Climate and wildfires in the North American boreal forest

Abstract

The area burned in the North American boreal forest is controlled by the frequency of mid-tropospheric blocking highs that cause rapid fuel drying. Climate controls the area burned through changing the dynamics of large-scale teleconnection patterns (Pacific Decadal Oscillation/El Niño Southern Oscillation and Arctic Oscillation, PDO/ENSO and AO) that control the frequency of blocking highs over the continent at different time scales. Changes in these teleconnections may be caused by the current global warming. Thus, an increase in temperature alone need not be associated with an increase in area burned in the North American boreal forest. Since the end of the Little Ice Age, the climate has been unusually moist and variable: large fire years have occurred in unusual years, fire frequency has decreased and fire–climate relationships have occurred at interannual to decadal time scales. Prolonged and severe droughts were common in the past and were partly associated with changes in the PDO/ENSO system. Under these conditions, large fire years become common, fire frequency increases and fire–climate relationships occur at decadal to centennial time scales. A suggested return to the drier climate regimes of the past would imply major changes in the temporal dynamics of fire–climate relationships and in area burned, a reduction in the mean age of the forest, and changes in species composition of the North American boreal forest.

1. Introduction

Lightning-caused wildfires are the main disturbance in the North American boreal forest. The boreal forest is a mosaic of stands of different ages since they last burned. Large fires account for over 85% of the total area burned but less than 5% of the fires. The landscape of the boreal forest is thus determined not by the numerous small fires but by the infrequent large fires (Johnson et al. 1998; Stocks et al. 2003). The frequency of large fire years, and thus area burned, in Canada has increased in the last four decades of the twentieth century (figure 1). Such increases have been attributed in part to global warming (Gillett et al. 2004; Skinner et al. 2006).

Figure 1.

Annual forested area burned in Canada for the period 1918–2005.

The northern latitudes are expected to suffer the largest temperature increases under a global change scenario (IPCC 2001). Based on this, general circulation models have been used to predict increases in area burned of up to 118% from the present in doubled and tripled CO₂ scenarios by the end of the twenty-first century (Gillett et al. 2004; Flannigan et al. 2005). Overall, the area burned in Canada is predicted to increase due to an increase in temperature, despite regional differences—for example predicted decreases in area burned for the eastern North American boreal forests (Flannigan et al. 1998; Carcaillet et al. 2001).

Besides the increase in area burned, the increase in temperatures associated with climate warming by increased CO₂ has been predicted to lengthen the fire season (e.g. Street 1989; Wotton & Flannigan 1993; Stocks et al. 1998; Flannigan & Wotton 2001; Flannigan et al. 2005) and also to increase its intensity due to the increased drying capacity of the air (Street 1989; Johnson 1993). The fire season (defined in terms of area burned and not number of fires) in the boreal forest of North America is relatively short and spans from May to August, starting earlier in the southern parts of the boreal forest and later further north. Typically, the area burned peaks between the second half of June and the first half of July, followed by a sharp reduction in August. Latitudinal changes have been related to the spring northward advance (and the progressive penetration of warm Pacific air into the continent) and the autumn southward advance of the polar front (Johnson 1992). In the western mixedwood boreal forest, changes in the length of the fire season during the last two centuries have been reported (Johnson et al. 1999), with longer fire seasons before the late nineteenth century, synchronous with a period of higher fire activity.

An increase in area burned and in the frequency of large fire occurrence would change the landscape age distribution and reduce the mean age of the boreal forest (e.g. Johnson & Fryer 1989; Johnson 1992). This would probably have implications for species distribution, for example, favouring species with earlier reproductive ages. Desponts & Payette (1993) found that a reduced interval between fires favours jack pine (Pinus banksiana Lamb.) over black spruce (Picea mariana Mill.).

A change in the area burned might also change the shapes of polygons within the landscape age mosaics, with shorter average fire intervals resulting in an age mosaic of younger, larger oblong-shaped polygons with irregular edges (Weir et al. 2000). Finally, an increase in the area burned would have consequences for the global carbon budget. Traditionally, higher fire activity in the boreal forest has been predicted to enhance carbon emissions and diminish (if not completely reverse) the role of the boreal forest as a carbon sink (Kasischke et al. 1995; Kurz & Apps 1999). Later studies have suggested different outcomes, depending on the frequency of fires, when taking into account changes in the albedo related to fires (Randerson et al. 2006).

However, to understand if global warming will increase the area burned in the boreal forest, it is essential that we understand the mechanisms that link weather and climate to area burned. Increases in temperature alone do not explain increases in area burned. In fact, most studies on fire frequency in the boreal forest of North America show that, despite significant increases in temperature since the end of the Little Ice Age, the frequency of fires decreased in the last 150 years (e.g. Bergeron & Brisson 1990; Flannigan et al. 1998; Lesieur et al. 2002). That is, an increase in temperature in the last 150 years did not increase the area burned, until the last decades of the twentieth century. This increase in recent decades is probably not due to fire suppression since large parts of the boreal forest are not subject to fire protection.

The objective of this paper is to describe what is known at present about the connection between large-area-burned years and the weather and climate that appear to cause these events. We do so in the following order:

1. The first part of the paper explains in detail the kind of wildfires that occur in the North American boreal forest and how these wildfires are related to climate.

2. Next, we address how climate has controlled large fire occurrence during the last decades of the twentieth century.

3. Then, the focus turns to what is known about the long-term past fire frequency (before fire reports), using cumulative time-since-fire distributions obtained from time-since-last-fire maps and palaeoecological studies based on lake sediments, macrofossils and ice cores.

4. Finally, the different time scales of the climate–wildfire system are considered to address what effect global warming might have on the North American boreal forest.

2. The relationship between fire and climate

In the boreal forest, crown fires (as opposed to surface fires) are the dominant fire behaviour (Johnson 1992; Johnson et al. 2001) and, generally, fuel moisture (weather) and not fuel accumulation are the primary determinant of fire behaviour (Bessie & Johnson 1995; Johnson et al. 2001; Keeley & Fotheringham 2001). This is because wildfires are propagated primarily by fine fuels which, in closed-canopy boreal forests, are relatively constant after canopy closure at 15 or 20 years.

The large area burned in the boreal forest of North America is related to the presence of a persistent (more than 10 days) mid-tropospheric (500 hPa) ridge (Schroeder et al. 1964; Newark 1975; Alexander et al. 1983; Fryer & Johnson 1988; Johnson & Wowchuk 1993); for example, large fire years in the Yukon Territory are characterized by persistent high pressures over this region (figure 2). Under such conditions, zonal air flow is blocked, enhancing meridional inflow of warm air and dominant air subsidence, both of which generate warmer and drier weather. These conditions result in rapid drying of fuels over large areas and high fire hazard (e.g. Johnson & Wowchuk 1993). Several authors have found empirically that the most fire-prone situations occur when the upper air ridge is located directly above or immediately upstream of the area burned (Street & Birch 1986; Skinner et al. 1999; Macias Fauria & Johnson 2006). Thus, the frequency of mid-tropospheric blocking highs relates to a large area burned in the boreal forest.

Figure 2.

Northern Hemisphere 500 hPa height anomaly chart for large fire years in Yukon Territory. H, high geopotential height anomalies; L, low geopotential height anomalies. Note the prominent mid-tropospheric blocking highs over the area of Yukon Territory (northwest Canada) .

Large fires usually occur during the breakdown of these persistent mid-tropospheric ridges and are associated with the occurrence of troughs (short waves) that generate strong surface winds, enhance lightning and do not produce enough rain to moisten the fuel (Flannigan & Harrington 1986; Street & Birch 1986).

(a) Large-scale fire–climate relationships

Johnson & Wowchuk (1993) described how large fire years in the southern Canadian Rocky Mountains were related to the positive phase of the Pacific North American (PNA) teleconnection (Dickson & Namias 1976; Namias 1978; Wallace & Gutzler 1981). The positive mode of the PNA consists of high 500 hPa heights centred over western Canada and low heights over the North Pacific and southeast USA (figure 3). This finding pointed out, for the first time, the importance of large-scale Northern Hemisphere climatic mechanisms in fire occurrence.

Figure 3.

Chart showing point correlations between Northern Hemisphere 500 hPa height and PNA index, based on spring–summer (March–August) anomalies. Contour interval is 0.2. Note the most prominent patterns of positive PNA—highs over western North America and lows in the Gulf of Alaska and southeastern USA. Data and software for the figure obtained from the NCEP/NCAR Reanalysis Project (Kalnay et al. 1996).

The main features of the normal climatological summer (fire season) flow in northern North America for the last half of the twentieth century (figure 4) are (i) a weak southward-extending trough located adjacent to the west coast of North America—the west coast trough (WCT), (ii) a parallel strong northward-extending ridge over western and northwestern North America, ranging from the west coast to approximately 110° W and extending from the mid-latitudes to Alaska—the western Canadian ridge (WCR), (iii) a strong southward-extending trough over northeastern North America—the Canadian Polar trough (CPT; Skinner et al. 1999) and, finally (iv) upper air height variability over the Arctic Ocean (figure 5, see below).

Figure 4.

Mean mid-tropospheric (500 hPa) Northern Hemispheric flow during the fire season (defined as May–August) for the period 1959–1999. The main circulation features are shown as vertical lines. From west to east these are: west coast trough (WCT), western Canadian ridge (WCR) and Canadian polar trough (CPT) .

Figure 5.

The Arctic Oscillation (AO). (a) Chart showing point correlations between Northern Hemisphere sea-level pressure (SLP) and Arctic Oscillation, based on the fire season (May–August) anomalies. Contour interval is 0.2. Note the most prominent patterns of positive AO—lows over the Arctic Ocean and a belt of highs at mid-latitudes, especially prominent in the north Atlantic. Data and software for the figure obtained from the NCEP/NCAR Reanalysis Project (Kalnay et al. 1996). (b) Summer (June–August) AO index, obtained from the National and Ocean and Atmospheric Administration (NOAA; http://www.cpc.noaa.gov/).

Large fires are usually linked to positive 500 hPa anomalies that correspond to amplifications (eastward, northward and northwestward) of an anomalously strong WCR and to an eastward shifted and sometimes weakened CPT (Skinner et al. 1999; Macias Fauria & Johnson 2006). The WCR can sometimes be displaced eastward by an eastward-displaced WCT (e.g. Macias Fauria & Johnson 2006).

Such patterns are related to two main large-scale climatic patterns that have their origin in Northern Hemispheric (and even global) atmosphere and atmosphere–ocean interactions: the Arctic Oscillation (AO) and the Pacific Decadal Oscillation (PDO)—El Niño Southern Oscillation (ENSO; Macias Fauria & Johnson 2006). The AO or northern annular mode (NAM; Trenberth & Paolino 1981; Wallace & Gutzler 1981; Thompson & Wallace 1998, 2000) is defined as the dominant pattern of sea-level pressure (SLP) variations north of 20° N and is characterized in its positive (negative) phase by negative (positive) SLP anomalies in the Arctic and positive (negative) anomalies at mid-latitudes (figure 5). The AO pattern is found to extend from the surface to the stratosphere, hence reflecting the strength of the polar vortex (Thompson & Wallace 1998; Serreze et al. 2000). The AO is closely related to the North Atlantic Oscillation (NAO; van Loon & Rogers 1978).

The PDO (figure 6) is the leading mode of monthly sea surface temperature (SST) in the North Pacific Ocean, north of 20° N (Hare 1996; Mantua et al. 1997; Zhang et al. 1997). PDO events persisted for 20–30 years during the twentieth century. Warm (cool) PDO phases were characterized by a strengthened (weakened) Aleutian low, enhancing (reducing) the advection of warmer air onto the west coast of North America, which caused positive (negative) temperature anomalies over northern North America with the largest anomalies located over central Canada (Minobe 1997; Mantua & Hare 2002). Warm PDO phases tend to produce persistent high-pressure anomalies over northern North America, especially over western Canada and Alaska (Zhang et al. 1997). The PDO is related to SST, precipitation and convection variability in the Indian Ocean and tropical Pacific Ocean, and its spatial climatic patterns are similar to the ENSO pattern (Mantua et al. 1997; Zhang et al. 1997).

Figure 6.

The Pacific Decadal Oscillation (PDO). (a) Chart showing point correlations between Northern Pacific Ocean Sea Surface Temperatures (SST) and the PDO Index, based on spring–summer (March–August) anomalies. Contour interval is 0.2. Note the most prominent patterns of positive PDO—high temperatures in front of the northwestern coast of North America and in the Gulf of Alaska, and low temperatures in the central North Pacific Ocean. Data and software for the figure obtained from the NCEP/NCAR Reanalysis Project (Kalnay et al. 1996). (b) Chart showing point correlations between Northern Hemisphere 500 hPa height and the PDO index, based on spring–summer (March–August) anomalies. Contour interval is 0.2. Note the most prominent patterns of positive PDO—highs over western North America and lows in the Gulf of Alaska. Also, note its similarity to the PNA pattern. Data and software for the figure obtained from the NCEP/NCAR Reanalysis Project (Kalnay et al. 1996). (c) Spring–summer PDO index obtained from the Joint Institute for the Study of the Atmosphere and Ocean (JISAO; http://tao.atmos.washington.edu/). Note the regime shifts in 1924, 1947 and 1976/1977 (arrows).

(b) Large-scale climate and large lightning fires in the boreal forest of North America, 1959–1999

Macias Fauria & Johnson (2006) showed how the large area burned in Canada and Alaska has been linked during the last four decades to the action and interaction of the PDO/ENSO and AO systems. As noted above, the climatic fingerprints of ENSO and PDO are very similar and can only be separated by the frequency bands at which they operate. The PDO is related to fire climate at low frequencies (decadal variability), whereas ENSO and especially AO are related to fire climate at much shorter time scales (months to a few years; table 1). The ENSO typically operates at scales of 2–6 years (Minobe 2000) and the PDO operates at longer frequencies. The interactions between interannual ENSO events (El Niño and La Niña) and interdecadal PDO suggested that typical El Niño patterns over North America are strong and consistent only during the positive (warm) phase of the PDO, whereas strong La Niña patterns over the continent are associated with the negative (cool) PDO phase (Gershunov & Barnett 1998).

Table 1.

(a) Correlations between the 7-year high- and low-pass filtered fire–climate indices and the summer Arctic Oscillation, spring–summer PDO and winter El Niño Southern Oscillation indices for the period 1959–1999 (from table 2 in Macias Fauria & Johnson 2006). Fire–climate indices were obtained from the dominant upper air Northern Hemisphere height anomaly patterns during large fire events over several regions in the boreal forest of North America. Type 1 (2) fire–climate indices are positively (negatively) correlated with PDO at low frequencies and with ENSO at high frequencies. Type 3 (4) fire–climate indices are positively (negatively) correlated with AO at high frequencies. Significances were computed using a Monte–Carlo significance test, which generated 1000 pairs of series with the same autoregressive structure as the original series; ^*p<0.1; ^**p<0.05; ^***p<0.01. (b) Correlations between total area burned in Canada and spring–summer PDO for the period 1918–2005. Filtered series and significances were computed as in a.

(a)	period 1959–1999	PDO spring–summer	Niño three winter
high-pass series	type 1	0.27*	0.76***
	type 2	−0.25*	−0.74***
low-pass series	type 1	0.72***	0.19
	type 2	−0.73***	−0.19

	period 1959–1999	AO summer
high-pass series	type 3	0.84***
	type 4	−0.82***
low-pass series	type 3	0.24
	type 4	−0.35

(b)	period 1918–2005	PDO spring–summer
area burned in Canada	unfiltered series	0.21*
	high-pass series	−0.06
	low-pass series	0.62**

The large area burned in western Canada, and much of the area burned in Alaska (e.g. Hess et al. 2001; Macias Fauria & Johnson 2006), occurs under positive PDO phases, with strong advection of warm Pacific air into the continent (figure 7). Moist, warm Pacific air west of the Canadian Rocky Mountains prevents large-fire weather conditions (some large fire years were also linked to positive PDO in British Columbia (e.g. 1985), but such events were rare during the period 1959–1999). Thus, area burned patterns in British Columbia are in anti-phase with those patterns east of the Rocky Mountains. Many large fire situations in eastern Canada were also directly related to a PDO situation with a WCR having an extreme eastward extension (figure 7).

Figure 7.

The geographical extent of PDO/ENSO and AO influences on large lightning fires during 1959–1999 in the North American boreal forest. Figure is based on results presented in Macias Fauria & Johnson (2006).

On the other hand, the positive AO has been shown to be related to frequent summer ridging over the eastern part of the North American boreal forest and a large area burned (figure 7). Some fires in Alaska and the northern parts of the Northwest Territories in Canada have been linked to negative AO situations, in which a southward extension of a high-pressure anomaly centred in the Arctic Ocean has affected these areas.

It is noteworthy that during 1959–1999, AO-related large fires in the eastern boreal forest (mainly Québec) only occurred under a positive PDO phase, suggesting an AO–PDO/ENSO interaction. However, this conclusion is based on a limited study period comprising only one cool and one warm regime of the PDO.

Thus, the increase in area burned for most of Canada in the last few decades is related to (i) a persistent positive PDO phase since 1976–1977 (figure 6) and (ii) the occurrence of positive AO situations, especially in the late 1980s and 1990s (figure 5). These conditions favoured weather prone to large fire occurrence over much of the boreal forest in North America. Neither of these two factors favoured a large area burned in British Columbia and only one (the positive phase of the PDO) favoured a large area burned in Alaska.

Using the longer (starting in 1918) area burned annual totals for Canada (Van Wagner 1988, 1991; figure 8), the link between the PDO and variability in area burned showed large and statistically significant correlations, especially (but not only) at decadal frequencies (Macias Fauria & Johnson 2006; figure 8; table 1). This relation was significant despite the probable underestimation of total area burned in the early part of the record due to a lack of data from the northern territories in Canada. Other studies (Girardin et al. 2006a; Skinner et al. 2006) have also pointed to the PDO as a major large-scale climatic pattern affecting large fires in the North American boreal forest. Skinner et al. (2006) found the summer severity rating index (cf. Van Wagner 1987) in the Canadian boreal forest to be related to global wintertime SSTs, in particular, to (i) a long-term increase in Southern Hemisphere SST, which they related to global warming, (ii) the Atlantic Multidecadal Oscillation (AMO; Delworth & Mann 2000; Kerr 2000) and (iii) the ENSO/PDO.

Figure 8.

Forested area burned in Canada, 1918–2005 (dashed line); spring–summer PDO index, 1900–2005 (continuous line). Thick lines represent the 7-year running means in order to highlight low frequency variability (black, PDO; grey, area burned). Note the overall high similarity between both series at low frequencies.

In conclusion, the area burned in the boreal forest of North America was explained as a function of large-scale climatic patterns that have been operating during the twentieth century and that are linked to the occurrence and distribution over the continent of mid-tropospheric height anomalies.

3. Past fire activity

(a) Fire frequency studies

Apart from historic fire records, we can explore past fire frequency by a time-since-last-fire map of a study area (figure 9) and by reconstructing fire frequency from the time-since-fire (or survivorship) distribution (e.g. Johnson & Van Wagner 1985; Johnson & Gutsell 1994). The time-since-fire distributions (figure 10a) have generally been found to fit a negative exponential curve A(t)=e^−(t/b), in which A(t) is the survivorship distribution; t is the time since last fire; and b is the fire cycle, defined as the time required to burn an area equal in size to the study area. The average fire frequency is defined as 1/b. A negative exponential distribution implies that the hazard of burning for a given stand is time independent; that is, the age of the stand does not increase or decrease the probability of a fire occurring (if fuel accumulates with stand age it does not increase the probability of fire in the boreal forest).

Figure 9.

Small portion of the time-since-fire map of Prince Albert National Park (Saskatchewan, Canada), corresponding to fig. 3 of Johnson et al. (1998; source: J. M. H. Weir).

Figure 10.

(a(i)) Example of exponential curve found to fit best most time-since-fire distributions (survivorship). A(t)=e^−(t/b), in which A(t) is the survivorship distribution; t is the time since last fire; and b is the fire cycle, defined as the time required to burn an area equal in size to the study area. It starts with the landscape unburned and describes the proportion remaining at different ages. (a(ii)) Semi-log time-since-fire distribution showing a straight line. (b) Time-since-fire distribution for the Boundary Waters Canoe Area (source: Heinselman 1973) indicating two changes in fire frequency (i.e. mixed distributions).

The time-since-fire distributions are generally mixed distributions (figure 10b). This is seen by the changes in slope when the survivorship distributions are plotted on semi-log scale (for methods of separating these mixed distributions, see Reed 1997, 1998, 2000, 2001; Reed et al. 1998; Reed & Johnson 1999). All fire frequency studies in the boreal forest have shown mixed distributions that imply changes in the fire frequency. Most of these studies have remarkably synchronous changes in fire frequency over vast areas (Masters 1990; Bergeron 1991; Johnson & Larsen 1991; Johnson & Miyanishi 1991; Reed 2000; Weir et al. 2000; Bergeron et al. 2004). The general pattern in all of these reports is that the fire cycle used to be much shorter in the past three centuries and that the twentieth century has generally had longer fire cycles.

The synchronous change of fire frequencies in most of the above studies over such large spatial scales has suggested these frequency changes to be due to large-scale climate changes and not to processes acting at a stand level. The changes in fire frequency in the 1700s and 1800s have been attributed to changes in global atmospheric circulation during the Little Ice Age (e.g. Johnson & Larsen 1991; Bergeron & Archambault 1993). Fire cycles were generally shorter during the Little Ice Age, which could be the result of more frequent occurrences of high-pressure systems over northern North America (Bergeron & Archambault 1993) than during the twentieth century.

The suggested change towards longer fire cycles in the mid-nineteenth century in the North American boreal forest (e.g. Masters 1990; Larsen 1997; Reed et al. 1998; Flannigan et al. 1998; Johnson et al. 1999; Bergeron et al. 2001; Lesieur et al. 2002) could be linked to a shift from reduced to enhanced PDO/ENSO activity (Swetnam & Betancourt 1990; Biondi et al. 2001). From ca 1790 to the mid-nineteenth century, the frequency and strength of ENSO switching are thought to have weakened (Anderson et al. 1992; Cleaveland et al. 1992). The PDO is also reported to have had a weak decadal variability during the eighteenth and early nineteenth centuries (Biondi et al. 2001; Gedalof et al. 2002; MacDonald & Case 2005). Taylor & Beaty (2005) proposed that during that period decadal scale relationships between the PDO and ENSO broke down. This would imply a different climate in the boreal forest over that period, with different atmosphere or atmosphere–ocean circulation patterns operating and, therefore, with a different frequency of occurrence and probably location of fire-prone summer blocking highs.

Thus, fire frequency is expected to change in a constantly changing climate (Johnson et al. 1998). In general, and consistent with a predicted increase in total area burned in the boreal forest of North America, fire frequency has been predicted to increase in a global change scenario (e.g. Kasischke et al. 1995; Flannigan et al. 2001, 2005); thus, the mean age of the forest as a whole is predicted to decrease. However, some reports have suggested fire frequency decreases in parts of eastern Canada associated with an increase in precipitation (Flannigan et al. 1998, 2001). Although the frequency of mid-tropospheric blocking highs is the causal mechanism for large fires to occur, none of these predictions has been based on such parameters.

Given the changing climate and the consequent changes in fire regimes over the last several centuries, it is impossible to talk about a ‘natural fire regime’ in any part of the boreal and subalpine forests of Canada and Alaska. Also, owing to this changing fire frequency, we cannot talk about a system in equilibrium, climax or stasis. Thus, there is no single ‘natural’ fire cycle for any part of the boreal forest. Instead, the forested landscape mosaic is a reflection of a dynamic fire cycle and carries the memory of different past fire cycles (Johnson et al. 1998).

(b) Palaeoecological studies

Palaeoecological studies using lake sediments, macrofossils and ice cores provide information on fire activity and climate conditions during the past millennia and suggest changes in large-scale climatic patterns that relate to past fires.

In a study of diatoms from lake sediments from the northern prairies of North America (Laird et al. 2003), abrupt changes in aridity were found, indicating that shifts in drought regimes on a decadal to multicentennial scale have been a common feature of this region for at least the last two millennia. Regardless of the exact timing of these shifts or the direction of these changes, the persistent abrupt nature of these events represents a scale of variation that is not yet well understood, but probably is related to changes in the shape and location of the jet stream and associated storm tracks (Laird et al. 2003). Schindler & Donahue (2006) found that the period since European settlement appears to have been the wettest century of the past two millennia in the western Canadian Prairie Provinces. The frequent, long periods of drought that characterized the last two millennia were largely absent in the twentieth century.

Likewise, in several studies using tree rings and lake sediments, Sauchyn et al. (2002) found that the climate of the twentieth century in the Canadian Plains was anomalous in terms of drought, as proxy data recorded an abrupt amelioration of climatic conditions near the start of the instrumental record period. While drought was frequent in the twentieth century, it tended to be of shorter duration than during past centuries. An abrupt shift in climatic variability from centennial to decadal scale was identified around the mid-nineteenth century (Sauchyn et al. 2003).

Using a ca 3100-year oxygen isotopic (δ¹⁸O) composition record from northern Minnesota, Tian et al. (2006) reported that drought events of greater magnitude than the aridity extremes of the twentieth century were not uncommon during the Late Holocene and were strongly related to the PDO. They also noted that approximately 90% of the drought variability values during the last three millennia were greater than in the twentieth century. An ice core time series from Mt Logan (Yukon, Canada) shows a shift in climate after the mid-nineteenth century with increased snow accumulation, paralleled by an increasing trend in temperature in northwestern North America that is associated with secular trends in the PDO (Moore et al. 2002). All such findings are in agreement with the general decrease in fire frequency in the North American boreal forest since the mid-nineteenth century reported in §3a of this paper.

Lake sediments on the west slope of the Canadian Rocky Mountains show highest fire frequencies during the thermal optimum in the Mid-Holocene and lower ones in the later cooler part of the Holocene, together with changes in vegetation composition (Hallett & Hills 2000). Lake sediments in Alaska show variable fire cycles at centennial scales linked to changes in climate and in species composition during the Holocene (Anderson et al. 2006), with longer fire cycles in the Late Holocene under a wetter climate.

Payette & Gagnon (1985) and Payette et al. (1989) proposed that fire, combined with cooler climate and the consequent poor conditions for germination in northern Québec during the Late Holocene, caused a shift from closed-canopy forest to the patchy distribution of forest stands which characterizes the modern northeastern Canadian forest tundra. Such a mechanism was proposed to have worked in periods as recent as the large fire years of the 1950s (Sirois & Payette 1991). In a study of pollen and charcoal analysis in Québec, Carcaillet et al. (2001) found that thermal optima in the region were associated with moist climate and reduced fires, whereas colder conditions brought abundant northwestern dry air masses that favoured shorter fire cycles. Under such conditions, they proposed that an eventual global warming would probably be associated with increased moisture in the area and, thus, with a less severe fire regime.

The results of these palaeoecological studies portray a constantly changing climate and environment at all time scales from interannual to centennial. Moreover, these studies document an important shift in the mid-nineteenth century, associated with moister climatic conditions that were possibly related to changes in the ENSO/PDO system, which affected fire frequencies all across the boreal forest of North America. Thus, these studies suggest that increases in temperature do not necessarily produce more fire activity, as seen in major fire activity during periods of the Little Ice Age.

The relationship between climate, fire frequency and vegetation composition is complex. Climate change modifies the fire frequency that, together with climate change itself, enhances shifts in vegetation composition. Vegetation composition can, in turn, enhance changes in fire frequency through changes from open to closed canopy or vice versa. Thus, these three parameters are intimately linked.

4. Discussion and conclusions

Changes in area burned due to the current global warming need to be linked to changes in the occurrence of blocking highs during the fire season. The dynamics of large-scale climatic patterns (ENSO/PDO and/or AO) are important in controlling such frequencies and are thus strongly related to the area burned in the North American boreal forest (Macias Fauria & Johnson 2006; Skinner et al. 2006). Studies on the influence of anthropogenic climate change on the AO and PDO remain inconclusive: while there is some evidence that the increasing greenhouse gases will lead to a more positive AO (Shindell et al. 1999; Hoerling et al. 2001), observations and climate models cannot provide a definitive answer on this yet, and have failed to predict the shift of the AO to a neutral state since the late 1990s (e.g. Overland & Wang 2005). As for the PDO, too little is known so far about its mechanisms to state whether its dynamics are being affected by global warming (e.g. Mantua & Hare 2002).

These teleconnections operate at different time scales from interannual to centennial. Strong climatic shifts in the past centuries in northern North America (e.g. Laird et al. 2003) have been associated with changes in the strength and coupling of the PDO/ENSO climatic system in this area (e.g. Hessl et al. 2004; Tian et al. 2006).

Fire reports, time-since-fire map fire frequency studies and palaeoecological studies report past changes in fire frequency. Moreover, many studies have noted changes of the dominant pattern of fire variability and fire–climate relationships from centennial/decadal to interannual time scales, both in the boreal forest of North America (Girardin et al. 2006a,b) and in other areas of North America (Taylor & Beaty 2005).

Here, we propose a simple idea of how such changes in the temporal scale of the fire–climate relationships might take place (figure 11). Within the centennial/decadal climate variability, climate remains under a very dry regime for long periods (e.g. Sauchyn et al. 2002, 2003; Laird et al. 2003; Schindler & Donahue 2006). Under such conditions, it is logical to think that persistent blocking highs are the usual climate pattern over the area. Thus, fire-prone conditions happen very frequently and year-to-year variability is not very important: decadal to centennial variability dominates fire–climate relationships. On the other hand, under periods with a more variable climate or with a climate ‘locked’ into a moist period, large-area-burned conditions only occur in particular years or clusters of years of extreme fire weather, and thus interannual to decadal variability is the dominant feature in the fire–climate relationships.

Figure 11.

Plot showing an idealized climate variation over time. The y-axis represents the frequency of mid-tropospheric blocking highs during the fire season (i.e. fire hazard); large fire years are expected in the squared area, whereas low fire activity is expected in the waved area. During period (a), the climate remains locked in a dry phase and large fire weather occurs very frequently; decadal to centennial scales in the fire–climate relationships are expected. During period (b), the climate remains locked in a wet phase and large fire years occur very infrequently; however, in period (c), the climate enters a highly variable state and both wet and dry years occur. In periods (b) and (c), interannual climate variation is expected to be the most prominent source of large fire variability.

It is important to remember that for large fire occurrence in the closed boreal forest, 10–15 days of persistent drought are needed to efficiently dry the fuel, followed by the occurrence of lightning (e.g. Nash & Johnson 1996). Thus, when referring to centennial, decadal and even interannual links between fire and climate, we refer to the frequency of occurrence of persistent blocking highs (i.e. changes in the probability of a given fire-prone synoptic-scale atmospheric arrangement to occur) and not to a generally dry or wet fuel over periods of years to centuries. Lightning is assumed to be abundant enough in most areas to not be limiting.

Most studies on fire frequency in the boreal forest of North America identify a change towards longer frequencies during the mid-nineteenth century, which has been attributed to climatic change (Masters 1990; Bergeron 1991; Johnson & Larsen 1991; Johnson & Miyanishi 1991; Reed 2000; Weir et al. 2000; Bergeron et al. 2004). Girardin et al. (2004) suggested that, associated with a northward displacement of the jet stream, subtropical Atlantic air masses have penetrated more frequently in Québec during this period, causing a moister climate and thus less fires. Laird et al. (2003) also proposed jet stream displacement to be associated with major climate shifts in more western areas of the continent. Changes in the temporal scale of fire–climate relationships have also been reported during this time (e.g. Girardin et al. 2006a,b). The period starting in the second half of the nineteenth century up to the present has been one of generally wet conditions and high interannual climatic variability, especially in eastern Canada but also in western Canada and Alaska. The centuries before this appear to have had more constantly fire-prone conditions (prolonged droughts) with a decadal to centennial coupling between fire and climate. The boreal forest during this period often had a younger landscape age mosaic and a different pattern of species composition than in recent centuries (e.g. Ritchie & Yarranton 1978; Lynch et al. 2004).

The last century and a half has thus seen fewer fires than before and a significant and generalized increase in temperatures marking the end of the Little Ice Age. That is, there is an unclear relationship between warmer temperatures and increased values of area burned. Both the Little Ice Age (ca AD 1500–1850) and the Medieval Warm Period (ca 1100–1400) showed great variability of arid and wet periods (Luckman 2000; Tian et al. 2006), with variable associated fire regimes. Johnson et al. (1990), Johnson & Larsen (1991), Reed (2000) and Weir et al. (2000) found changes in fire frequency within the Little Ice Age. Thus, a warmer world need not necessarily be considered a world with more fire activity in the boreal forest of North America.

It is important to clearly distinguish between a warmer world and the frequency of droughts; these two climate characteristics have occurred simultaneously at some periods but are not intrinsically related. Thus, there is no reason to predict higher values of area burned in a warmer world based solely on an increase in temperature (Flannigan & Harrington 1988; Flannigan et al. 1998; Bergeron et al. 2001).

Inter-regime climate fluctuations represent a substantial portion of the climate variability in North America, largely affected by the PDO/ENSO regime fluctuations (e.g. Tian et al. 2006). Some studies have hypothesized the possibility of such centennial, large spatial- and temporal-scale droughts reoccurring as a result of climate change (e.g. Schindler & Donahue 2006) or even as a part of normal climate variability (e.g. Laird et al. 2003). So far, there is no reason to think that such conditions should not come back. As stated above, such climatic conditions would represent a completely different set of fire–climate relationships and change the average age of the forest and its species composition.

Finally, most climatological studies on teleconnection patterns in the Northern Hemisphere are based on the cold season (e.g. Wallace & Gutzler 1981; Zhang et al. 1997; Thompson & Wallace 2001; Mantua & Hare 2002). Some recent studies have focused on seasons other than winter for the study of the AO and NAO teleconnections using novel methods (e.g. Portis et al. 2001; Ogi et al. 2004). Given that many biological processes take place during summer, there is a clear need for more research on the relationships and mechanisms between teleconnections and climate during the warm season in the Northern Hemisphere.