- Email: [email protected]
Fuel fragmentation and fire size distributions in managed and unmanaged boreal forests in the province of Saskatchewan, Canada
Forest Ecology and Management 376 (2016) 148–157
Contents lists available at ScienceDirect
Forest Ecology and Management journal homepage: www.elsevier.com/locate/foreco
Fuel fragmentation and fire size distributions in managed and unmanaged boreal forests in the province of Saskatchewan, Canada Veiko Lehsten a,⇑, William de Groot b, Florian Sallaba a a b
Department of Physical Geography and Ecosystem Science, Lund University, Sweden Canadian Forest Service, Natural Resources Canada, Sault Ste. Marie, Ontario, Canada
a r t i c l e
i n f o
Article history: Received 5 February 2016 Received in revised form 2 June 2016 Accepted 6 June 2016 Available online 14 June 2016 Keywords: Forest fires Burned area Saskatchewan Boreal Shield Saskatchewan Boreal Plain Characteristic fire size Climate fire relationship Wildfires Fire size distribution Forest management Climate-forest fire linkages
a b s t r a c t Forest fires are an important disturbance factor of boreal forests, annually burning about 0.5% of the forested area in Canada. Wildfire regimes are influenced by climate and a number of studies project an increase in wildfire activity with climate change. Another factor influencing wildfires is human intervention (fire suppression), and one factor that has rarely been assessed is fuel fragmentation. Studies evaluating the effect of forest fire suppression concluded that in areas with strong suppression effort the burned area as well as the fire size decreased. Here we evaluate wildfire distributions over the last three decades for two areas that differ mainly in their level of forest management and fire suppression: the Boreal Shield (unmanaged) and the Boreal Plain regions (intensively managed) in the Canadian Province of Saskatchewan. We calculate a fuel fragmentation index and relate fire sizes and burned areas to fire weather. We use the concept of the characteristic fire size (CFS); hence we analyze how much burned area is contributed to the total burned area per fire size class. Both areas show a uni-modal distribution of the CFS, indicating that the majority of burned area was contributed by medium sized fires (Boreal Shield 6.39 104 ha, the Boreal Plain 8.79 104 ha). Burned area as well as fuel fragmentation is lower in the managed forest compared to the unmanaged area. The fuel fragmentation index constantly increased since the 1980s in both regions. Despite the large efforts of fire suppression in the Boreal Plains, the CFS is slightly larger in this managed region. Neither the burned area nor the fire size could be linked statistically to the weather conditions, at the time of the fire (using the Canadian Fire Weather Index). We argue that the high fragmentation over the last decades have decreased the burned area. The slightly higher characteristic fire size in the managed area might be explained by the considerably lower fragmentation, counteracting fire suppression efforts. Fuel fragmentation is likely to decrease over the next decades due re-growth. Though a strong link between fire weather and burned area at the fine scale of this study could not be detected we expect that a decrease in fragmentation in combination with an increase in fire prone weather conditions (as expected for the future) might increase the risk of large fires in both areas. We suggest that future fire risk analysis should include an assessment of the effect of fuel fragmentation. Ó 2016 Elsevier B.V. All rights reserved.
1. Introduction Wildland fire has been a prevalent natural disturbance across the North American boreal forest region for millennia (Girardin et al., 2013). The North American boreal fire regime is characterized by infrequent high intensity crown fires (de Groot et al., 2013) and the annual area burned in Canada has been episodic ⇑ Corresponding author at: Department of Physical Geography and Ecosystem Science, Lund University, Sölvegatan 12, SE-23623 Lund, Sweden. E-mail address: [email protected] (V. Lehsten). http://dx.doi.org/10.1016/j.foreco.2016.06.014 0378-1127/Ó 2016 Elsevier B.V. All rights reserved.
and variable, ranging from 300,000 to 7.5 M ha per anno. Standreplacing crown fires currently burn an average of 2–3 M ha (0.5% of total forest area) each year in Canada (Stocks et al., 2002) with typical fire cycles of 75–150 years. Large fires account for most of the annual area burned in the Canadian boreal region, creating a landscape pattern of large areas of young forest (established since the most recent fires) with small patches of older forest embedded, which represent unburned islands and the remnants of older fires (Johnson et al., 1998; Weir et al., 2000). The median size of unburned islands increases sharply with fire size, and the frequency of unburned islands (per 100 ha burned)
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
is greatest for smaller classes of large fires (201–2000 ha) (Eberhart and Woodard, 1987). Fire shape also becomes more irregular with increasing fire size such that the largest fires create the greatest amount of edge relative to area burned (Eberhart and Woodard, 1987). Although some studies have been done on the spatial pattern of individual fires in the boreal region, there has been very limited research conducted on the pattern of fire distribution on the landscape and effect of fuel fragmentation on fire spread. To reach a rational use of forest resources preserving long term ecosystem functionality it has been suggested that forest practices should minimize the differences between managed and unmanaged forests (e.g. Franklin, 1993; Hunter, 1993). Ecosystem-based forest management should favor interventions that lead to landscape compositions and structures close to their counterparts in natural forests (DeLong and Tanner, 1996) including disturbance pattern. Therefore a large interest in past, present and future burned areas has resulted in a number of studies focus on the relationship between climate and burned area, or projecting climate into the future and translating climatic changes into changes in burned areas (for a comprehensive review see Girardin et al., 2013). These studies have also shown that there is considerable uncertainty in the link between climate and burned area (ibid). Though recent increases in North American fire activity have been attributed, at least in part, to a changing climate and increasing temperature (Gillett and Weaver, 2004; Westerling et al., 2006). Amiro et al. (2004) have shown that if the climate is evaluated on a per fire basis, no trends could be detected in fire weather components within the last four decades. Additional difficulties arise from the influence of large fires which result in an over-proportional amount caused by very few fires. Together with changes in reporting and suppression rates, as well as short term climatic variations, linkages between fire and climate might be hard to detect especially at a landscape scales (Girardin et al., 2013). One aspect which has received very little attention in the past is the effect of fragmentation on the fire regime. When analyzing burned area and fire size distributions, Lehsten et al. (2014) found that fire size as well as burned area in Canada increased continuously from the 1960s until the 1980s and remained at a very high level in the 1990s before both characteristics dropped sharply in the 2000s. Given the decrease in the 2000s at a national scale, this raises questions about the nature of the link between temperatures and burned area, which was assumed by Gillett and Weaver (2004) to be a linear relationship, using data from 1960 until 2000. This recent decrease in burned area was hypothesized to be the result of fuel fragmentation resulting from the high fire activities in the 1980s and 1990s (Lehsten et al., 2014). If this is the case, this decrease in burned area would only last a short time period and a re-growth of forests (together with a warmer climate) would result in an increased fire activity in the following decades. Analyzing temporal effects of fuel fragmentation is challenging for a number of reasons. Similar to other fire parameters, fuel fragmentation can be expected to be dependent on the forest type, the climatic conditions and the level of forest management, including forest protection. Additionally, data covering large spatial and temporal scales are needed to allow collection of a sufficiently large and representative distribution of small and large fires in order to detect if they are causing fragmentation. Directly linked to fuel fragmentation is the fire size distribution, as a shift towards large fires (with a constant burned total area) would result in a decrease of fuel fragmentation. Hence both fire-size and fuel fragmentation should be analyzed simultaneously. Cumming (2005) has evaluated the effect of fire management and concluded that fire suppression efforts have decreased both the burned area as well as the ratio of large to small fires. However, if the total burned area decreases, this will also decrease fuel fragmentation, which can
149
potentially favor the development of large and thereby costly fires compared to a landscape with less fire suppression effort. In this study we investigate the effect of forest management (fire suppression) on fuel fragmentation and fire size distribution in two areas which are relatively similar in their ecological attributes apart from their forest management history. We will also analyze the temporal development of fuel fragmentation and its relation to climatic conditions.
2. Methods 2.1. Study area Our study areas are the Boreal Shield and Boreal Plain eco-zones (Ecological Stratification Working Group, 1995) within the Province of Saskatchewan, Canada (we will use the term Shield and Plain in the following to account for the Boreal Shield and Boreal Plain of the province Saskatchewan rather than the whole ecozone which is larger). We used the classification scheme of the National Ecological Framework for Canada (http://sis.agr.gc. ca/cansis/nsdb/ecostrat/gis_data.html) to assign the areas (Fig. 1). Eco-zones are ecological areas based on terrestrial characteristics accounting for surface forms, soils, faunal realms, vegetation and macro climates (Marshall et al., 1999). The Shield is dominated by coniferous (83%), scattered mixedwood forest (8%), and by lakes and rivers (8%), while the Plain consists of lower amounts of coniferous forest (43%) and more mixedwood forest (16%) and broadleaved forest (14%), but additionally includes rangelands (20%) and a lower amount of lakes and rivers (5%). All other land cover classes are below 1%. The total area of the Shield is 1.83 107 ha. Though the Plain is only marginally smaller (1.7 107 ha), its burnable forest area is ca. 22% smaller (sum of all forest types: Shield 1.68 107 ha, Plain 1.31 107 ha). Drying of forest floor organic soils in the boreal shield is generally quicker than in the boreal plains. This is because forest floor fuel loads are higher in the boreal plains than in the boreal shield for black spruce (Picea mariana), jack pine (Pinus banksiana) and aspen (Populus tremuloides) (Letang and de Groot, 2012), which represent the majority of the forested area in those regions. Therefore, the lower forest floor fuel loads in the boreal shield promote faster drying due to shallower, or less densely compacted, organic soils. The Shield is - until today - experiencing essentially a natural fire regime, and is classified as a fire management observation zone where values at risk are assessed with the intent to allow for fire in ecological processes (Government of Saskatchewan., 2014). On the other hand, the Plain has been used as high value commercial forest over the last decennia and is classified as either a modified or full response fire management zone with the object to control and suppress fires. We therefore decided to use the two areas as representatives of boreal forests with relatively similar ecological conditions but different fire management history.
2.2. Fire data We used data from the Canadian National Fire Database (CNFDB)1 (Stocks et al., 2002). We used the polygon data provided to create yearly fire maps of the Shield and the Plain. The CNFDB is a collection of forest fire polygons from Canadian provincial, territorial and national fire management agencies. We filtered out all fires that were located on non-forested areas, fires below 100 ha and used only fire data ranging from 1960 to 2010 since the fire data for earlier times and smaller fires are not considered reliable (Stocks 1
http://cwfis.cfs.nrcan.gc.ca/ha/nfdb.
150
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Fig. 1. Land cover map of the Boreal Shield and the Boreal Plain within Saskatchewan (based on Palko et al., 1995).
et al., 2002). The annual burned areas for the two areas are displayed in Fig. 2. 2.2.1. Fire sizes While the most common metric to report fire activity is total burned area (e.g. Stocks et al., 2002), the size of fires and the num-
ber and size of unburned patches (defining the amount of fire edge and distance to edge within the burned area) are also important fire characteristics affecting seeding distances and the ability to regenerate (Greene et al., 1999). The fire size distribution is also of economic and modelling interest for fire management planning, since fire fighting activity and evacuation activity are based on the
3
18
x 10
Annual burned area [ha]
16
Saskatchewan Boreal Shield
14
Saskatchewan Boreal Plain
12 10 8 6 4 2 0 1960
1970
1980
1990
2000
Fig. 2. Annual burned areas for the Saskatchewan Boreal Shield and Boreal Plain area.
2010
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
assessment of expected number, intensity, and sizes of fires, affecting long term fire management strategy and decision making (Ward and Mawdsley, 2000). A number of different metrics of the fire size distribution have been used in the past. These include the proportion of large fires (Cumming, 2005) or the slope of the linear fit of the fire size to fire number relationship (in log-log) space (Malamud et al., 2005). The interpretation of the proportion of large fires at an ecosystem level is problematic since it requires a threshold value to be set. It is additionally problematic since small fires have a higher omission error and in terms of total burned area they are relatively unimportant. We will elaborate on the use of the proportion of large fires compared to the statistic which we apply in the discussion section. With respect to the second mentioned fire metric, Lehsten et al. (2014) showed that for the boreal forest biome the assumption of a power law (a log-linear relationship required to calculate the slope parameter) is not valid, and (Hantson et al., 2016) could show that only in 2 out of 8 cases the power law assumption was not rejected. Since we were interested in the amount of total burned area that each fire size class contributed . Therefore, we used the concept of the characteristic fire size introduced in Lehsten et al. (2014) to assess whether forest (and forest fire) management has changed the fire size - fire area relationship. For the boreal region it was shown that wildfire sizes follow an unimodal distribution if the fires are binned according to their size (in equal sized bins at a logarithmic scale) and the number of fires in each fire size bin is multiplied with the average size of the fires in the corresponding bin. This results in a plot of the contribution of each fire size class to the total burned area. If this procedure results in a unimodal distribution, a normal density distribution is fitted to the data and the characteristic fire size is the expected value of this distribution (for details see Lehsten et al., 2014). We compare the characteristic fire size of the two areas to assess whether fire management has led to a decrease in fire size as it has been hypothesized in the literature. A detailed example of how the characteristic fire size is calculated is given in the supplementary material S.1. 2.2.2. Fuel data We considered all forest types as burnable fuel. Although coniferous forests have flammable needle fuels that typically promote crown fires during the summer months, broadleaved forests can burn in summer by understory fire during dry years, and they are also well-known to support surface fires in the spring and autumn when cured herbaceous vegetation dries out underneath the leafless trees. To estimate fuel fragmentation by wildfires we assumed that burned areas are not able to support wildfire for 20 years. This re-fueling time reflects the average minimum natural fire cycle (10–35 years) in the southern boreal region of Western Canada found by Weir et al. (2000) and corresponds with general observation of the time required for burned areas in the boreal to revegetate enough to become susceptible to fire again. Though we consider all burned areas to be equal with respect to fuel reduction and hence regrowth of fuels, Kafka et al. (2001) showed that even a single fire can have very different impacts in different parts of its burned area depending on surface material, stand age and stand composition. We used the Vegetation and Land Cover map from Palko et al. (1995) to create a binary land cover mask excluding all areas unsuitable for fires (all non-forested areas). By overlaying this land cover map with the CNFDB fire polygons in the study area we generated a fuel map. All forested areas which had not burned in the preceding 20 years were considered burnable. We rasterized the yearly fuel maps together with the land cover mask using the Albers Equal-Area Conic Projection to guarantee a constant spatial resolution of 1 km by 1 km for each raster cell (one cell equals 100 ha).
151
2.2.3. Fire fragmentation To assess fuel fragmentation, we needed a statistic which would mimic the growth of a fire. A review of the existing statistical measures did not result in a suitable implementation being available. For example the ‘largest patch size’ provided by FRAGSTATS (McGarial and Marks, 1995) and used in Loureiro et al. (2002) has the disadvantage that even a few unburnable cells would stop the growth of a fire, while other statistics which could be used to estimate fire size increased the burned area even if only a few cells in the flaming front would be available. Therefore we introduced a simple process simulating the spread of a fire. Since the majority of fires in the Canadian Boreal forests are caused by lightning strikes (Stocks et al., 2002) we tested whether the distribution of flashes have an influence on the estimated fragmentation using a coarse scale flash rate map (based on the LISOTD dataset http://thunder. msfc.nasa.gov/data/) as well as a fine scale flash map where we filtered out all cloud to ground strikes of the investigation area (a spatially summarized dataset from the Canadian Lightning Detection Network (CLDN); Burrows and Kochtubajda, 2010; Kochtubajda and Burrows, 2010). These two datasets were used to initialize the process simulating fire spread and compared to results with lightning ignition following an equal distribution over the whole area. These tests have shown that there are only very marginal differences between an initialization of the process following an observed lightning distribution and an initialization following an equal distribution of lightning strikes (data not shown). As our study purpose was to analyze the general landscape feature of fuel fragmentation and not a particular fire-lightning relationship, we therefore decided to continue the remaining analysis with an equal distribution of lightning strikes initializing the process. We ran the process separately on the annual fuel maps as well as on the land cover map to assess the proportion of fragmentation caused by wildfires and the proportion caused by landscape fragmentation (e.g. non-forested areas). Since the effect of fires on the fuel map lasted for 20 years, we started the process in the year 1979 to ensure that we have a sufficiently complete representation of fuel fragmentation in our dataset (the fire data starts in the year 1960). Initial tests showed that a stable value for the fragmentation can be obtained by seeding 5 104 flashes, hence each year we seeded this number of flashes across the fuel map based on a uniformly random distribution. The prevailing wind direction in the study area during the fire season (April to September) was from south west, so we simulated a fire spread in north east (NE) direction for all fires. For comparison purposes, additional simulations using other wind directions (i.e., west wind, north east wind) resulted in qualitatively similar results (data not shown). We simulated the potential fire development over ten steps. First we tested whether the lighting hit a non-burnable cell which would stop the process. Otherwise, the process tested whether the fire could spread into the surrounding cells in NE direction. This spreading was simulated in steps (with the first step being the spread from the ignition cell to the neighboring cells in a north east direction) as long as two conditions were fulfilled: At least 2⁄stepnumber-2 cells neighboring the current burned area are burnable. The burnable cells must be adjacent to each other in order to ensure a closed fire front. A visualization of the process is given in the supplementary material Fig. S.4. In case a rule is not fulfilled, the process stops and the total number of cells burned by the process are counted. After the simulation of each potential lightning strike we calculate the fragmentation index as 1 minus the proportion of burned cells (relative to the maximum total area of 121 km2).
152
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Hence a higher number of burned cells correspond to a lower fuel fragmentation. The maximum simulation size for the process was chosen since we assume that fires above this size are relatively unsusceptible to fuel fragmentation at an one km scale, and this size is still well below the characteristic fire size for Canadian forests of more than 104 ha found by Lehsten et al. (2014). We estimate a mean annual fragmentation and analyzed both the temporal change of the fragmentation index as well as its distribution for the two study areas. The process always uses the original fuel map, hence a calculated potential fire does not influence the results of the next seeded fire. If a subsequent fire would be (by chance) seeded in the same cell it would result in the same potential fire size. 2.3. Fire Weather Index Since fire requires certain weather conditions, we included this in our analysis. Here we use the Fire Weather Index (FWI) component of the Canadian Forest Fire Weather Index (FWI) System (Van Wagner, 1987) over the timespan from 1980 to 2009. The FWI is a general indicator of head fire intensity and flame length (Van Wagner, 1987). Through the structure of the FWI System, information from all other components is input to the FWI component, and as such, it is the most commonly-used component in the FWI System to indicate general landscape-level fire danger (Wotton, 2009). Head fire intensity and difficulty in controlling wildfire increase sharply with increasing FWI value (Van Wagner, 1987; Williams, 1959). We acquired data from the global fire weather database developed by Field et al. (2015), which has a spatial resolution of 0.5 by 0.667 degree longitude latitude. We extracted the daily FWI values from May to September (since these are the months with the highest fire activity, see: Lehsten et al., 2014) and plotted their distribution for each decade. 3. Results
the Saskatchewan Boreal Shield is 6.39 104 ha, while the characteristic fire size for the Boreal Plain is 8.79 104 ha. Hence, the Plain has a larger characteristic fire size despite the history of active fire management. The correlation coefficients (R-squares) between the fitted normal distribution and the data displayed in Fig. 3 is 0.96 for the Shield and 0.96 for the Plain. The decadal changes of the burned area are listed in Table 1. While Lehsten et al. (2014) found that even decadal distributions of the CFS followed a normal distribution rather closely, the decadal distribution in the two investigation areas of this study are more irregular caused by the low number of fires (compared to the whole of Canada in Lehsten et al., 2014). The plots of the decadal fire sizes are displayed in the supplementary material Fig. S.3. The largest burned area per decade in the Shield was reached in the 1980s, whereas the Plain reached its largest decadal burned area in the 1990s. The lowest burned area was recorded in the 1990s in the Shield and in the 2000s in the Plain. 3.2. Fuel fragmentation The fragmentation index for the Shield increased steadily from the 1980s to the 1990s, decreased in the early 2000s, and strongly increased afterwards to the highest value in 2010 (Fig. 4). In the Plain, the fragmentation index was rather stable in the 1980s and the first part of the 1990s, with a strong increase in 1996 followed by a slight increase afterwards. The increasing trend over the last three decades is highly significant (p < 0.001) for both areas. Both regions have a different history of forest management including not only fire suppression but also forest harvesting. Our analysis did not include the influence of forest harvesting
Table 1 Decadal changes in total burned area for the Saskatchewan Boreal Shield and the Saskatchewan Boreal Plain. Please also note that the Boreal Shield is ca 28% larger than the Boreal Plain. Time period
3.1. Fire sizes and burned area The differences in the total burned areas between the two areas are clearly visible in the evaluation of the characteristic fire size (Fig. 3). The sum of the bars in Fig. 3 represents the average annual burned area. The characteristic fire size between 1960 and 2009 for
Total burned area [ha]
1980s 1990s 2000s 1980–2009
Shield
Plain
4.0 106 2.7 106 3.0 106 9.8 106
1.2 106 1.6 106 0.74 106 3.6 106
4
Contribution to annual burned area
8
x 10
7
Saskatchewan Boreal Shield Saskatchewan Boreal Plain
ha a 5 4 3 2 1 0 102
103
104
[ha]
106
Fire size class Fig. 3. Characteristic fire sizes for the Boreal Plain and the Boreal Shield of the province of Saskatchewan for the time period from 1980 to 2009. Please note that the Boreal Shield is ca 28% larger than the Boreal Plain.
153
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Saskatchewan Boreal Shield
Fragmentation
0.7 0.6 0.5 0.4 0.3 0.2 1980
1990
2000
2010
Years Saskatchewan Boreal Plain
Fragmentation
0.7 0.6 0.5 0.4 0.3 0.2 1980
1990
2000
2010
Years Fig. 4. Fuel fragmentation for the Saskatchewan Boreal Shield and the Saskatchewan Boreal Plain. Solid line: annual fragmentation, dashed line: trend line, dash-dot line: background fragmentation caused by non-forested land cover (e.g. lakes and fields).
because typical forest cutblock size in Saskatchewan is 20–75 ha (Work et al., 2003), which is smaller than a single cell in our analysis. Forest harvesting is therefore assumed to have only a limited effect on fuel fragmentation and was not included in this study. Both areas also differ in their distribution of fire fragmentation (Fig. 5). The unmanaged Boreal Shield has very few nonfragmented sub-areas, and a large proportion of highly fragmented areas following an approximately linear relationship. The managed Boreal Plain shows a local maximum of areas with medium sized fragmentation, but the global maximum of the distribution is at the same fragmentation level as for the unmanaged area.
3.3. Weather We assess the weather over the last three decades by displaying the distribution of the FWI. Provincial fire management agencies calibrate the FWI scale using local fire and weather data, so scale calibration varies between provinces but in general a value of 30 + is considered extreme in Canada (Van Wagner, 1987). Due to the wide range of values in the dataset we used, the FWI scale is calibrated in the following way for analysis purposes in this study: 0–11.9: low to moderate; 12.0–24.9 high; 25–49.9 very high; 50– 74.9 severe; 75–99.9 extreme; above 100 catastrophic.
Fig. 5. Distribution of fragmentation for the same grid cells (0.5 by 0.66 longitude/latitude).
154
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Number of days and gridcells
Boreal Shield 100000
1980s 1990s 2000s
10000 1000 100 10 0
low to moderate
high
very high
severe
extreme
catastrophic
Fire weather index
Number of days and gridcells
Boreal Plain 100000
1980s 1990s 2000s
10000 1000 100 10 0
low to moderate
high
very high
severe
extreme
catastrophic
Fire weather index Fig. 6. Categories of Fire Weather Index values in the decades 1980s, 1990s and 2000s. Note that the number of days of within a FWI category is plotted with a log scale.
In Fig. 6 we list how often a grid cell of 0.5 degree longitude and 0.66 degree latitude (resolution of the global fire weather database) was within a certain FWI range. The area contained a total of 106 grid cells in the FWI dataset within the Shield from Field et al. (2015) while 96 grid cells covered the Plain. The FWI values are relatively similar over time (on a log scale) for low to very high fire risk (FWI from 0 to 49.0) For FWI values above 49.9, representing severe to catastrophic conditions, the 1980s had the most extreme fire potential, followed by the 2000s and the 1990s in both investigated areas. The Plain (which is more southern) had a markedly higher number of days per grid cell in all categories above high, and it had the only recorded values in the extreme and catastrophic categories.
3.4. Fire weather versus burned area and fragmentation The link between the FWI and burned area and fragmentation is very weak at the fine scale of this investigation. Fig. 7 relates the percentage of annual area burned, the fragmentation index and the 95th percentile of the FWI (over each year) to each other for the two investigation areas. Each grid cell within this analysis has a size of 0.5 by 0.66 degree latitude/longitude as this is the original resolution of the weather dataset (Field et al., 2015).
4. Discussion Our study of fire size distribution and the fragmentation in the two areas in the province of Saskatchewan has demonstrated that forest fire suppression has not decreased the characteristic fire size and that while fire suppression has led to a lower burned area, it also resulted in a lower fuel fragmentation compared to the area without fire suppression. The link between fire weather and burned area was very weak, which was probably a result of the small area that we investigated.
4.1. Relevance of forest fires The effect of fire management is of high interest to optimize commercial forest use for the coming decades given the amount of forest resources at stake and the cost of fire suppression. A sustainable forest use can only be attained if all aspects of forest ecology are taken into account (e.g. Franklin, 1993; Hunter, 1993) and the forest maintains its natural fire protection in the form of fuel fragmentation, which acts as a dampening factor that can potentially decrease the chances of large fire development after times of high fire activity. The natural forest fuel fragmentation level is hard to assess since we currently have only reliable spatially explicit data covering the last few decades at a national level (Stocks et al., 2002, CNFDB2). A number of proxies have been used to assess forest fire activity in Canada including stand age maps, charcoal records or tree ring densities which were used in correlative approaches to estimate burned area proportions. The review by Girardin et al. (2013) concluded that current fire activity is larger in the majority of the area while in the southwestern and –eastern locations a decrease of fire activity has occurred compared to the pre-industrial era using simulated climate data and fire history proxies. However these results are based on very few data points and therefore do not allow a translation into a Canadian-wide fragmentation estimate for pre-industrial conditions. Additionally, these are only estimates of fire activities. How changes in fire activity influenced fire size distribution, which is crucial for estimating fuel fragmentation, has not been investigated so far.
4.2. Fire size distributions in managed and unmanaged forests Cumming (2005) found that fire suppression has not only decreased the total burned area but also the proportion of large fires which are causing the majority of damage and costs. Our 2
Canadian National Fire Database, http://cwfis.cfs.nrcan.gc.ca/ha/nfdb.
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
155
Fig. 7. The relationship between FWI, fragmentation and proportion burned area (in %) and their distributions. Note that the burned area is plotted on a logarithmic scale. All burned area proportions smaller than 0.01% have been set to 0.01% to allow the logarithmic display. Large numbers indicate the correlation coefficient.
results however show that the characteristic fire size slightly increased in the managed forest. Even though our study areas are relatively similar in their ecological and climatological characteristics, a certain part of the larger characteristic fire size could still be caused by the slightly more fire prone weather as indicated by the higher number of days with FWI ratings above the severe category. Apart from this, there is a more fundamental difference in the ratio between large and small fires which was used as the test statistic for Cumming (2005) and the characteristic fire size. Theoretical considerations show that if one assumes the suppressed fires to develop into a representative sample of the total fire size distribution, the characteristic fire size should be unaffected if only fires larger than the size which is typically attacked are analyzed. The characteristic fire size is estimated based on the contribution of the fires of different sizes, and since small sized fires have a very low contribution to the total burned area, their influence on the characteristic fire size is limited. Therefore, even if fire suppression is effective in decreasing the ratio of large fires versus small fires, the CFS of an area will be largely unaffected. This property of the CFS to be relatively independent of fire suppression efforts allows to better compare CFSs between different ecoregions with varying degrees of fire management likewise as it was done in Lehsten et al. (2014). For fire suppression to have a substantial effect on the characteristic fire size, a considerably larger reduction in total burned area would be required than was found in our study. Our
results show that forest management in conjunction to fire suppression has probably not affected the fire size distribution. We use the term probably because our study implicitly assumes that without management both areas would result in a similar CFS, given a relatively similar climate and ecotype. Since large fires cause the majority of costs, a fire suppression scheme will tend to shift the distribution to smaller fires. The slightly larger CFS of the Plain compared to the Shield region indicates that the majority of burned area is contributed by larger fires compared to the Shield region. Hence fire suppression has not succeeded in shifting the fire distribution towards smaller fires. This is an indication that the decreased fragmentation in the area with fire suppression has resulted in somewhat larger fires on average, though the reduction in numbers still results in a considerable decrease in total burned area. The slight increase in CFS in the managed area might well be the result of the lower fragmentation of the forested area, combined with the slightly more fire prone weather conditions. 4.3. Differences in fuel fragmentation We used a very simplistic and rather conservative fire spread algorithm, requiring a high level of non-fragmented fuel beds in the landscape for a fire to increase in size for our assessment. We focused this analysis on the establishment phase of a forest fire. The assessment of the fire spread is stopped after the fire has
156
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
reached a size of ca 12,000 ha. This is well below the characteristic fire size in this area, which according to Lehsten et al. (2014) is around 47,000 ha. We consider all fires reaching this size to not be limited in spread by forest fragmentation. Our comparison between the two study areas showed that the fragmentation in the Shield is considerably higher than in the Plain as a result of fire suppression. This finding ignores the effect of harvesting on fuel fragmentation. Given that, despite the fact that the landscape fragmentation of the Plain is 50% higher than the landscape fragmentation of the Shield (caused by land uses different than forests), the total fragmentation of the Shield is still 40–50% higher than the total fragmentation of the Plain, it is very unlikely that the fragmentation caused by forest harvesting is compensating for this. Over the investigation period, the fragmentation of both study areas is more or less continuously increasing, accumulating the legacy of increasing high fire activities with time. This increased fire fragmentation in the last decade, which was climatologically comparable to the 1980s, might have led to the lower burned areas compared to the 1980s in both the Plain and the Shield. The lower burned area in the 2000s, however will lead to a regrowth of the fuel bed and this will very likely lead to a lower fragmentation of the fuel in the 2010s increasing the risk of large fires. We therefore suggest that when evaluating the risk of forest fires, the fuel fragmentation of the landscape should be evaluated as well as the expected climatic conditions. 4.4. Fuel fragmentation estimation We used a very simplistic and rather conservative fire spread algorithm, requiring a high level of non-fragmented fuel beds in the landscape for a fire to increase in size for our assessment. We are aware that a suite of fire models exist which are designed to simulate fire spread in fragmented landscapes such as the BEHAVE fire modelling system (http://www.firelab.org/project/behaveplus) or PROMETHEUS (http://www.firegrowthmodel.ca/prometheus/ software_e.php). To apply any of these models would require not only the spatial information of which areas have burned before, but also information on climate variables, fuel loads and terrain at a resolution typically much finer than our 1 km2 resolution. However, our aim was not to do a complete evaluation of fire spread in the real landscape, but to evaluate landscape fragmentation. We focused in this analysis on the establishment phase of a forest fire. The assessment of the fire spread was stopped after the fire had reached a size of ca 120 km2. This is well below the characteristic fire size in this area, which according to Lehsten et al. (2014) is around 470 km2. We consider all fires reaching this size to be not susceptible to be extinguished by forest fragmentation anymore. 4.5. Fire weather, climate and fire Though a number of studies have, contradictory to our study, successfully demonstrated the link between fire activity and fire weather (Parisien et al., 2011) or climate (with climate indicating long term weather parameters; Gillett and Weaver, 2004; Westerling et al., 2006) and even used this link in projecting forest fires into the future, we could not detect a link between climate and fire. Previous studies also showed weak relationships between provincial area burned and both fire weather and FWI System components (Harrington et al., 1983; Flannigan and Harrington, 1988). The link between fire and climate is typically demonstrated at a coarse scale, while we choose to perform our study at a finer scale which allowed us to evaluate the effect of fire management without confounding this effect with different eco-zones. However such a link may not be detectable at a fine scale for a number of reasons including high spatial and inter-annual variability (Girardin et al.,
2013). A main reason for the high variability may be the fact that boreal forests are typically ignition limited, which can add a high degree of unpredictability in fire location and occurrence. By contrast, in a study of fires in Africa, Lehsten et al. (2010) found a correlation between burned area and biophysical conditions (including vegetation and climate) with a coefficient of determination of 0.7, which was only possible because the African fires are not ignition limited but productivity (fuel) limited. In a boreal forest however, not only fire prone weather conditions are required to develop a fire. Additionally an ignition also needs to be happening at the right time and place. Unfortunately, it is currently not possible to directly relate all fires to the lightning which caused them. Though lightning detection accuracy can be up to 1–2 km for about 95% of the time with detection efficiency of 80–90% but dropping off sharply in northern areas (Wotton and Martell, 2005). However, lightning detection accuracy can be as low as 3–10 km with estimated detection efficiency of 70% (Flannigan and Wotton, 1991). Analyzing fire ignition pattern is challenged because the vast majority of ignited fires are not developing above the minimum size at which they are recorded with a sufficiently high accuracy. Small fires might also smolder for several days or even weeks before the weather is suitable and it reaches a place where the fuel conditions are sufficient to develop into a detectable fire (Wotton and Martell, 2005). Thus given the current status of the data availability where the starting date of the fires is only a rough guess and the lightning data is at a very coarse temporal and spatial resolution, a more refined analysis between climate and fire activity is hard to perform at a fine scale. Even a study performed at a coarse scale has shown that the trend which was detected in the burned area could not be found in elements of the FWI (Amiro et al., 2004). 4.6. Fire size A number of studies used a linear relationship in log-log terms to relate the fire size to the number of fires and some of them found that in doing so the importance of very large fires was underrepresented (e.g. Cumming, 2001). To account for this effect, the distribution was truncated and parameters were estimated only for the linearly increasing (in log-log space) part of the function (for a review of fire-size fire-number distributions see Cui and Perera, 2008). Lehsten et al. (2014) analyzed the fire-size firenumber distribution of the global boreal biome by multiplying the fire size with the number of fires for each bin (with fire size bins equally spaced on a logarithmic scale). This resulted in a display of the amount of burned area contributed to the total by each fire size class. The distribution had a strong unimodal shape showing that for the boreal biome the assumption of a linear distribution in log-log space of fire-sizes and fire-number is not suitable. That study was performed at a coarse scale and whether the unimodal distribution is also detectable at a finer scale was not clear. This study showed that even at a considerably finer scale the area contribution by fires of different sizes has a clear unimodal distribution, with medium sized fires (at a log scale) contributing the largest proportion of the burned area. However, the decrease in the areal extent has also strongly increased the variability (caused mainly by inter-annual variability), leading to a highly variable CFS distribution at a decadal resolution, compared to the decadal resolved CFSs calculated for the whole Canadian forest in Lehsten et al. (2014). 5. Conclusions The study of two ecological similar forest regions with different fire management history has shown that fuel fragmentation is
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
markedly decreased by fire suppression and has continuously increased over the last three decades, which might be the reason why the burned area in the 2000s was lower than the burned area in the 1980s despite comparable fire prone weather conditions. Given that a regrowth of vegetation will decrease the fuel fragmentation, if fire suppression maintains the currently rather low fire levels, this poses a potential danger for the development of large (and hence expensive) fires. We suggest that the link between fires (sizes and burned area) and fuel fragmentation should be further investigated and that an evaluation of future fire risks should not only include climatic conditions but also fuel fragmentation. Acknowledgement We would like to acknowledge Canadian Forest Service. 2013. Canadian National Fire Database – Agency Fire Data. Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre, Edmonton, Alberta. http://cwfis.cfs.nrcan.gc.ca/ha/nfdb AVHRR Land Cover Data, Canada. The work of Veiko Lehsten has been supported by the Swedish Research Council (FORMAS) via the BIODIVERSA project EC21C. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foreco.2016.06. 014. References Amiro, B.D., Logan, K.A., Wotton, B.M., Flannigan, M.D., Todd, J.B., Stocks, B.J., Martell, D.L., 2004. Fire weather index system components for large fires in the Canadian boreal forest. Int. J. Wildl. Fire 13, 391–400. http://dx.doi.org/10.1071/ WF03066. Burrows, W.R., Kochtubajda, B., 2010. A decade of cloud to ground lightning in Canada: 1999–2008. Part 1: Flash density and occurrence. Atmos. Ocean 48, 177–194. http://dx.doi.org/10.3137/AO1118.2010. Cui, W., Perera, A.H., 2008. What do we know about forest fire size distribution, and why is this knowledge useful for forest management? Int. J. Wildl. Fire 17, 234– 244. http://dx.doi.org/10.1071/WF06145. Cumming, S.G., 2005. Effective fire suppression in boreal forests. Can. J. For. Res. 35, 772–786. Cumming, S.G., 2001. A parametric model of the fire-size distribution. Can. J. For. Res. 31, 1297–1303. http://dx.doi.org/10.1139/cjfr-31-8-1297. de Groot, W.J., Cantin, A.S., Flannigan, M.D., Soja, A.J., Gowman, L.M., Newbery, A., 2013. A comparison of Canadian and Russian boreal forest fire regimes. For. Ecol. Manage. 294, 23–34. http://dx.doi.org/10.1016/j.foreco.2012.07.033. DeLong, S.C., Tanner, D., 1996. Managing the pattern of forest harvest: Lessons from wildfire. Biodivers. Conserv. 5, 1191–1205. http://dx.doi.org/10.1007/ BF00051571. Eberhart, K.E., Woodard, P.M., 1987. Distribution of residual vegetation associated with large fires in Alberta. Can. J. For. Res. 17, 1207–1212. http://dx.doi.org/ 10.1139/x87-186. Field, R.D., Spessa, A.C., Aziz, N.A., et al., 2015. Development of a global fire weather database. Nat. Hazards Earth Syst. Sci. 15, 1407–1423. Flannigan, M.D., Harrington, J.B., 1988. A study of the relation of meteorological variables to monthly provincial area burned by wildfire in Canada (1953–80). J. Appl. Meteorol. 27, 441–452. Flannigan, M.D., Wotton, B.M., 1991. Lightning-ignited forest-fires in Northwestern Ontario. Can. J. For. Res. 21, 277–287. Franklin, J., 1993. Preserving biodiversity: species, ecosystems, or landscapes? Ecol. Appl. 3, 202–205. http://dx.doi.org/10.2307/1941820. Gillett, N.P., Weaver, A.J., 2004. Detecting the effect of climate change on Canadian forest fires. Geophys. Res. Lett. 31. http://dx.doi.org/10.1029/2004GL020876. Girardin, M.P., Ali, A.A., Carcaillet, C., Gauthier, S., Hély, C., Le Goff, H., Terrier, A., Bergeron, Y., 2013. Fire in managed forests of eastern Canada: risks and options. For. Ecol. Manage. 294, 238–249. http://dx.doi.org/10.1016/j.foreco.2012.07.005.
157
Government of Saskatchewan, 2014. Saskatchewan’s Wildfire Management Strategies. Greene, D.F., Zasada, J.C., Sirois, L., Kneeshaw, D., Morin, H., Charron, I., Simard, M.-J., 1999. A review of the regeneration dynamics of North American boreal forest tree species. Can. J. For. Res. 29, 824–839. Hantson, S., Pueyo, S., Chuvieco, E., 2016. Global fire size distribution: from power law to log-normal. Int. J. Wildl. Fire 4, 403–412. Harrington, J.B., Flannigan, M.D., Van Wagner, C.E., 1983. A study of the relation of components of the fire weather index to monthly provincial area burned by wildfire in Canada 1953–1980. Ontario. Hunter, M.L., 1993. Natural fire regimes as spatial models for managing boreal forests. Biol. Conserv. 65, 115–120. http://dx.doi.org/10.1016/0006-3207(93) 90440-C. Johnson, E.A., Miyanishi, K., Weir, J.M.H., 1998. Wildfires in the Western Canadian Boreal Forest: landscape patterns and ecosystem management. J. Veg. Sci. 9, 603–610. http://dx.doi.org/10.2307/3237276. Kafka, V., Gauthier, S., Bergeron, Y., 2001. Fire impacts and crowning in the boreal forest: Study of a large wildfire in western Quebec. Int. J. Wildl. Fire 10, 119– 127. http://dx.doi.org/10.1071/WF01012. Kochtubajda, B., Burrows, W.R., 2010. A decade of cloud to ground lightning in Canada: 1999–2008. Part 2: Polarity, multiplicity and first-stroke peak current. Atmos. Ocean 48, 195–209. Lehsten, V., de Groot, W.J., Flannigan, M., George, C., Harmand, P., Balzter, H., 2014. Wildfires in boreal ecoregions: evaluating the power law assumption and intraannual and interannual variations. J. Geophys. Res. Biogeosci. 119, 14–23. http://dx.doi.org/10.1002/2012JG002252. Lehsten, V., Harmand, P., Palumbo, I., Arneth, A., 2010. Modelling burned area in Africa. Biogeosciences 7, 3199–3214. http://dx.doi.org/10.5194/bg-7-31992010. Letang, D.L., de Groot, W.J., 2012. Forest floor depths and fuel loads in upland Canadian forests. Can. J. For. Res. 42, 1551–1565. Loureiro, C., Fernandes, P., Botelho, H., 2002. Optimizing prescribed burning to reduce wildfire propagation at the landscape scale. In: Forest Fire Research and Wildland Fire Safety: Proceedings of IV International Conference on Forest Fire Research 2002 Wildland Fire Safety Summit, Luso, Coimbra, Portugal, 18–23 November 2002, p. 74. Malamud, B.D., Millington, J.D.A., Perry, G.L.W., 2005. Characterizing wildfire regimes in the United States. Proc. Natl. Acad. Sci. USA 102, 4694–4699. http://dx.doi.org/10.1073/pnas.0500880102. Marshall, I.B., Wiken, E.B., Hirvonen, H., 1999. Terrestrial Ecoprovinces of Canada. Ottawa. McGarial, K., Marks, B.J., 1995. FRAGSTAT: Spatial Pattern Analysis Program for Quantifying Landscape Structure. United States Department of Agriculture, Pacific Northwest Research Station. Palko, S., St-Laurent, L., Huffman, T., Unrau, T., 1995. The Canada vegetation and land cover: a raster and vector data set for GIS applications – uses in agriculture. In: GIS’95 Conference Proceedings 2. GIS World Inc, Fort Collins, CO, USA, pp. 613–619. Parisien, M.-A., Parks, S.A., Krawchuk, M.A., Flannigan, M.D., Bowman, L.M., Moritz, M.A., 2011. Scale-dependent controls on the area burned in the boreal forest of Canada, 1980–2005. Ecol. Appl. 21, 789–805. Stocks, B.J., Mason, J.A., Todd, J.B., Bosch, E.M., Wotton, B.M., Amiro, B.D., Flannigan, M.D., Hirsch, K.G., Logan, K.A., Martell, D.L., Skinner, W.R., 2002. Large forest fires in Canada, 1959–1997. J. Geophys. Res. 108, 8149. http://dx.doi.org/ 10.1029/2001JD000484. Van Wagner, C.E., 1987. Development and Structure of the Canadian Forest FireWeather Index System, Report to the Canadian Forest Service, Ottawa, Canada. Ward, P.C., Mawdsley, W., 2000. Fire management in the boreal forests of Canada. In: Kasischke, E.S., Stocks, B.J. (Eds.), Fire, Climate Change, and Carbon Cycling in the Boreal Forest. New York, pp. 66–84. Weir, J.M.H., Johnson, E.A., Miyanishi, K., 2000. Fire frequency and the spatial age mosaic of the mixed-wood boreal forest in Western Canada. Ecol. Appl. 10, 1162–1177. Westerling, A.L., Hidalgo, H.G., Cayan, D.R., Swetnam, T.W., 2006. Warming and earlier spring increase western U.S. forest wildfire activity. Science 313, 940– 943. Williams, D.E., 1959. Fire Season Severity rating, Technical note. http://dx.doi.org/ 10.1038/180270a0. Work, T.T., Spence, J.R., Voney, J.A., Morgantini, L.E., Innes, J.L., 2003. Integrating biodiversity and forestry practices in western Canada. For. Chron. 75, 906–916. Wotton, B.M., 2009. Interpreting and using outputs from the Canadian Forest Fire Danger Rating System in research applications. Environ. Ecol. Stat. 16, 107–131. http://dx.doi.org/10.1007/s10651-007-0084-2. Wotton, B.M., Martell, D.L., 2005. A lightning fire occurrence model for Ontario. Can. J. For. Res. 35, 1389–1401. http://dx.doi.org/10.1139/x05-071.
Contents lists available at ScienceDirect
Forest Ecology and Management journal homepage: www.elsevier.com/locate/foreco
Fuel fragmentation and fire size distributions in managed and unmanaged boreal forests in the province of Saskatchewan, Canada Veiko Lehsten a,⇑, William de Groot b, Florian Sallaba a a b
Department of Physical Geography and Ecosystem Science, Lund University, Sweden Canadian Forest Service, Natural Resources Canada, Sault Ste. Marie, Ontario, Canada
a r t i c l e
i n f o
Article history: Received 5 February 2016 Received in revised form 2 June 2016 Accepted 6 June 2016 Available online 14 June 2016 Keywords: Forest fires Burned area Saskatchewan Boreal Shield Saskatchewan Boreal Plain Characteristic fire size Climate fire relationship Wildfires Fire size distribution Forest management Climate-forest fire linkages
a b s t r a c t Forest fires are an important disturbance factor of boreal forests, annually burning about 0.5% of the forested area in Canada. Wildfire regimes are influenced by climate and a number of studies project an increase in wildfire activity with climate change. Another factor influencing wildfires is human intervention (fire suppression), and one factor that has rarely been assessed is fuel fragmentation. Studies evaluating the effect of forest fire suppression concluded that in areas with strong suppression effort the burned area as well as the fire size decreased. Here we evaluate wildfire distributions over the last three decades for two areas that differ mainly in their level of forest management and fire suppression: the Boreal Shield (unmanaged) and the Boreal Plain regions (intensively managed) in the Canadian Province of Saskatchewan. We calculate a fuel fragmentation index and relate fire sizes and burned areas to fire weather. We use the concept of the characteristic fire size (CFS); hence we analyze how much burned area is contributed to the total burned area per fire size class. Both areas show a uni-modal distribution of the CFS, indicating that the majority of burned area was contributed by medium sized fires (Boreal Shield 6.39 104 ha, the Boreal Plain 8.79 104 ha). Burned area as well as fuel fragmentation is lower in the managed forest compared to the unmanaged area. The fuel fragmentation index constantly increased since the 1980s in both regions. Despite the large efforts of fire suppression in the Boreal Plains, the CFS is slightly larger in this managed region. Neither the burned area nor the fire size could be linked statistically to the weather conditions, at the time of the fire (using the Canadian Fire Weather Index). We argue that the high fragmentation over the last decades have decreased the burned area. The slightly higher characteristic fire size in the managed area might be explained by the considerably lower fragmentation, counteracting fire suppression efforts. Fuel fragmentation is likely to decrease over the next decades due re-growth. Though a strong link between fire weather and burned area at the fine scale of this study could not be detected we expect that a decrease in fragmentation in combination with an increase in fire prone weather conditions (as expected for the future) might increase the risk of large fires in both areas. We suggest that future fire risk analysis should include an assessment of the effect of fuel fragmentation. Ó 2016 Elsevier B.V. All rights reserved.
1. Introduction Wildland fire has been a prevalent natural disturbance across the North American boreal forest region for millennia (Girardin et al., 2013). The North American boreal fire regime is characterized by infrequent high intensity crown fires (de Groot et al., 2013) and the annual area burned in Canada has been episodic ⇑ Corresponding author at: Department of Physical Geography and Ecosystem Science, Lund University, Sölvegatan 12, SE-23623 Lund, Sweden. E-mail address: [email protected] (V. Lehsten). http://dx.doi.org/10.1016/j.foreco.2016.06.014 0378-1127/Ó 2016 Elsevier B.V. All rights reserved.
and variable, ranging from 300,000 to 7.5 M ha per anno. Standreplacing crown fires currently burn an average of 2–3 M ha (0.5% of total forest area) each year in Canada (Stocks et al., 2002) with typical fire cycles of 75–150 years. Large fires account for most of the annual area burned in the Canadian boreal region, creating a landscape pattern of large areas of young forest (established since the most recent fires) with small patches of older forest embedded, which represent unburned islands and the remnants of older fires (Johnson et al., 1998; Weir et al., 2000). The median size of unburned islands increases sharply with fire size, and the frequency of unburned islands (per 100 ha burned)
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
is greatest for smaller classes of large fires (201–2000 ha) (Eberhart and Woodard, 1987). Fire shape also becomes more irregular with increasing fire size such that the largest fires create the greatest amount of edge relative to area burned (Eberhart and Woodard, 1987). Although some studies have been done on the spatial pattern of individual fires in the boreal region, there has been very limited research conducted on the pattern of fire distribution on the landscape and effect of fuel fragmentation on fire spread. To reach a rational use of forest resources preserving long term ecosystem functionality it has been suggested that forest practices should minimize the differences between managed and unmanaged forests (e.g. Franklin, 1993; Hunter, 1993). Ecosystem-based forest management should favor interventions that lead to landscape compositions and structures close to their counterparts in natural forests (DeLong and Tanner, 1996) including disturbance pattern. Therefore a large interest in past, present and future burned areas has resulted in a number of studies focus on the relationship between climate and burned area, or projecting climate into the future and translating climatic changes into changes in burned areas (for a comprehensive review see Girardin et al., 2013). These studies have also shown that there is considerable uncertainty in the link between climate and burned area (ibid). Though recent increases in North American fire activity have been attributed, at least in part, to a changing climate and increasing temperature (Gillett and Weaver, 2004; Westerling et al., 2006). Amiro et al. (2004) have shown that if the climate is evaluated on a per fire basis, no trends could be detected in fire weather components within the last four decades. Additional difficulties arise from the influence of large fires which result in an over-proportional amount caused by very few fires. Together with changes in reporting and suppression rates, as well as short term climatic variations, linkages between fire and climate might be hard to detect especially at a landscape scales (Girardin et al., 2013). One aspect which has received very little attention in the past is the effect of fragmentation on the fire regime. When analyzing burned area and fire size distributions, Lehsten et al. (2014) found that fire size as well as burned area in Canada increased continuously from the 1960s until the 1980s and remained at a very high level in the 1990s before both characteristics dropped sharply in the 2000s. Given the decrease in the 2000s at a national scale, this raises questions about the nature of the link between temperatures and burned area, which was assumed by Gillett and Weaver (2004) to be a linear relationship, using data from 1960 until 2000. This recent decrease in burned area was hypothesized to be the result of fuel fragmentation resulting from the high fire activities in the 1980s and 1990s (Lehsten et al., 2014). If this is the case, this decrease in burned area would only last a short time period and a re-growth of forests (together with a warmer climate) would result in an increased fire activity in the following decades. Analyzing temporal effects of fuel fragmentation is challenging for a number of reasons. Similar to other fire parameters, fuel fragmentation can be expected to be dependent on the forest type, the climatic conditions and the level of forest management, including forest protection. Additionally, data covering large spatial and temporal scales are needed to allow collection of a sufficiently large and representative distribution of small and large fires in order to detect if they are causing fragmentation. Directly linked to fuel fragmentation is the fire size distribution, as a shift towards large fires (with a constant burned total area) would result in a decrease of fuel fragmentation. Hence both fire-size and fuel fragmentation should be analyzed simultaneously. Cumming (2005) has evaluated the effect of fire management and concluded that fire suppression efforts have decreased both the burned area as well as the ratio of large to small fires. However, if the total burned area decreases, this will also decrease fuel fragmentation, which can
149
potentially favor the development of large and thereby costly fires compared to a landscape with less fire suppression effort. In this study we investigate the effect of forest management (fire suppression) on fuel fragmentation and fire size distribution in two areas which are relatively similar in their ecological attributes apart from their forest management history. We will also analyze the temporal development of fuel fragmentation and its relation to climatic conditions.
2. Methods 2.1. Study area Our study areas are the Boreal Shield and Boreal Plain eco-zones (Ecological Stratification Working Group, 1995) within the Province of Saskatchewan, Canada (we will use the term Shield and Plain in the following to account for the Boreal Shield and Boreal Plain of the province Saskatchewan rather than the whole ecozone which is larger). We used the classification scheme of the National Ecological Framework for Canada (http://sis.agr.gc. ca/cansis/nsdb/ecostrat/gis_data.html) to assign the areas (Fig. 1). Eco-zones are ecological areas based on terrestrial characteristics accounting for surface forms, soils, faunal realms, vegetation and macro climates (Marshall et al., 1999). The Shield is dominated by coniferous (83%), scattered mixedwood forest (8%), and by lakes and rivers (8%), while the Plain consists of lower amounts of coniferous forest (43%) and more mixedwood forest (16%) and broadleaved forest (14%), but additionally includes rangelands (20%) and a lower amount of lakes and rivers (5%). All other land cover classes are below 1%. The total area of the Shield is 1.83 107 ha. Though the Plain is only marginally smaller (1.7 107 ha), its burnable forest area is ca. 22% smaller (sum of all forest types: Shield 1.68 107 ha, Plain 1.31 107 ha). Drying of forest floor organic soils in the boreal shield is generally quicker than in the boreal plains. This is because forest floor fuel loads are higher in the boreal plains than in the boreal shield for black spruce (Picea mariana), jack pine (Pinus banksiana) and aspen (Populus tremuloides) (Letang and de Groot, 2012), which represent the majority of the forested area in those regions. Therefore, the lower forest floor fuel loads in the boreal shield promote faster drying due to shallower, or less densely compacted, organic soils. The Shield is - until today - experiencing essentially a natural fire regime, and is classified as a fire management observation zone where values at risk are assessed with the intent to allow for fire in ecological processes (Government of Saskatchewan., 2014). On the other hand, the Plain has been used as high value commercial forest over the last decennia and is classified as either a modified or full response fire management zone with the object to control and suppress fires. We therefore decided to use the two areas as representatives of boreal forests with relatively similar ecological conditions but different fire management history.
2.2. Fire data We used data from the Canadian National Fire Database (CNFDB)1 (Stocks et al., 2002). We used the polygon data provided to create yearly fire maps of the Shield and the Plain. The CNFDB is a collection of forest fire polygons from Canadian provincial, territorial and national fire management agencies. We filtered out all fires that were located on non-forested areas, fires below 100 ha and used only fire data ranging from 1960 to 2010 since the fire data for earlier times and smaller fires are not considered reliable (Stocks 1
http://cwfis.cfs.nrcan.gc.ca/ha/nfdb.
150
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Fig. 1. Land cover map of the Boreal Shield and the Boreal Plain within Saskatchewan (based on Palko et al., 1995).
et al., 2002). The annual burned areas for the two areas are displayed in Fig. 2. 2.2.1. Fire sizes While the most common metric to report fire activity is total burned area (e.g. Stocks et al., 2002), the size of fires and the num-
ber and size of unburned patches (defining the amount of fire edge and distance to edge within the burned area) are also important fire characteristics affecting seeding distances and the ability to regenerate (Greene et al., 1999). The fire size distribution is also of economic and modelling interest for fire management planning, since fire fighting activity and evacuation activity are based on the
3
18
x 10
Annual burned area [ha]
16
Saskatchewan Boreal Shield
14
Saskatchewan Boreal Plain
12 10 8 6 4 2 0 1960
1970
1980
1990
2000
Fig. 2. Annual burned areas for the Saskatchewan Boreal Shield and Boreal Plain area.
2010
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
assessment of expected number, intensity, and sizes of fires, affecting long term fire management strategy and decision making (Ward and Mawdsley, 2000). A number of different metrics of the fire size distribution have been used in the past. These include the proportion of large fires (Cumming, 2005) or the slope of the linear fit of the fire size to fire number relationship (in log-log) space (Malamud et al., 2005). The interpretation of the proportion of large fires at an ecosystem level is problematic since it requires a threshold value to be set. It is additionally problematic since small fires have a higher omission error and in terms of total burned area they are relatively unimportant. We will elaborate on the use of the proportion of large fires compared to the statistic which we apply in the discussion section. With respect to the second mentioned fire metric, Lehsten et al. (2014) showed that for the boreal forest biome the assumption of a power law (a log-linear relationship required to calculate the slope parameter) is not valid, and (Hantson et al., 2016) could show that only in 2 out of 8 cases the power law assumption was not rejected. Since we were interested in the amount of total burned area that each fire size class contributed . Therefore, we used the concept of the characteristic fire size introduced in Lehsten et al. (2014) to assess whether forest (and forest fire) management has changed the fire size - fire area relationship. For the boreal region it was shown that wildfire sizes follow an unimodal distribution if the fires are binned according to their size (in equal sized bins at a logarithmic scale) and the number of fires in each fire size bin is multiplied with the average size of the fires in the corresponding bin. This results in a plot of the contribution of each fire size class to the total burned area. If this procedure results in a unimodal distribution, a normal density distribution is fitted to the data and the characteristic fire size is the expected value of this distribution (for details see Lehsten et al., 2014). We compare the characteristic fire size of the two areas to assess whether fire management has led to a decrease in fire size as it has been hypothesized in the literature. A detailed example of how the characteristic fire size is calculated is given in the supplementary material S.1. 2.2.2. Fuel data We considered all forest types as burnable fuel. Although coniferous forests have flammable needle fuels that typically promote crown fires during the summer months, broadleaved forests can burn in summer by understory fire during dry years, and they are also well-known to support surface fires in the spring and autumn when cured herbaceous vegetation dries out underneath the leafless trees. To estimate fuel fragmentation by wildfires we assumed that burned areas are not able to support wildfire for 20 years. This re-fueling time reflects the average minimum natural fire cycle (10–35 years) in the southern boreal region of Western Canada found by Weir et al. (2000) and corresponds with general observation of the time required for burned areas in the boreal to revegetate enough to become susceptible to fire again. Though we consider all burned areas to be equal with respect to fuel reduction and hence regrowth of fuels, Kafka et al. (2001) showed that even a single fire can have very different impacts in different parts of its burned area depending on surface material, stand age and stand composition. We used the Vegetation and Land Cover map from Palko et al. (1995) to create a binary land cover mask excluding all areas unsuitable for fires (all non-forested areas). By overlaying this land cover map with the CNFDB fire polygons in the study area we generated a fuel map. All forested areas which had not burned in the preceding 20 years were considered burnable. We rasterized the yearly fuel maps together with the land cover mask using the Albers Equal-Area Conic Projection to guarantee a constant spatial resolution of 1 km by 1 km for each raster cell (one cell equals 100 ha).
151
2.2.3. Fire fragmentation To assess fuel fragmentation, we needed a statistic which would mimic the growth of a fire. A review of the existing statistical measures did not result in a suitable implementation being available. For example the ‘largest patch size’ provided by FRAGSTATS (McGarial and Marks, 1995) and used in Loureiro et al. (2002) has the disadvantage that even a few unburnable cells would stop the growth of a fire, while other statistics which could be used to estimate fire size increased the burned area even if only a few cells in the flaming front would be available. Therefore we introduced a simple process simulating the spread of a fire. Since the majority of fires in the Canadian Boreal forests are caused by lightning strikes (Stocks et al., 2002) we tested whether the distribution of flashes have an influence on the estimated fragmentation using a coarse scale flash rate map (based on the LISOTD dataset http://thunder. msfc.nasa.gov/data/) as well as a fine scale flash map where we filtered out all cloud to ground strikes of the investigation area (a spatially summarized dataset from the Canadian Lightning Detection Network (CLDN); Burrows and Kochtubajda, 2010; Kochtubajda and Burrows, 2010). These two datasets were used to initialize the process simulating fire spread and compared to results with lightning ignition following an equal distribution over the whole area. These tests have shown that there are only very marginal differences between an initialization of the process following an observed lightning distribution and an initialization following an equal distribution of lightning strikes (data not shown). As our study purpose was to analyze the general landscape feature of fuel fragmentation and not a particular fire-lightning relationship, we therefore decided to continue the remaining analysis with an equal distribution of lightning strikes initializing the process. We ran the process separately on the annual fuel maps as well as on the land cover map to assess the proportion of fragmentation caused by wildfires and the proportion caused by landscape fragmentation (e.g. non-forested areas). Since the effect of fires on the fuel map lasted for 20 years, we started the process in the year 1979 to ensure that we have a sufficiently complete representation of fuel fragmentation in our dataset (the fire data starts in the year 1960). Initial tests showed that a stable value for the fragmentation can be obtained by seeding 5 104 flashes, hence each year we seeded this number of flashes across the fuel map based on a uniformly random distribution. The prevailing wind direction in the study area during the fire season (April to September) was from south west, so we simulated a fire spread in north east (NE) direction for all fires. For comparison purposes, additional simulations using other wind directions (i.e., west wind, north east wind) resulted in qualitatively similar results (data not shown). We simulated the potential fire development over ten steps. First we tested whether the lighting hit a non-burnable cell which would stop the process. Otherwise, the process tested whether the fire could spread into the surrounding cells in NE direction. This spreading was simulated in steps (with the first step being the spread from the ignition cell to the neighboring cells in a north east direction) as long as two conditions were fulfilled: At least 2⁄stepnumber-2 cells neighboring the current burned area are burnable. The burnable cells must be adjacent to each other in order to ensure a closed fire front. A visualization of the process is given in the supplementary material Fig. S.4. In case a rule is not fulfilled, the process stops and the total number of cells burned by the process are counted. After the simulation of each potential lightning strike we calculate the fragmentation index as 1 minus the proportion of burned cells (relative to the maximum total area of 121 km2).
152
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Hence a higher number of burned cells correspond to a lower fuel fragmentation. The maximum simulation size for the process was chosen since we assume that fires above this size are relatively unsusceptible to fuel fragmentation at an one km scale, and this size is still well below the characteristic fire size for Canadian forests of more than 104 ha found by Lehsten et al. (2014). We estimate a mean annual fragmentation and analyzed both the temporal change of the fragmentation index as well as its distribution for the two study areas. The process always uses the original fuel map, hence a calculated potential fire does not influence the results of the next seeded fire. If a subsequent fire would be (by chance) seeded in the same cell it would result in the same potential fire size. 2.3. Fire Weather Index Since fire requires certain weather conditions, we included this in our analysis. Here we use the Fire Weather Index (FWI) component of the Canadian Forest Fire Weather Index (FWI) System (Van Wagner, 1987) over the timespan from 1980 to 2009. The FWI is a general indicator of head fire intensity and flame length (Van Wagner, 1987). Through the structure of the FWI System, information from all other components is input to the FWI component, and as such, it is the most commonly-used component in the FWI System to indicate general landscape-level fire danger (Wotton, 2009). Head fire intensity and difficulty in controlling wildfire increase sharply with increasing FWI value (Van Wagner, 1987; Williams, 1959). We acquired data from the global fire weather database developed by Field et al. (2015), which has a spatial resolution of 0.5 by 0.667 degree longitude latitude. We extracted the daily FWI values from May to September (since these are the months with the highest fire activity, see: Lehsten et al., 2014) and plotted their distribution for each decade. 3. Results
the Saskatchewan Boreal Shield is 6.39 104 ha, while the characteristic fire size for the Boreal Plain is 8.79 104 ha. Hence, the Plain has a larger characteristic fire size despite the history of active fire management. The correlation coefficients (R-squares) between the fitted normal distribution and the data displayed in Fig. 3 is 0.96 for the Shield and 0.96 for the Plain. The decadal changes of the burned area are listed in Table 1. While Lehsten et al. (2014) found that even decadal distributions of the CFS followed a normal distribution rather closely, the decadal distribution in the two investigation areas of this study are more irregular caused by the low number of fires (compared to the whole of Canada in Lehsten et al., 2014). The plots of the decadal fire sizes are displayed in the supplementary material Fig. S.3. The largest burned area per decade in the Shield was reached in the 1980s, whereas the Plain reached its largest decadal burned area in the 1990s. The lowest burned area was recorded in the 1990s in the Shield and in the 2000s in the Plain. 3.2. Fuel fragmentation The fragmentation index for the Shield increased steadily from the 1980s to the 1990s, decreased in the early 2000s, and strongly increased afterwards to the highest value in 2010 (Fig. 4). In the Plain, the fragmentation index was rather stable in the 1980s and the first part of the 1990s, with a strong increase in 1996 followed by a slight increase afterwards. The increasing trend over the last three decades is highly significant (p < 0.001) for both areas. Both regions have a different history of forest management including not only fire suppression but also forest harvesting. Our analysis did not include the influence of forest harvesting
Table 1 Decadal changes in total burned area for the Saskatchewan Boreal Shield and the Saskatchewan Boreal Plain. Please also note that the Boreal Shield is ca 28% larger than the Boreal Plain. Time period
3.1. Fire sizes and burned area The differences in the total burned areas between the two areas are clearly visible in the evaluation of the characteristic fire size (Fig. 3). The sum of the bars in Fig. 3 represents the average annual burned area. The characteristic fire size between 1960 and 2009 for
Total burned area [ha]
1980s 1990s 2000s 1980–2009
Shield
Plain
4.0 106 2.7 106 3.0 106 9.8 106
1.2 106 1.6 106 0.74 106 3.6 106
4
Contribution to annual burned area
8
x 10
7
Saskatchewan Boreal Shield Saskatchewan Boreal Plain
ha a 5 4 3 2 1 0 102
103
104
[ha]
106
Fire size class Fig. 3. Characteristic fire sizes for the Boreal Plain and the Boreal Shield of the province of Saskatchewan for the time period from 1980 to 2009. Please note that the Boreal Shield is ca 28% larger than the Boreal Plain.
153
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Saskatchewan Boreal Shield
Fragmentation
0.7 0.6 0.5 0.4 0.3 0.2 1980
1990
2000
2010
Years Saskatchewan Boreal Plain
Fragmentation
0.7 0.6 0.5 0.4 0.3 0.2 1980
1990
2000
2010
Years Fig. 4. Fuel fragmentation for the Saskatchewan Boreal Shield and the Saskatchewan Boreal Plain. Solid line: annual fragmentation, dashed line: trend line, dash-dot line: background fragmentation caused by non-forested land cover (e.g. lakes and fields).
because typical forest cutblock size in Saskatchewan is 20–75 ha (Work et al., 2003), which is smaller than a single cell in our analysis. Forest harvesting is therefore assumed to have only a limited effect on fuel fragmentation and was not included in this study. Both areas also differ in their distribution of fire fragmentation (Fig. 5). The unmanaged Boreal Shield has very few nonfragmented sub-areas, and a large proportion of highly fragmented areas following an approximately linear relationship. The managed Boreal Plain shows a local maximum of areas with medium sized fragmentation, but the global maximum of the distribution is at the same fragmentation level as for the unmanaged area.
3.3. Weather We assess the weather over the last three decades by displaying the distribution of the FWI. Provincial fire management agencies calibrate the FWI scale using local fire and weather data, so scale calibration varies between provinces but in general a value of 30 + is considered extreme in Canada (Van Wagner, 1987). Due to the wide range of values in the dataset we used, the FWI scale is calibrated in the following way for analysis purposes in this study: 0–11.9: low to moderate; 12.0–24.9 high; 25–49.9 very high; 50– 74.9 severe; 75–99.9 extreme; above 100 catastrophic.
Fig. 5. Distribution of fragmentation for the same grid cells (0.5 by 0.66 longitude/latitude).
154
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
Number of days and gridcells
Boreal Shield 100000
1980s 1990s 2000s
10000 1000 100 10 0
low to moderate
high
very high
severe
extreme
catastrophic
Fire weather index
Number of days and gridcells
Boreal Plain 100000
1980s 1990s 2000s
10000 1000 100 10 0
low to moderate
high
very high
severe
extreme
catastrophic
Fire weather index Fig. 6. Categories of Fire Weather Index values in the decades 1980s, 1990s and 2000s. Note that the number of days of within a FWI category is plotted with a log scale.
In Fig. 6 we list how often a grid cell of 0.5 degree longitude and 0.66 degree latitude (resolution of the global fire weather database) was within a certain FWI range. The area contained a total of 106 grid cells in the FWI dataset within the Shield from Field et al. (2015) while 96 grid cells covered the Plain. The FWI values are relatively similar over time (on a log scale) for low to very high fire risk (FWI from 0 to 49.0) For FWI values above 49.9, representing severe to catastrophic conditions, the 1980s had the most extreme fire potential, followed by the 2000s and the 1990s in both investigated areas. The Plain (which is more southern) had a markedly higher number of days per grid cell in all categories above high, and it had the only recorded values in the extreme and catastrophic categories.
3.4. Fire weather versus burned area and fragmentation The link between the FWI and burned area and fragmentation is very weak at the fine scale of this investigation. Fig. 7 relates the percentage of annual area burned, the fragmentation index and the 95th percentile of the FWI (over each year) to each other for the two investigation areas. Each grid cell within this analysis has a size of 0.5 by 0.66 degree latitude/longitude as this is the original resolution of the weather dataset (Field et al., 2015).
4. Discussion Our study of fire size distribution and the fragmentation in the two areas in the province of Saskatchewan has demonstrated that forest fire suppression has not decreased the characteristic fire size and that while fire suppression has led to a lower burned area, it also resulted in a lower fuel fragmentation compared to the area without fire suppression. The link between fire weather and burned area was very weak, which was probably a result of the small area that we investigated.
4.1. Relevance of forest fires The effect of fire management is of high interest to optimize commercial forest use for the coming decades given the amount of forest resources at stake and the cost of fire suppression. A sustainable forest use can only be attained if all aspects of forest ecology are taken into account (e.g. Franklin, 1993; Hunter, 1993) and the forest maintains its natural fire protection in the form of fuel fragmentation, which acts as a dampening factor that can potentially decrease the chances of large fire development after times of high fire activity. The natural forest fuel fragmentation level is hard to assess since we currently have only reliable spatially explicit data covering the last few decades at a national level (Stocks et al., 2002, CNFDB2). A number of proxies have been used to assess forest fire activity in Canada including stand age maps, charcoal records or tree ring densities which were used in correlative approaches to estimate burned area proportions. The review by Girardin et al. (2013) concluded that current fire activity is larger in the majority of the area while in the southwestern and –eastern locations a decrease of fire activity has occurred compared to the pre-industrial era using simulated climate data and fire history proxies. However these results are based on very few data points and therefore do not allow a translation into a Canadian-wide fragmentation estimate for pre-industrial conditions. Additionally, these are only estimates of fire activities. How changes in fire activity influenced fire size distribution, which is crucial for estimating fuel fragmentation, has not been investigated so far.
4.2. Fire size distributions in managed and unmanaged forests Cumming (2005) found that fire suppression has not only decreased the total burned area but also the proportion of large fires which are causing the majority of damage and costs. Our 2
Canadian National Fire Database, http://cwfis.cfs.nrcan.gc.ca/ha/nfdb.
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
155
Fig. 7. The relationship between FWI, fragmentation and proportion burned area (in %) and their distributions. Note that the burned area is plotted on a logarithmic scale. All burned area proportions smaller than 0.01% have been set to 0.01% to allow the logarithmic display. Large numbers indicate the correlation coefficient.
results however show that the characteristic fire size slightly increased in the managed forest. Even though our study areas are relatively similar in their ecological and climatological characteristics, a certain part of the larger characteristic fire size could still be caused by the slightly more fire prone weather as indicated by the higher number of days with FWI ratings above the severe category. Apart from this, there is a more fundamental difference in the ratio between large and small fires which was used as the test statistic for Cumming (2005) and the characteristic fire size. Theoretical considerations show that if one assumes the suppressed fires to develop into a representative sample of the total fire size distribution, the characteristic fire size should be unaffected if only fires larger than the size which is typically attacked are analyzed. The characteristic fire size is estimated based on the contribution of the fires of different sizes, and since small sized fires have a very low contribution to the total burned area, their influence on the characteristic fire size is limited. Therefore, even if fire suppression is effective in decreasing the ratio of large fires versus small fires, the CFS of an area will be largely unaffected. This property of the CFS to be relatively independent of fire suppression efforts allows to better compare CFSs between different ecoregions with varying degrees of fire management likewise as it was done in Lehsten et al. (2014). For fire suppression to have a substantial effect on the characteristic fire size, a considerably larger reduction in total burned area would be required than was found in our study. Our
results show that forest management in conjunction to fire suppression has probably not affected the fire size distribution. We use the term probably because our study implicitly assumes that without management both areas would result in a similar CFS, given a relatively similar climate and ecotype. Since large fires cause the majority of costs, a fire suppression scheme will tend to shift the distribution to smaller fires. The slightly larger CFS of the Plain compared to the Shield region indicates that the majority of burned area is contributed by larger fires compared to the Shield region. Hence fire suppression has not succeeded in shifting the fire distribution towards smaller fires. This is an indication that the decreased fragmentation in the area with fire suppression has resulted in somewhat larger fires on average, though the reduction in numbers still results in a considerable decrease in total burned area. The slight increase in CFS in the managed area might well be the result of the lower fragmentation of the forested area, combined with the slightly more fire prone weather conditions. 4.3. Differences in fuel fragmentation We used a very simplistic and rather conservative fire spread algorithm, requiring a high level of non-fragmented fuel beds in the landscape for a fire to increase in size for our assessment. We focused this analysis on the establishment phase of a forest fire. The assessment of the fire spread is stopped after the fire has
156
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
reached a size of ca 12,000 ha. This is well below the characteristic fire size in this area, which according to Lehsten et al. (2014) is around 47,000 ha. We consider all fires reaching this size to not be limited in spread by forest fragmentation. Our comparison between the two study areas showed that the fragmentation in the Shield is considerably higher than in the Plain as a result of fire suppression. This finding ignores the effect of harvesting on fuel fragmentation. Given that, despite the fact that the landscape fragmentation of the Plain is 50% higher than the landscape fragmentation of the Shield (caused by land uses different than forests), the total fragmentation of the Shield is still 40–50% higher than the total fragmentation of the Plain, it is very unlikely that the fragmentation caused by forest harvesting is compensating for this. Over the investigation period, the fragmentation of both study areas is more or less continuously increasing, accumulating the legacy of increasing high fire activities with time. This increased fire fragmentation in the last decade, which was climatologically comparable to the 1980s, might have led to the lower burned areas compared to the 1980s in both the Plain and the Shield. The lower burned area in the 2000s, however will lead to a regrowth of the fuel bed and this will very likely lead to a lower fragmentation of the fuel in the 2010s increasing the risk of large fires. We therefore suggest that when evaluating the risk of forest fires, the fuel fragmentation of the landscape should be evaluated as well as the expected climatic conditions. 4.4. Fuel fragmentation estimation We used a very simplistic and rather conservative fire spread algorithm, requiring a high level of non-fragmented fuel beds in the landscape for a fire to increase in size for our assessment. We are aware that a suite of fire models exist which are designed to simulate fire spread in fragmented landscapes such as the BEHAVE fire modelling system (http://www.firelab.org/project/behaveplus) or PROMETHEUS (http://www.firegrowthmodel.ca/prometheus/ software_e.php). To apply any of these models would require not only the spatial information of which areas have burned before, but also information on climate variables, fuel loads and terrain at a resolution typically much finer than our 1 km2 resolution. However, our aim was not to do a complete evaluation of fire spread in the real landscape, but to evaluate landscape fragmentation. We focused in this analysis on the establishment phase of a forest fire. The assessment of the fire spread was stopped after the fire had reached a size of ca 120 km2. This is well below the characteristic fire size in this area, which according to Lehsten et al. (2014) is around 470 km2. We consider all fires reaching this size to be not susceptible to be extinguished by forest fragmentation anymore. 4.5. Fire weather, climate and fire Though a number of studies have, contradictory to our study, successfully demonstrated the link between fire activity and fire weather (Parisien et al., 2011) or climate (with climate indicating long term weather parameters; Gillett and Weaver, 2004; Westerling et al., 2006) and even used this link in projecting forest fires into the future, we could not detect a link between climate and fire. Previous studies also showed weak relationships between provincial area burned and both fire weather and FWI System components (Harrington et al., 1983; Flannigan and Harrington, 1988). The link between fire and climate is typically demonstrated at a coarse scale, while we choose to perform our study at a finer scale which allowed us to evaluate the effect of fire management without confounding this effect with different eco-zones. However such a link may not be detectable at a fine scale for a number of reasons including high spatial and inter-annual variability (Girardin et al.,
2013). A main reason for the high variability may be the fact that boreal forests are typically ignition limited, which can add a high degree of unpredictability in fire location and occurrence. By contrast, in a study of fires in Africa, Lehsten et al. (2010) found a correlation between burned area and biophysical conditions (including vegetation and climate) with a coefficient of determination of 0.7, which was only possible because the African fires are not ignition limited but productivity (fuel) limited. In a boreal forest however, not only fire prone weather conditions are required to develop a fire. Additionally an ignition also needs to be happening at the right time and place. Unfortunately, it is currently not possible to directly relate all fires to the lightning which caused them. Though lightning detection accuracy can be up to 1–2 km for about 95% of the time with detection efficiency of 80–90% but dropping off sharply in northern areas (Wotton and Martell, 2005). However, lightning detection accuracy can be as low as 3–10 km with estimated detection efficiency of 70% (Flannigan and Wotton, 1991). Analyzing fire ignition pattern is challenged because the vast majority of ignited fires are not developing above the minimum size at which they are recorded with a sufficiently high accuracy. Small fires might also smolder for several days or even weeks before the weather is suitable and it reaches a place where the fuel conditions are sufficient to develop into a detectable fire (Wotton and Martell, 2005). Thus given the current status of the data availability where the starting date of the fires is only a rough guess and the lightning data is at a very coarse temporal and spatial resolution, a more refined analysis between climate and fire activity is hard to perform at a fine scale. Even a study performed at a coarse scale has shown that the trend which was detected in the burned area could not be found in elements of the FWI (Amiro et al., 2004). 4.6. Fire size A number of studies used a linear relationship in log-log terms to relate the fire size to the number of fires and some of them found that in doing so the importance of very large fires was underrepresented (e.g. Cumming, 2001). To account for this effect, the distribution was truncated and parameters were estimated only for the linearly increasing (in log-log space) part of the function (for a review of fire-size fire-number distributions see Cui and Perera, 2008). Lehsten et al. (2014) analyzed the fire-size firenumber distribution of the global boreal biome by multiplying the fire size with the number of fires for each bin (with fire size bins equally spaced on a logarithmic scale). This resulted in a display of the amount of burned area contributed to the total by each fire size class. The distribution had a strong unimodal shape showing that for the boreal biome the assumption of a linear distribution in log-log space of fire-sizes and fire-number is not suitable. That study was performed at a coarse scale and whether the unimodal distribution is also detectable at a finer scale was not clear. This study showed that even at a considerably finer scale the area contribution by fires of different sizes has a clear unimodal distribution, with medium sized fires (at a log scale) contributing the largest proportion of the burned area. However, the decrease in the areal extent has also strongly increased the variability (caused mainly by inter-annual variability), leading to a highly variable CFS distribution at a decadal resolution, compared to the decadal resolved CFSs calculated for the whole Canadian forest in Lehsten et al. (2014). 5. Conclusions The study of two ecological similar forest regions with different fire management history has shown that fuel fragmentation is
V. Lehsten et al. / Forest Ecology and Management 376 (2016) 148–157
markedly decreased by fire suppression and has continuously increased over the last three decades, which might be the reason why the burned area in the 2000s was lower than the burned area in the 1980s despite comparable fire prone weather conditions. Given that a regrowth of vegetation will decrease the fuel fragmentation, if fire suppression maintains the currently rather low fire levels, this poses a potential danger for the development of large (and hence expensive) fires. We suggest that the link between fires (sizes and burned area) and fuel fragmentation should be further investigated and that an evaluation of future fire risks should not only include climatic conditions but also fuel fragmentation. Acknowledgement We would like to acknowledge Canadian Forest Service. 2013. Canadian National Fire Database – Agency Fire Data. Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre, Edmonton, Alberta. http://cwfis.cfs.nrcan.gc.ca/ha/nfdb AVHRR Land Cover Data, Canada. The work of Veiko Lehsten has been supported by the Swedish Research Council (FORMAS) via the BIODIVERSA project EC21C. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foreco.2016.06. 014. References Amiro, B.D., Logan, K.A., Wotton, B.M., Flannigan, M.D., Todd, J.B., Stocks, B.J., Martell, D.L., 2004. Fire weather index system components for large fires in the Canadian boreal forest. Int. J. Wildl. Fire 13, 391–400. http://dx.doi.org/10.1071/ WF03066. Burrows, W.R., Kochtubajda, B., 2010. A decade of cloud to ground lightning in Canada: 1999–2008. Part 1: Flash density and occurrence. Atmos. Ocean 48, 177–194. http://dx.doi.org/10.3137/AO1118.2010. Cui, W., Perera, A.H., 2008. What do we know about forest fire size distribution, and why is this knowledge useful for forest management? Int. J. Wildl. Fire 17, 234– 244. http://dx.doi.org/10.1071/WF06145. Cumming, S.G., 2005. Effective fire suppression in boreal forests. Can. J. For. Res. 35, 772–786. Cumming, S.G., 2001. A parametric model of the fire-size distribution. Can. J. For. Res. 31, 1297–1303. http://dx.doi.org/10.1139/cjfr-31-8-1297. de Groot, W.J., Cantin, A.S., Flannigan, M.D., Soja, A.J., Gowman, L.M., Newbery, A., 2013. A comparison of Canadian and Russian boreal forest fire regimes. For. Ecol. Manage. 294, 23–34. http://dx.doi.org/10.1016/j.foreco.2012.07.033. DeLong, S.C., Tanner, D., 1996. Managing the pattern of forest harvest: Lessons from wildfire. Biodivers. Conserv. 5, 1191–1205. http://dx.doi.org/10.1007/ BF00051571. Eberhart, K.E., Woodard, P.M., 1987. Distribution of residual vegetation associated with large fires in Alberta. Can. J. For. Res. 17, 1207–1212. http://dx.doi.org/ 10.1139/x87-186. Field, R.D., Spessa, A.C., Aziz, N.A., et al., 2015. Development of a global fire weather database. Nat. Hazards Earth Syst. Sci. 15, 1407–1423. Flannigan, M.D., Harrington, J.B., 1988. A study of the relation of meteorological variables to monthly provincial area burned by wildfire in Canada (1953–80). J. Appl. Meteorol. 27, 441–452. Flannigan, M.D., Wotton, B.M., 1991. Lightning-ignited forest-fires in Northwestern Ontario. Can. J. For. Res. 21, 277–287. Franklin, J., 1993. Preserving biodiversity: species, ecosystems, or landscapes? Ecol. Appl. 3, 202–205. http://dx.doi.org/10.2307/1941820. Gillett, N.P., Weaver, A.J., 2004. Detecting the effect of climate change on Canadian forest fires. Geophys. Res. Lett. 31. http://dx.doi.org/10.1029/2004GL020876. Girardin, M.P., Ali, A.A., Carcaillet, C., Gauthier, S., Hély, C., Le Goff, H., Terrier, A., Bergeron, Y., 2013. Fire in managed forests of eastern Canada: risks and options. For. Ecol. Manage. 294, 238–249. http://dx.doi.org/10.1016/j.foreco.2012.07.005.
157
Government of Saskatchewan, 2014. Saskatchewan’s Wildfire Management Strategies. Greene, D.F., Zasada, J.C., Sirois, L., Kneeshaw, D., Morin, H., Charron, I., Simard, M.-J., 1999. A review of the regeneration dynamics of North American boreal forest tree species. Can. J. For. Res. 29, 824–839. Hantson, S., Pueyo, S., Chuvieco, E., 2016. Global fire size distribution: from power law to log-normal. Int. J. Wildl. Fire 4, 403–412. Harrington, J.B., Flannigan, M.D., Van Wagner, C.E., 1983. A study of the relation of components of the fire weather index to monthly provincial area burned by wildfire in Canada 1953–1980. Ontario. Hunter, M.L., 1993. Natural fire regimes as spatial models for managing boreal forests. Biol. Conserv. 65, 115–120. http://dx.doi.org/10.1016/0006-3207(93) 90440-C. Johnson, E.A., Miyanishi, K., Weir, J.M.H., 1998. Wildfires in the Western Canadian Boreal Forest: landscape patterns and ecosystem management. J. Veg. Sci. 9, 603–610. http://dx.doi.org/10.2307/3237276. Kafka, V., Gauthier, S., Bergeron, Y., 2001. Fire impacts and crowning in the boreal forest: Study of a large wildfire in western Quebec. Int. J. Wildl. Fire 10, 119– 127. http://dx.doi.org/10.1071/WF01012. Kochtubajda, B., Burrows, W.R., 2010. A decade of cloud to ground lightning in Canada: 1999–2008. Part 2: Polarity, multiplicity and first-stroke peak current. Atmos. Ocean 48, 195–209. Lehsten, V., de Groot, W.J., Flannigan, M., George, C., Harmand, P., Balzter, H., 2014. Wildfires in boreal ecoregions: evaluating the power law assumption and intraannual and interannual variations. J. Geophys. Res. Biogeosci. 119, 14–23. http://dx.doi.org/10.1002/2012JG002252. Lehsten, V., Harmand, P., Palumbo, I., Arneth, A., 2010. Modelling burned area in Africa. Biogeosciences 7, 3199–3214. http://dx.doi.org/10.5194/bg-7-31992010. Letang, D.L., de Groot, W.J., 2012. Forest floor depths and fuel loads in upland Canadian forests. Can. J. For. Res. 42, 1551–1565. Loureiro, C., Fernandes, P., Botelho, H., 2002. Optimizing prescribed burning to reduce wildfire propagation at the landscape scale. In: Forest Fire Research and Wildland Fire Safety: Proceedings of IV International Conference on Forest Fire Research 2002 Wildland Fire Safety Summit, Luso, Coimbra, Portugal, 18–23 November 2002, p. 74. Malamud, B.D., Millington, J.D.A., Perry, G.L.W., 2005. Characterizing wildfire regimes in the United States. Proc. Natl. Acad. Sci. USA 102, 4694–4699. http://dx.doi.org/10.1073/pnas.0500880102. Marshall, I.B., Wiken, E.B., Hirvonen, H., 1999. Terrestrial Ecoprovinces of Canada. Ottawa. McGarial, K., Marks, B.J., 1995. FRAGSTAT: Spatial Pattern Analysis Program for Quantifying Landscape Structure. United States Department of Agriculture, Pacific Northwest Research Station. Palko, S., St-Laurent, L., Huffman, T., Unrau, T., 1995. The Canada vegetation and land cover: a raster and vector data set for GIS applications – uses in agriculture. In: GIS’95 Conference Proceedings 2. GIS World Inc, Fort Collins, CO, USA, pp. 613–619. Parisien, M.-A., Parks, S.A., Krawchuk, M.A., Flannigan, M.D., Bowman, L.M., Moritz, M.A., 2011. Scale-dependent controls on the area burned in the boreal forest of Canada, 1980–2005. Ecol. Appl. 21, 789–805. Stocks, B.J., Mason, J.A., Todd, J.B., Bosch, E.M., Wotton, B.M., Amiro, B.D., Flannigan, M.D., Hirsch, K.G., Logan, K.A., Martell, D.L., Skinner, W.R., 2002. Large forest fires in Canada, 1959–1997. J. Geophys. Res. 108, 8149. http://dx.doi.org/ 10.1029/2001JD000484. Van Wagner, C.E., 1987. Development and Structure of the Canadian Forest FireWeather Index System, Report to the Canadian Forest Service, Ottawa, Canada. Ward, P.C., Mawdsley, W., 2000. Fire management in the boreal forests of Canada. In: Kasischke, E.S., Stocks, B.J. (Eds.), Fire, Climate Change, and Carbon Cycling in the Boreal Forest. New York, pp. 66–84. Weir, J.M.H., Johnson, E.A., Miyanishi, K., 2000. Fire frequency and the spatial age mosaic of the mixed-wood boreal forest in Western Canada. Ecol. Appl. 10, 1162–1177. Westerling, A.L., Hidalgo, H.G., Cayan, D.R., Swetnam, T.W., 2006. Warming and earlier spring increase western U.S. forest wildfire activity. Science 313, 940– 943. Williams, D.E., 1959. Fire Season Severity rating, Technical note. http://dx.doi.org/ 10.1038/180270a0. Work, T.T., Spence, J.R., Voney, J.A., Morgantini, L.E., Innes, J.L., 2003. Integrating biodiversity and forestry practices in western Canada. For. Chron. 75, 906–916. Wotton, B.M., 2009. Interpreting and using outputs from the Canadian Forest Fire Danger Rating System in research applications. Environ. Ecol. Stat. 16, 107–131. http://dx.doi.org/10.1007/s10651-007-0084-2. Wotton, B.M., Martell, D.L., 2005. A lightning fire occurrence model for Ontario. Can. J. For. Res. 35, 1389–1401. http://dx.doi.org/10.1139/x05-071.