GEB220.fm Page 133 Friday, April 20, 2001 3:35 PM

Global Ecology & Biogeography (2001) 10, 133 –146

R E S E A R C H A RT I C L E

Regional-scale spatial patterns of fire in relation to rainfall gradients in sub-tropical mountains, NW Argentina Blackwell Science, Ltd

HÉCTOR RICARDO GRAU Department of Geography, Campus Box 260, University of Colorado, 80309, Boulder, Colorado, U.S.A. E-mail: [email protected]

ABSTRACT 1 Spatial patterns of burns are described using Landsat TM images from the sub-tropical mountains of north-west Argentina, over a span of 6 degrees of latitude, and a precipitation range from 250 to 1300 mm / yr. Burns were discriminated easily from unburnt vegetation, mainly by using infrared spectral bands from images taken at the end of the fire season of 1986. 2 Nineteen sampling units were defined on the basis of geographical proximity and relatively homogeneous rainfall as inferred from topography, and they were characterized in terms of percentage of burnt area and burn size distribution during one fire season. Regression and Correspondence Analysis were used to assess the relationship between rainfall and spatial descriptors of fire regime. 3 Burnt size area was log-normally distributed with most fires in the small-size classes. Of a total of 643 burns, the five largest (more than 2000

INTRODUCTION Disturbance regimes are important influences on ecosystems. Spatial parameters of disturbance regime (extent, size distribution) and the related temporal parameters (frequency, return interval) strongly influence structure and functioning of ecosystems (Miller, 1982; Pickett & White, 1985). Fire has long been recognized as a key component of disturbance regimes in grasslands, shrub-

Correspondence: Héctor Ricardo Grau, Laboratorio de Investigaciones Ecológicas de las Yungas, Casilla de Correo 34 (4107) Yerba Buena, Tucumán, Argentina.

hectares each) represented about 30% of the total burnt area. 4 Percentage of burnt area, density of burns per unit area, and skewness of the burn-size frequency distribution showed a unimodal pattern along the rainfall gradient, peaking between 700 and 900 mm/ yr. Mean and maximum burn size showed a negative but weak correlation with rainfall. The first axis of a Correspondence Analysis ordination of sampling units, on the basis of different descriptors of spatial patterns of fire, was significantly correlated with the rainfall of the sampling unit. 5 The results suggest that climate is an important factor controlling fuel conditions and therefore fire regime at the spatial scale of this study, which includes different mountain ranges spanning ≈ 700 km. Key words Andean ecosystems, climatic gradient, Correspondence Analysis, disturbance regime, fire, Landsat TM, NW Argentina, remote sensing, sub-tropical mountains.

lands and open woodlands (Braun-Blanquet, 1950; Sauer, 1950). Spatial and temporal patterns of fire affect important ecological characteristics including floristic and life-form composition (Noble & Slatyer, 1980; Christensen, 1985; Glenn & Collins, 1992), diversity (Huston, 1994; Collins et al., 1995), productivity (Knapp & Seastadt, 1986; Hobbs et al., 1991; DeBano et al., 1998), and risk of invasion by exotic plants (Hobbs & Huenneke, 1992). Consequently, understanding the fire regime is a key factor for the management and conservation of these ecosystems. Fire regime is controlled by anthropogenic, climatic and topographic factors. Human activities

© 2001 Blackwell Science Ltd. http://www.blackwell-science.com/geb

133

GEB220.fm Page 134 Friday, April 20, 2001 3:35 PM

134

H. R. Grau

directly influence ignition sources (Pyne, 1993) and indirectly control fuel availability through grazing of domestic animals, which is the dominant type of land use in grasslands and open woodlands (e.g. Adámoli et al., 1990; Savage & Swetnam, 1990; Archer, 1994). In Neotropical mountains, besides the obvious climatic influence on seasonal patterns of fire, the fire regime at longer temporal scales is considered to be mostly controlled by human influences, according to descriptions both in NW Argentine mountains (Molinillo & Vides-Almonacid, 1989; Brown, 1995) and in lower latitude (Ecuador, Peru and Bolivia) montane ecosystems (Ellenberg, 1979; Kessler, 1995). On the other hand, climate controls fuel availability and moisture, and lightning, which is the most important natural ignition source; and topography influences patterns of fire spread at the landscape scale by affecting landscape configuration. Since grasslands and shrublands are typically located in sub-mesic and semi-arid environments, precipitation should have an important effect on fire regimes by controlling both fuel production and fuel dessication (Rothermel, 1983; Bond & van Wilgen, 1996). For relatively long and steep precipitation gradients, fire frequency is expected to follow a unimodal pattern, peaking at intermediate values of rainfall: reduced fire frequency should be expected in dry areas as a consequence of low fuel production and in wet areas as a consequence of high fuel moisture (Christensen, 1993; Bond & van Wilgen, 1996). Such a pattern, for example, is observed at a global scale, where fires are more common in tropical and sub-tropical savannas than in semi-arid vegetation or humid forests (Dwyer et al., 1998); at a continental scale, along long rainfall gradients of North America, Europe and Australia (Trabaud et al., 1993); and at a regional scale, along gradients from rain forests to semi-arid open woodlands (Kitzberger et al., 1997; Veblen et al., 1999). Regional descriptions of fire regime, however, remain scarce in many regions, including the tropical and sub-tropical Andes. The effects of variation in disturbance regime along geographical gradients is superimposed with effects of the climatic gradient on the structure and dynamics of vegetation (Harmon et al., 1984; Veblen et al., 1992). In tropical and

sub-tropical Andean ecosystems, fire has been hypothesized as having large-scale effects on the distribution of major vegetation types along environmental gradients, including the location of treelines and forest patches (Ellenberg, 1979; Grau, 1985; Lægard, 1992; Young, 1993; Grau & Brown, 1995; Kessler, 1995). However, quantitative measurements of fire consequences have been limited to local assessments of shortterm effects on herbaceous and shrub-dominated plant communities (e.g. Williamson et al., 1986; Horn, 1989; Verweij & Budde, 1992; Keating, 1998). The lack of studies of fire effects at landscape and regional scales can be attributed partially to the absence of quantitative descriptions of fire regimes at large spatial scales. This study describes the spatial patterns of fire at a regional scale in the sub-tropical montane zone of North-western Argentina using Landsat TM images. The region is characterized by substantial variation in precipitation, which provides the opportunity to study the variation in fire regime along rainfall gradients at a spatial scale of several hundred kilometers. The relationship between spatial patterns of fire (extent, size distribution) are explored and rainfall inferred from topographic features, such as mountain range elevation and exposure to easterly winds.

STUDY AREA The study area includes mountain ranges located between 24°10′ and 29°50 ′ S, and between 65°30′ and 67°20 ′ W in the provinces of Salta, Tucumán, Catamarca and La Rioja, north-western Argentina (Fig. 1). The area is a complex of orographic features of different geological origin, east of the sub-tropical Andean plateau or ‘Puna’. Despite their differing geologies, the mountain ranges are all orientated in a north– south direction, producing the same general patterns of orographic precipitation. The northwest sector of the study area (San Lorenzo, Salta, Escoipe, Alemania, El Cajón, Carahuasi and Cumbres Calchaquíes ranges) is considered part of the ‘Cordillera Oriental’ or ‘Precordillera’; the north-east sector (Metán, Rosario de la Frontera, Candelaria, Medina and El Nogalito ranges) is part of the ‘Sierras Subandinas’; and the southern sector (Tafí, Aconquija, Santa Ana, Narvaez, Balcozna, Ancasti, Ambato and

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 135 Friday, April 20, 2001 3:35 PM

Fire and rainfall gradients in sub-tropical mountains

Fig. 1 Location of the study area and sampling units in NW Argentina. Sampling unit numbers correspond to Table 1.

Velazco ranges) is part of the ‘Sierras Pampeanas’ (Aceñolaza & Toselli, 1981). Annual rainfall is highly variable as well as fairly predictable at a regional scale due to topography. Since orographic precipitation is a major

135

contribution to total rainfall, annual precipitation can be predicted from exposure to humid easterly winds and elevation of the mountain range. According to Bianchi & Yañez (1992), the mountain ranges studied cover a range of annual precipitation from 250 mm to 1300 mm (Table 1). These authors generated a map of annual rainfall isolines using an iterative approach, which combines instrumental records from about 50 meteorological stations, inferences of elevational trends in rainfall due to saturation curves and temperature environmental lapse rate, and distribution of major vegetation units that were used to test predictions of the model and redraw rainfall lines. Since the northern portion of the study area (Salta and Tucuman) includes a larger number of instrumental records, it is probable that the accuracy of the predicted rainfall would be lower in the southern portion of the study area (Catamarca and La Rioja). Despite the differences in rainfall among mountain ranges, seasonality of rainfall for the whole region follows the same pattern since it is controlled by continental-scale atmospheric circulation. Rainfall follows a monsoonal regime with distinctive wet (November– May) and dry (June – October) seasons (Prohaska, 1976; Rao et al., 1996). Plant phenology clearly follows the climatic seasonality. During dry season months, both grasslands and deciduous forests have low proportions of photosynthetic tissues, making them readily identifiable in Landsat satellite images (Bell, 1991). Although water balance would be a more direct measure of fuel productivity and moisture, only annual rainfall was considered because: (1) there are no long-term records of temperature and no maps of temperature for the study area that could be used to estimate evapotranspiration, and (2) the area where fires occur in the different mountain ranges coincides roughly in terms of elevational range, which is the major geographical control of temperature (a minimum of 70% of the area of each mountain range is located between 1800 and 2500 m elevation). Therefore, within the margins of errors of this study, rainfall can reliably be considered as the main source of differences in water balance. The vegetation of the study area is dominated by grasslands and patches of shrublands. Open woodlands, mostly dominated by Alnus

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

Percentage of burned area

Density

Skewness of burn size distribution

Topography

Mean burn size

Maximum burn size

Maximum elevation

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19.

0.4 3.0 14.0 8.7 6.7 2.3 1.7 0.6 0.9 4.8 7.7 7.4 1.0 1.8 8.8 3.5 1.6 6.6 1.3

0.017 0.058 0.035 0.011 0.011 0.010 0.080 0.017 0.007 0.132 0.009 0.009 0.025 0.080 0.065 0.069 0.099 0.007 0.006

2.397 2.581 2.237 2.822 3.869 2.566 4.198 2.165 1.785 3.951 4.210 3.562 0.954 1.807 5.046 4.148 1.947 1.947 2.558

ES Sev ES Top Top ES ES WS ES Sev V ES ES Sev Top Sev Top V Top

22.5 51.1 551.2 169.8 57.8 226.3 19.3 32.1 117.2 35.6 159.3 210.3 4.4 22.5 133.3 49.0 15.4 106.2 218.1

70 272 658 136 116 2100 462 120 136 148 584 421 40 140 160 146 38 549 176

5500 3500 4500 1300 1800 3500 4500 4500 3800 2500 4500 3800 3200 2500 3300 4000 2700 4000 3500

Aconquija Alemania Ambato Ancasti Balcozna Cajon CalchE CalchW Carahuasi Escaba Estancias Humaya Mala-mala Medina Metan-Ro Salta SantaAna Tafi Velazco

Rainfall (mm/ yr) 1300 600 400 700 900 250 700 250 500 800 900 600 1000 1000 800 800 1100 600 300

Total area (hectares) 49 600 52 000 117 955 109 532 32 064 90 920 166 097 23 501 73 859 26 031 37 067 118 641 16 000 45 395 78 731 129 689 14 716 25 838 115 988

H. R. Grau

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

Sampling unit

GEB220.fm Page 136 Friday, April 20, 2001 3:35 PM

136 Table 1 Geographic characteristics and parameters descriptors of spatial patterns of fire in the 19 sampling units. Topographic characteristics include: ES = east-facing slope; WS = west-facing slope; Sev = Several mountains considered together; V = valley; Top = Top and both east and west slopes. Annual rainfall is derived from regional rainfall maps by Bianchi & Yanez (1992; see text)

GEB220.fm Page 137 Friday, April 20, 2001 3:35 PM

Fire and rainfall gradients in sub-tropical mountains acuminata H.B.K. and occasionally by Polylepis australis Bitt, are typical of the most humid sites. Grasslands are dominated by different Poaceae species, such as Festuca hieronymi Hackel, Deyeuxia poligama (Griseb) Parodi, and Stipa eriostachia H.B.K., and shrublands are dominated by Asteraceae genera (Baccharis and others). The upper vegetation limit (4100 – 4300 m) is reached in the higher mountain ranges (Salta, Calchaquies, Carahuasi, Cajon, Tafí, Aconquija and Ambato), where grasslands are replaced by the barren alpine zone. The lower limit of grasslands and open woodlands varies according to the rainfall gradient. In the humid sampling units (Salta, Metán, Medina, Mala-Mala, Aconquija, Santa Ana, Escaba, Ancasti and Balcozna) the lower limit is defined by the presence of mostly evergreen-dominated montane forests, or bamboo (Chusquea lorentziana Griseb) thickets at the wettest sites. At drier sites, the low-elevation limit of grasslands is desert vegetation or irrigated agriculture (Cabrera, 1976). Grazing by introduced domestic animals is widespread throughout the study area. Cattle and sheep are the most abundant species, followed by goats and llamas. Range management varies from small-scale intensive grazing around villages and puestos (livestock stations) to transhumance pastoralism covering extensive areas and long elevational gradients (Millones, 1982; Solbrig, 1984; Molinillo, 1993). Herds of naturalized burros and of native guanacos (Camelidae) and taruca deer (Cervidae) are relatively abundant in the more remote places (Cabrera & Willink, 1980). Fire is used to promote pasture regrowth and eliminate unpalatable woody species and, to a lesser extent, for driving game such as peccaries in grasslands, woodlands and bamboolands (Molinillo & Vides-Almonacid, 1989). This information, however, is based on general observation and interviews with local people. There have been no quantitative assessments of anthropogenic or natural ignition sources. The majority of fires occur during the dry season (May – October) when there is little lightning, so most fires are assumed to be ignited by people. When the rainy season starts and temperatures are warm enough, recently burnt areas are revegetated by fast growing annuals and resprouting herbaceous plants (Molinillo & Vides-Almonacid, 1989).

137

METHODS Definition of sampling units Nineteen sampling units were defined (Fig. 1, Table 1) according to geographical and precipitation characteristics determined from the map of annual rainfall created by Bianchi & Yañez (1992). Mountain ranges defined the sampling units. However, some mountain ranges were joined into one unit or divided into two in order to have units of comparable size and homogeneous rainfall. One mountain range (Cumbres Calchaquies) was divided into two sampling units (east and west slope) on the basis of differential precipitation regimes. In five cases: Metán– Rosario de la Frontera (Metán-Ro), Medina– Candelaria–El Nogalito (Medina), Sierra del Campo–Cerro Quico–Sierra de La Merced (Balcozna), Cumbre de Santa Ana–Silleta de Escaba (Santa Ana) and Narvaez–Los Pinos (Escaba), two or three mountain chains were analysed jointly as the same unit, due to their small size, short distance between them, and similar inferred precipitation. Three sampling units include several relatively small ranges: ‘Mala-mala’ includes all the east slopes of the Cumbres de Tafi and south of the Nuñorcos ranges; ‘Salta’ includes several chains on the San Lorenzo, Rio del Toro, and Escoipe basins; and ‘Alemania’ includes several ranges in the La Viña, Osma and El Carril basins. In the low-elevation ranges (e.g. Medina, Ancasti and Balzona), sampling units included the top of the mountain range and slopes with different aspects, while in the high-elevation ranges (i.e. Ancasti, Aconquija, Cumbres Calchaquies and El Cajon) the sampling units were mostly dominated by eastern-slope exposures. Additionally, two valleys (‘Estancias’ and ‘Tafí’) were analysed as sampling units. In these valleys, the fire regime was studied for the different slopes surrounding the valley, but excluding the valley bottom, where there has been significant agricultural and hydroelectric development.

Image processing Five contiguous Landsat TM images were used (path 231, rows 077 – 081), taken on November 13, 1986. This date was chosen because it: (1) includes most fires that occurred during the dry season,

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 138 Friday, April 20, 2001 3:35 PM

138

H. R. Grau

which ends at this time of the year; (2) maximizes the spectral differences between burnt sites, because vegetation at this time of the year has not yet started to regrow; and (3) minimizes cloud cover and atmospheric haze. Only the northern sector of the Ambato range had significant cloud cover, and it was excluded from the analysis. Atmospheric haze was only significant in band 1 (blue). Precipitation throughout the region in 1986 was close-to-average (Bianchi & Yañez, 1992). For example, in a sample of 10 meteorological stations spanning the latitudinal range of the study area and having several decades of rainfall records (San Salvador de Jujuy, La Mendieta, Cerrillos, Campo Quijano, Rosario de la Frontera, Famailla, Rio Seco, Monteros, Concepcion and Potrero del Clavillo), 1986 rainfall was within mean ± SD in all cases, and in six cases it was within mean ± 0.5 SD. Five deviations from the mean were positive and five negative, and no trend in the deviations along the latitudinal gradient was evident. Landsat images are composed of seven spectral bands. Bands 1, 2 and 3 are located in the visible spectrum, bands 4, 5 and 7 in the reflected infrared, and band 6 in the thermal infrared portion of the spectrum (Sabins, 1987). Burns have been discriminated previously from surrounding vegetation using Landsat images in other environments (Minnich, 1983; Chuvieco & Congleton, 1988; Kushla & Ripple, 1988; García & Caselles, 1991; Pereira & Setzer, 1993). The main differences in spectral signature between burns and non-burned vegetation occurs in bands 4, but there are also important differences in bands 3, 5 and 7 (Pereira & Setzer, 1993). There is also an overall lower reflectance in all the bands due to the high absorption of ashes and burnt vegetal tissues (Minnich, 1983; Pereira & Setzer, 1993). Similarly, in this study, all dominant unburnt vegetation types differed significantly from burnt sites in terms of spectral signature, particularly in the infrared bands 4 and 7. Evergreen forests and bamboo thickets showed the spectral characteristics of green vegetation (e.g. Pereira & Setzer, 1993) with peaks in values of digital numbers (DN) in the near-infrared bands, and with low DN in band 3 (red) and band 7 (Fig. 2a). Grasslands showed a spectral signature of vegetation with a low proportion of

Fig. 2 Spectral signature in bands 3, 4 and 7 of Landsat TM of the main cover types in the study area expressed in digital number means. (a) Vegetation with abundant photosynthetic tissues; (b) vegetation with abundant dead organic material; and (c) surfaces without significant vegetation cover.

photosynthetic tissues (i.e. Graetz & Gentle, 1982), consisting of relatively high DN values in band 3 and in band 7, and low DN values in band 4 in comparison with evergreen vegetation (Fig. 2b). Alnus woodlands showed a spectral signature intermediate between evergreen vegetation and dry grasslands (Fig. 2b). Burns showed lower reflectance in all bands for all vegetation types. They had the lowest DN values in band 4 (normally very high in vegetation), which was as low as in bands 2 and 3; DN values in band 7 were high (Fig. 2c). The only types of surface with a spectral signature similar to the burnt sites (but much brighter) were areas with bare ground such as landslides, river banks, severely eroded areas near population centres, and rocky areas above the vegetation limit (Fig. 2c). Based on these differences in spectral signatures, burns were identified using colour composite images (RGB = 7-4-3). On these images, burns appear as distinct red-coloured areas while non-burned vegetation appears in different tones of yellow and green (Fig. 3a).

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 139 Friday, April 20, 2001 3:35 PM

Fire and rainfall gradients in sub-tropical mountains

139

Fig. 3 Patterns of burns and vegetation in the central region of the study site (eastern slope of Aconquija range). (a) Colour composite (R-G-B = 7-4-3), burns appear red-coloured. (b) Supervised classification using Principal Components 2 – 4, based on five bands of the Landsat image (excluding bands 1 and 6). Blue = burnt sites; green = deciduous forests, including Alnus open woodlands; yellow = bamboo lands; red = upper evergreen forest; pale blue = lower evergreen forest; orange = coniferous forest; maroon = tall grasslands and shrublands; cyan = short-grasslands; sea green = landslides, riverbanks and upper elevation unvegetated zones. The scene covers an area of approximately 50 by 20 km.

Discrimination of burns and other cover types can be improved slightly by combining the different TM bands using Principal Components Analysis, which generates new uncorrelated ‘bands’ and reduces the effects of topography in the spectral signature (Schowengerdt, 1997). Supervised classification (Schowengerdt, 1997), excluding the first principal component, which is mostly controlled by topography, discriminated burnt sites, grasslands, bamboo thickets and deciduous and evergreen forests (Fig. 3b). However, as can be seen by comparing Fig. 2.3a and b, the colour composite 7-4-3 shows burns (red) in complete agreement with the supervised

classification image. In addition, it was the most time-efficient way to map burns across the complete scene. Two accessible sites on the eastern slope of the Aconquija range that had burned recently according to the 1986 image, and an August 1989 image of the same area, were located in the field. At the two sites, fire dates were corroborated by dating fire scars in Alnus acuminata using dendroecological techniques based on annual tree-ring counts (Arno & Sneck, 1977). Also, for the east slope of the Aconquija range, the 1986 image was compared with images taken in August 1989, and in February 1986. None of

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 140 Friday, April 20, 2001 3:35 PM

140

H. R. Grau

the areas classified as burned sites in the October 1986 image are visible on the preceding and subsequent images, supporting the assumption that burns in the 1986 image only reflect fires that occurred during the 1986 dry season (i.e. fires that occurred during the previous fire season should already be re-vegetated). Landslides and river banks were visible in the images of the three dates. They could be discriminated clearly from burns on the basis of shape and topographic location and, in general, they were much smaller. The only places where bare ground could be confounded with burns were on areas located close to human settlements, probably as a consequence of overgrazing and soil erosion. Therefore, such areas were excluded from the analysis. Burns were identified visibly in the colourcomposite image. All burns and the total area of each sampling unit were measured manually, using the computer mouse and the ‘region of interest’ function of the ENVI 3.0 (1998) image analysis program. The same program was used for all the image analyses in this study. The minimum mapping unit was 20 pixels. Area was computed on the basis of number of pixels, assuming a pixel size of 30 × 30 m (Sabins, 1987).

Fire spatial parameters and data analysis In order to describe the spatial patterns of fire, the following parameters were derived: (1) percentage of burnt area over total area for each sampling unit; (2) density of fires expressed in number of burns per km2; (3) skewness of the size distribution of burns; (4) mean size of burns; and (5) maximum burn size. Size distribution is presented as square root- and log-transformed area frequency – distribution histograms. The pairwise relationships between the fire regime parameters and rainfall are presented in the form of scatterplots with the linear or quadratic curves that best fit the data. Correspondence Analysis (CA; ter Braak, 1987) was used to assess if the spatial descriptors of fire, considered together, were related to the rainfall gradient. CA was preferred as a multivariate technique over linear methods (Principal Components Analysis, for example), because CA assumes a unimodal response of the variables to the underlying environmental gradient,

as was the case for three of four studied variables in relation to precipitation (see Results). The different parameters (percentage of burn area, density of fires, skewness of the size distribution, mean size of burns and maximum burn size) show different spatial characteristics of the fire regime, but they were partially correlated among themselves. Mean size and maximum size were highly correlated (r 2 = 0.75), therefore only mean size was used in the CA. Instead, the three other variables were only partially correlated ( r 2 < 0.5), which implies that each variable contains additional information on the spatial pattern of fire, but cannot be treated as an independent variable (for example, in a multiple regression model). Correspondence analysis was performed using a matrix of sites by fire-descriptors (which included four variables: density, percentage, skewness and mean size). Study sampling units were ordered by CA. In order to assess whether the spatial pattern of fire as a whole (including the different descriptors) can be predicted from precipitation, assuming a unimodal response to the dominant environmental variable, the scores of each sampling unit in the first CA axis were correlated with precipitation of each sampling unit using Kendall’s coefficient of rank correlation (Sokal & Rohlf, 1995). CA was performed with PC-Ord 3.0 (1997 ), and other analyses with SPSS 6.1 (1995).

RESULTS In total, 643 burns were identified over a total surveyed area of 1326 000 ha. The percentage of burned area per sampling unit ranged from 0.4 to 14%. The largest percentage of burned area occurred at the Ambato sampling unit (Table 1). Size distribution of burns was highly skewed (Fig. 4a), with the great majority of burns in the small-size classes. The general pattern of size distribution followed a lognormal curve (Fig. 4b). More than 85% of the mapped burns were < 200 ha, while only five burns were > 2000 ha. Three of them, including the largest burn (> 4000 ha) were located in the Ambato range. Large-sized burns comprised most of the total burnt area. Considering all the study area, burns of > 200 ha comprised around 75% of the total burnt area. At most sampling units,

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 141 Friday, April 20, 2001 3:35 PM

Fire and rainfall gradients in sub-tropical mountains

141

rainfall gradient. For the three parameters, the coefficient of determination of the best quadratic fit was around 20% and significance levels were lower than 0.2 (Fig. 5a – c). The Ambato range appeared as an outlier, having the highest percentage of burnt area, although its precipitation is relatively low (Fig. 5b, Table 1). Mean burn size showed a generally decreasing trend with precipitation (r = – 0.53), but it was influenced strongly by the high values of the Cajon range (Fig. 5d, Table 1). The first two axes of the CA plot showed a clear relationship between the sampling units’ ordination scores and the precipitation gradient. Drier sampling units had low first axis and intermediate second axis values, while moister sampling units ranked high in the first axis and were more widespread along the second axis (Fig. 6). The first axis site scores showed a highly significant correlation with the rainfall of each site (Kendall’s Tau = 0.55, P = 0.001).

DISCUSSION

Fig. 4 Distribution of burn sizes in hectares: (a) square root-transformed; (b) log-transformed.

the percentage of burnt area was associated with burns in the large size classes. The only two exceptions were Escaba Basin and EastCalchaquíes range, where most of the burnt area was due to burns between 20 and 200 ha. The Alemania and Salta sampling units also had a large percentage of burned area in the 20– 200 ha class, although their percentage was lower than that in the 200 – 2000 ha class (Table 2). These four sampling units are characterized by a high density of relatively small fires and, consistently, with relatively high values of skewness (Table 1). When plotted against precipitation, the percentage of burnt area, fire density and skewness at each site showed a unimodal pattern, peaking in sampling units with intermediate annual precipitation (i.e. 700 – 900 mm / year, Fig. 5a – c). The highest variability in these parameters also occurred in the intermediate portion of the

The size distribution of burns (Fig. 4, Table 2) was characterized by a large proportion of small fires, and a small number of large fires that accounted for a large proportion of the burned area. While the number of fires is controlled mainly by ignition sources (Pyne, 1993), the size of the burned area is controlled mainly by fuel characteristics (Rothermel, 1983; Bond & van Wilgen, 1996). These results suggest that, in addition to ignition sources, factors controlling fuel availability, moisture and spatial distribution are affecting fire regime in these ecosystems. Climate (particularly rainfall and water balance) and grazing are likely to be the two most important factors controlling fuel loads at the spatial scale of this study. Size distribution, percentage of burned area and density of burns showed a unimodal response to the rainfall gradient, peaking between 700 and 900 mm of inferred annual rainfall (Fig. 5). This pattern is consistent with observations at global (Dwyer et al., 1998), continental (Christensen, 1993; Trabaud et al., 1993) and regional scales in temperate latitudes (Kitzberger et al., 1997; Veblen et al., 1999). Such a pattern can be interpreted as the consequence of the climatic control over fuel production and fuel moisture: it

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 142 Friday, April 20, 2001 3:35 PM

142

H. R. Grau

Table 2 Total area (in hectares) included in burns of different size classes in the different sampling units Sampling unit

Size range (ha) > 20

Aconquija Alemania Ambato Ancasti Balcozna Cajon CalchE CalchW Carahuasi Escaba Estancias Humaya Mala-mala Medina Metan-Ro Salta SantaAna Tafi Velazco Total %

20 – 200

200 – 2000

45 110 100 193 175 28 799 0 29 173 43 195 18 180 206 419 30 79 22

157 608 934 952 262 491 1754 28 195 714 757 1136 0 652 1084 1808 109 667 216

0 866 3767 8367 1262 1568 267 165 479 357 2066 2933 0 0 2862 2283 0 954 1287

2844 4.5

12 524 19.6

29 483 46.1

is expected that dry sampling units have lower fire frequencies due to low fuel productivity whereas wet sampling units have low fire frequencies due to high fuel moisture. The coefficient of determination of fire descriptors from rainfall was approximately 20%, implying a remaining 80% of unexplained variance. Interannual fire variability, local patterns of rainfall anomalies, differences in range management and differences in landscape configuration are variables likely to be significant additional sources of variation in fire regime. The study of landscapescale factors such as local vegetation patterns controlled by land use, topography and successional patterns should be addressed in future research. Outlier sampling units with percentage of burnt area higher than expected as a function of rainfall (e.g. Ambato and Cajón) were characterized by individual fires of relatively large size. The importance of large fires in defining the fire regime emphasizes again the role of fuel availability over ignition sources, since great areas can potentially be burned by one or a few

> 2000

Sum

0 0 11 734 0 0 0 0 0 0 0 0 4569 0 0 2781 0 0 0 0

202 1584 16 535 9512 1699 2087 2820 193 703 1244 2866 8833 18 832 6933 4510 139 1700 1525

19 084 29.8

63 935 100

ignition sources under favourable conditions and landscape configuration of the fuels. This also emphasizes the potential role of inter-annual climatic variability in controlling fuel quantity and quality, a factor that has a major effect on largescale fires (Swetnam, 1993; Veblen et al., 1999). Although the statistical significance of the correlation between individual fire parameters and inferred rainfall was low, the CA ordination of sampling units combining the different fire parameters (Fig. 6) was correlated with rainfall at highly significant probability levels. These results imply that, at the regional scale, climate is an important factor controlling fire regime. The fact that rainfall shows a strong correlation with the ordination scores that combine the different fire descriptors (percentage of burn area, density and size distribution), emphasizes that the interactions between different components of the fire regime needs to be explored further in order to understand climate – fire regime relationships. Most previous descriptions of fire–vegetation interactions in Neotropical mountains assume

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 143 Friday, April 20, 2001 3:35 PM

Fire and rainfall gradients in sub-tropical mountains

143

Fig. 6 Distribution of sampling units along the first two axes of the Correspondence Analysis, performed using a matrix of sites by fire regime descriptors. Sampling unit numbers correspond to Table 1. Sampling units are presented in three annual rainfall categories: black circles = 600 mm or less; open circles = 650 – 950 mm; inverted triangles = 1000 mm or more.

Fig. 5 Scatterplots of parameters of spatial fire regime vs. average annual rainfall (mm) in each sampling unit. (a) Density of fires. (b) Percentage of burned area. (c) Skewness of burn size distribution. (d) Mean size of individual fires. Values of r 2 and lines correspond to best quadratic fits for a – c and linear fit for d.

that fire is controlled mostly by human influences, particularly ignition sources (e.g. Ellenberg, 1979; Molinillo & Vides-Almonacid, 1989; Lægard, 1992; Young, 1993; Brown, 1995; Kessler, 1995)

and, although the climatic influence on fire regimes along climatic gradients is recognized implicitly, it has not been quantified previously. This study documents that climate is an important factor controlling fire regime. The interaction between anthropogenic and climatic controls of fire emerges as a key research objective for predicting and managing fire in low-latitude Andean ecosystems. In addition to the quantitative assessment of ignition sources (both natural and anthropogenic), the study of the relationship between climate, fire and grazing, and its influence on fuel availability (Adamoli et al., 1990; Savage & Swetnam, 1990; Archer, 1994), should be prioritized. Fire has been recognized as an important ecological factor controlling species composition and diversity (Braun-Blanquet, 1950; Sauer, 1950; Noble & Slatyer, 1980; Miller, 1982; Christensen, 1985; Hobbs & Huenneke, 1992; Huston, 1994; Collins et al., 1995) and ecosystem function (Knapp & Seastadt, 1986; Hobbs et al., 1991; Christensen, 1993; DeBano et al., 1998). At large spatial scales, the effects of disturbances are superimposed with the effects of climate in controlling vegetation gradients (Harmon et al., 1984; Veblen et al., 1992). However, previous studies on the fire–vegetation relationship in the

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 144 Friday, April 20, 2001 3:35 PM

144

H. R. Grau

high-elevation Neotropics have been limited to local-scale and short-term experiments and observations (e.g. Williamson et al., 1986; Horn, 1989; Verweij & Budde, 1992; Keating, 1998). This study shows the potential importance of fire and fire–climate interactions affecting vegetation patterns at large spatial scales. This study assessed patterns during only 1 year. In order to gain a better understanding of fire regime, multi-year studies would be desirable. A multi-year study would allow for descriptions of the relationship between fire regime and climate variation at different temporal scales along the rainfall gradient, and the quantification of the range of inter-annual variability of the fire regime. The latter is important in determining how representative these results are. One important question emerging from this study is how important are unusual events such as very large fires, and what are their effects on the statistics of the fire regime? For example, if large fires, such as the ones observed in the Ambato range, occur in other areas, they would significantly increase the mean percentage burn and decrease the fire return interval. Overall, this study shows the importance of climate in controlling fire regimes at regional scales in Neotropical montane ecosystems, and the potential of remote sensing for the study of the fire–climate–vegetation relationship.

ACKNOWLEDGMENTS This work was supported by grants from the National Geographic Society (USA), International Fund for Science (IFS — Sweden), and Fondo para la Ciencia y la Técnica (FONCYT — Argentina). Image processing was conducted in the Center for the Study of the Earth from Space (CSES) of the University of Colorado at Boulder, directed by Alex Goetz. The manuscript was improved by the comments from Tom Veblen, Carol Wessman, Phil Keating, Lori Daniels, Kathleen Farley, Juan Manuel Morales, Robert Whittaker and two anonymous reviewers.

REFERENCES Aceñolaza, F.G. & Toselli, A. (1981) Geología del Noroeste Argentino, p. 212. Universidad Nacional de Tucumán, Argentina.

Adámoli, J., Sennhauser, E., Acero, J.M. & Rescia, A. (1990) Stress and disturbance: vegetation dynamics in the dry chaco region of Argentina. Journal of Biogeography, 17, 491– 500. Archer, S. (1994) Woody plant encroachment into southwestern grasslands and savannas: rates, patterns and proximate causes. Ecological implications of livestock herbivory in the West (ed. by M. Vavra, W. Laycock & R. Piepper), pp. 13 – 60. Society of Range Management, Denver. Arno, S.F. & Sneck, K.M. (1977) A method for determining fire history in coniferous forests of the mountain west, p. 28. USDA Forest Service General Technical Report INT- 42. Bell, D.A. (1991) Distribución de los bosques de aliso del cerro, Alnus acuminata (Betulaceae), en la provincia de Tucuman, Argentina. Boletin Sociedad Argentina de Botanica, 27, 21–30. Bianchi, A.R. & Yañez, C. (1992) Las Precipitaciones Del Noroeste Argentino, p. 383. Instituto Nacional de Tecnología Agropecuaria, Salta, Argentina. Bond, W.J. & van Wilgen, B. (1996) Fire and plants, p. 215. Chapman & Hall, London. Braun-Blanquet, J. (1950) Pflanzensoziologie, p. 444. Springer, Berlin. Brown, A.D. (1995) Las selvas de montaña del noroeste de Argentina: problemas ambientales e importancia de su conservación. Investigación, Conservación Y Desarrollo En Selvas Subtropicales de Montaña (ed. by A.D. Brown & H.R. Grau), pp. 9 –18. Universidad Nacional de Tucumán, Tucumán, Argentina. Cabrera, A.L. (1976) Regiones fitogeográficas de Argentina, p. 85. ACME, Buenos Aires, Argentina. Cabrera, A.L. & Willink, A. (1980) Biogeografía de America Latina, p. 122. Organization of American States, Washington DC. Christensen, N.L. (1985) Shrubland fire regimes and their evolutionary consequences. The ecology of natural disturbances and patch dynamics (ed. by S.T.A. Pickett & P.S. White), pp. 85 –100. Academic Press, Orlando. Christensen, H.L. (1993) Fire regimes and ecosystem dynamics. Fire in the environment: the ecological, atmospheric and climatic importance of vegetation fires (ed. by P.J. Crutzen & J.G. Goldammer), pp. 233 – 244. Wiley & Sons, Vienna. Chuvieco, E. & Congleton, R.G. (1988) Mapping and inventory of forest fires from digital processing of TM data. Geocarto International, 4, 41–53. Collins, S.L., Glenn, S.M. & Gibson, D.J. (1995) Experimental analysis of intermediate disturbance and initial floristic composition: decoupling cause and effect. Ecology, 76, 486 – 492. DeBano, L., Neary, D.G. & Ffolliot, P.F. (1998) Fire effects on ecosystems, p. 402. Wiley & Sons, Inc., New York.

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 145 Friday, April 20, 2001 3:35 PM

Fire and rainfall gradients in sub-tropical mountains Dwyer, E., Gregoire, J.M. & Malingreau, J.P. (1998) A global analysis of vegetation fires using satellite images: spatial and temporal dynamics. Ambio, 27, 175 –181. Ellenberg, H. (1979) Man’s influence on tropical mountain ecosystems in South America. Journal of Ecology, 67, 401– 416. ENVI. 3.0. (1998) Image visualizing software. Research Systems Inc., Boulder, Colorado. García, M.J.L. & Caselles, V. (1991) Mapping burns and natural reforestation using thematic mapper data. Geocarto International, 6, 31– 37. Glenn, S.M. & Collins, S.L. (1992) Effects of scale and disturbance on rates of inmigration and extinction of species in praires. Oikos, 63, 273– 280. Graetz, R.D. & Gentle, M.R. (1982) The relationship between reflectance in the Landsat wavebands and the composition of an Australian semi-arid rangeland. Photogrammetric Engineering and Remote Sensing, 60, 183 –191. Grau, A. (1985) La expansión del Aliso del cerro (Alnus acuminata H.B.K. subsp. acuminata) en el noroeste de Argentina. Lilloa, 36, 237– 247. Grau, H.R. & Brown, A.D. (1995) Patterns of tree species diversity along latitudinal and altitudinal gradients in the Argentinean sub-tropical montane forests. Biodiversity and conservation of Neotropical montane forests (ed. by S.P. Churchill, H. Balslev, E. Forero & J.L. Luteyn), pp. 295 – 300. New York Botanical Garden, New York. Harmon, M.E., Bratton, S.P. & White, P.S. (1984) Disturbance and vegetation response in relation to environmental gradients in the Great Smoky Mountains. Vegetatio, 55, 129 –139. Hobbs, R.J. & Huenneke, L.F. (1992) Disturbance, diversity and invasion: implications for conservation. Conservation Biology, 6, 324–336. Hobbs, N.T., Schimel, D.S., Owensby, C.E. & Ojima, D.J. (1991) Fire and grazing in tallgrass praire: contingent effects on nitrogen budgets. Ecology, 72, 1374 –1382. Horn, S.P. (1989) Postfire vegetation development in the Costa Rican paramos. Madroño, 36, 93 –114. Huston, M.A. (1994) Biological diversity: the coexistence of species in changing landscapes, p. 681. Cambridge University Press, Cambridge. Keating, P. (1998) Effects of anthropogenic disturbances on Paramo vegetation in Podocarpus National Park, Ecuador. Physical Geography, 19, 221– 238. Kessler, M. (1995) Present and potential distribution of Polylepis (Rosaceae) forests in Bolivia. Biodiversity and conservation of Neotropical montane forests (ed. by S.P. Churchill, H. Balslev, E. Forero & J.L. Luteyn), pp. 281– 294. New York Botanical Garden, New York. Kitzberger, T., Veblen, T.T. & Villalba, R. (1997) Climatic influences on fire regimes along a rain

145

forest-to-xeric woodland gradient in northern Patagonia, Argentina. Journal of Biogeography, 24, 35 – 47. Knapp, A.K. & Seastadt, T.R. (1986) Detritus accumulation limits productivity of tallgrass praire. Bioscience, 36, 662 – 668. Kushla, J.D. & Ripple, W.J. (1988) Assessing wildfire effects with Landsat thematic mapper data. International Journal of Remote Sensing, 19, 2493 – 2507. Lægard, S. (1992) Influence of fire in the grass páramo vegetation of Ecuador. Paramo: an Andean ecosystem under human influence (ed. by H. Balslev & J.L. Luteyn), pp. 151–170. Academic Press, London. Miller, T.E. (1982) Community diversity and interactions between size and frequency of disturbances. American Naturalist, 120, 533 –536. Millones, J. (1982) Patterns of land use and associated environmental problems of the central Andes. Mountain Research and Development, 2, 49 – 64. Minnich, R.A. (1983) Fire mosaics in southern California and northern Baja California. Science, 219, 1287 –1294. Molinillo, M.F. (1993) Is traditional pastoralism the cause of erosive processes in mountain environments? The case of the Cumbres Calchaquies in Argentina. Mountain Research and Development, 13, 189 – 202. Molinillo, M.F. & Vides-Almonacid, R. (1989) Uso de los recursos naturales en el bosque de Aliso de la provincia de Tucumán. Avances Invest INGEMA, 1, 1– 20. Noble, I.R. & Slatyer, R.O. (1980) The use of vital attributes to predict successional changes in plant communities subject to recurrent disturbance. Vegetatio, 43, 5 –12. PC-Ord. (1997) Multivariate analysis of ecological data, Version 3.0 for Windows. MJM Software Desing, Gleneden Beach, Oregon. Pereira, M.C. & Setzer, A.W. (1993) Spectal characteristics of fire scars in Landsat-5 TM images of Amazonia. International Journal of Remote Sensing, 14, 2061– 2078. Pickett, S.T.A. & White, P.S. (eds) (1985) The ecology of natural disturbances and patch dynamics, p. 490. Academic Press, Orlando. Prohaska, F. (1976) The climate of Argentina, Paraguay and Uruguay. World survey of climatology, Vol. 12 (ed. by W. Schwerdtfeger), pp. 13 –112. Pyne, S.J. (1993) Keeper of the flame: a survey of anthropogenic fire. Fire in the environment: the ecological, atmospheric and climatic importance of vegetation fires (ed. by P.J. Crutzen & J.G. Goldammer), pp. 245 – 266. Wiley & Sons, Vienna. Rao, V.B., Cavalcanti, I.F.A. & Hada, K. (1996) Annual variation of rainfall over Brazil and water vapor characteristics over South America.

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146

GEB220.fm Page 146 Friday, April 20, 2001 3:35 PM

146

H. R. Grau

Journal of Geophysical Research, 101, 26539 – 26551. Rothermel, R.C. (1983) How to predict the spread and intensity of forest and range fires, p. 21. USDA Forest Service General Technical Report INT-143. Sabins, F. (1987) Remote sensing: principles and interpretation, p. 449. Freeman, New York. Sauer, C.O. (1950) Grassland climax, fire and man. Journal of Range Management, 3, 16 – 21. Savage, M. & Swetnam, T.W. (1990) Early 19th century decline following sheep pasturing in a Navajo Ponderosa pine forest. Ecology, 71, 2374 – 2378. Schowengerdt, R.A. (1997) Remote sensing: models and methods for image processing, p. 521. Academic Press, San Diego. Sokal, R.R. & Rohlf, F.J. (1995) Biometry: the principles and practice of statistics in biological sciences, 3rd edn, p. 887. Freeman, New York. Solbrig, O. (1984) State of knowledge report on Andean ecosystems, Vol. 4: The Southern Andes and Sierras Pampeanas. Mountain Research and Development, 4, 1–198. SPSS (1995) SPSS for Windows, Release 6.0. Norusis SPSS Inc., Chicago. Swetnam, T. (1993) Fire history and climate change in giant Sequoia groves. Science 262, 885 – 889. ter Braak, C.J.F. (1987) Ordination. Data analysis in community and landscape ecology (ed. by R.G.H. Jongman, C.J.F. ter Braak & O.F.R. van

Tongeren), pp. 91–173. Cambridge University Press, Cambridge. Trabaud, L.V., Christensen, N.L. & Gill, A.M. (1993) Historical biogeography of fire in temperate and Mediterranean ecosystems. Fire in the environment: the ecological, atmospheric and climatic importance of vegetation fires (ed. by P.J. Crutzen & J.G. Goldammers), pp. 277– 298. Wiley & Sons, Vienna. Veblen, T.T., Kitzberger, T. & Lara, A. (1992) Disturbance and forest dynamics along a transect from Andean rain forest to Patagonian shrubland. Journal of Vegetation Science, 3, 507– 520. Veblen, T.T., Kitzberger, T., Villalba, R. & Donnegan, J. (1999) Fire history in northern Patagonia: the roles of humans and climatic variation. Ecological Monographs, 69, 47 – 67. Verweij, P.A. & Budde, P.E. (1992) Burning and grazing gradients in páramo vegetation: initial ordination analysis. Paramo: an Andean ecosystem under human influence (ed. by H. Balslev & J.L. Luteyn), pp. 177–195. Academic Press, London. Williamson, G.B., Schatz, G.E., Avlarado, A., Redhead, C.S., Stam, A.C. & Sterner, R.W. (1986) Effects of repeated fires on tropical paramo vegetation. Tropical Ecology, 27, 62 – 69. Young, K.R. (1993) Tropical treelines: changes in forest structure and regeneration between two Peruvian timberline margins. Arctic and Alpine Research, 25, 167 –174.

© 2001 Blackwell Science Ltd, Global Ecology & Biogeography, 10, 133 –146