scapes. In 1940 they differed only slightly but by 1970 differed radically in structure and pattern. In 1940, LTC had only 30% denser patches and 15% more initial forested area, and both areas had similar patch size hierarchies as indicated by LPIs and ranked size distributions (Table 3; Figure 4). Thirty years later a single large post-fire forest patch accounted for 68% of the total area of LTC (Table 3). In contrast, although forests are the dominant cover class in the LTI landscape, forest patches remain relatively disjunct, as evident from the relatively low LPI, and are surrounded to a greater extent by a shrubland vegetation (Table 3). Forest patch proximity Land cover transitions from 1940 to 1970 were influenced by the original (1940) proximity of each site

to a forest patch (i.e., seed source) in 1940 as indicated by deviations from patterns expected under a distance-independent hypothesis (Figure 5). In both areas shrub-to-forest transitions within 10 m of forest patches occur more often than expected due to chance (P < 0.001) and occur less frequently than expected beyond c. 11.5 m from a forest patch. Given that most forest patches were immediately bordered by shrubland rather than grassland in 1940, transitions from grassland to forest at distances of less than c. 10 m are not greater than expected (P > 0.001). However, the influence of proximity of grassland sites to forest patches is manifest at intermediate distances; grassland-to-forest transitions occurred more frequently than expected due to chance (P < 0.001) at intermediate distances of c. 11.5 to 23 m in LTC and at distances of 12 to 33 m in LTI. The declining influence of seed sources for arboreal species is suggested by

10

Figure 4. Ranked size distributions of the 40 largest patches of forest, shrubland and grassland in 1940 (open circles) and 1970 (solid circles) in area LTC and area LTI.

Table 4. Cross-classification matrix of the 1940 and 1970 cover types in area LTC and area LTI at Lake Traful. Values are percentages of total area (21.33 ha for LTC and 45.67 ha for LTI) of cover types that remained stable (diagonal), changed to physiognomically higher cover types (upper right cells) and changed to physiognomically lower cover types (lower left cells). Forest

Shrubland

Grassland

Forest Shrubland Grassland

19.1 2.9 0.2

43.5 19.6 4.4

3.7 3.7 2.9

Total

22.2

67.5

10.3

Forest Shrubland Grassland

6.9 0.9 0.1

36.3 25.4 2.4

4.3 15.8 7.9

Total

7.9

64.1

28.0

Total

1940 State Area LTC 1970 state

66.3 26.2 7.5

Area LTI 1970 state

47.5 42.1 10.4

11 more frequent than expected (P < 0.001) grasslandto-shrubland transitions at distances greater than 14 and 16.5 m from 1940 forest patches in LTC and LTI, respectively (Figure 5). The tendency for grassland to remain stable is significant (P < 0.001) at distances greater than c. 14 and 19 m from 1940 forest patches in LTC and LTI, respectively. The likelihood of shrublands remaining stable is not strongly associated with distance from 1940 forest patches, except for the increased tendency for stability of shrublands at c. 7.5 to 11.5 m (P < 0.001) in LTI. Overall, despite different initial patch sizes and configurations in 1940, the relationships of cover type transitions to proximity of arboreal seed sources in forest patches are remarkably similar in both landscapes.

Discussion Changes in vegetation patterns that have occurred during the twentieth century at both a regional scale (e.g., over 100 000s of ha) and at a local landscape scale (e.g, over tens of ha) must be considered in the context of marked changes in fire regimes associated with permanent white settlement of the region. Near the ecotone of xeric woodlands of Austrocedrus with the steppe, fire frequency dramatically declined at the end of the 19th century coincident with the demise of the Native American population who frequently set fires for hunting purposes (Veblen and Lorenz 1988; Veblen et al. 1992a; Kitzberger et al. 1997). In the Nothofagus- and Austrocedrus-dominated forests further west, fire occurrence peaked in 1890 to 1910 in relation to forest-clearing activities by white settlers. Since c. 1910 fire frequency has declined in both regions as a consequence of fewer intentional fires and more effective fire suppression (Veblen et al. 1992a; Kitzberger et al. 1997). During the twentieth-century period of dramatically reduced fire frequency, there has been a regionally extensive tendency for the percentage of treedominated cover types to increase at the expense of cover types such as recently burned sites, grasslands and shrublands. Even allowing for co-registration errors between the historical and modern vegetation maps (as well as some inaccuracies inherent in the historical map due to the unavailability of aerial photography in 1913), the magnitude and inter-site consistencies of the changes in land cover types indicate a major shift in vegetation patterns. This shift in vegetation patterns is consistent with regionally extensive data

on tree population age structures (Veblen and Lorenz 1987; Kitzberger 1994; Villalba and Veblen 1997) and comparisons of historical with modern landscape photographs (Veblen and Lorenz 1988). This regional trend reflects both tree regeneration at sites of burned forests and succession from grassland or shrubland to forest cover. Neither N. dombeyi nor Austrocedrus sprout following fire (Veblen et al. 1995; Veblen et al. 1996), and the capacity of N. dombeyi to store viable seed in soil or litter for more than a year is nil, and that of Austrocedrus is low. In general in the Austrocedrus forest type, seed banks play a relatively unimportant role in vegetation dynamics (Raffaele and Gobbi 1996). Overall, transition from recent burn or shrubland to either the N. dombeyi or Austrocedrus forest cover depends on seed dispersal and successful seedling establishment. Stand structure data indicate that at mesic sites N. dombeyi and Austrocedrus regenerate vigorously after fire, either as monospecific or mixed stands (Veblen and Lorenz 1987, 1988). Although the focus of this study is on vegetation changes associated with reduced fire frequency, it is important to recognize that the northern Patagonian landscape has also been subject to heavy impacts from introduced livestock and cervids during the twentieth century (Martín et al. 1985; Veblen et al. 1992b). Livestock numbers in the region peaked during the 1930 to 1940 period (Ericksen 1971) and locally probably impeded the afforestation of some grassland and shrubland areas. Although the major tree species are relatively resistant to browsing once they reach sapling stages, exceptionally heavy cattle pressure during early post-fire recovery can locally impede tree regeneration and instead result in turfs of mostly exotic species or shrublands of spiny shrubs and dwarfed trees (Veblen et al. 1992b; De Pietri 1992). Despite the strong regional trend for former shrubland areas to succeed to forest, large areas of shrubland remained stable, and there have been some localized transitions from forest to shrubland cover (Table 2). Some such sites may have formerly supported forest, but due either to fire- or livestock-induced erosion, site quality may have declined, or continued heavy impacts of livestock may have inhibited development of a forest cover. The success of post-fire regeneration of N. pumilio in the subalpine zone is highly variable. Some stands have typical post-fire age structures, but we also have observed many high elevation areas of N. pumilio forest that still show no tree regeneration several decades after a fire (Veblen et al. 1996). In some cases the lack

12

Figure 5. Observed frequencies of cover-type transitions in area LTC (A) and area LTI (B) based on 1940-to-1970 cell transitions (solid bars) and transition frequencies expected by chance independent of distance from the nearest remnant patch (crosshatched bars). Z-statistic values (circles) above the dashed line indicate significant differences between observed and expected values (P < 0.001).

13 of regeneration might be due to heavy livestock pressure, but many of the sites are on steep slopes at high elevations where there is no current evidence of any livestock. Fire-induced erosion or climatic variability may explain some of these post-fire regeneration failures. Thus, although fire is less frequent in subalpine N. pumilio forests, as indicated both by fire history data (Veblen et al. 1992a) and by relatively few transitions from post-fire shrublands or recent burns to N. pumilio forests (Table 2), when fire does occur it can have a long lasting influence in the subalpine habitat. In contrast to the low predictability of the consequences of fire in the subalpine zone of N. pumilio forests, in the montane zone of N. dombeyi and Austrocedrus forests and woodlands the stand-level and landscape-scale consequences of fire are much more predictable. By examining changes on aerial photographs from 1940 and 1970, we identified several major spatial patterns of vegetation change during this period of reduced fire frequency. Due to co-registration errors, we emphasize the general trends rather than minor quantitative differences. In the submesic area, remnant forest patches resulting from widespread early twentieth century burning expanded and in many cases coalesced into continuous forest. Thus, although forest fragmentation is considered a common trend under increasing human influences on wild landscapes elsewhere (e.g., Harris 1984), in protected areas of northern Patagonia the reverse of forest fragmentation has been the norm over the past c. 70 years. Concomitantly, shrubland and grassland areas have become more disjunct and in many cases have been completely replaced by trees. Many areas that were grasslands in 1940 were replaced by shrublands in 1970, and only areas relatively isolated from tree seed sources (i.e., by distances of at least 14 to 19 m) remained stable. During this period of reduced fire frequency there was a shift in dominance from species with short life spans and re-sprouting capacity (e.g., shrubs) towards longer-lived species and obligate seed-dispersers (e.g., Austrocedrus and N. dombeyi) as predicted by Noble and Slatyer’s (1980) model of vegetation dynamics based on vital attributes. The relatively restricted seed dispersal ability of N. dombeyi and Austrocedrus may also explain the presence of extensive shrublands of fire-resprouting species (e.g., N. antarctica, Maytenus boaria and Lomatia hirsuta) where today small numbers of N. dombeyi and/or Austrocedrus are slowly invading. Similarly, dendroecological studies in nearby more mesic stands have demonstrated a

shift during long fire-free periods from the short-lived, post-fire resprouting N. antarctica towards the longlived, non-sprouting N. dombeyi (Veblen and Lorenz 1987). The results of seed-source proximity analyses indicate that configuration of remnant forest patches plays an important role in subsequent changes in landscape pattern. Post-fire forest regeneration, at least over a period of less than a single tree generation, is highly limited to the first c. 25 m from remnant forest patches (Figure 5) probably due to limited seed dispersal by N. dombeyi and Austrocedrus. In general in northern Patagonia, the extensive even-aged post-fire cohorts of N. dombeyi and N. dombeyi-Austrocedrus (Veblen and Lorenz 1987; Veblen et al. 1992a) must have developed within the relatively short range of remnant seed-bearing trees (i.e., radii of c. 50 m for the tallest trees). For example, in submesic habitats grasslands were significantly less invaded by arboreal species when located c. 20 m or more from forest patches (Figure 5). Shrublands, in contrast to arboreal vegetation, did not show any distance limitation when replacing grasslands which is consistent with the vigorous resprouting capacity of most of the xerophyllous shrub species (Gobbi et al. 1995), many of which were probably present but subdominant in the 1940 grasslands. The dependence of N. dombeyi and Austrocedrus on sexual reproduction and their relatively limited seed dispersal ability, results in vegetation patterns that closely follow isolines of distance from seed sources. Thus, resulting landscape patterns are highly sensitive to initial patterns of forest patches. The area included in the 1940–1970 aerial photograph comparison has long supported one of the densest cattle populations for this habitat type within Nahuel Huapi National Park (Veblen et al. 1992a). Thus, the stability of grasslands could be explained by a combination of isolation from tree seed sources and livestock pressure. A similar but more extreme situation occurs in central Chile where matorral patches have maintained their size for three decades due to heavy grazing, unfavorable establishment conditions outside patches, and enhanced entrapment of windand bird-dispersed seeds (Fuentes et al. 1984). Another potential influence on the rate and nature of vegetation change in northern Patagonia is climatic variation. For example, the 1940 to 1970 period was generally above average in precipitation compared to dry periods such as the 1910s or the 1980s (Villalba 1995). Successful establishment of Austrocedrus at more xeric sites near the forest-steppe ecotone appears

14 to be dependent on multi-year periods of above average spring-summer moisture (Villalba and Veblen 1997; Kitzberger et al. unpublished). Given, the high degree of climatic variability at time scales of a few years to several decades during the twentieth century (Villalba and Veblen 1997), climatic variability probably has influenced the rate and perhaps the nature of vegetation type transitions since the major changes in fire regimes about 1910. Reduced fire frequency in northern Patagonia clearly has permitted a regionally extensive transition from early to late stages of post-fire succession. The changes in landscape structure and composition associated with the changes in fire regimes caused by white settlement also have altered the potential for future fire occurrence, spread and intensity. Fuel loads and configurations continue to change as the vegetation responds to alterations in fire regimes. For example, the massive burning of mesic Nothofagus dombeyi forests has resulted in vast areas of forest that are currently self-thinning so that there is an abundance of intermediate-sized fuels on forest floors. Similarly, flammability and potential fire intensity have probably resulted from the reduction in fire frequency in N. antarctica woodlands and shrublands where reduced fire occurrence may explain the abundance of senescent stands of this short-lived species which is characterized by crown ‘dieback’ (Veblen and Lorenz 1988). Potential fire spread in submesic areas of formerly disjunct remnant patches has changed as trees regenerate and forest patches coalesce so that the horizontal continuity of fuels increases. Similarly, near the steppe ecotone, formerly open woodlands of Austrocedrus have been replaced by relatively dense stands during c. 80 years of reduced fire frequency. Thus, even if fewer human-set fire frequencies maintain a relatively low fire frequency, the increased connectivity of fire-susceptible vegetation types probably has created a much greater potential for high rates of fire spread (Turner et al. 1989). Analogous to the pattern documented for other landscapes (e.g., Agee 1994; Covington and Moore 1994) twentieth-century reduction in fire frequency in northern Patagonia has profoundly altered landscape patterns so that potentials for intensity and spread of fires have increased.

Aknowledgements For supporting this research we thank the National Science Foundation, National Geographic Society and

the Council on Research and Creative Work of the University of Colorado. For research assistance we thank J. Grosfeld and G. Rep, and for drawing Figure 1 we thank D. Lorenz. For facilitating our work in Argentine National Parks we thank C. Martín and M. Mermoz. We thank L. Daniels for commenting on the manuscript.

References Agee, J.K. 1994. Fire ecology of Pacific Northwest forests. Island Press, Washington, D.C. Baker, W.L. 1989. Landscape ecology and nature reserve design in the Boundary Waters CanoeArea, Minnesota. Ecology 70: 23–35. Baker, W.L. 1992. Effects of settlement and fire suppression on landscape structure. Ecology 73: 1879–1887. Barros, V., Cordón, V., Moyano, C., Méndez, R., Forquera, J. and Pizzio, O. 1983. Cartas de precipitación de la zona oeste del las provincias de Rio Negro y Neuquén. Unpublished report, Facultad de Ciencias Agrarias, Universidad Nacional del Comahue, Cinco Saltos. Burns, B. R. 1993. Fire-induced dynamics of Araucaria araucanaNothofagus antarctica forestin the southern Andes. Journal of Biogeography 20: 669–685. Covington, W.W. and Moore, M. 1994. Southwestern Ponderosa forest structure. Journal of Forestry 92: 39–47. De Pietri, D.E. 1992. Alien shrubs in a national park: can they help in the recovery of natural degraded forest? Biological Conservation 62: 127–130. Eastman, J.R. 1990. IDRISI. A grid-based geographic analysis system. Graduate School of Geography, Clark University, Worcester, MA. Ericksen, W. 1971. Betriebsformen und probleme der viehwirthschaft am rande der Argentinische südkordillere. Z. Ausl. Landwirtschaft 10: 24–27. Fuentes, E.R., Otaiza, R.D., Alliende, M.C., Hoffmann, A. and Poiani, A. 1984. Shrub clumps of the Chilean matorral vegetation: structure and possible maintenance mechanisms. Oecologia 62: 405–411. Gobbi, M., Puntieri, J., and Calvelo, S. 1995. Post-fire recovery and invasion by alien plant species in a South American woodlandsteppe ecotone. In Plant Invasions: General Aspects and Special Problems. pp. 105–115. Edited by Pysek, P., Prach, K., Rejmanek, M., and Wade, M. Academic Publishing, Amsterdam. Harris, L.D. 1984. The fragmented forest. Island biogeography and the presevation of biotic diversity. University of Chicago Press, Chicago. Heinselman, M.L. 1973. Fire in the virgin forest of the Boundary Waters Canoe Area, Minnesota. Quaternary Research 3: 329–82. Johnson, E.A. and Van Wagner, C.E. 1985. The theory and use of two fire history models. Canadian Journal of Forest Research 15: 214–220. Kitzberger, T. 1994. Fire regime variation along a northern Patagonian forest-steppe ecotone: stand and landscape response. Ph.D. Thesis. Department of Geography, University of Colorado, Boulder. Kitzberger, T., Steinaker, D.F. and Veblen, T.T. Unpublished. Spatial and temporal establishment opportunities for Austrocedrus chilensis at the northern Patagonian forest-steppe transition. Unpublished manuscript.

15 Kitzberger, T., Veblen, T.T. and Villalba, R. 1997. Climatic influences on fire regimes along a rainforest-to-xeric woodland gradient in northern Patagonia, Argentina. Journal of Biogeography 23: 35–47. Legendre, L. and Legendre, P. 1983. Numerical ecology. Elsevier, New York. Martín, C., Mermoz, M., and Gallopín G. 1985. Impacto de la ganadería en la cuenca del Río Manso Superior. Parte I. Bosque de ñire con laura. Report, Administración de Parques Nacionales, Buenos Aires. McBride, J.R. 1983. Analysis of tree rings and fire scars to establish fire history. Tree-Ring Bulletin 43: 51–67. McGarrigal, K. and Marks, B.J. 1993. FRAGSTATS. Spatial Pattern Analysis Program for Quantifying Landscape Structure. Unpublished software. Department of Forest Science, Oregon State University. Mermoz, M. and Martín, C. 1986. Mapa de la vegetación del Parque y Reserva Nacional Nahuel Huapi. Unpublished report. Administración de Parques Nacionales, Buenos Aires. Raffaele, E. and Gobbi, M.E. 1996. Seed bank composition and variability in Austrocedrus chilensis forest sites in Patagonia, Argentina. International Journal of Ecology and Environmental Science 22: 59–72. Story, M. and Congalton, R.G. 1986. Accuracy assessment: a user’s perspective. Photogrammetric Engineering and Remote Sensing 52: 397–399. Turner, M.G. and Romme, W.H. 1994. Landscape dynamics in crown fire ecosystems. Landscape Ecology 9: 59–77. Turner, M.G., Gardner, R.H., Dale, V.H. and O’Neill, R.V. 1989. Predicting the spread of disturbance across heterogeneous landscapes. Oikos 55: 121–129. Van Wagner, C.E. 1978. Age-class distribution and the forest cycle. Canadian Journal of Forest Research 8: 220–227.

Veblen, T.T., Burns, B.R., Kitzberger, T., Lara, A. and Villalba, R. 1995. The ecology of the conifers of southern South America. In Ecology of the Southern Conifers. pp. 120–155. Edited by Enright, N. and Hill, R. Melbourne University Press, Parkville, Australia. Veblen, T.T., Donoso, C., Kitzberger, T. and Rebertus, A.J. 1996. Ecology of southern Chilean and southern Argentinean Nothofagus forests. In Ecology and Biogeography of Nothofagus Forests. pp. 293–353. Edited by Veblen, T.T., Hill, R.S. and Read, J. Yale University Press, New Haven. Veblen, T.T, Kitzberger, T. and Lara, A. 1992a. Disturbance and vegetation dynamics along a transect from rain forest to Patagonian shrublands. Journal of Vegetation Science 3: 507–520. Veblen, T.T. and Lorenz, D.C. 1987. Post-fire stand development of Austrocedrus-Nothofagus forests in Patagonia. Vegetatio 71: 113–126. Veblen, T.T. and Lorenz, D.C. 1988. Recent vegetation changes along the forest/steppe ecotone in northern Patagonia. Annals of the Association of American Geographers 78: 93–111. Veblen, T.T. and Markgraf, V. 1988. Steppe expansion in Patagonia? Quaternary Research 30: 331–338. Veblen, T.T., Mermoz, M., Martin, C. and Kitzberger, T. 1992b. Ecological impacts of introduced animals in Nahuel Huapi National Park, Argentina. Conservation Biology 6: 71–83. Villalba, R and Veblen, T.T. 1997. Regional patterns of tree population age structures in northern Patagonia: climatic and distrubance influences. Journal of Ecology 85: 113–124. Villalba, R. 1995. Climatic influences on forest dynamics along the forest-steppe ecotone in northern Patagonia. Ph.D. Thesis. Department of Geography, University of Colorado, Boulder. Willis, B. 1914. Northern Patagonia: Character and resources. Vol. I. Ministry of Public Works. Buenos Aires.