Influence of tree species on continental differences …...Influence of tree species on continental...

Influence of tree species on continentaldifferences in boreal fires and

climate feedbacksBrendan M. Rogers, Amber J. Soja, Michael L. Goulden, James T. Randerson

SUPPLEMENTARY INFORMATIONDOI: 10.1038/NGEO2352

NATURE GEOSCIENCE | www.nature.com/naturegeoscience

© 2015 Macmillan Publishers Limited. All rights reserved

http://dx.doi.org/10.1038/ngeo2352

1

Supplementary Methods

Overview

Here we present details on the data sources and processing approaches used to quantify

fire intensity, immediate fire severity, and longer-term burn severity across the circumpolar

boreal zone. As described in the main text, analyses were performed on MODIS-based satellite

products at different time scales. Because of its temporal coverage and multiple spectral bands

designed to monitor the biosphere, the MODIS instrument aboard the Terra and Aqua satellites is

currently one of the best tools for studying fire and vegetation processes at moderate-resolution.

Using MODIS products allowed us to achieve a high level of spatial and temporal consistency

with a number of decade-long datasets. While it is possible to evaluate multi-decade responses in

boreal North America using national fire scar databases1, this is not yet possible for Eurasia.

Domain

Most analyses were implemented on the native MODIS '500 m' (463 m nadir) resolution of

the surface reflectance data also used to generate the MCD64A1 burned area product. MODIS

active fire and land surface temperature products use a different set of spectral bands that have a

‘1 km’ (927 m nadir) resolution. Boreal pixels were defined by climate and vegetation type for

pixels north of 40°N. Following ref. 2, we delineated the boreal-temperate boundary using an

upper threshold of 3 °C mean annual temperature. Long-term (1950 - 2000) mean climate data

were taken from ref. 3 at 30 arc-second resolution (926 m × roughly 400 m at domain latitudes,

available from http://www.worldclim.org/current) and re-gridded to 500 m using a nearest-

neighbor approach. We excluded areas under human management (croplands, cropland/natural

vegetation mosaic, and urban types), considering only pixels of non-grassland natural vegetation

from the level 5 MODIS MCD12Q1 product (available from

http://e4ftl01.cr.usgs.gov/MOTA/MCD12Q1.051/) with IGBP land cover classification4.

The Eurasian boreal domain is roughly twice as large the North American (2.04 times

larger using our classification), and, because of larger climatic gradients, boreal forests within

Eurasia are functionally more diverse. We therefore divided Eurasia into more coherent regions

for analysis. Northwest Eurasia is characterized by its relatively maritime climate, dark taiga

ecosystems, and human land use. Northeast Eurasia is home to large expanses of Siberian larch

forests. Southern Eurasia contains more human-affected mixed forests that primarily burned in


2

mid-to-late spring (Supplementary Fig. 1)5–8: April and may fires accounted for 70% of annual

burned area in Southern Eurasia, whereas over 90% of the burned area occurred between June

and August for other regions. The border between Southern and Northwest Eurasia was based on

the southern-central taiga ecoregion divide from ref. 9. We used a line that roughly followed the

patterns of spring burning (pre-June 1) to delineate Southern and Northeast Eurasia. The two

northern regions were separated by a border that follows the light-dark taiga transition from ref.

9, and generally corresponds to the transition between the West Siberian Plain and the Central

Siberian Plateau.

A sensitivity analysis indicated that further classification using lower-bound thresholds of

tree cover (MOD44B, available from http://e4ftl01.cr.usgs.gov/MOLT/MOD44B.005/)10 did not

qualitatively change our results, but excluded many fires in the sparse forests of the Russian Far

East. By not using tree cover, our analyses may include some fires in tundra ecosystems,

although these likely comprise only a small fraction of burned area11. It is also uncertain to what

degree our analyses include or omit fires in peatlands, which cover approximately 25 – 30% of

the circumpolar boreal biome12. Satellite detection of peatland fires is difficult; as a result,

estimates of burned area from boreal peatlands vary widely, but are generally less than 20% of

the total13,7,14–16.

Processing of satellite products

We used six remote sensing-derived products to quantify instantaneous fire intensity,

immediate fire severity, and longer-term burn severity: fire radiative power (FRP), spring albedo,

tree cover, normalized burn ratio (NBR), normalized difference vegetation index (NDVI), and

land surface temperature (LST) (Supplementary Table 1, 2). In all cases, values relative to North

America are reported for Eurasian regions. 95% confidence intervals for regional ratios were

derived using a bootstrap technique. Sample populations were generated from individual year

cohorts. These were then resampled 1 000 times, using the same number of samples as the

original number of years and allowing for repeats. Ratios were calculated from each resample,

and confidence intervals from the population of ratios.

FRP (fire intensity) was quantified using daily 1 km thermal anomalies/active fires of all

quality during 2003 - 2013 from the Aqua (MYD14A1, available from

http://e4ftl01.cr.usgs.gov/MOLA/MYD14A1.005/) and Terra (MOD14A1, available from


3

http://e4ftl01.cr.usgs.gov/MOLT/MOD14A1.005/) satellites17, considering only 1 km pixels that

contained at least two 500 m boreal mask pixels. We restricted our analyses to begin in 2003 as

this is the first full year of coverage from the Aqua satellite. Using only Aqua or only Terra data

from the same time period produced similar results with respect to continental differences (mean

FRP across Eurasia was 51% of the North American mean using all data, 50% with just Aqua,

and 53% with just Terra). While consistent with previous work18,17, our mean FRP differences

are larger than ref. 17: mean FRP in Eurasia was 49% lower than North America in our study vs.

34% in the latter. These differences are likely due to different temporal analysis periods and

region boundaries.

All fire severity metrics were analyzed for burned pixels from the MCD64A1 burned area

dataset that were contained in our 500 m boreal mask, and were calculated using values from one

year pre-fire and one season to one year post-fire. We characterized increases in spring albedo

(dAlbedo) one year after fire using observations spanning day of year (DOY) 49 through 81 from

the MCD43A3 white sky shortwave albedo product19 (available from

http://e4ftl01.cr.usgs.gov/MOTA/MCD43A3.005/) for 2001 - 2012 fires. Spring albedo is

typically higher after fires in high latitudes as trees and tall shrubs are partially combusted, killed,

and some fall over, thereby exposing more snow-covered surfaces to sunlight20–22. We used

pixels derived from full inversions of best and good quality, as well as magnitude inversions with

at least seven observations (BRDF band quality 1, 2, and 3), taken from the MCD43A2 albedo

quality product (available from http://e4ftl01.cr.usgs.gov/MOTA/MCD43A2.005/). Southern

Eurasia displayed a relatively high level of inter-annual variability in spring albedo due to

variable snow cover: the coefficient of variation for spring albedo was 0.094 for Southern

Eurasia vs. 0.015, 0.026, and 0.036 for Northeast Eurasia, Northwest Eurasia, and North

America, respectively, for all boreal pixels during 2000 - 2013. To reduce the resultant

variability and improve the accuracy of our metric, years one and two pre-fire, and one and two

post-fire, were averaged together to calculate dAlbedo in Southern Eurasia, when available.

The MODIS tree cover product (MOD44B) provides annual estimates of percent tree

canopy cover over 5 m in height. Multi-temporal metrics are derived from high-resolution

training data and monthly reflectance composites from bands 1-7, NDVI, and LST. The metrics

are chosen to represent important salient features of vegetation phenology, and include time-

integrated means, amplitudes, and ranks of annual composites. Regression trees are bias-adjusted


4

and compiled to produce a global annual dataset at '250 m' (232 m nadir) of percent tree,

herbaceous, and bare ground cover23. Percent tree cover is thus primarily capturing live tree

cover, not dead, and is a useful complement to spring albedo as both are related to stem area, leaf

area, and the fractional coverage of trees. We calculated relative decreases in tree cover (dTree)

from the MODIS 250 m vegetation continuous fields product (MOD44B)23,10, averaged to 500

m:

dTree = 1- TCpost-fireTCpre-fire

x 100% (1)

Because change ratios can be artificially inflated by low-density stands, we considered only

pixels with at least 19% pre-fire tree cover. This threshold adequately separates forest from

tundra in North America1. We chose to use relative, instead of absolute, changes as our metric

because the proportion of trees killed is more relevant to fire ecology and modeling. This

decision primarily affected comparisons with Northeast Eurasia as it contains more open, sparse

stands24 (Supplementary Fig. 2): mean pre-fire tree cover was 23% in Northeast Eurasia

compared to 36%, 35%, and 37% for Northwest Eurasia, Southern Eurasia, and North America,

respectively.

The differenced normalized burn ratio (dNBR) utilizes changes to near- and shortwave-

infrared reflectance bands and is generally sensitive to landscape charring, increased soil

exposure, and loss of vegetation25,26. Decreases in NDVI (dNDVI) indicate destruction to

photosynthetic tissue27,28,25. We calculated these metrics in similar ways for fires during 2001 -

2012. dNBR was calculated using the MCD43A3 product. As in ref. 29:

NBR = ρ2- ρ7ρ2+ ρ7

(2)

dNBR = NBRpre-fire – NBRpost-fire (3)

NDVI values were obtained from the MOD13A1 (available from

http://e4ftl01.cr.usgs.gov/MOLT/MOD13A1.005/) and MYD13A1 (available from

http://e4ftl01.cr.usgs.gov/MOLA/MYD13A1.005/) vegetation indices products30, and dNDVI


5

was calculated similar to equation 3. Only the highest-quality observations from summer (DOY

177 - 209) were used for both metrics. In locations where burning occurred early in the season,

mainly Southern Eurasia, the landscape experienced two growing seasons by the summer of year

one post-fire. For these pixels, using a post-fire value from the following summer notably

decreased dNBR and dNDVI compared to using the year-of-fire summer value. We therefore

used the year-of summer value for all fires occurring in the spring (pre-June 1), which mostly

acted to increase mean fire severity in Southern Eurasia.

Summertime land surface temperature is elevated after fires because of multiple effects

on the energy budget: decreased roughness length (resulting in less turbulent energy transfer),

reduced transpiration, and lower albedo because of surface char31,32. Increases in summer land

surface temperature (dLST) were quantified using DOY 177 - 209 from the 1 km Aqua satellite

product (MYD11A2, available from http://e4ftl01.cr.usgs.gov/MOLA/MYD11A2.005/)33,34 for

fires during 2003 - 2012. We chose to use Aqua data because its daytime overpass time occurs at

approximately 1:30 pm, which was expected to provide greater temperature contrasts compared

to the 10:30 am overpass time of Terra. We used 1 km pixels that contained at least two boreal

mask pixels and at least one 500 m burn pixel from the MCD64A1 dataset. As with dNBR and

dNDVI, post-fire values for dLST were selected from the year-of-fire for spring fires, and from

one year post-fire for all others.

Longer-term responses

Fires during 2001 - 2008 were used to characterize longer-term responses of spring

albedo and tree cover. We first derived post-fire trajectories for each variable using fires between

2001 and 2005. For every fire year, variables were tracked from the pre-fire year through

maximum post-fire year for pixels that contained a valid pre-fire year and at least one valid post-

fire year data point. Collections of trajectories were then averaged to produce one coherent time

series, similar to ref. 22. Because these variables may display temporal trends, we normalized

burned trajectories by control trajectories, derived in the same manner using all available pixels

for each fire year. Because of the relatively high inter-annual variability in spring albedo in

Southern Eurasia, we used a running three-year average filter for spring albedo in each pixel for

this region when applicable. We quantified uncertainty in the trajectories by deriving them


6

separately from each fire year between 2001 and 2005 (Fig. 3). Trajectories for Eurasia were

calculated by averaging the separate Eurasian regional trajectories, weighted by total burned area.

These longer-term responses of spring albedo (dAlbedolt) and tree cover (dTreelt) were

used to quantify longer-term burn severity. Based on our trajectories, we chose to average post-

fire years 5 - 12 for spring albedo and 5 - 7 for tree cover from fires during 2001 - 2008 to derive

dAlbedolt and dTreelt. The averaging window for tree cover represented a compromise between

capturing the continual decline in Northeast Eurasia and excluding the rebound in tree cover

apparent in other regions (Supplementary Fig. 2). For every pixel with a valid pre-fire value and

at least one valid post-fire value, dAlbedolt was calculated as:

dAlbedolt = albedoyrs 5-12 - albedopre-fire (4)

where albedoyrs 5-12 represents the mean spring albedo during years 5 - 12 post-fire. As with other

albedo metrics, years one and two pre-fire were averaged for albedopre-fire in Southern Eurasia.

dTreelt was calculated in the same manner as dTree (equation 1), except years 5 - 7 post-fire

were used instead of year 1 post-fire.

Surface shortwave forcing

We quantified the surface shortwave forcing (SSF) due to albedo changes during the first 11

years after fire for each region using monthly albedo trajectories and solar insolation. For each

region, albedo trajectories were computed as above for every month of the year using fires

between 2001 and 2005. We computed mean hourly diurnal cycles of solar insolation for every

month at 0.5° using CRU-NCEP reanalysis data35 from 2000 - 2010. For every 0.5° grid cell in

every month during years 1 - 11 after a fire, solar absorption was calculated by multiplying the

mean insolation with its region-specific absorptivity (1 - albedo). We derived the SSF by

subtracting the pre-fire from the post-fire value, and monthly values were then averaged to create

an annual mean forcing. Grid cells were then weighted by the spatial distribution of burned area

within each region to compute region-specific annual mean forcing (Fig. 3). SSF trajectories and

their uncertainties were quantified in a similar manner to spring albedo (above). We calculated

uncertainties for decade-total SSF using the same bootstrap technique on separate SSF

trajectories from each fire year.


7

Derived metrics

Satellite products were transformed and combined to derive metrics more relevant for fire

ecology and modeling: an index of crown scorch, an index of live vegetation destruction, and

percent tree mortality (conceptual flowchart shown in Supplementary Fig. 3). Crown scorch is

much greater in crown fires compared to surface fires and is strongly correlated with tree

mortality36. We employed FRP, dAlbedo, and dTree to estimate crown scorch because,

collectively, these products provide information on the intensity of a fire and the degree of tree

combustion and/or immediate mortality (Supplementary Table 1). Under certain conditions, the

highest-intensity portion of a fire (flaming front) may pass through a given pixel relatively

quickly and therefore be missed by Aqua or Terra. The satellites may also register active fires

with low radiative energy from residual smoldering or ground fires that may persist for days to

weeks37,38. Instead of mean FRP, we therefore chose to use the maximum annual FRP for each

pixel within a 500 m buffer. As stated in the main text, FRP is the product of fireline intensity

and fire line length, and as such is influenced by the length of the fire front, its intensity and rate

of spread, and any residual burning in areas the fire front has passed through39–41. For the

purpose of a crown scorch index we were ultimately interested in fireline intensity, as it has been

shown to correlate linearly with scorch height when raised to a power of 2/342,43. Our final

measure of FRP for this crown scorch index at a given pixel was thus equal to (maximum annual

FRP)2/3 within a 500 m buffer (the exponent also helps account for the highly right-tailed

distribution of FRP values). We were unable to account for the other dynamics of fire behavior

on FRP that may not directly relate to crown scorch, except note that all of these (length of fire

front, rate of spread, and residual burning) are generally higher in high-intensity crown fires

compared to surface fires44,37.

We calculated an index of crown scorch for all MCD64A1 500 m burned pixels between

2003 and 2009 that contained valid dAlbedo, dTree, and maximum FRP values (Supplementary

Table 1, Supplementary Fig 4.). The three metrics were normalized linearly between zero and

one, although dAlbedo and dTree were allowed to be negative at individual pixels to preserve

regional contrasts. Upper bounds for each variable corresponded to the highest 95th percentile of

the four regions (typically North America). We set a lower bound at the lowest 5th percentile

(typically Southern Eurasia). Normalized values were then averaged to produce the crown scorch


8

index. Due to the lack of validation data, we did not attempt to actually separate crown from

surface fires.

We created an index of live vegetation destruction from dNBR, dNDVI, and dLST.

Together, these metrics provide information about fire-induced losses in photosynthesizing and

transpiring vegetation, reductions in upper-layer soil water and roughness length, and the degree

of landscape charring (Supplementary Table 1). As with our index of crown scorch, burned

pixels during 2003 - 2012 that contained valid data for these three products were used. Values

were normalized between zero and one in a manner similar to dAlbedo and dTree (above), and

averaged together to produce an index. We did not attempt to correlate this index with carbon

emissions for several reasons. Both dNBR and dNDVI are known to display non-linear

correlations with properties that influence combustion: NDVI is known to saturate with

increasing vegetation27, and dNBR and is known to saturate with increasing fire severity45,46.

Additionally, our index is not necessarily expected to capture soil organic matter combustion,

which can constitute a high fraction of total emissions47,46. Finally, non-model derived data on

large-scale carbon emissions are currently lacking.

Total relative tree mortality is an essential variable for fire and dynamic vegetation

modeling, and can have a large impact on post-fire carbon fluxes. We quantified total relative

tree mortality using a similar index-based method with dAlbedolt and dTreelt. The MODIS tree

cover product was useful for this assessment because it represents the fractional coverage of trees.

However, its accuracy over boreal forests and burned landscapes has not been fully explored.

Boreal forest spring albedo, on the other hand, may not correlate linearly with tree cover, but it is

a relatively direct, less processed and less parameter-dependent observation. Because of the

expected non-linear and regionally-dependent correlation between spring albedo and tree cover,

we first derived functions relating the two for each region based on all pixels containing valid

values for both (Supplementary Fig. 5). This had the benefit of not only resolving non-linear

relationships, but also of accounting for regional differences due, potentially, to species

composition and leaf habit (deciduous vs. evergreen).

For each pixel, spring albedo from the pre-fire year and the mean from the post-fire years

were transformed into tree cover using the above relationships. Relative decreases in tree cover

were estimated from these albedo-transformed values using the same approach as for dTreelt.

These two independent metrics of decreased relative tree cover were then normalized to values


9

between zero and one and averaged in a similar manner to the other derived metrics. The

resultant index was multiplied by a scalar to arrive at actual percent tree mortality, with an upper

threshold of 100%. We chose a multiplier that resulted in a mean North American tree mortality

of 80%. Although no regional or continental estimates of fire-induced tree mortality are available

for North America or Eurasia, mortality is assumed to be near-complete in North America48. Ref.

49 estimated that because of spatial aggregation effects, AVHRR-derived 1 km burn pixels

contained on average 58% burn scars and 42% unburned islands or perimeter areas. Assuming

this property scales linearly with resolution, 500 m MODIS pixels would contain approximately

80% burn scars and 20% unburned fragments. Our estimate of 80% mortality across North

America therefore theoretically accounts for unburned fractions in the moderate-resolution

imagery, assuming near-complete mortality at the fire-affected plot level. We note that temporal

consideration is especially important here, as estimates of immediate mortality in northern

Eurasia (45% for Northeast Eurasia and 38% for Northwest) were substantially less than decade-

integrated totals (62% for Northeast Eurasia and 68% for Northwest).

We estimated uncertainty levels for all of our derived metrics (Fig. 2, Supplementary

Table 2). In each case, we successively derived the metric using only one of its associated

satellite products. 95% confidence intervals were then calculated from the regional means

relative to North America.

Supporting observations

We analyzed a number of independent datasets to validate our conclusions on fire

behavior and effects. These datasets included smoke plume heights, fire spread rates, fire sizes,

fire weather indices, and lightning flash frequencies. Smoke plume heights were taken directly

from the MISR Plume Height Project50,51, including the Alaska Summer 2009, Arctic Research

of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS) Canada 2008,

North America 2001 – 2008, and Siberia 2001 – 2008 campaigns (85% of plumes located in the

Northeast Eurasia domain). We calculated mean plume heights for boreal North America and

Eurasia using wind-corrected data flagged as good or fair quality over 0.5° grid cells whose land

fraction was composed of at least 90% boreal pixels (Supplementary Fig. 6). This produced 2

213 data points for North America and 1 470 for Eurasia. Results indicated that mean smoke

plume heights were approximately 14 ± 3% higher in North America, while the fraction of


10

plumes above 1.7 km was substantially larger (29% in North America vs. 15% in Eurasia) (Table

1, Supplementary Fig. 6). We did not expect differences in smoke plume height to be

quantitatively similar to our intensity and severity metrics because of the complex non-linear

processes affecting smoke height52. Additionally, we did not account for differences in

atmospheric boundary layer heights and/or dynamics, which could influence the data. However,

given that high-intensity fires generate strong convective air movement and generally inject

smoke into higher layers of the atmosphere53–56, this corroborated our finding that high-intensity

fires were more prevalent in North America.

Higher-intensity crown fires travel faster57,37 and are therefore expected to result in larger

individual fire sizes compared to surface fires. To test the hypothesis that North American fires

are bigger and spread quicker, we calculated mean fire sizes and spread rates between 2001 and

2012 from the MCD64A1 burned area dataset, which includes estimates of fire locations and

dates. We first aggregated pixels into individual fire events in a manner similar to ref. 58. To

calculate fire sizes, adjacent burn pixels (including a 500 m buffer) were assumed to correspond

to the same event if they occurred within 15 days of each other. While shorter temporal windows

may be necessary in other biomes, boreal forest fires can survive as low-intensity ground fires

for several days or weeks, flaring up to become surface or crown fires when meteorological and

fuel continuity conditions are adequate for spread37. Visual inspection indicated this was an

appropriate time frame, and a sensitivity analysis with a five-day window and no 500 m buffer

did not produce qualitatively different results: mean fire sizes were 59%, 42%, and 25% relative

to North America with a 15 day window, vs. 77%, 63%, and 50% with a 5 day window and no

buffer for Northeast, Southern, and Northwest Eurasia, respectively. Because the data were right-

tailed and highly non-normal, we report means and medians without confidence statistics (Table

1).

We used the above fire event database to quantify fire spread rates in three different ways.

Across the boreal zone, as is true in many other biomes, the majority of fires are relatively small

and contribute little to overall burned area: 90% of the fires accounted for only 11% of burned

area in the boreal MCD64A1 dataset. Because the errors in spread rates calculated from

moderate-resolution imagery are also likely high for these small fires, we only considered large

fires for our analysis. These were defined to be fires covering at least six 500 m pixels (129 ha),

as this was the smallest fire size that, when including all larger fires, accounted for at least 95%


11

of the burned area in each region. We first used temporal and spatial characteristics for

individual fire events to calculate area- and length-based spread rates. Fire duration was defined

as the time between minimum and maximum burn dates within a given fire. Fire spread rates

were then calculated as:

Sa= Adt

(5)

Sl = Ldt

(6)

where Sa is the area-based fire spread, A is fire area, dt is fire duration, Sl is the length-based fire

spread, and L is fire length calculated as the maximum distance between pixels with minimum

and maximum burn dates.

Because of uncertainties in defining individual fire events, we also calculated pixel-level

spread rates using an approach similar to ref. 59. Spread rate for an individual pixel was defined

by the time and distance from surrounding previously-burned pixels with the smallest temporal

and spatial differences, constrained by a maximum separation of 15 days and 9 km. Pixels that

burned on the first day of a fire event, or were not surrounded by previously burned pixels, were

defined as ignition centers. Spread rates for ignition pixels were calculated in a similar manner

using surrounding pixels that burned subsequently. Due to the non-normality of the data, we

report geometric means and their confidence intervals analyzed from log-transformations. North

America displayed slightly but significantly faster spread rates with all methods (Table 1).

Although spread rates derived from the above methods were averaged by fire event, weighting

by burned area did not change the results (i.e. Eurasia displayed slower spread rates in all cases).

This was also true when using a maximum separation of 4 days in the pixel-based algorithm, as

in Loboda and Csiszar59, which was performed to account for smoldering fires that were not

actively spreading across the landscape (means of 560 and 531 m d-1 for North America and

Eurasia, respectively).

We assessed fire weather over the burning areas of North America and Eurasia with the

Canadian Fire Weather Index (FWI) System60 (Supplementary Fig. 7). The Canadian FWI has

been used operationally for decades and, more recently, to assess fire weather severity in boreal


12

North America and Eurasia61–64. Climate variables at 0.5° from CRU-NCEP reanalysis data35

during 2000 - 2010 were temporally interpolated using the Community Earth System Model’s

data atmosphere model65. At each grid cell, daily values of temperature, wind, relative humidity,

and cumulative precipitation were extracted at local noon, determined by the hour of maximum

solar insolation. We calculated the daily fine fuel moisture code (FFMC), duff moisture code,

(DMC), drought code (DC), initial spread index (ISI), buildup index (BUI), and fire weather

index as in ref. 66. Annual initial conditions were determined using a method from Turner and

Lawson67: FFMC was set to 85, DMC to 6, and DC to 15, and calculations started when the

mean daily temperature was at least 6°C for three consecutive days. We report means during the

fire season for each region weighted by the spatial distribution of burned area. Only active grid

cells, i.e. locations that experienced at least three consecutive days with noon-time temperatures

exceeding 6°C, were used in calculations. Confidence intervals were based on time series of

annual means (Table 1). Fire season was defined as the three consecutive months of highest

mean burned area. This corresponded to June through August for North America (93% of burned

area), Northeast Eurasia (95% of burned area), and Northwest Eurasia (94% of burned area), and

to April through June for Southern Eurasia (80% of burned area).

Taken together, ISI and BUI capture the important weather-based controls on fire

behavior. ISI characterizes the potential rate of fire spread using wind speed and fine fuel

moisture, and has been related to crown fuel consumption68,69. BUI signifies the amount of

generic dry ground fuel available for combustion. We found that ISI was generally higher in

Eurasia compared to North America during the fire season because of higher temperatures and

lower humidity in Southern and Northeast Eurasia during the fire season (Table 1,

Supplementary Fig. 7). When averaged across the Eurasian continent, fire-season BUI values

were nearly identical to those in North America (Table 1). Our results indicated that fire weather

was not more severe in the fire-prone forests of North America compared to Eurasia during our

study period, and was therefore unlikely driving differences in fire intensity and severity.

While the above analysis was done on fire season means, we also compared fire weather

indices extracted from the day of burning provided by the MCD64A1 dataset. Results from this

analysis are similar but show even stronger support for the hypothesis that fire weather was

similar or more severe in Eurasia. Mean BUI in North America was 38.5 compared to 43.0 in

Eurasia (48.9 in Northeast, 34.8 in Southeast, and 21.4 in Northwest Eurasia), and mean ISI was


13

2.62 in North America compared to 3.44 in Eurasia (3.18 in Northeast, 4.08 in Southeast, and

2.07 in Northwest Eurasia).

To compare frequencies of natural ignition sources, we utilized mean monthly lighting

flash data at 0.5° from the Lightning Imaging Sensor/Optical Transient Detector product

(LIS/OTD, available from http://thunder.nsstc.nasa.gov/data/index.html)70. High-latitude

observations were derived from the OTD sensor on the Microlab-1 satellite between April 3,

1995 and March 23, 2000, which includes intra-cloud and cloud-to-ground lightning flashes

adjusted for view time. We quantified mean annual flash rates (flashes km-2 y-1) for the fire-

prone forests in each region by weighting grid cell values by the spatial distribution of burned

area (Table 1). In all regions, at least 98% of lightning flashes occurred between May and

September, and at least 88% occurred between June and August. Histograms of lightning flashes

were generated using 0.5° grid cells that contributed to 95% of the burn area within each region.

We used these grid cell populations to construct 95% confidence intervals for regional means

and ratios relative to North America, using the same bootstrap technique as above.

Results indicated that boreal Eurasia experienced approximately 35 ± 6% more lightning

flashes than North America (1.83 ± 0.04 vs. 1.36 ± 0.06 flashes km-2 y-1), with Southern Eurasia

experiencing the highest frequency (2.71 ± 0.08 flashes km-2 y-1). Although human ignitions now

constitute a substantial fraction of all ignitions, especially in Southern Eurasia, most burned area

in boreal North America and northern Eurasia is due to large fires from natural lightning

ignition71,13,72,7,73. Our analysis implied that natural lightning ignitions may have been more

common in Eurasia during the course of evolution and community assembly, and therefore

played a role in selecting fire resister species.

Model Comparisons

We compared a number of our observations to two global fire models: the Global Fire

Emissions Database version 3, including contributions from small fires (GFED)74,75, and the

Community Land Model version 4.5 (CLM)76 (Supplementary Table 3). GFED is used as one of

the primary datasets on contemporary global fire emissions, and CLM is employed for numerous

terrestrial modeling studies. Residing within the Community Earth System Model (CESM)65,

CLM is also used to estimate future climate changes for different socioeconomic scenarios.

CESM is one of the models examined in the fifth phase of the Coupled Model Intercomparison


14

Project (CMIP5)77 and in the Intergovernmental Panel on Climate Change (IPCC) reports78,79. By

design, GFED is driven and constrained by more observational data than CLM, although both

utilize a generic plant functional type (PFT) structure to simulate ecosystem properties. We

hypothesized that because the intensity and severity differences we observed were likely due to

species-level controls, the models would be unable to capture continental differences using

generic PFTs.

GFED utilizes a version of the CASA biogeochemical model to simulate global fire

emissions using satellite-derived inputs of precipitation, temperature, absorbed

photosynthetically active radiation, vegetation continuous fields, land cover and ecoregion

classifications, burned area, and active fires74. We compared regional differences in mean fuel

combustion (kg C (combusted) m-2 (burned area)) from GFED to our index of vegetation

destruction. Although we did not expect our index to capture combustion of soil organic matter,

and GFED simulates soil organic matter combustion in boreal forests, the model-data

comparison of regional contrasts was still revealing. Our index of vegetation destruction

incorporates spectral information related to the destruction of photosynthesizing and transpiring

vegetation, landscape charring, and decreased roughness lengths, all properties related to carbon

emissions. We considered all GFED 0.25° grid cells whose land fractions were composed of at

most 10% grass or agricultural areas and at least 50% boreal pixels (using our classification

above). Fuel combustion was then averaged between 2001-2010, weighted within each region by

the spatial distribution of burned area. Although we found substantial differences in combustion-

related indices between North America and Eurasia in the satellite record, GFED simulated

nearly identical relative combustion across the regions (Supplementary Table 3).

We used CLM with its carbon-nitrogen (CN) and biogeochemistry (BGC) models on a 1°

grid resolution (approximately 0.9° latitude × 1.25° longitude), driven by reanalysis climate data

from ref. 80. This configuration simulates the exchange of energy, momentum, water, and carbon

between the land and atmosphere on a half-hourly time step. CLM’s fire model81 mechanistically

simulates fire occurrence, spread, and ecosystem impact. To initialize CLM’s carbon and

nitrogen pools, we spun the model up for 600 years using the accelerated decomposition

routine82, and a further 300 years under normal conditions, using repeating year 2000 driver

datasets. We then conducted two branch runs: a control simulation with fire turned off, and a

prescribed burn in which 5% of every grid cell burned on July 1, followed by 12 years with no


15

fire. A 5% burn at this resolution (approximately 2 500 ha) is within the range of large boreal fire

events. This prescribed burn was employed instead of using the model’s native fire occurrence

for two reasons: (1) we found biases in the spatial distributions of boreal burning, and (2) we

were able directly track the post-fire recovery of all variables. To calculate changes in the fire-

affected portion of forest grid cells, we differenced the control and experimental simulations for

boreal forest PFTs (boreal trees and shrubs) and divided the anomaly by the prescribed burn

fraction. Reported values were averaged within each region, scaled by the observed spatial

distribution of burned area from the MCD64A1 dataset.

Our comparisons with CLM focused on combustion, dAlbedo, and tree mortality.

Combustion was defined by the quantity of carbon combusted per unit burned area. The current

CLM fire model does not combust soil organic matter, enabling more direct comparisons to our

index of vegetation destruction. dAlbedo was calculated in the same manner as our reported

observations (equation 1). Note that without employing a dynamic vegetation module, fire-

affected grid cells in this model configuration immediately began to recover after disturbance

without any change in PFT composition. We therefore do not report comparisons for longer-term

burn severity variables such as dAlbedolt and dTreelt. Comparisons to our tree mortality metric

were made using the percent reduction in tree PFT carbon pools one year after fire. As

hypothesized, results indicated that CLM was unable to capture the observed continental

differences in severity (Supplementary Table 3). To assess the impact of scale, we also simulated

a 50% burn in each grid cell using the same methods as above. While vegetation combustion and

tree mortality in the fire-affected portion of grid cells were identical between the two burn

prescriptions, modeled dAlbedo was slightly altered (0.025 vs. 0.030 in North America, and

0.048 vs. 0.044 in Eurasia, with the 5% and 50% burns, respectively). This did not, however,

qualitatively affect the regional comparisons (Eurasia dAlbedo was 90% higher than North

America with the 5% burn, and 48% higher with the 50% burn).

Although the above models are driven by different data streams and are generally used

for different applications, they both simulate ecosystem properties using a generic PFT approach.

This approach has a multitude of benefits for mechanistic global modeling, but was unable to

capture our observed continental contrasts in fire and burn severity. This supports our hypothesis

that the differences are due to species-level controls on fuel structure because of varying fire

strategies100, and implies that future model developments are needed to simulate boreal forest fire


16

impacts on the land surface and atmosphere.

Species effects

We utilized several datasets on forest species distributions to quantify the influence of

particular fire strategies (as defined by collections of species traits) on fire dynamics. We used

the Fuel Characteristic Classification System (FCCS, available from

http://www.fs.fed.us/pnw/fera/fccs/maps.shtml)83 fuelbeds aggregated to 1 km for Alaska,

National Forest Inventory-based 250 m species distribution maps for Canada84, and a 1 km

genus-level forest distribution map from the International Institute for Applied Systems Analysis

(IIASA, available from http://webarchive.iiasa.ac.at/Research/FOR/russia_cd/download/)85 for

Russia. Note that in the latter, some categories are individual species (e.g. ‘cedar’ represents

Pinus sibirica and ‘fir’ represents Abies sibirica), while others are genus-level (e.g. ‘spruce’

represents Picea abies, P. obovata, and P. ajanensis, and ‘larch’ represents Larix gmelinii, L.

sukaczewii, L. sibirica, and all sub-species). Russia constituted the majority of the Eurasian

domain (91%) and its burned area (95%). A dominant species/genus map was produced from

each source at the native MODIS 500 m resolution using a maximum coverage approach, first

categorized by vegetated/non-vegetated, forest/non-forest, conifer/deciduous, and finally

species/genus. In Canada, mixed forests were defined as those containing between 33% and 67%

conifer and deciduous broadleaf species. This was mainly done for consistency across North

America because the FCCS database in Alaska includes mixed forest classes in its base layer.

Although the majority of tree species or genera are not mapped individually in global land cover

maps, larch (Larix spp.) are distinctly represented by boreal deciduous needleleaf trees. Because

the IIASA map appeared to classify many larch forests as non-forest (shrubs or non-vegetated

land), we also labeled pixels classified as deciduous needleleaf in the Global Land Cover 2000

(GLC2000)86 product as larch forests. This increased the proportion of larch in forested lands of

Russia from 42% to 52%.

Conifer species/genera were categorized into fire strategy types according plant traits

derived from the literature (Supplementary Table 4). In general, fire ‘avoiders’ exhibit no

conspicuous adaptations to surviving or completing a life cycle in fire-prone environments.

Avoiders tend to occupy wetter environments, where fires are relatively infrequent. As such,

annual burned area in these forests is low (Fig. 4, Supplementary Fig. 8), but the avoiders are


17

prone to stand-replacing crown fires under the right conditions5,87,88,55. Fire ‘embracers’ display

traits that facilitate crown fires, such as the retention of low-lying branches, and traits that aid in

post-fire regeneration, such as (semi-)serotinous cones. In contrast, crown fires are suppressed by

the lack of ladder fuels in resister forests due to tree self-pruning. Resisters also tend to have

thick bark, especially toward the tree base, that protects against cambium kill from intense

surface fires89,90,55.

We did not focus on individual deciduous broadleaf tree species, which we aggregated

into one category for comparison purposes. These trees, typically classified as fire ‘endurers’ or

‘invaders’, are largely a function of post-fire successional dynamics91,92,87,1 and often occur in

forests mixed with evergreen conifers92–94. The influence of deciduous broadleaf species on the

fire regimes in both North America and Eurasia was expected to be qualitatively similar, but

limited given the small amount of deciduous broadleaf burned area: deciduous broadleaf/mixed

forests contributed to 9% of forested (8% of total) burned area in North America and 12% of

forested (8% of total) burned area in Eurasia. Deciduous broadleaf/mixed forests were also less

likely to burn: annual burn probability was 0.27% for deciduous broadleaf/mixed vs. 0.57% for

conifers in North America, and 0.28% for deciduous broadleaf vs. 0.41% for conifers in Eurasia.

These likely represent upper bounds because of sub-pixel heterogeneity and the likelihood of

more extensive burning in the conifer component of mixed pixels.

We quantified how our derived fire characteristics (crown scorch, vegetation destruction,

and tree mortality) varied as a function of conifer species/genera and fire strategy (Fig. 4,

Supplementary Fig. 8) beginning in year 2002 for North America and 2001 for Eurasia (based on

the species maps’ reference years). 95% confidence intervals for severity metrics were generated

from annual time series. Results confirmed our hypothesis that fire strategy was a dominant

control on severity patterns; while severity in avoider and embracer forests was similar, it was

consistently lower in resister forests, which exhibited similar metrics to deciduous broadleaf

stands. An analysis using raw satellite products confirmed these results (Supplementary Table 5):

resisters exhibited lower intensity/severity metrics than embracers and avoiders with all metrics

except for FRP and dNBR in Eurasia. Regarding the latter, it should be noted that when using

only ‘extended’ analysis for dNBR (i.e. not using year-of fire post-fire values in locations with

spring fires), dNBR was similar between Eurasian resisters (0.240 ± 0.034) and avoiders (0.245

± 0.037). We note that understory vegetation in the relatively open resister larch canopies of


18

Siberia likely inflated summer-based severity metrics in this region, including vegetation

destruction (Fig. 4).

To further explore the drivers, we used several statistical modeling approaches that

considered predictors of conifer fire strategy and fire weather variables at the time of burning for

individual pixels, including vapor pressure deficit (VPD), wind speed (WS), initial spread index

(ISI), buildup index (BUI), and fire weather index (FWI) (derived from the Canadian fire

weather indices using reanalysis climate data, described above). We first utilized a random

forests model, as implemented in the ‘randomForest’ package95 in R96 derived from Breiman97.

500 regression trees were built using two candidate variables at each tree split. Because of

computational limitations, we conducted 20 simulations using 20 000 random observation pixels.

Variable importance was assessed using the mean decrease in accuracy, or mean percent increase

in mean square error (%IncMSE), when one candidate variable was randomly permuted in the

‘out-of-bag’ data (the proportion withheld for error estimation). This approach suggested that fire

strategy was a substantially more influential explanatory variable than fire weather for crown

scorch (r2 = 0.45, mean ratio of %IncMSE for fire strategy to fire weather variable was 3.2 for

VDP, 3.2 for WS, 3.2 for BUI, 5.3 for ISI, and 5.6 for FWI), vegetation destruction (r2 = 0.29,

mean ratio of %IncMSE for fire strategy to fire weather variable was 1.3 for VDP, 1.5 for WS,

1.7 for BUI, 2.4 for ISI, and 2.7 for FWI), and tree mortality (r2 = 0.30, mean ratio of %IncMSE

for fire strategy to fire weather variable was 2.7 for VDP, 1.5 for WS, 3.0 for BUI, 4.0 for ISI,

and 4.4 for FWI).

Because variable importance is known to become inflated for highly correlated

predictors98, we also conducted random forest simulations using the ‘party’ package99 in R,

which utilizes conditional inference trees as base learners and contains a conditional variable

importance routine that adjusts for correlations between predictor variables98,100. Due to

computational limitations, we conducted 20 simulations using 750 random observation pixels.

Results were qualitatively similar, although fire strategy emerged as an even stronger predictor

compared to fire weather for crown scorch (mean ratio of %IncMSE for fire strategy to fire

weather variable was 45 for VDP, 68 for WS, 51 for BUI, 266 for ISI, and 923 for FWI),

vegetation destruction (mean ratio of %IncMSE for fire strategy to fire weather variable was 1.9

for VDP, 13 for WS, 8.3 for BUI, 11 for ISI, and 36 for FWI), and tree mortality (mean ratio

of %IncMSE for fire strategy to fire weather variable was 6.7 for VDP, 19 for WS, 6.4 for BUI,


19

132 for ISI, and 75 for FWI).

Finally, we performed a multiple linear regression using all available pixels. The

importance of fire strategy and fire weather, collectively, was assessed by conducting the

regression with all predictor variables, with only fire strategy, and with only fire weather

variables. As with the random forests models, fire strategy emerged as the most influential

predictor for crown scorch (r2 = 0.313 with the full model, r2 = 0.307 with only fire strategy, and

r2 = 0.010 with only fire weather), vegetation destruction (r2 = 0.100 with the full model, r2 =

0.075 with only fire strategy, and r2 = 0.024 with only fire weather), and tree mortality (r2 =

0.173 with the full model, r2 = 0.172 with only fire strategy, and r2 = 0.063 with only fire

weather).

We believe that species effects on fire behavior and effects may in fact be greater than we

show here. This is due primarily to potential mapping errors in the vegetation products we used

and the presence of mixed stands. When less flammable trees (e.g., resisters, deciduous broadleaf

trees) are mixed with other flammable conifers, their crowns are frequently destroyed and the

trees are killed in high-intensity fires101,36. Additionally, FRP data are particularly variable and

sporadic; thus, while regional trends in FRP can be indicative of typical fire intensities,

characterizing specific pixels is more problematic.

Uncertainties and biases

Our analysis had a number of sources of uncertainty. Although there was no reason to

suspect differential biases over boreal North America versus Eurasia in the satellite products we

used, these datasets have well-documented limitations. For example, fire radiative power

displays a high degree of daily variability, may be highest during peak burning and fire spread

which occurs after the Terra and Aqua satellites' overpass102,103, and can be influenced by

residual burning after a fire front has passed38. It remains somewhat unclear what physical

characteristics are being measured by dNBR, and its effectiveness at mapping fire severity has

been reported with mixed success25,45. NDVI is particularly vulnerable to contamination from

clouds, atmospheric moisture, and variability in soil reflectance27,104. Both dNBR and NDVI

correlate saturate with properties that relate to fire severity, although, if anything, this suggests

our estimates of continental differences are conservative for these metrics. The MODIS tree

cover product has not been thoroughly tested on burned sites or in boreal forests, and the MODIS


20

land surface temperature product uses a transfer model parameterized by land cover type for the

estimate of emissivity (which may not be representative after a fire)33. In general, spectral indices

may not behave in exactly the same manner for different tree species or functional groups

included in our study area.

Our processing may introduce additional errors through choice of domain, regions,

seasons, and metrics. Differential snow dynamics may influence our spring albedo metric.

Although we attempted to account for the high interannual variability in snow cover in Southern

Eurasia, variations in snow depth, cover, timing, and decadal trends were not taken into account.

Moreover, post-fire changes in spring albedo may be smaller in Northeast Eurasia, home to large

expanses of deciduous needleleaf larch forests, in part because of leaf habit: the absence of

needles in spring may not as effectively mask the underlying snow cover before fire compared to

evergreen trees in other regions. We also recognize that some fires we sampled may have

occurred in tundra or boreal peatlands. It remains uncertain to what degree these fires impact

continental patterns in severity, although our use of vegetation datasets likely excluded most of

these fires from our analysis of tree species effects. Finally, our use of American larch as a

resister in North America is less robust than other species groups because (1) it is a weak resister

(American larch self-prunes and has high leaf moisture but is easily killed by fires), (2) the small

sample size of burned pixels (138 total), and (3) this species frequently occupies low-lying areas

with poor drainage where hydrological conditions may limit fire severity more directly than plant

traits.

Nonetheless, taken together, these metrics consistently indicated that fire intensities and

severities were considerably higher in North America during the observational period. Many of

our difference estimates were also likely to be conservative for several reasons. Firstly, the

MCD64A1 burned area product's mapping algorithm is based on spectral change detections in

the visible, near-infrared, and shortwave infrared reflectance bands105. By producing anomalies

below the algorithm's threshold, smaller or lower-intensity surface fires are more likely to be

omitted75. These fires are, however, often detected via thermal anomalies in the active fire

product. As an example, the ratio of active fire detections to MCD64A1 burned area in boreal

Eurasia was 62% higher than in North America during 2001 - 2012. If included, intensity and

severity metrics from these small un-detected surface fires would likely be low, which would

tend to further amplify the differences we report between the continents. Secondly, in order to


21

capture the strongest post-fire signal possible, we used year-of observations for summer-based

severity metrics in fires that occurred in the spring. This tended to amplify severity in Southern

Eurasia and decrease severity differences when compared to other regions. For example, using

‘extended’ analysis (year-before and year-after) in all pixels for these metrics increased the

contrast in vegetation destruction between the continents (Eurasia was 36% less than the North

American mean with original, and 44% less using extended analysis) and between resisters and

avoiders in Eurasia (avoiders were 8% more severe than resisters with original, and 19% more

using extended analysis).


22

Supplementary references

1. Rogers, B. M., Randerson, J. T. & Bonan, G. B. High-latitude cooling associated with landscape changes from North American boreal forest fires. Biogeosciences 10, 699–718 (2013).

2. Wolfe, J. A. Temperature parameters of humid to mesic forests of Eastern Asia and relation to forests of other regions of the Northern Hemisphere and Australasia: analysis of temperature data from more than 400 stations in Eastern Asia. (US Dep. Int., 1979).

3. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25, 1965–1978 (2005).

4. Friedl, M. A. et al. MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sens. Environ. 114, 168–182 (2010).

5. Furyaev, V. V., Wein, R. W. & MacLean, D. A. in The Role of Fire in Northern Circumpolar Ecosystems (eds. Wein, R. W. & MacLean, D. A.) 18, 221–234 (Wiley, 1983).

6. Valendik, E. N. in Fire in Ecosystems of Boreal Eurasia (eds. Goldammer, J. G. & Furyaev, V. V.) 129–138 (Springer Netherlands, 1996).

7. Shvidenko, A. Z. & Nilsson, S. in Fire, Climate Change, and Carbon Cycling in the Boreal Forest (eds. Kasischke, E. S. & Stocks, B. J.) 132–150 (Springer New York, 2000).

8. Furyaev, V. V., Vaganov, E. A., Tchebakova, N. M. & Valendik, E. N. Effects of fire and climate on successions and structural changes in the Siberian boreal forest. Eurasian J. For. Res. 2, 1–15 (2001).

9. Isachenko, A. G., Shlyapnikov, A. A., Robozertseva, O. D. & Filipetskaya, A. Z. Landscape Map of the USSR. Gen. Minist. Geod. Cartogr. USSR Mosc. 4 Plates (1988).

10. Townshend, J. R. G. et al. Vegetation Continuous Fields MOD44B, 2000 - 2010 Percent Tree Cover, Collection 5. (Univ. of Maryland, 2011).

11. Mack, M. C. et al. Carbon loss from an unprecedented Arctic tundra wildfire. Nature 475, 489–492 (2011).

12. Flannigan, M., Stocks, B., Turetsky, M. & Wotton, M. Impacts of climate change on fire activity and fire management in the circumboreal forest. Glob. Change Biol. 15, 549–560 (2009).

13. Korovin, G. N. in Fire in Ecosystems of Boreal Eurasia (eds. Goldammer, J. G. & Furyaev, V. V.) 112–128 (Springer Netherlands, 1996).

14. Benscoter, B. W. & Wieder, R. K. Variability in organic matter lost by combustion in a boreal bog during the 2001 Chisholm fire. Can. J. For. Res. 33, 2509–2513 (2003).

15. Turetsky, M. R., Amiro, B. D., Bosch, E. & Bhatti, J. S. Historical burn area in western Canadian peatlands and its relationship to fire weather indices. Glob. Biogeochem. Cycles 18, GB4014 (2004).

16. Turquety, S. et al. Inventory of boreal fire emissions for North America in 2004: Importance of peat burning and pyroconvective injection. J. Geophys. Res.-Atmospheres 112, D12S03 (2007).

17. Giglio, L., Csiszar, I. & Justice, C. O. Global distribution and seasonality of active fires as observed with the Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) sensors. J. Geophys. Res.-Biogeosciences 111, G02016 (2006).

18. Wooster, M. J. & Zhang, Y. H. Boreal forest fires burn less intensely in Russia than in North America. Geophys. Res. Lett. 31, L20505 (2004).


23

19. Schaaf, C. B. et al. First operational BRDF, albedo nadir reflectance products from MODIS. Remote Sens. Environ. 83, 135–148 (2002).

20. Amiro, B. D. et al. The effect of post-fire stand age on the boreal forest energy balance. Agric. For. Meteorol. 140, 41–50 (2006).

21. Lyons, E. A., Jin, Y., Randerson, J. T. & Hall, C. Changes in surface albedo after fire in boreal forest ecosystems of interior Alaska assessed using MODIS satellite observations. J Geophys Res 113, G02012 (2008).

22. Jin, Y. et al. The influence of burn severity on postfire vegetation recovery and albedo change during early succession in North American boreal forests. J. Geophys. Res.-Biogeosciences 117, G01036 (2012).

23. Carroll, M. et al. in Land Remote Sensing and Global Environmental Change (eds. Ramachandran, B., Justice, C. O. & Abrams, M. J.) 725–745 (Springer New York, 2011).

24. Sofronov, M. A., Volokitina, A. V., Kajimoto, T. & Uemura, S. The ecological role of moss-lichen cover and thermal amelioration of larch forest ecosystems in the northern part of Siberia. Eurasian J. For. Res. 7, 11–19 (2004).

25. Lentile, L. B. et al. Remote sensing techniques to assess active fire characteristics and post-fire effects. Int. J. Wildland Fire 15, 319–345 (2006).

26. Miller, J. D. et al. Calibration and validation of the relative differenced Normalized Burn Ratio (RdNBR) to three measures of fire severity in the Sierra Nevada and Klamath Mountains, California, USA. Remote Sens. Environ. 113, 645–656 (2009).

27. Huete, A. et al. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 83, 195–213 (2002).

28. Isaev, A. S. et al. Using remote sensing to assess Russian forest fire carbon emissions. Clim. Change 55, 235–249 (2002).

29. Loboda, T., O’Neal, K. J. & Csiszar, I. Regionally adaptable dNBR-based algorithm for burned area mapping from MODIS data. Remote Sens. Environ. 109, 429–442 (2007).

30. Carroll, M. L., DiMiceli, C. M., Sohlberg, R. A. & Townshend, J. R. G. 250m MODIS Normalized Difference Vegetation Index, Collection 5. (Univ. of Maryland, 2004).

31. Chambers, S. D. & Chapin III, F. S. Fire effects on surface-atmosphere energy exchange in Alaskan black spruce ecosystems: Implications for feedbacks to regional climate. J. Geophys. Res. 107, 8145 (2002).

32. Liu, H., Randerson, J. T., Lindfors, J. & Chapin III, F. S. Changes in the surface energy budget after fire in boreal ecosystems of interior Alaska: An annual perspective. J Geophys Res 110, D13101 (2005).

33. Wan, Z. Collection-5 MODIS Land Surface Temperature Products User’s Guide. 30 (ICESS, Univ. of California, 2007).

34. Wan, Z. New refinements and validation of the MODIS Land-Surface Temperature/Emissivity products. Remote Sens. Environ. 112, 59–74 (2008).

35. Wang, A., Barlage, M., Zeng, X. & Draper, C. S. Comparison of land skin temperature from a land model, remote sensing, and in situ measurement. J. Geophys. Res.-Atmospheres 119, 3093–3106 (2014).

36. Hely, C., Flannigan, M. & Bergeron, Y. Modeling tree mortality following wildfire in the southeastern Canadian mixed-wood boreal forest. For. Sci. 49, 566–576 (2003).

37. Ryan, K. C. Dynamic interactions between forest structure and fire behavior in boreal ecosystems. Silva Fenn. 36, 13–39 (2002).


24

38. Barrett, K. & Kasischke, E. S. Controls on variations in MODIS fire radiative power in Alaskan boreal forests: Implications for fire severity conditions. Remote Sens. Environ. 130, 171–181 (2013).

39. Kaufman, Y. J., Kleidman, R. G. & King, M. D. SCAR-B fires in the tropics: Properties and remote sensing from EOS-MODIS. J. Geophys. Res.-Atmospheres 103, 31955–31968 (1998).

40. Wooster, M. J., Zhukov, B. & Oertel, D. Fire radiative energy for quantitative study of biomass burning: derivation from the BIRD experimental satellite and comparison to MODIS fire products. Remote Sens. Environ. 86, 83–107 (2003).

41. Smith, A. M. S. & Wooster, M. J. Remote classification of head and backfire types from MODIS fire radiative power and smoke plume observations. Int. J. Wildland Fire 14, 249–254 (2005).

42. Van Wagner, C. E. Height of crown scorch in forest fires. Can. J. For. Res. 3, 373–378 (1973).

43. Rothermel, R. C. Predicting behavior and size of crown fires in the northern Rocky Mountains. 46 (USDA Forest Service, 1991).

44. Van Wagner, C. E. Conditions for the start and spread of crown fire. Can. J. For. Res. 7, 23–34 (1977).

45. French, N. H. F. et al. Using Landsat data to assess fire and burn severity in the North American boreal forest region: an overview and summary of results. Int. J. Wildland Fire 17, 443–462 (2008).

46. Rogers, B. M. et al. Quantifying fire-wide carbon emissions in interior Alaska using field measurements and Landsat imagery. J. Geophys. Res. Biogeosciences 119, 1608–1629 (2014).

47. Boby, L. A., Schuur, E. A. G., Mack, M. C., Verbyla, D. & Johnstone, J. F. Quantifying fire severity, carbon, and nitrogen emissions in Alaska’s boreal forest. Ecol. Appl. 20, 1633–1647 (2010).

48. Kasischke, E. S. et al. Evaluation of the composite burn index for assessing fire severity in Alaskan black spruce forests. Int. J. Wildland Fire 17, 515–526 (2008).

49. Fraser, R. H. et al. Validation and calibration of Canada-wide coarse-resolution satellite burned-area maps. Photogramm. Eng. Remote Sens. 70, 451–460 (2004).

50. Diner, D. J. et al. Quantitative studies of wildfire smoke injection heights with the Terra Multi-angle Imaging SpectroRadiometer. in Proceedings of SPIE (ed. Hao, W. M.) 7089, 708908 (2008).

51. Nelson, D., Lawshe, C., Diner, D. & Kahn, R. MISR Plume Height Project. (2013). at <http://www-misr.jpl.nasa.gov/getData/accessData/MisrMinxPlumes/>

52. Martin, M. V. et al. Smoke injection heights from fires in North America: analysis of 5 years of satellite observations. Atmospheric Chem. Phys. 10, 1491–1510 (2010).

53. Lavoue, D., Liousse, C., Cachier, H., Stocks, B. J. & Goldammer, J. G. Modeling of carbonaceous particles emitted by boreal and temperate wildfires at northern latitudes. J. Geophys. Res.-Atmospheres 105, 26871–26890 (2000).

54. Stocks, B. J. et al. Crown fire behaviour in a northern jack pine-black spruce forest. Can. J. For. Res. 34, 1548–1560 (2004).

55. Wirth, C. in Forest Diversity and Function (eds. Scherer-Lorenzen, D. M., Körner, P. D. C. & Schulze, P. D. E.-D.) 309–344 (Springer Berlin Heidelberg, 2005).


25

56. Sofiev, M., Ermakova, T. & Vankevich, R. Evaluation of the smoke-injection height from wild-land fires using remote-sensing data. Atmospheric Chem. Phys. 12, 1995–2006 (2012).

57. Bourgeau-Chavez, L. L., Alexander, M. E., Stocks, B. J. & Kasischke, E. S. in Fire, climate change, and carbon cycling in the boreal forest (eds. Kasischke, E. S. & Stocks, B. J.) 111–131 (Springer, 2000).

58. Archibald, S., Lehmann, C. E. R., Gomez-Dans, J. L. & Bradstock, R. A. Defining pyromes and global syndromes of fire regimes. Proc. Natl. Acad. Sci. U. S. A. 110, 6442–6447 (2013).

59. Loboda, T. V. & Csiszar, I. A. Reconstruction of fire spread within wildland fire events in Northern Eurasia from the MODIS active fire product. Glob. Planet. Change 56, 258–273 (2007).

60. Van Wagner, C. E. Development and structure of the Canadian Forest Fire Weather Index System. 35 (Canadian Forestry Service Headquarters, 1987).

61. Flannigan, M. D. & Van Wagner, C. E. Climate change and wildfire in Canada. Can. J. For. Res. 21, 66–72 (1991).

62. De Groot, W. J. et al. A comparison of Canadian and Russian boreal forest fire regimes. For. Ecol. Manag. 294, 23–34 (2013).

63. De Groot, W. J., Flannigan, M. D. & Cantin, A. S. Climate change impacts on future boreal fire regimes. For. Ecol. Manag. 294, 35–44 (2013).

64. Flannigan, M. et al. Global wildland fire season severity in the 21st century. For. Ecol. Manag. 294, 54–61 (2013).

65. Gent, P. R. et al. The Community Climate System Model Version 4. J. Clim. 24, 4973–4991 (2011).

66. Van Wagner, C. E. & Pickett, T. L. Equations and FORTRAN program for the Canadian Forest Fire Weather Index System. 18 (Canadian Foresty Service, Petawa National Forestry Institute, 1985).

67. Turner, J. A. & Lawson, B. D. Weather in the Canadian forest fire danger rating system. A user guide to national standards and practices. 73 (Environment Canada, Pacific Forest Research Centre, 1978).

68. Stocks, B. J. Fire Behavior in Immature Jack Pine. Can. J. For. Res. 17, 80–86 (1987). 69. Stocks, B. J. Fire Behavior in Mature Jack Pine. Can. J. For. Res. 19, 783–790 (1989). 70. Cecil, D. J., Buechler, D. E. & Blakeslee, R. J. Gridded lightning climatology from

TRMM-LIS and OTD: Dataset description. Atmospheric Res. 135, 404–414 (2014). 71. Viereck, L. A. Wildfire in the taiga of Alaska. Quat. Res. 3, 465–495 (1973). 72. Conard, S. G. & Ivanova, G. A. Wildfire in Russian boreal forests - Potential impacts of

fire regime characteristics on emissions and global carbon balance estimates. Environ. Pollut. 98, 305–313 (1997).

73. Stocks, B. J. et al. Large forest fires in Canada, 1959-1997. J. Geophys. Res.-Atmospheres 108, 8149 (2003).

74. Van der Werf, G. R. et al. Global fire emissions and the contribution of deforestation, savanna, forest, agricultural, and peat fires (1997–2009). Atmos Chem Phys 10, 11707–11735 (2010).

75. Randerson, J. T., Chen, Y., Werf, G. R. van der, Rogers, B. M. & Morton, D. C. Global burned area and biomass burning emissions from small fires. J. Geophys. Res. 117, G04012 (2012).


26

76. Oleson, K. W. et al. Technical Description of version 4.5 of the Community Land Model (CLM). (NCAR Earth System Laboratory, Climate and Global Dynamics Division, 2013).

77. Taylor, K. E., Stouffer, R. J. & Meehl, G. A. An Overview of CMIP5 and the Experiment Design. Bull. Am. Meteorol. Soc. 93, 485–498 (2012).

78. IPCC. in Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Solomon, S. D. et al.) (Cambridge University Press, 2007).

79. IPCC. in Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Stocker, T. F. et al.) (Cambridge University Press, 2013).

80. Qian, T., Dai, A., Trenberth, K. E. & Oleson, K. W. Simulation of global land surface conditions from 1948 to 2004. Part I: forcing data and evaluations. J. Hydrometeorol. 7, 953–975 (2006).

81. Li, F., Zeng, X. D. & Levis, S. A process-based fire parameterization of intermediate complexity in a dynamic global vegetation model (vol 9, pg 2761, 2012). Biogeosciences 9, 4771–4772 (2012).

82. Thornton, P. E. & Rosenbloom, N. A. Ecosystem model spin-up: Estimating steady state conditions in a coupled terrestrial carbon and nitrogen cycle model. Ecol. Model. 189, 25–48 (2005).

83. Ottmar, R. D., Sandberg, D. V., Riccardi, C. L. & Prichard, S. J. An overview of the Fuel Characteristic Classification System - Quantifying, classifying, and creating fuelbeds for resource planning. Can. J. For. Res.-Rev. Can. Rech. For. 37, 2383–2393 (2007).

84. Beaudoin, A. et al. Mapping attributes of Canada’s forests at moderate resolution through kNN and MODIS imagery. Can. J. For. Res. 44, 521–532 (2014).

85. Stolbovoi, V. & McCallum, I. Land Resources of Russia. (International Institute for Applied Systems Analysis and the Russian Academy of Science, 2002).

86. Bartholome, E. & Belward, A. S. GLC2000: a new approach to global land cover mapping from Earth observation data. Int. J. Remote Sens. 26, 1959–1977 (2005).

87. Furyaev, V. V. in Fire in Ecosystems of Boreal Eurasia (eds. Goldammer, J. G. & Furyaev, V. V.) 168–185 (Springer Netherlands, 1996).

88. Granström, A. in Fire in Ecosystems of Boreal Eurasia (eds. Goldammer, J. G. & Furyaev, V. V.) 445–452 (Springer Netherlands, 1996).

89. Uemura, S., Tsuda, S. & Hasegawa, S. Effects of fire on the vegetation of Siberian taiga predominated by Larix dahurica. Can. J. For. Res. 20, 547–553 (1990).

90. Wein, R. W. & MacLean, D. A. in The Role of Fire in Northern Circumpolar Ecosystems (eds. Wein, R. W. & MacLean, D. A.) 1–18 (Wiley, 1983).

91. Dyrness, C. T., Viereck, L. A. & Van Cleve, K. in Forest ecosystems in the Alaskan taiga (eds. Van Cleve, K., Chapin III, F. S., Flanagan, P. W., Viereck, L. A. & Dyrness, C. T.) 74–86 (Springer-Verlag, 1986).

92. Viereck, L. A., Cleve, K. V. & Dyrness, C. T. in Forest Ecosystems in the Alaskan Taiga (eds. Cleve, K. V., III, F. S. C., Flanagan, P. W., Viereck, L. A. & Dyrness, C. T.) 22–43 (Springer New York, 1986).

93. Bergeron, Y. & Dansereau, P. R. Predicting the composition of Canadian southern boreal forest in different fire cycles. J. Veg. Sci. 4, 827–832 (1993).

94. Johnstone, J. F. & Kasischke, E. S. Stand-level effects of soil burn severity on postfire regeneration in a recently burned black spruce forest. Can. J. For. Res. 35, 2151–2163 (2005).


27

95. Liaw, A. & Wiener, M. Package ‘randomForest’: Brieman and Cutler’s random forests for classification and regression. (CRAN Repository, 2013).

96. R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, 2012. (R Foundation for Statistical Computing, 2012).

97. Breiman, L. Random forests. Mach. Learn. 45, 5–32 (2001). 98. Strobl, C., Boulesteix, A.-L., Zeileis, A. & Hothorn, T. Bias in random forest variable

importance measures: Illustrations, sources and a solution. BMC Bioinformatics 8, 25 (2007). 99. Hothorn, T. & Strobl, C. Package ‘party’: A Laboratory for Recursive Partytioning.

(CRAN Repository, 2014). 100. Strobl, C., Malley, J. & Tutz, G. An introduction to recursive partitioning: rationale,

application, and characteristics of classification and regression trees, bagging, and random forests. Psychol. Methods 14, 323–348 (2009).

101. Hely, C., Bergeron, Y. & Flannigan, M. D. Effects of stand composition on fire hazard in mixed-wood Canadian boreal forest. J. Veg. Sci. 11, 813–824 (2000).

102. Mu, M. et al. Daily and 3-hourly variability in global fire emissions and consequences for atmospheric model predictions of carbon monoxide. J. Geophys. Res.-Atmospheres 116, D24303 (2011).

103. Zhang, X., Kondragunta, S., Ram, J., Schmidt, C. & Huang, H.-C. Near-real-time global biomass burning emissions product from geostationary satellite constellation. J Geophys Res 117, D14201 (2012).

104. Epting, J., Verbyla, D. & Sorbel, B. Evaluation of remotely sensed indices for assessing burn severity in interior Alaska using Landsat TM and ETM+. Remote Sens. Environ. 96, 328–339 (2005).

105. Giglio, L., Loboda, T., Roy, D. P., Quayle, B. & Justice, C. O. An active-fire based burned area mapping algorithm for the MODIS sensor. Remote Sens. Environ. 113, 408–420 (2009).

106. Horton, K. W. The ecology of lodgepole pine in Alberta. 29 (Department of Northern Affairs and National Resources, Forestry Branch, Forest Research Division, 1956).

107. Beaufait, W. R. Some effects of high temperatures on the cones and seeds of jack pine. For. Sci. 6, 194–8 (1960).

108. Day, R. Stand Structure, Succession, and Use of Southern Albertas Rocky Mountain Forest. Ecology 53, 472–478 (1972).

109. Heinselman, M. L. Fire in the virgin forests of the Boundary Waters Canoe Area, Minnesota. Quat. Res. 3, 329–382 (1973).

110. Archibold, O. W. Buried viable propagules as a factor in postfire regeneration in northern Saskatchewan. Can. J. Bot. 57, 54–58 (1979).

111. Archibold, O. Seed Input into a Postfire Forest Site in Northern Saskatchewan. Can. J. For. Res.-Rev. Can. Rech. For. 10, 129–134 (1980).

112. Carroll, S. & Bliss, L. Jack Pine - Lichen Woodland on Sandy Soils in Northern Saskatchewan and Northeastern Alberta. Can. J. Bot.-Rev. Can. Bot. 60, 2270–2282 (1982).

113. Viereck, L. A. in The Role of Fire in Northern Circumpolar Ecosystems (eds. Wein, R. W. & Maclean, D. A.) 201–220 (Wiley, 1983).

114. Mccune, B. Ecological Diversity in North-American Pines. Am. J. Bot. 75, 353–368 (1988).


28

115. Taylor, K. L. & Fonda, R. W. Woody fuel structure and fire in subalpine fir forests, Olympic National Park, Washington. Can. J. For. Res. 20, 193–199 (1990).

116. Barrett, S., Arno, S. & Key, C. Fire Regimes of Western Larch - Lodgepole Pine Forests in Glacier-National-Park, Montana. Can. J. For. Res. 21, 1711–1720 (1991).

117. Arseneault, D. Impact of fire behavior on postfire forest development in a homogeneous boreal landscape. Can. J. For. Res. 31, 1367–1374 (2001).

118. Schwilk, D. W. & Ackerly, D. D. Flammability and serotiny as strategies: correlated evolution in pines. Oikos 94, 326–336 (2001).

119. Michaletz, S. T., Johnson, E. A., Mell, W. E. & Greene, D. F. Timing of fire relative to seed development may enable non-serotinous species to recolonize from the aerial seed banks of fire-killed trees. Biogeosciences 10, 5061–5078 (2013).

120. Nikolov, N. & Helmisaari, H. in A systems analysis of the global boreal forest (eds. Shugart, H. H., Leemans, R. & Bonan, G. B.) 13–84 (Cambridge Universitty Press, 1992).

121. Makoto, K. et al. Regeneration after forest fires in mixed conifer broad-leaved forests of the Amur region in Far Eastern Russia: The relationship between species specific traits against fire and recent fire regimes. Eurasian J. For. Res. 10, 51–58 (2007).

122. Alexander, H. D. et al. Carbon accumulation patterns during post-fire succession in Cajander larch (Larix cajanderi) forests of Siberia. Ecosystems 15, 1065–1082 (2012).

123. Whelan, R. J. The ecology of fire. (Cambridge University Press, 1995). 124. Keeley, J. E. Fire intensity, fire severity and burn severity: a brief review and suggested

usage. Int. J. Wildland Fire 18, 116–126 (2009).


29

Supplementary Table 1. Description of satellite products used to quantify fire intensity, fire severity, and burn severity.

Variable Information about fire behavior or ecosystem impacts

Product name(s)

Satellite Days of year

Years1

Fire radiative power

Rate of energy emitted by combustion at time of satellite overpass

MOD14A1, MYD14A1

Aqua & Terra

1 - 365 2003 – 2013

Spring albedo Tree combustion or fire-induced

mortality that increases snow exposure (ref. 22)

MCD43A3 Aqua & Terra

49 - 81 2000 – 2013

Tree cover Fire-induced losses of trees greater than

5 m in height MOD44B Terra annual 2000 – 2010

dNBR Landscape charring, loss of live

vegetation, and decreased water in the canopy and upper soil layer (ref. 25,26)

MCD43A3 Aqua & Terra

177 - 209 2000 – 2013

NDVI Loss of photosynthetic vegetation (ref.

27,28,25) MOD13A1, MYD13A1

Aqua & Terra

177 - 209 2000 – 2013

Land surface temperature

Fire-induced decreases in roughness length, transpiration, and summer surface albedo due to char (ref. 31,32)

MYD11A2 Aqua 177 - 209 2002 - 2013

1Acquisition years. Except for fire radiative power, fire years were less because of the requirement for at least one pre-fire and one post-fire year


30

Supplementary Table 2. Absolute and relative values of satellite products and derived metrics.

Fire process Metric North America1

Eurasia Northeast Eurasia

Southern Eurasia

Northwest Eurasia

Satellite products2 fire intensity fire radiative

power (MW) 198.2 ± 11.4 101.6 (51 ± 4%) 111.8 (56 ± 5%) 103.6 (52 ± 4%) 70.2 (35 ± 3%)

fire severity dAlbedo3 0.113 ± 0.008 0.017 (15 ± 6%) 0.016 (14 ± 11%) 0.015 (14 ± 28%) 0.032 (28 ± 6%)

dTree4 (%) 34.6 ± 3.4 21.1 (61 ± 13%) 22.5 (65 ± 16%) 20.0 (58 ± 22%) 21.8 (63 ± 15%) dNBR 0.387 ± 0.022 0.251 (65 ± 9%) 0.305 (78 ± 8%) 0.145 (37 ± 9%) 0.228 (59 ± 6%) dNDVI5 0.159 ± 0.011 0.115 (72 ± 9%) 0.138 (87 ± 13%) 0.070 (44 ± 12%) 0.141 (89 ± 13%) dLST6 (°C) 4.75 ± 1.28 2.79 (59 ± 20%) 3.01 (63 ± 27%) 1.86 (39 ± 19%) 5.74 (121 ± 34%)

burn severity dAlbedolt

7 0.163 ± 0.008 0.057 (35 ± 8%) 0.068 (42 ± 17%) 0.043 (27 ± 10%) 0.069 (42 ± 9%) dTreelt

8 (%) 42.8 ± 6.0 26.9 (63 ± 17%) 27.5 (64 ± 21%) 24.8 (58 ± 17%) 48.2 (113 ± 23%) Derived metrics9 fire intensity crown scorch10 0.385 0.142 (37 ± 28%) 0.124 (32 ± 36%) 0.161 (42 ± 24%) 0.148 (38 ± 22%) fire severity vegetation

destruction10 0.496 0.318 (64 ± 5%) 0.381 (77 ± 12%) 0.197 (40 ± 4%) 0.428 (86 ± 33%)

burn severity tree mortality (%) 80.0 46.6 (58 ± 5%) 62.4 (78 ± 28%) 33.7 (42 ± 25%) 68.0 (85 ± 42%) 1Uncertainties represent 95% confidence intervals derived from cohorts of individual years 2Values in parentheses are percent relative to North America, with 95% confidence intervals generated from a bootstrap technique for mean regional ratios 3Increase in spring albedo 4Relative decrease in tree cover 5Decrease in NDVI 6Increase in summer land surface temperature 7Increase in spring albedo years 5 - 12 after a fire 8Relative decrease in tree cover years 5 - 7 after a fire 9Values in parentheses are percent relative to North America, with 95% confidence intervals based on calculating the derived metrics separately from each of their individual components 10Index varies between zero and one


31

Supplementary Table 3. Comparisons of fire severity and burn severity between observations and two global fire models.

Metric Source North America

Eurasia1 Northeast Eurasia1

Southern Eurasia1

Northwest Eurasia1

vegetation destruction

observations 0.496 0.318 (64%) 0.381 (77%) 0.197 (40%) 0.428 (86%) GFED2,3 2.58 2.55 (99%) 2.63 (102%) 2.53 (98%) 2.00 (77%) CLM2 2.56 2.13 (83%) 1.61 (63%) 2.83 (110%) 2.72 (106%)

dAlbedo observations 0.113 0.017 (15%) 0.016 (14%) 0.015 (14%) 0.032 (28%) CLM 0.025 0.048 (190%) 0.046 (185%) 0.051 (205%) 0.039 (155%)

tree mortality (%)4

observations 80.0 46.6 (58%) 62.4 (78%) 33.7 (42%) 68.0 (85%) CLM 34.7 33.7 (97%) 33.7 (97%) 33.3 (96%) 35.6 (103%)

1Values in parentheses are percent relative to North America 2Modeled values are combustion (kg C m-2) 3GFED combustion includes organic soils, and is therefore total biomass combustion per unit area 4Tree mortality for boreal forests in the GFED model is prescribed to be a constant 60%


32

Supplementary Table 4. Plant-adapted fire traits categorized into general fire strategies Continent Fire

strategy Species Coverage

(%)1 Burned area (%)2

Fire traits

North America3

avoider

white spruce (Picea glauca)

6.1 9.6 aerial seed banks that may somewhat survive fire, wind-dispersed seeds that germinate on burned ground

subalpine fir (Abies lasiocarpa)

2.8 0.5 wind-dispersed seeds that germinate on burned ground, highly flammable

balsam fir (Abies balsamea)

0.7 0.0007 no fire strategy traits

embracer

black spruce (Picea mariana)

58.2 64.8 some vegetative resprouting, semi-serotinous cones, retention of low-lying branches, dense stands with moss/lichen floor, early reproductive maturity, highly flammable

jack pine (Pinus banksiana)

2.7 11.2 some vegetative resprouting, seritonous cones, loosely-packed needles on fine twigs, early reproductive maturity, rapid growth, preference for mineral soils, accumulation of flammable forest floor debris, litter, and lichen

lodgepole pine (Pinus contorta)

5.3 3.4 seritonous cones, aggressive seed colonizer, early reproductive maturity, rapid growth

resister American larch (Larix

laricina) 0.4 0.01 high leaf moisture, self-pruning

Eurasia4

avoider

Siberian pine (Pinus sibirica) 5.8 3.5 no fire strategy traits

Norway spruce (Picea abies)

10.6 2.2

no fire strategy traits

Siberian spruce (Picea obovata)


Yeddo spruce (Picea ajanensis)


Siberian fir (Abies sibirica) 2.2 0.5 no fire strategy traits

resister

Dahurian larch (Larix gmelinii)

51.5 70.2

thick bark, thick litter layer, high leaf moisture, long-range wind-dispersed seeds that germinate on burned ground, rapid growth, self-pruning, low resin content

Russian larch (Larix sukaczewii)

some vegetative resprouting, thick bark, high leaf moisture, wind-dispersed seeds that germinate on burned ground, early reproductive maturity, rapid growth, self-pruning, low resin content

Siberian larch (Larix sibirica)

thick bark, high leaf moisture, wind-dispersed seeds that germinate on burned ground, high seedling density, self-pruning, low resin content

Scots pine (Pinus sylvestris)

15 11.7 thick bark, early reproductive maturity, wind-dispersed seeds that germinate on burned ground, self-pruning

1Percent of total forested area in the vegetation datasets, including contributions from deciduous broadleaf/mixed forests 2Percent of total mean annual burned area from forests according to the vegetation datasets, including contributions from deciduous broadleaf/mixed forests 3Information on fire traits for North American tree species was gathered from ref.’s 106–118, 55, and 119 4Information on fire traits for Eurasian tree species was gathered from ref.’s 90, 89, 120, 55, and 121


33

Supplementary Table 5. Effects of fire strategy on intensity and severity from raw satellite products. North America Eurasia

Metric Deciduous Broadleaf

Resisters Avoiders Embracers Deciduous Broadleaf

Resisters Avoiders

Fire intensity

Fire Radiative Power (MW)

153.5 ± 11.6 191.1 ± 49.9 195.3 ± 15.8 216.2 ± 11.6 97.8 ± 7.0 102.3 ± 7.8 91.3 ± 9.1

Fire Severity

dAlbedo 0.097 ± 0.023 0.108 ± 0.057 0.139 ± 0.017 0.118 ± 0.011 0.012 ± 0.028 0.020 ± 0.006 0.035 ± 0.014 dTree (%) 21.9 ± 9.8 15.3 ± 13.4 40.3 ± 7.4 35.4 ± 3.0 14.8 ± 6.3 22.4 ± 4.2 31.4 ± 3.9 dNBR 0.230 ± 0.062 0.228 ± 0.038 0.402 ± 0.024 0.406 ± 0.022 0.154 ± 0.033 0.271 ± 0.029 0.240 ± 0.038 dNDVI 0.092 ± 0.026 0.061 ± 0.024 0.153 ± 0.014 0.168 ± 0.013 0.085 ± 0.028 0.127 ± 0.013 0.151 ± 0.024 dLST (°C) 2.77 ± 1.50 3.07 ± 1.35 4.12 ± 1.53 5.08 ± 1.27 1.81 ± 0.97 2.98 ± 0.54 3.20 ± 0.69

Burn Severity

dAlbedolt 0.137 ± 0.036 0.126 ± 0.044 0.194 ± 0.032 0.171 ± 0.013 0.061 ± 0.031 0.074 ± 0.027 0.112 ± 0.022 dTreelt (%) 44.9 ± 8.1 27.4 ± 15.6 52.6 ± 2.5 42.9 ± 4.2 25.2 ± 3.3 25.6 ± 10.3 47.7 ± 6.4


34

Supplementary Figure 1. Mean (a) annual burned area, (b) annual burn fraction, and (c) monthly burned area by region. In the legend, North America is represented by NA, Northeast Eurasia by NEEU, Southern Eurasia by SEU, and Northwest Eurasia by NWEU. Mean burned area was calculated from MCD64A1 pixels that were contained within our boreal domain. Note that many of these contain tundra vegetation, such that the burn fraction of boreal forests is likely higher than shown in (b).


35

Supplementary Figure 2. Post-fire trajectories of tree cover aggregated from 2001 - 2005 fires. In the legend, North America is represented by NA, Northeast Eurasia by NEEU, Southern Eurasia by SEU, and Northwest Eurasia by NWEU. Tree cover began to rebound 5 - 8 years post-fire in most regions, likely due to short-stature saplings that did not affect spring albedo (Fig. 3). The exception was Northeast Eurasia, which may require several decades for tree re-growth due to the cold and harsh environment7,122.


36

Supplementary Figure 3. Conceptual flowchart for derived fire intensity, fire severity, and burn severity metrics. Fire intensity describes fire behavior, including the temperature and heat released at the flaming front123. This process is represented by fire radiative power and our derived crown scorch metric, as higher crown scorch is associated with more intense combustion. Fire severity characterizes immediate fire effects on the environment, and is represented here by vegetation destruction. We employed post-fire tree mortality as a measure of burn severity, which is defined by the degree to which an ecosystem has changed due to a fire38,124.


37

Supplementary Figure 4. Maps of derived metrics aggregated to 1° for (a) crown scorch, (b) vegetation combustion, and (c) total relative tree mortality.


38

Supplementary Figure 5. Regional relationships between tree cover and spring albedo used to quantify tree mortality. Logistic curves were fit to points with albedo greater than 0.2. Linear regressions were used below this threshold: regional-specific for Northeast (NEEU) and Northwest Eurasia (NWEU), and domain-wide for North America (NA) and Southern Eurasia (SEU). Error bars represent standard errors.


39

Supplementary Figure 6. Histograms of regional (a) digitized smoke plume heights and (b) annual lightning flashes. In the legend, North America is represented by NA, Eurasia by EU, Northeast Eurasia by NEEU, Southern Eurasia by SEU, and Northwest Eurasia by NWEU.


40

Supplementary Figure 7. Monthly mean local noon 2-meter (a) air temperature, (b) precipitation, (c) vapor pressure deficit, (d) wind speed, (e) initial spread index (ISI), and (f) buildup index (BUI), and histograms of (g) ISI and (h) BUI during the fire season for each region during 2000 - 2010. In the legend, North America is represented by NA, Northeast Eurasia by NEEU, Southern Eurasia by SEU, and Northwest Eurasia by NWEU.


41

Supplementary Figure 8. Fire characteristics by forest species groups for (a) North America and (b) Eurasia, organized by deciduous broadleaf trees, mixed forests, and finally conifers ordered by increasing contribution to burned area. Colors represent fire strategy. All metrics except for crown scorch and vegetation destruction are plotted as percentages on the left-hand axis. Error bars represent 95% confidence intervals derived from populations of annual means. On the x-axis, Am. larch = American larch, Decid. broad. = deciduous broadleaf, Other con. = other conifers, Sib. fir = Siberian fir, and Sib. pine = Siberian pine. Note that contributions from non-forested areas, 16% of total burned area in North America and 28% in Eurasia according to the vegetation maps, are not included here, and that only Russia is included for Eurasia (95% of its burned area).


Influence of tree species on continental differences …...Influence of tree species on continental...

Documents

Transcript of Influence of tree species on continental differences …...Influence of tree species on continental...