International Journal of Wildland Fire, 2002, 11, 33–39

ENSO as a forewarning tool of regional fire occurrence in northern Patagonia, Argentina Thomas Kitzberger Laboratorio Ecotono, Departamento de Ecología, Universidad Nacional del Comahue, Quintral 1250, 8400, Bariloche, Argentina. Telephone: +54 2944 422193; fax: +54 2944 422111; email: [email protected]

Abstract. Composite series of ENSO indices recorded over 36 months preceding major fire years in four National Parks in northern Patagonia were compared with series of these indices for individual years over the period 1950–1996 by means of an additive temporal phase coherence index. Logistic regressions of the dichotomous variable high vs low regional fire occurrence against the coherence index gained highest significant classificatory power using an index based on SST anomaly data between January of year –3 to August of year –1. Thus, warnings of extreme fire seasons could be declared as early as 3 months before the full fire season starts (i.e. early September). A regional fire season readiness index is proposed based on the periodicity of the Southern Oscillation, strong links with climate at particular regions of the globe, and empirically derived climatic controls on fine fuel buildup and coarse fuel desiccation. This long-range alerting tool could help decision-makers prepare preventative measures to mitigate the effects of large, high intensity wildfire seasons. However, it should be used with caution given that differences in timing in the onset of ENSO events and instability in teleconnection patterns could change climatic sequences, differentially affecting fire susceptibility. Additional keywords: ENSO; regional fire occurrence; El Niño; forecasting; Patagonia. Thom asKitzbegrer

TWFh0om10as41Ktizberger ENSO-basedfire index

Short summary. A regional fire season readiness index is proposed based on the periodicity of the Southern Oscillation, strong links with climate at particular regions of the globe, and empirically derived climatic controls on fine fuel buildup and coarse fuel desiccation developing up to 3 years before the fire season. By comparing a mean behavior of the ENSO system during extreme fire years in northern Patagonia and the behavior during each year in the period 1954–1996, an index is constructed that is capable of giving a warning as early as 3 months before the full fire season . This index could be used as a long-range alerting tool that could help decision-makers prepare preventative measures to mitigate the effects of large, high intensity wildfire seasons in northern Patagonia or other ENSO-sensitive regions.

Introduction Under extreme climatic conditions, temperate as well as tropical forested regions are exposed to very large wildfires and very active fire seasons. Many of these climatic anomalies worldwide are linked to the phenomenon of El Niño–Southern Oscillation (ENSO) (Leighton and Wirawan 1986; Swetnam and Betancourt 1990; Nicholls 1992; Meggers 1994; Kitzberger and Veblen 1997; Brown 1998; Laurance 1998). Anomalous coupled oceanographic–atmospheric conditions which recur periodically over the tropical Pacific Ocean and the atmospheric interactions (teleconnections) with extratropical regions constitute the single largest source of climatic variability on a global scale (Díaz and Markgraf 1992). © IAWF 2002

Climate variability in southern South America is strongly influenced by the strength and latitudinal position of the south-east Pacific subtropical anticyclone blocking the flow of moist Pacific air into the continent by steering westerly cyclonic storms southwards. Small interannual deviations in latitude, particularly during its spring–summer southward expansion, are accompanied by substantial anomalies in regional precipitation and temperature (Pittock 1980). The degree of anomalous tropical Pacific convection related to ENSO controls the strength of the south-east Pacific high pressure cell (Aceituno 1988), thus indirectly influencing climate over large portions of southern South America. Coincident with El Niño events (the warm or negative phase of the Southern Oscillation) and a weak, northerly displaced 10.1071/WF01041

1049-8001/02/010033

34

anticyclone, winter–spring precipitation is abundant over mid-latitudes of the Pacific coast of South America and summer temperature is above average. Conversely cold La Niña, related to a strong, southerly spread anticyclone is associated with the opposite conditions (Aceituno 1988; Kiladis and Díaz 1989). The hydrological cycle in northern Patagonia is mostly dominated by cool season (winter and early spring) precipitation, and variations in this seasonal moisture are correlated with southern-oscillation indices (SOI), sea surface temperatures (SST) and tree-ring width indices (Villalba 1990, 1994; Villalba and Veblen 1997). Tree-ring fire reconstructions based on sampling and analyses of hundreds of widely distributed fire-scarred trees that span 3–4 centuries back in time indicate that fires burned synchronously over wide areas of northern Patagonia when the previous 1–2 year period had reduced cool season moisture, generally preceded by an increased moisture during the previous 3–4 years (Kitzberger et al. 1997; Veblen et al. 1999). Increased amplitudes of the Southern Oscillation can lead to an increased fire activity over ENSO-sensitive regions by at least two different mechanisms. First, increased fine fuel production in previous seasons and years appears to be quite important to regional fire. A pattern of 1 to several years of El Niño conditions leads to fuel buildup via both increased plant productivity and decreased fire activity. Second, La Niña conditions following El Niño events result in drying of the accumulated fuels, leading to more successful fire ignitions and widespread burning (Baisan and Swetnam 1990; Swetnam and Betancourt 1990; Kitzberger and Veblen 1997). It is encouraging, however, that the onset and relatively predictable development of ENSO events precede the affected fire season by months to years, because this offers a window of opportunity for planning and action. Preventative measures and fire fighting readiness might reduce the occurrence and mitigate the effects of large, high intensity fire events in the future, but only if decision makers are alerted to the hazards beforehand. This paper proposes an objective empirically derived ENSO index-based forecasting methodology for predicting fire season activity. Study area The study area comprises four major national parks (Lanín, Nahuel Huapi, Lago Puelo or Los Alerces National parks) that spread over a N–S distance of approximately 400 km and from the Andean continental divide towards the eastern Patagonian plains in northern Patagonia, Argentina (~39°S to 43°S; 71°W to 72°W). The total protected area is approximately 1.4 million ha (Bruno and Martin 1982) and encompasses a steep precipitation gradient. Along the gradient major vegetation types changes from mesic ~40 m tall Nothofagus dombeyi dominated evergreen forests, to

Thomas Kitzberger

mixed evegreen-conifer (N. dombeyi–Austrocedrus chilensis) forests and Nothofagus antarctica shrublands, which eastward open up into pure A. chilensis conifer woodlands surrounded by a matrix of Patagonian steppe. In the northern park a second conifer, Araucaria araucana, is a dominant component of the forest, developing mixed stands with Nothofagus as well as pure forests and woodlands. Spatially, fires regimes change from high intensity large fires in the wet forest, mixed high intensity and surface fires in woodlands, and grassland fires in the steppe. Temporal changes in fire regimes in northern Patagonia have been linked to both changes in human occupation and climatic variation (Veblen and Lorenz 1988; Veblen et al. 1992; Kitzberger et al. 1997). Previous studies have shown an abrupt rise in fire frequency in the mid–19th Century coincident with increased Native American use of the area. These studies also have related the abundance of ~100-year old forest cohorts to massive burning of European colonists in the late 1800s, and increases in woodland densities during the present century with a sharp decline in surface fires formerly set by Native American hunters (Veblen and Lorenz 1987; Veblen and Markgraf 1988; Kitzberger and Veblen 1997). Despite the fact that most modern fires are set by humans (lightning accounts for 15% of all ignitions and 21% of the area burned; Bruno and Martin 1982), previous research based on instrumental as well as pre-instrumental records has demonstrated strong controls of interannual climatic variablity (e.g. ENSO-related variability) on regional fire spread and occurrence (Kitzberger et al. 1997; Kitzberger and Veblen 1997; Veblen et al. 1999). Methods The temporal pattern of monthly climatic anomalies associated with high regional fire occurrence were investigated by compositing indices of the Southern Oscillation and Sea Surface Temperature over a 36month moving window preceding the year of extensive fires in northern Patagonia. The mean anomaly in the Southern Oscillation Index (SOI; standardized sea level pressure difference between Tahiti and Darwin, Australia) and the mean sea surface temperature (SST) anomaly for Niño regions 1 and 2 (between 80°–90°W and 0°–10°S) in the eastern tropical Pacific were composited for extensive fire years (> 2000 ha burned 1950–1996, n = 14) based on fire reports of the Argentinean National Park Service (Bruno and Martin 1982; Administración de Parques Nacionales, unpublished data). An index was developed to measure the degree in which monthly ENSO conditions that precede a particular fire year are in phase (coherent) or out of phase with the mean composite ENSO behavior that lead to extensive fire years in northern Patagonia. High/low coherence between the composite and the individual year monthly series should indicate high/low regional fire risk for that particular year. Let ci be the composite value for each of the 36 monthly ENSO indices (SST or SOI) preceding years of extensive fire and xki and be each raw ENSO index of the 36 monthly series for an individual year k. For a particular year k, the product ci xki and its sum over i months takes increasing positive values when ci and xki are of the same sign and depart from 0 (when they are in phase). This situation is indicative of a high fire risk for that particular year. Conversely, when ci and xki are of opposite sign during individual months (or during longer periods), the

35

Fire Season

0.75

SST SOI

0.50 0.25 0.00

–1.00 J FMAM J J A SOND J FMAM J J A SOND J FMAM J J A SOND J FM

year –3

year –2

year –1

year 0

Fig. 1. Mean composite of standardized sea surface temperatures (SST) in the eastern equatorial Pacific (Niño Region 1+2; solid line) and Southern Oscillation Index (SOI; dotted line) calculated for a moving window of 36 months preceding the onset of the fire season centred on years of extensive fire (> 2000 ha burned annually in four National Parks; 1950–1996; n = 14). The shaded area is the fire season of the fire-event year (i.e. year 0). Fire reports are based on Bruno and Martin (1982) and Administración de Parques Nacionales (unpublished data). Arrows indicate the three forecasting times (May, August and December of year –1) analysed in this study.

product (and its sum over i months) takes increasing negative values, indicating that the composite and the individual series are out of phase or uncoherent and a relatively low fire risk is expected. Thus, a coherence index Ik of the form: k

=Σc x i

ki

was calculated for each of the k years between 1950 and 1996. To make this (or any) fire risk index useful for warning systems it should be readily calculable well before the fire season. Therefore, three forecasting times were tested:

䊉

2 1 0 –1 –2

A very early forecasting in May (year –1); this is in late fall before the fire season; In August (year –1); this is late winter before the fire season; and In December (year –1), right at the beginning of the fire season (high fire season is December (year –1 to March (year 0; Fig. 1).

Thus, calculations of Ik were done summing January (year –3) to either May, August or December of year –1 to obtain IkMay, IkAug, and IkDec, respectively. Non-linear threshold-type functions between Ik and annual area burned led to the use of logistic regression. This procedure is used when the dependent variable can be classified into discrete outcomes, in this case years with high fire incidence (1) and years with low fire incidence (0). The model describes the probability of a year k of belonging to population 1 (high fire incidence, PI) as:

P = 1 / {1 + exp[–( b + b I )]}, I

3

IAug = –3.65 40 ha burned

2 1 0 –1 –2

J FMAM J J A SOND J FMAM J J A SOND J FMAM J J A SOND J FM

year –3

year –2

year –1

year 0

Fig. 2. Standardized sea surface temperatures in the eastern equatorial Pacific (Niño Region 1+2; SST) of 36 months (solid line) preceding a active fire year (1975, a) and a year of low fire activity (1981, b); and mean composite of SST standardized anomalies of 36 months preceding fires seasons of extensive fire (> 2000 ha burned annually in four National Parks; 1950–1996; n = 14; dashed line). Bars indicate standardized deviations from the long-term mean monthly precipitation based on the Bariloche Airport weather station (1905–1996). Annual area burned within four National Parks and coherence index IkAug are indicated for both individual years.

Results

i

I

䊉

IAug= 8.94 3049 ha burned

1981 Fire season –0.50 –0.75

䊉

1975 Fire season

3

–0.25

Standard Deviation Units

Standard Deviation Units

1.00

Standard Deviation Units

ENSO-based fire index

0

1

where b0 and b1 are coefficients of the model. To select the model with highest predictive power, logistic regressions were performed for indices constructed from SST and SOI anomaly series with three forecasting points in May, August and December of year –1. Estimation was performed minimizing the least squares loss function (sum of squared deviations of the observed values for the dependent variable from those predicted by the model).

Composites based on averages of years of most extensive wildfire in northern Patagonia between 1950 and 1996 are characterized by mean SOI and SST anomalies approximately +0.5 s.d. and –0.5 s.d., respectively, during a period of 18–20 months preceding the fire season (i.e. until fall of year –2), corresponding to cold La Niña conditions in the tropical eastern Pacific (Fig. 1; Appendix I). A reversal of similar magnitude in the pattern is detected during year –3, corresponding to previous warm El Niño conditions. Smallest deviations from 0 are evident during the summers (JFM) of all years, coincident with the season in which a reversal in climatic teleconnection for northern Patagonia has been detected (i.e. drier warmer conditions during El Niño; Aceituno 1988; Kitzberger and Veblen, in press). Previous analyses indicate that years of very little annual area burned show the reversed composite pattern of El Niño conditions during 1–2 years previous to the fire season with a reversal to the preceding La Niña stage by year –3 (Kitzberger and Veblen 1997; Fig. 2b). As examples, the evolution of standardized SST prior to the very active 1975 fire season shows a relatively tight coherence with the composite series, with above average SST (El Niño conditions) during year –3, followed by a sudden drop in SST during almost the entire 2 years preceding the fire season, confirmed by a relatively high coherence index value of 8.94 (Fig. 2). In contrast, the 1981

36

Thomas Kitzberger

2

16000

15

Fire ocurrence (>250 ha/year)

1.0 14000

Area burned (ha)

12000

10000

8000

0.8 0.6 0.4 0.2 2

25

0.0 6000

–15

–10

–5

0

5

10

15

IAug

4000

2000

0 –15

–10

–5

0

5

10

15

IAug Fig. 3. Annual area burned (ha) within four National Parks as a function of the coherence index IkAug. The vertical dashed line corresponds to the value of IkAug =1.41, which is the cutoff value where the probability of occurrence of years with > 250 ha burned is 0.5 (based on logistic regression).

season characterized by very little fire activity shows almost a reversed pattern with below-average SST during year –3 and above- or near-average SST during the 2 years preceding the fire season and a low coherence value of –3.65 (Fig. 2). A plot of all coherence indices of the entire period as a function of the annual area burned shows a pronounced steplike function with an IkAug threshold value of approximately 1.4 (Fig. 3). Annual area burned for coherence values below the threshold fall generally below 500 ha with many values close to 0 whereas, at coherence values above 1.4, area burned quickly increases to several thousand hectares (Fig. 3). Similar relations appeared for IkMay and IkDec. This relationship suggests that the annual area burned increases

Fig. 4. Fitted logistic regression of fire occurrence (a binary variable that takes the value of 1 when years have > 250 ha burned within four National Parks and 0 when the annual area burned is < 250 ha) as a function of the coherence index IkAug. The vertical dotted line corresponds to the value of IkAug =1.41, which is the cutoff value where the probability of occurrence of years with > 250 ha burned is 0.5. Solid points indicate years correctly classified and open dots correspond to years incorrectly classified by the model, based on the 1.41 cutoff value. Numbers indicate their corresponding frequencies.

abruptly when monthly ENSO indices mimic at least in the trend (not necessarily the intensity) the pattern reversal from El Niño to La Niña conditions from year –3 to years –2 and –1 evidenced in the composite series (Fig. 1). Logistic regressions performed between wildfire incidence (0: < 250 ha burned, 1: > 250 ha burned) and coherence indices indicate highest predictive for IkAug (Table 1; Fig. 4). Earlier (May) and later (December) forecasting times gave poorer fits to the models. Alternative criteria to define fire occurrence (150 ha and 500 ha thresholds) as well as coherence indices constructed to include also year –4, resulted in all cases in models with lower explanatory power. The model that best explained (53.9% of the variance) was when the incidence of fire was defined by a 250 ha threshold and when the coherence index was calculated from year –3 to August preceding the fire season as the forecasting time (Table 1; Appendix II). Taking the 0.5 probability as the classification threshold, this model

Table 1. Percentage variance explained by logistic regressions between the incidence of fire (0: < 250 ha burned, 1: > 250 ha burned) and coherence indices Ik constructed from SST and SOI anomaly series starting January year –3 and ending at three forecasting times: May year –1 (Ik May), August year –1 (IkAug), and December year –1 (IkDec) Regressed data span the period 1950–1996. Goodness of the model fit to the data was computed with least squares (LS) loss function, computing the variance explained (VE%) b0 SST SOI

LS LS

–2.024 –0.614

IkMay b1 1.257 0.191

VE%

b0

47.3% 20.4%

–2.334 –0.609

IkAug b1 1.656 0.196

VE%

b0

IkDec b1

VE%

53.9% 21.7%

–2.730 –0.586

2.032 0.188

51.2% 21.4%

Standard Deviation Units

Standard Deviation Units

ENSO-based fire index

37

1983 Fire season 4 3

a

IAug = 0.38 275 ha burned

2 1 0 –1 –2 1955 Fire season

3 2

IAug = 0.28

b

737 ha burned

1 0 –1 –2 –3 J FMAM J J A SOND J FMAM J J A SOND J FMAM J J A SOND J FM

Standard Deviation Units

year –3

year –2

year –1

year 0

1972 Fire season

3

c

IAug = 8.53 20 ha burned

d

IAug = 6.89 3 ha burned

2 1 0 –1 –2

Standard Deviation Units

1968 Fire season 3 2 1 0 –1 –2 J FMAM J J A SOND J FMAM J J A SOND J FMAM J J A SOND J FM

year –3

year –2

year –1

year 0

Fig. 5. Fire years misclassified by the model. Standardized anomalies in sea surface temperature in the eastern equatorial Pacific (Niño Region 1+2; SST) of 36 months (solid line) preceding two years predicted to be of low fire activity but which actually had > 250 ha burned (a and b) and two years predicted to be of high fire activity but that had <250 ha burned (c and d); and mean composite of SST standardized anomalies of 36 months preceding fires seasons of extensive fire (> 2000 ha burned annually in four National Parks; 1950–1996; n = 14; dashed line). Bars indicate standardized deviations in monthly precipitation based on the Bariloche Airport weather station (1905–1996). Annual area burned within four National Parks and coherence index IkAug are indicated for individual years.

misclassified 4 out of 43 years (Fig. 4). A contingency analysis of the classificatory power of this threshold (I Aug = 1.41) was highly significant (χ2 = 28.7, 1 d.f., P < 0.00001). Two years (1955 and 1983) that the model predicted to be of low fire incidence (< 250 ha) had actually 737 and 275 ha burned, respectively (Fig. 5a and 5b), and two years predicted to be of high fire occurrence (1968 and 1972) had 3 and 20 ha burned, respectively (Fig. 5c and 5d).

Discussion This model suggest that it is possible to assign probabilities of occurrence of high vs low wildfire seasons in northern Patagonia based on the evolution of ENSO anomalies during the 36 months that precede the fire season. In addition, warnings of extreme fire seasons could be declared as early as 3 months before the full fire season starts (i.e. early September). However, considerable care should be taken with the use of this index as the model misclassified almost 10% of the cases. Several reasons explain the anomalous predictions. First, ENSO events differ in timing and, because the seasonal climatic signal switches (e.g. moist cool springs to warm dry summers during warm phases), lagged events may induce a different temporal sequence of climatic anomalies, which in turn could influence fire proneness (Veblen et al. 1999). This is the case of the relatively active 1983 fire season, predicted to be a season with low fire occurrence. In this case a very strong El Niño event developed during mid 1982, producing a relatively dry and hot fire season. Had the warm event initiated earlier in 1982, moist cool winters and springs may have prevented the occurrence of widespread fires. As a matter of fact the 1984 fire season, located at the end of the warm event, had less fire activity (30 ha burned). If the cold event is strong enough, a single year of cold phase conditions may suffice to create fire prone conditions, as was the case prior to the 1955 fire season (Fig. 5b). Second, not all events produce the same climatic teleconnection pattern (Villalba 1994). Therefore, some cold events predicted to relate to drier winters–springs could have anomalous behaviors. This is possibly the case prior to the 1972 and 1968 fire seasons during which, despite the permanent cold ENSO state, winters and springs were interrupted by anomalous high precipitation that possibly reduced the risk of fire during the subsequent summer (Fig. 5c and 5d). Third, the model assumes that, over the study region (four National Parks), ignition sources are not limiting, thus the amount of area annually burned is a function of fuel moisture. Given the high interannual variability in lightning occurrence (Kitzberger and Veblen, in press) this may not be the case during some years. However, this limitation may be averaged out by an increasing human use of these areas (tourism, houses, cities). Finally, the use of this index assumes that the ENSO–climate relationships are stable over time and in the future. The historical record shows that tropical–extratropical teconnection patterns in general have changed in different multi-decadal periods (Villalba and Veblen 1997), which could in turn modify long-term regional fire regimes (Swetnam and Betancourt 1998). For instance, the climatic signal of the Southern Oscillation in southern South America weakened and virtually disappeared during the 1930–1940 period (Villalba and Veblen 1998). Despite the high

38

variability in the timing of events and instability of the ENSO–climate relation, long-term data based on regional networks of fire scars and historical documentary records of El Niño events (1520–1929) revealed that years of widespread fire in northern Patagonia coincided either with the warm summer related to the onset of the El Niño events or occurred the year that preceded the onset of El Niño events; or, in other words, dry spring conditions related to late stages of La Niña events (Veblen et al. 1999). Unfortunately the length of both the fire records as well as the ENSO indices prevented an independent test of these relationships based on a reasonably long validation period. Therefore, until new cases are accumulated to produce statistical validation, this analysis should be considered tentative. In contrast to short-term fire danger indices which predict individual fire occurrence and behavior based on the current state of fuels as a function of measured past and present climatic conditions, this index predicts in a coarser dichotomous fashion the severity of the fire season. This approach takes advantage of the relatively predictable periodicity and high temporal autocorrelation of the Southern Oscillation, strong links with climate at particular regions, and the fact that fine fuel buildup and coarse fuel moisture conditions depend in a time-lagged fashion on past climatic conditions. Therefore, rather than providing fire readiness at particular sites and during particular times of the fire season, this index could be used as an long-range alerting tool that could help decision makers prepare preventative measures to mitigate the effects of large, high intensity wildfire seasons in northern Patagonia or other ENSO-sensitive regions elsewhere. Acknowledgements This research was supported by the National Science Foundation, Universidad Nacional del Comahue and CONICET of Argentina. For helpful comments on the manuscript I thank T.T. Veblen. References Aceituno P (1988) On the functioning of the Southern Oscillation in the South American sector. Part 1. Surface climate. Monthly Weather Review 116, 505–524. Baisan CH, Swetnam TW (1990) Fire history on a desert mountain range: Rincon Mountain Wilderness, Arizona, U.S.A. Canadian Journal of Forest Research 20, 1559–1569. Brown N (1998) Out of control: fires and forestry in Indonesia. TREE 13, 41. Bruno J, Martín G. (1982) Los incendios forestales en los Parques Nacionales. Unpublished report, Administración de Parques Nacionales, Buenos Aires. Díaz HF, Markgraf V (Eds) (1992) ‘El Niño: Historical and paleoclimatic aspects of the Southern Oscillation.’ (Cambridge University Press: Cambridge)

Thomas Kitzberger

Kiladis GN, Díaz HF (1989) Global climatic anomalies associated with extremes in the Southern Oscillation. Journal of Climate 2, 1069–1090. Kitzberger T, Veblen TT (1997) Influences of humans and ENSO on fire history of Austrocedrus chilensis woodlands in northern Patagonia, Argentina. Ecoscience 4, 508–520. Kitzberger T, Veblen TT, 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. Kitzberger T, Veblen TT (in press) Influences of climate on fire in northern Patagonia, Argentina. In ‘Fire and climatic changes in temperate ecosystems of the Western Americas’. (Eds TT Veblen, W Baker, G Montenegro and TW Swetnam) (Springer: New York) Laurance WF (1998) A crisis in the making: responses of Amazonian forests to land use and climate change. TREE 13, 411–415. Leighton M, Wirawan N (1986) Catastrophic drought and fire in Borneo tropical rain forest associated with the 1982–1983 El Niño Southern Oscillation event. In ‘Tropical rain forest and the world atmosphere’. (Ed. GT Prance) pp. 75–102 (Westview Press: Boulder) Meggers BJ (1994) Archeological evidence for the impact of megaNiño events on Amazonia during the past two millennia. Climatic Change 28, 321–338. Nicholls N (1992) Historical El Niño/Southern Oscillation variability in the Australasian region. In ‘El Niño: Historical and paleoclimatic aspects of the Southern Oscillation’. (Eds HF Díaz and V Markgraf) pp. 151–173. (Cambridge University Press: Cambridge) Pittock AB (1980) Patterns of climatic variation in Argentina and Chile. I. Precipitation, 1931–60. Monthly Weather Review 108, 1347–1361. Swetnam TW, Betancourt JL (1990) Fire–southern oscillation relations in the southwestern United States. Science 249, 1017–1020. Swetnam TW, Betancourt JL (1998) Mesoscale disturbance and ecological response to decadal climatic variability in the American Southwest. Journal of Climate 11, 3128–3147. Veblen TT, Lorenz DC (1987) Post-fire stand development of Austrocedrus–Nothofagus forests in Patagonia. Vegetatio 73, 113–126. Veblen TT, Lorenz DC (1988) Recent vegetation changes along the forest/steppe ecotone in northern Patagonia. Annals of the Association of American Geographers 78, 93–111. Veblen TT, Markgraf V (1988) Steppe expansion in Patagonia? Quaternary Research 30, 331–338. Veblen TT, 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 TT, Kitzberger T, Villalba R, Donnegan J (1999) Fire history in northern Patagonia: the roles of humans and climatic variation. Ecological Monographs 69, 47–67. Villalba R (1990) Climatic fluctuations in northern Patagonia during the last 1000 years as inferred from tree-ring records. Quaternary Research 34, 346–360. Villalba R (1994) Tree-ring and glacial evidence for the Medieval Warm Epoch and the Little Ice Age in southern South America. Climatic Change 26, 183–197. Villalba R, Veblen TT (1997) Spatial and temporal variation in tree growth along the forest–steppe ecotone in northern Patagonia. Canadian Journal of Forest Research 27, 580–597. Villalba R, Veblen TT (1998) Influences of large-scale climatic variability on episodic mortality at the forest–steppe ecotone in northern Patagonia. Ecology 79, 2624–2640.

ENSO-based fire index

39

Appendix I. Mean composite of sea surface temperature (SST) standardized anomalies in the eastern equatorial Pacific (Niño Region 1+2) and Southern Oscillation Index (SOI) calculated for a moving window of 36 months preceding the onset of the fire season centered on years of extensive fire (> 2000 ha burned annually in four National Parks; 1950–1996; n = 14) Month Year –3 January February March April May June July August September October November December Year –2 January February March April May June July August September October November December Year –1 January February March April May June July August September Otober November December

SST

SOI

–0.165 0.090 0.330 –0.040 0.181 0.188 0.245 0.298 0.342 0.356 0.203 0.260

–0.367 –0.667 –0.533 –0.333 –0.578 –0.711 –0.500 –0.444 –0.444 –0.478 –0.033 –0.078

0.223 –0.031 –0.320 –0.562 –0.424 –0.457 –0.328 –0.624 –0.374 –0.328 –0.332 –0.347

–0.333 0.089 –0.267 –0.056 0.033 –0.067 0.222 0.444 0.467 –0.000 0.556 0.389

–0.233

0.789

–0.057 –0.313 –0.641 –0.543 –0.452 –0.378 –0.283 –0.087 –0.174 –0.217 –0.246

0.178 0.133 0.311 0.522 0.256 0.400 –0.122 0.211 0.411 –0.078 0.011

Appendix II. Probability of occurrence of years with > 250 ha burned over four National Parks in northern Patagonia, Argentina as a function of coherence values based on SSTs and a cutoff month of August year –1 IAug

PI

IAug

PI

–5.55 –4.20 –4.15 –2.75 –1.35 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0

0.0000 0.0001 0.0001 0.0010 0.0100 0.1000 0.1189 0.1374 0.1582 0.1816 0.2075 0.2360 0.2672 0.3009 0.3368 0.3747 0.4143 0.4550 0.4962 0.5376 0.5784 0.6182 0.6564 0.6928 0.7269

2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4.0 4.1 4.2 5.6 7.0 8.4

0.7585 0.7875 0.8139 0.8377 0.8590 0.8779 0.8946 0.9092 0.9220 0.9331 0.9427 0.9510 0.9582 0.9644 0.9696 0.9741 0.9780 0.9813 0.9841 0.9865 0.9885 0.9900 0.9990 0.9999 1.0000

http://www.publish.csiro.au/journals/ijwf