Wildfires can result in losses of 15% – 35% of aboveground biomass and 37% – 70% of ground layers 1. The pre-fire spatial variations in the C stocks in ground layers and the variations in the fraction of C consumed during burning both contribute to the uncertainty of C emission estimates 2. Neff et al. 3 believe that the uncertainty is also attributed to the C density of deeper organic layers and the depth to which fires may reach.

It has been reported that soil drainage condition affects fire frequency and severity 45, vegetation recovery 6, and rates of organic matter decomposition 7 in the boreal region where thicker ground layer and higher soil C stocks are usually associated with poorer drained soils 45. In other words, soil drainage conditions directly affect organic C accumulation as indicated by the thickness of the ground layer. On the other hand, soil drainage conditions determine the fractions of the ground layer consumed 8 and total fire emissions 49, even though fire weather conditions and seasonal thawing of soil or permafrost likely play an important role. For example, Neff et al. 3 estimated losses of up to 50% of ground layer C on poorly-drained sites, while Stocks and Kauffman 10 and French et al. 11 reported the losses of 60% to 90% on well-drained sites in Alaska. Because there are 40 – 60% of Alaska land consists of poorly drained soils 12, it is critical to understand how fire-induced C emissions are related to soil drainage classes.

There are a few studies to quantify ground fuels in Alaska 1314, but there has been a lack of a spatially-explicit linkage of firescars to the information about ground layers and soil drainage conditions over the Alaskan boreal forests, while uncertainties associated with burning of ground layers which are commonly found in boreal forests and peatlands may be mainly responsible for the variation in fire-induced C emission estimate. According to Kasischke et al. 15, if only both ground fuel and burned area are considered to be responsible for the total variation in C emissions in the boreal forests, 63% are resulted from the uncertainty in ground fuels and 37% are attributed to the uncertainty in burned area, suggesting that the information about variations in ground fuels is essential for C emission estimation.

The year 2004 was the largest fire year in Alaska since 1950, and an estimate of C emissions for that year will help evaluate the contribution of wildfires in boreal forests to the global C budget. We took the Alaskan Yukon River Basin (YRB) as the study area because there was 83% of the total burned area across Alaska occurred within this area in 2004. The estimation of the C emissions from the wildfires in 2004 was based on the assumption that the level of fire-induced C release under a given fire weather condition depends on fire severity, land cover type, the thickness and C density of the ground layer, while these variables are supposedly associated with soil drainage conditions.

The total C emission from the 2004 wildfires across the YRB is about 6.7 times the annual anthropogenic emissions (12.3 Tg C) from Alaska in 2003 17, or equivalent to 61% of the annual biomass C photosynthesized over the study area if the average NPP was about 266 g Cm^-2 in 2004 18.

Generally, our estimate of C emissions from the 2004 fire year is comparable with those reported for previous wildfire years in Alaska. For example, Kasischke et al.19 reported an average C release of 2.88 kg C m^-2 from 1990/1991 fires. Kasischke and Bruhwiler 20 estimated a range of total C emissions from 1.77 to 3.72 kg C m^-2 burned for the 1998 boreal forest fires across the Western North America by setting three different burn severity class scenarios (low, medium and high) and assuming each class had a different set of fractions of C consumed. Kasischke et al. 21 also simulated a range of 1.17 to 4.22 kg C m^-2 burned from the 1990 boreal forest fires in the central YRB.

The consumed C fractions of both aboveground biomass and the ground layer are very close to the average values reported previously. The C density of pre-fire ground layer expressed as kg C m^-2 per cm depth of the ground layer, our estimate of 0.34 kg C m^-2 per cm depth is comparable to the 0.36 kg C m^-2 per cm depth derived from the data of Kasischke et al. 21, but higher than the 0.21 kg C m^-2 per cm depth reported by Kane et al. 22. As presented in Table 1, 73% of all consumed C was contributed by the combustion of the ground layer. It has been realized that the consumed portion of the ground layer (or reduction in the depth by burning) depends on the burn severity and ground wetness at burning time 42324. Historical fire records indicate that about 20% more burned area was associated with well drained soils in 2004 than in 1990/1991. Again, the consumed fraction of aboveground biomass varies with vegetation type 2125. About 60% of all burned area in 2004 was associated with the shrub-related land cover types. However, the error associated with the Alaska vegetation class dataset could partially contribute to the uncertainty of the estimate due not only to the coarse spatial resolution but also to the quality of remote sensing data and interpretation. Note that the firescars data from the Alaska BLM Wildland Fire Dataset 26 were collected based on the reporting system. Such reporting systems may come up with uncertainties due to inaccuracy in determining perimeters and missing data, which consequently could contribute some to the uncertainty of our estimates.

Another important contributor to the uncertainty of the C release estimate is burn severity. Data are scarce for verifying burn severity of the Alaskan fires in 2004. As a case study, we examined the 1999's firescars in the Yukon-Charley National Preserve. We estimated the C release from the fires in 1999 based on the moderate burn severity scenario. We estimated C release from the 2004 fires again, using the burn severity map (with low, moderate and high classes) to define the consumed C fractions for aboveground biomass and ground organic layers. The burn severity map was generated using the differenced normalized burn ratio (dNBR) approach that is suggested by the U.S. Wildland Fire Leadship Council for the Monitoring Trends in Burn Severity project. No significant difference was observed, although the method using three burn severity classes resulted in a higher estimate than did the method with an assumption that all fires were moderate severity.

According to French et al. 11, large fire years usually lead to a high burn severity; and more burned area associated with well drained soils generally have a higher consumed C fraction of ground layers. Therefore, both together, as characterized by the 2004 wildfires across the YRB, certainly lead to a high C release. In this study we accounted for the contribution of the ground layer to the total C emission by spatially relating individual firescars to the pre-fire land cover types and also defining the consumed C fractions in association with both the soil drainage classes of the firescars and the C stocks in the ground layer. This can partially trade off some uncertainty associated with burn severity quantification. Note that the uncertainty of the estimate might be also attributed to the errors associated with the STATSGO database 16 from which the pre-fire C stocks of the ground layer were derived. Unfortunately, the soil database is the only data source available on the state scale even though its spatial resolution is not fine enough for the presented study. Meanwhile, the STATSGO approach ignored the fact that the upper part of the ground layer has a lighter C density than does the lower part (by about 20% to 30%), implying that the consumed C fraction and the total C emission based on the C density averaged for the ground layer were certainly overestimated. Assuming the C density of the upper part of the ground layer is 20% smaller than that of the lower part, the total C emission would be about 68.7 Tg and the contribution of the ground layer to the total C emission would be reduced to 70%. Such an estimate is close to that (70.5 Tg C) made by Ottmar et al. 27 in which they classified all fuels across the YRB into three FCCS fuelbed types (i.e. birch and aspen, black spruce, and willow/alder shrubland) to run their model CONSUME 3.0.

As indicated in Table 1, the ground layer along with the litter/lichen/moss layer was estimated to contribute as high as 83% to the total C emission (about 6 cm top organic soil layer was destroyed), implying the importance of ground fuel management to the control of wildfires and mitigation of C emissions. More importantly, the consequences and impacts of such predominant contribution can be profound on the boreal ecosystem services and future global change. First, about 77% of all C stock is stored in ground layers and soils in the whole boreal region, and the C emissions from historical wildfires in this region might have been largely underestimated by previous approaches. Second, large wildfires are associated with the boreal forests that have large areas of permafrost 2829. The fire-induced C emissions from the combustion of the ground layer could enhance degradation of permafrost. It has been documented that the fire- and warming-induced collapse of ecosystem structure and permafrost has damaged infrastructure, altered discharge pathways of surface water, disrupted native subsistence cultures, and aided the spread of invasive species 3031. An increase in the frequency and extent of boreal forest wildfires in response to the warming climate will lead to additional deterioration of boreal ecosystem functioning.

To more accurately predict potential C emissions from wildfires, additional pre- and post-fire sampling studies are needed to quantify ground layers over large areas for monitoring changes in both ground layers and permafrost with various wildfire regimes.

The predominant contribution of ground organic layers to the total C emission implies the importance of ground fuel management to the control of wildfires and mitigation of C emissions. The magnitude of the total C emission depends on fire extent, while the C loss in kg C m-2 burned is affected strongly by ground organic layer and soil drainage condition. The significant reduction in the ground layer by large fires may result in profound impacts on boreal ecosystem services with an increase in feedbacks between wildfires and climate change.

The author(s) declare that they have no competing interests.

ZT designed this study, prepared data, analyzed results, and drafted the manuscript. LLT and ZZ supervised this research and made comments on the manuscript formulation. SL gave technical support and critical thoughts on the manuscript finalization. SMH assisted in processing images for burning severity mapping. All authors read and approved the final manuscript.