Boreal Forest Wildfire and Climate Linked Drivers of Carbon and Nitrogen Loss

The boreal landscape covers large portions of the earth’s land area and stores a significant percentage of its terrestrial carbon (C). Increased emissions due to climate change amplified fire frequency, size and intensity threaten to remove elements such as C and nitrogen (N) from forest soil and vegetation at rates faster than they accumulate. This may result in large areas within the region becoming a net source of greenhouse gases creating a positive feedback loop with a changing climate. Estimates of per area fire emissions are regionally limited and knowledge of their relation to climate and ecosystem properties 5 is sparse. This study sampled 50 separate Swedish wildfires from 2018 providing quantitative estimates of C and N loss due to fire along a climate gradient. Mean annual precipitation had strong positive effects on total fuel, which was the strongest driver for increasing C and N losses, while mean annual temperature (MAT) had greater influence on both preand postfire fuel bulk and chemical properties which had mixed effects on C and N losses. Significant fire induced loss of C occurred in the 50 plots comparable to estimates in similar Eurasian forests but approximately a quarter of those found in typical North American 10 boreal wildfires. N loss was insignificant though large proportions were collected from lower soil layers to a surface layer of char in proportion to increased MAT. These results reveal the variability of C and N losses between global regions and across local climate conditions and a need to better incorporate these factors into models to improve estimates of global emissions of C and N due to fire in future climate scenarios. Additionally, this study demonstrated the linkage between climate and the chemical transformation of residual soil fuel and discusses its potential for altering C and N dynamics in postfire recovery. 15


Introduction
Worldwide, boreal forests account for a net C sink into plants and soil of 0.31 ± 0.19 Pg of C per year, equivalent to 27.3 ± 16.7% of the planet's terrestrial C sink (Tagesson et al., 2020). This sink plays a pivotal role in the greenhouse gas content of the atmosphere (Lemprière et al., 2013). Low temperatures and waterlogged soil conditions slow decomposition 20 of centuries of litter additions resulting in the build up of thick layers of soil organic material where the majority of C is stored (Malhi et al., 1999;Rapalee et al., 1998). The balance of C transfer between atmospheric and terrestrial stocks on the by poorly constrained free parameters such as total fuel load (French et al., 2004;Soja et al., 2004). This field sampling has been regionally limited and biased towards a few high intensity burn complexes in North America which may in turn bias global emission estimates (van Leeuwen et al., 2014). For example, the Eurasian boreal region is dominated by relatively fire resistant 60 vegetation that promotes lower intensity fire (Rogers et al., 2015;de Groot et al., 2013a) and C loss (0.88 kgC/m 2 ) (Ivanova et al., 2011) than that in typical (Walker et al., 2020) North American wildfire (3.3 kgC/m 2 ) (Boby et al., 2010). Though Eurasia contains over 70% of the boreal global land area (de Groot et al., 2013a), and about 50% (20 Mha/yr) of its yearly burnt area (Rogers et al., 2015), wildfire emissions from this region are severely under sampled in the field (van Leeuwen et al., 2014). Additionally, estimates of N loss from boreal wildfire are rare in all regions despite its well recognized role as a 65 limiting nutrient and evidence of its immediate removal in percentages similar to C (Boby et al., 2010). Lastly, boreal wildfire research has often focused on individual or small groups of fires located relatively near to each other, with little information about the representativeness of the observations or context of the results within the broader spectrum of fire impacts across the wider region, especially those relating to variation in climate. This knowledge gap has thus far been addressed with conglomerated studies spanning different fire seasons, ecosystem types and research methodologies (Walker et al., 2020;Gaboriau 70 et al., 2020). Therefore, widely replicated, simultaneous and systematic field measurements of fire processes in under sampled regions with particular attention to climate are needed to derive more robust, generalizable conclusions about boreal forest responses to wildfire.
This study sampled 50 separate fire complexes spanning broad gradients of mean annual temperature (MAT) and precipitation (MAP) which ignited in Sweden during summer 2018 (Fig. 1). Analysis intended to distinguish the effects of climate 75 on fire induced changes in C and N stocks with direct, fine scale measurements and little loss of generality thereby providing insight into both local processes and valuable, globally comparable data from an under sampled region. Space-for-time substitution (De Frenne et al., 2013) along with a control-impact design provided insight into the possible future conditions of Fennoscandian forests in a changing climate and fire regime. Specifically, it was hypothesized that: 1. Fire significantly reduced and spatially rearranged both C and N stocks. 80 2. Loss of C and N stocks and their transfer to charred material were related to prefire total fuel amount, composition and arrangement amongst forest compartments.
3. A direct relation between climate variables and fire induced C and N stock changes exists and is mediated by long term ecosystem properties as well as time-of-fire processes which are represented by the extent of charring of residual fuel.
2 Materials and methods 85 2.1 Experimental design and field site selection 50 burnt plots were selected from a pool of 325 fires identified during the summer 2018 period which were mapped from remotely sensed data and provided by the Swedish Forest Agency (Skogsstyrelsen). Each 20 × 20 m 2 plot was located within distinct burn scars (greater than 2 km separation) to reduce potential for pseudoreplication or spatial autocorrelation (Bataineh et al., 2006) and allow for increased spread across the climate gradients (Schweiger et al., 2016). Remote sensed data was 90 taken as the average pixel value within a 20 m diameter circle centered on the plot with GIS analysis utilizing QGIS (QGIS Development Team, 2019), ArcGIS (Esri Inc., 2019) and the pandas Python 3 package (Wes McKinney, 2010). The first constraints on site selection were to avoid wetland or steeply sloping areas using prefire, topo-edaphic derived soil moisture data (TEM) provided by the Swedish Environmental Protection Agency (Naturvårdsverket) (Naturvårdsverket, 2018) and elevation and slope data provided within the ArcGIS software environment. This restricted the study to non-wetland ecosystem 95 types, which tend to have markedly differing ecosystem functioning than relatively dry forested regions, and to retain focus on climate driven effects and their space-for-time substitution by reducing the effects of exogenous variables such as topography on models. Next, sites with postfire salvage logging were omitted using recent visual imagery along with Swedish Forest Agency records. Burnt plot selection was then filtered to maximize the spread over latitude, MAT and MAP gradients. Sentinel-2 infrared imagery taken during the time of fire assisted in delineating the exact locations of burnt plots by placing them where 100 there had been a strong and consistent infrared signal that was well within the mapped final fire boundaries. This gave greater certainty of strong development of sustained fire. Manual assessment of visual and infrared imagery was performed through the brandkarta web application provided by the Swedish Forest Agency.
Prefire properties of each burnt plot were estimated by measurements from a single identically sized adjacent control plot centered between approximately 15 and 150 m outside the fire boundaries (100 plots total, i.e. 50 plot pairs). To reduce 105 mismatch between control and prefire burnt plot properties the following restrictions were placed on control plot selection.
Control plot locations were selected to minimize elevation, slope and TEM differences from the adjacent burnt plot. Swedish Forest Agency data regarding tree species, overstory biomass, and basal area (collected during 2014) were used to best match properties of control and burnt plot pairs. Stand appearance and age were examined with historic, visual images provided by the Swedish National Land Survey (Lantmäteriet) verifying time since last disturbance had been at least 30 years for plot pairs 110 and that stand structure between plot pairs appeared physically connected over this period.
Due to their documented effects on emissions, long and short term approximations of moisture were introduced as exogenous variables to models in order to test the ability of the study design to isolate variation in C and N stock losses to the effects of climate. Long term moisture was represented by the TEM used in plot selection while short term moisture balance used the Standardized Precipitation-Evapotranspiration Index (SPEI) over the period January to June 2018 (i.e. spei06 2018-06) to 115 capture the desiccation process leading up to the fire season. This variable was also compared to summer 2018 anomalies in temperature and precipitation, i.e. the difference in the 2018 June, July, and August average of these values from those during the same months averaged over the period from 1961 to 2017.

Sampling
Site visits occurred approximately 1 year postfire over the dates August 5 to August 20 in 2019. Sampling and analysis were 120 broken into six forest compartments. These were the four soil layers of mineral, duff, moss/litter and char as well as the two aboveground compartments of the understory and overstory vegetation. The organic layer was considered the grouping of the duff, moss/litter and char layers. Each compartment was further sorted by weight into characteristic features to form compartment compositional variables (CCVs) which were used in regression to test for relationships between compartment composition and the quantity and quality of fuel loading as well as C and N loss. Samples were acquired for all four soil layers. Four mineral soil samples were taken using a 3 cm diameter corer at four 135 corners of a square each 15 m from the plot center. Where feasible, 10 cm vertical mineral cores were taken, however in shallower layers a minimum depth of 5 cm each was collected. Duff samples were taken near the mineral cores by excavating four soil volumes (at least 400 cm 3 each) and trimming the mineral and moss/litter layers off the bottom and top of the volumes respectively. Duff and mineral soils were kept frozen until portions were freeze dried for separate analysis. Moss/litter samples were collected at approximately equal intervals along the soil profile transects in a 553 cm 3 steel container with attention to 140 preservation of the natural in situ volume. Char was similarly collected in a 112 cm 3 container. On the upper surface of the char layer were small portions of dry, unburnt material, much of which may be new additions of litter to the forest floor. This material was discarded from the char collection and was not included in stock estimates.

Vegetation
Individual tree bole diameter (sampled at 130 cm height above the forest floor) and species were recorded on site for all trees 145 of at least 5 cm diameter at measurement height. If a fallen tree was charred only on its lower (in standing orientation) portions it was deemed standing during fire ignition and its measurements were included if its base was within plot boundaries. In burnt plots, the percentage of brown and black of each tree canopy was visually estimated and approximated as 0%, 25%, 50%, 75%, or 100%. Overstory biomass was calculated by entering bole diameters into allometric equations for Scots pine (Pinus sylvestris), Norway spruce (Picea abies), silver birch (Betula pendula), and downy birch (Betula pubescens) (Marklund, 1987).

150
The equations provided CCVs derived from biomass of stem wood, stem bark, living branches, dead branches, and stump for all species. Roots ≥ 5 cm diameter, roots < 5 cm diameter and needle biomass were additionally provided for pine and spruce.
When testing the influence of overstory vegetation on C and N loss bole diameters from the burnt plots were used and not the adjacent control. C and N stock estimates for overstory were not included in analysis and its measurements were only used to assess its role in controlling C and N stocks in all other compartments. In all plots understory was clearly distinguished from overstory by pronounced height differences and samples were taken from control plots by cutting all non-moss, non-tree plant material at the surface of the soil from within four 40 × 40 cm 2 patches. Patches were chosen for their representativeness of plant abundance and composition for the portion of the plot that was vegetated, which was always all non bare rock surface. These values were applied to a visual estimate of non bare rock surface area of the burnt plots as an approximation of its prefire understory coverage. CCVs for understory were determined 160 by sorting the sampled understory plant material and measuring dried weights of the functional groups graminoid, forb, shrub, and pteridophyte.

Sample processing
All samples were dried at 40 • C for at least 3 days. Mineral and duff samples were sieved to 2 mm and 4 mm respectively.
CCVs for these two layers were formed by weights of these coarse and fine fractions. Dry moss/litter samples were weighed 165 and visual estimates for percentage volume of needles, broad leaves, woody material, moss and lichen were multiplied by total weight to form CCVs. Bulk density of each soil layer per plot was calculated as the total dry weight of its samples divided by their total volume on collection. All samples were pulverized, except the mineral soil where only the fine earth fraction (< 2 mm) was analyzed (C and N content was set to 0 for the coarse fraction), and run through a Costech ECS4010 elemental analyzer to produce ratios of C and N weight to sample total weight (C R and N R respectively). Duff and mineral 170 layer elemental weight ratios were recalculated by the sum of C or N in each of their fine and coarse fractions and divided by total compartment weight. The C:N ratio for each ecosystem compartment was calculated by dividing its total weight of C by total weight N.

Data analysis
Data was stored in comma-separated value files with minimal redundancy. Calculations were performed with custom written 175 Python 3 code using the pandas library. The measurable properties used in stock calculation within soil compartments are the depth, bulk density and C R or N R . Total C and N stocks per soil compartment were calculated as a product of these properties using the equation where subscript Z is substituted with C or N for reference to C or N stocks, d is the soil layer depth in meters, ρ is the layer 180 bulk density (kg m −3 ) and Z R is C R or N R . Understory compartment stock calculations were performed with the equation where m is the sampled mass in kilograms, A is sampled area (m 2 ), Z R is C R or N R , and F is the estimated fractional vegetation coverage of the 20 × 20 m 2 plot.
Changes between control and burnt plots were first calculated by subtracting control plot values of a variable from those of 185 its burnt pair thereby forming a single distribution of 50 elements for statistical testing. When C and N stocks were described as losses their distribution was negated. These distributions were approximated as normal and unless otherwise noted all confidence intervals were constructed at the 95% level using the formula wherex is the sample mean, z is always 1.96 for the 95% interval, σ its standard deviation and n the sample size (always 50).

190
Significance of differences between control and burnt plots was deemed to be when their interval did not include zero.
Simple regression was performed using the stats.linregress method from SciPy (Virtanen et al., 2020) providing significance (p), correlation (r), and slope (b). Multiple regression was carried out with the OLS method in the Python 3 statsmodels package (Seabold and Perktold, 2010) with models evaluated in order of increasing Akaike information criterion. Standardized regression coefficients (β) were produced by normalizing all variables (converting to z scores) before regression. CCVs were 195 assessed in regression models both using original variables and their principal components produced by the PCA method in statsmodels. The effects of C and N stock arrangement amongst forest compartments was tested by entering the per plot ratios of the sums of different combinations of compartment stocks into regression analyses. Here, variables are considered correlated when p values from simple regression are less than 0.05.

200
Bar plot confidence intervals of the mean were bootstrapped at 95% and n = 10,000 using the seaborn.barplot method. Additional Python 3 packages that assisted in data exploration were NumPy (Harris et al., 2020) and pingouin (Vallat, 2018).

C and N stock losses and rearrangement
Averaging across all sites sampled, fire caused significant rearrangement of C and N stocks particularly through changes in soil 205 depth and bulk density which increased the mass per volume of both C and N. Significant decreases in total C stocks in burnt plots from their paired controls were observed, but changes in N were insignificant. Fire clearly transferred large amounts of C and N from lower soil layers to the highly nitrogenous surface layer of char. Organic layer C R was unaffected by fire, however strong changes in N R were measured resulting in an overall significant increase in the C:N ratio of this layer. Mean values and confidence intervals of the changes from control to burnt plots highlighted in this section for all compartments are found in 210 Table 1 with soil specific properties found in Table 2.
The largest total loss of C in burnt plot compartments due to fire was from the duff layer (Fig. 2a). About three quarters of the moss/litter C was removed from burnt plots, comprising about half as much as the total amount of C that was removed from the duff layer. Understory C removal due to fire was near complete but had a relatively small contribution to overall elemental stocks and their changes. Of the average amount of C lost from these three compartments, 54.3% was found in the averaged 215 char layer and only 0.19% in the increased C found in burnt plot mineral layers which themselves had no significant overall change in C between control and burnt plots.
Fire rearranged N significantly despite having no overall effect on its total amount (Fig. 2b). Similar percentages of N as C were lost from the duff, moss/litter and understory compartments with 100.8% of their averaged amount of lost N found in the char layer and 5.1% in increased burnt plot mineral layer which itself had no significant change in total N.

220
Disproportionate changes of C and N from control to burnt plots caused significant decreases in the C:N ratio in all compartments except the duff layer which was unchanged (Fig. 2c). The low C:N ratio in the char layer (29.78 ± 1.70) made a strong contribution to the overall reduction in this value in burnt plot organic layers compared to control.
The duff layer C R and N R did not change significantly, though the moss/litter layer showed a significant increase in both values in burnt plots compared to their paired controls (Fig. 3c, 3d). These layers together with the char layer C R (0.498 ± 0.0190) 225 and N R (0.0173 ± 0.00108) resulted in the organic layer having no change in C R but a significant increase in N R in burnt plots. The mineral layer experienced significant decreases of C R and N R in burnt plots both overall and in the fine fraction C R (-0.0210 ± 0.0145, -33.7%) and N R (-0.000509 ± 0.000415, -24.1%) compared to the controls.
Fire had a strong effect on reducing soil layer depths with removal of nearly the entire moss/litter layer and about one third of the duff thickness (Fig. 3a). Together with the formation of the char layer and insignificant mineral layer depth changes, fire 230 removed about a quarter of total soil depth and nearly 40% of the organic layer depth in burnt plots. Fire induced increases in bulk density of the soil layers counteracted C and N loss due to these depth changes (Fig. 3b). Bulk density of both the duff and moss/litter layers increased significantly and, along with producing a dense char layer, fire had a strong densifying effect on the organic layer.
To quantify the relative contribution of fire induced changes in organic layer depth, bulk density and elemental weight ra-235 tios on organic layer C and N losses they were linearly combined and entered into multiple regression. The C loss regression produced a model of fit of R 2 = 0.865 and standardized regression coefficients for changes in depth (β = -0.670), bulk density (β = -0.633) and C R (β = -0.583). N loss produced a model fit of R 2 = 0.777 and coefficients for loss of depth (β = -0.599), bulk density (β = -0.398) and N R (β = -0.382). This shows that changes of these variables due to fire all had a strong effect on stock loss estimates. Measured change in organic layer depth is the strongest determinant of losses of N. However, for C bulk 240 density and elemental weight ratios are nearly as important as depth.

Forest level drivers of fire induced C and N loss
The strongest correlator to total C and N losses among long term ecosystem properties was total paired control plot C (p < 0.001, r = 0.703, b = 0.744) and N stocks (p < 0.001, r = 0.585, b = 0.574), respectively. An even stronger correlation was found between control plot organic layer C and N stocks (here abbreviated C O and N O ) and estimated losses of C (p < 0.001, r = 0.736, 245 b = 0.762) and N (p < 0.001, r = 0.653, b = 0.665) from this compartment. Due to this increased explanatory power, and because the majority of fire affected stocks were located there, the focus of analysis was placed on the organic layer. Variables used in regression with percentage changes in C and N stocks tended to have less explanatory power than total stock losses and also were sensitive to outliers having erratic changes in model fit with removal of data points. Therefore only total stock losses were assessed. were deemed to be confounding or lacked supporting causal mechanism and were at a high risk of omitted-variable bias. were correlated with MAT but not the total mass of prefire fuel. This means that warmer regions produced larger amounts of lower C R material irrespective of total fuel amount. In a multiple regression using C O , the organic layer C:N ratio, MAT, MAP and total char layer C production to explain C O loss, direct effects of MAT lost significance and overall model fit was improved (Fig. 5a). This suggested that, while controlling for C O and the organic layer C:N ratio, C O loss from this layer was reduced by MAT through the creation of char. Similarly, N O loss is further explained by additions of char layer N to the climate model but a large direct effect of MAT remains (Fig. 5b). The organic layer C:N ratio in the N model was able to replace the direct effect of MAT, however with decreased model fit and inflation of variables which is suggestive of a confounding influence of the organic layer C:N ratio on MAT and N O loss. Again, CCVs and fuel arrangement could not improve either model.

Trends in ecosystem C and N stocks
Significant overall reduction in C stocks were found in burnt plots relative to their paired control, with the largest removals from the duff layer. Averaged total C loss was relatively low at 0.815 ± 0.652 kgC/m 2 (15.6%) compared to estimates from inland Alaskan black spruce stands (3.3 kgC/m 2 ) (Boby et al., 2010) but was comparable to averaged losses from Scots pine stands in Siberia (0.992 kgC/m 2 ) (Ivanova et al., 2011). However, N stocks were not significantly different overall in burnt 310 plots compared to controls. This contrasts with the Alaskan study which estimated percentage removal from soils of N (49.8%) to be similar to C (52.9%) at an average loss of 0.09 kgN/m 2 (Boby et al., 2010). Averaging over the 50 burnt plots, N was clearly removed in large amounts from the duff and moss/litter layers but its transfer into a highly nitrogenous char layer prevented differences when considering the overall soil profile. The char layer was likely largely produced by fire interacting with the understory and moss/litter layer, however averaged char layer C and N stocks were greater than losses from the two 315 10 https://doi.org/10.5194/bg-2021-178 Preprint. Discussion started: 9 July 2021 c Author(s) 2021. CC BY 4.0 License.
layers combined suggesting large contributions also from the duff layer. In burnt plots with residual moss/litter an upwards mixing of mobilized duff C and N may have occurred due to simultaneous effects of heating throughout the depth of the fuel bed. Because the char layer was conglomerated and completely blackened it is unlikely that material was incorporated postfire.
However, material may have been added from downward movement of overstory components during the time of fire. The above mentioned study in Alaskan black spruce forests, which are known for their great extent of canopy damage (Walker et al., 2020;Boby et al., 2010), showed C and N loss from the canopy to be about an order of magnitude lower than losses from soil while also assuming that losses from the tree bole are negligible and that a large fraction of these overstory losses were released to the atmosphere (Boby et al., 2010). In the current study, low levels of overstory damage and its lack of correlation with char layer mass suggests that the large majority of C and N stock changes between control and burnt plots were captured within the sampled soil and understory compartments.

325
The lack of change in total N stock due to fire is consistent with available evidence from existing study in Fennoscandian forests where fire had only slight effects on total N over extended periods (Palviainen et al., 2017). N losses in non-boreal forests have been related to fuel temperature during time of fire with lower intensity fires transferring a greater proportion of pools of organic N into soil ammonium and nitrate rather than removing N in gaseous forms (Neary et al., 1999). Laboratory studies have linked the amount of N transferred from organic to inorganic forms during heat exposure to both applied temperature 330 and fuel type (Gundale and DeLuca, 2006;Makoto et al., 2011). Therefore, the N cycle in boreal systems may be highly dependent on active fire properties, fuel type and resulting fuel transformation and the greater N losses in Alaska compared to Eurasia could be explained by its dissimilar fuel and the characteristically more intense fires across the North American boreal zone (de Groot et al., 2013a;Wooster and Zhang, 2004). Fire intensity and temperature may also be related to C losses and overall fuel transformation, so it is of interest to compare remote measurements (i.e. satellite data) of these time-of-fire 335 properties to on-site measured ecosystem changes. This can lead to a more complete predictive understanding of wildfire in the entire Fennoscandian region and beyond.
In addition to removals, C and N was densified by fire in the organic layer due to significant drops in depth and increased bulk density in burnt plots. Mean C R dropped in duff between plot pairs driven most likely by the increased ratio of incombustible inorganic material to remaining organic material. However, increased variability of duff C R in burnt plots contributed to 340 the insignificance of this change and appeared to be related to extreme volume reductions which reduced the duff layer to exceptionally ashy, rocky material in some plots. C R and the C:N ratio in the char layer measured here are surprisingly low compared to known measurements showing char to be highly carbonaceous with C:N ratios often well above 100 (Hart and Luckai, 2014). Fire induced structural change of fuel has been shown to play a strong role in N retention where highly porous char material adsorbs inorganic N preventing its leaching loss from the system until its reuptake into organic forms by 345 plants or microbes (Makoto et al., 2012). This sorptive power has been observed to fade over the interval between fire events suggesting newly produced char is required for this retention effect (Zackrisson et al., 1996). The high nitrogen content of the char layer may therefore be due to adsorption of fire mineralized N and act as a steady source of bioavailable nutrients to plant and microbial communities during succession. This study employed coarse-scale sampling of char based on soil horizon identification and separation in the field and a more rigorously defined assessment of char production in all layers may provide 350 more detailed relations to climate and soil processes.

Climate linked effects of fire
Both MAP and MAT had significant direct relations to total C and N removals from plots with the strongest mediator being estimated prefire C and N stocks in the organic layer. MAP had a stronger effect on the build up of control plot fuel, namely through a positive correlation with total organic layer depth. MAT affected C and N losses through increasing bulk density, 355 reducing C R and increasing N R thus reducing the C:N ratio in the organic layer in control plots, suggesting warmer conditions had a fuel conditioning effect through greater decomposition of organic soils (Callesen et al., 2007). When controlling for control plot organic layer C and N stocks and their ratios using multiple regression, MAT had a direct negative effect on C and N losses from this layer. This direct effect was largely mediated by incorporation of measures of fire induced fuel transformation into the models, i.e. production of char layer C or N. These models suggest that warmer regions tended to conserve larger pools 360 of fire affected fuel as charred material rather than release it from the ecosystem either in gaseous or dissolved forms. This fuel transformation in turn may have extended effects on C and N turnover through links to nutrient availability and the biotic community which in turn affect process rates such as primary production and soil decomposition (Schmidt and Noack, 2000).

Considering representativeness and prediction of future wildfire impacts
Ignition probability and fire propagability may relate to the analyzed drivers in this study with ignition and propagation more 365 likely in stands with greater and more flammable fuel loads. As a result, C and N losses might have been underestimated by burnt plots being biased to a greater prefire fuel load than their paired controls (systematic error) rather than these differences being approximately random (random error). This was evident in the fact that N losses were centered near 0 by their mean yet had a strong correlation with control plot total N despite the improbability that fire actually increased total N within nearly half of the burnt plot sample pool. Contrarily, control plots were biased towards higher TEM, which was observed to be related to 370 greater fuel loading in measured control plots, though the magnitude of this bias did not increase along gradients of climate or fuel loading nor did attempts to correct for it significantly affect C and N loss estimates. Therefore, control plot matching appears to have been performed adequately within the scope of data collected with any potential bias coming from unknown parameters. Further investigation of these parameters is merited in order to improve control plot matching methodology and better constrain emission estimates in this region.

375
Variation of many ecosystem properties could not be statistically linked to C and N losses, though that does not mean they are without role in determining extent of emissions and ecosystem change. Cross regional boreal wildfire study has shown variables such as the abundance of spruce to strongly affect C emissions across regions, but within a single region the abundance may be too homogeneous to show an effect (Walker et al., 2020). Furthermore, ratios of fine to heavy fuel loads have been manipulated in experimental burns to produce varying fire severity (Alexander et al., 2018;Ludwig et al., 2018). Accordingly, CCVs and 380 fuel arrangement amongst compartments in this study may have been simply too homogeneous to produce significant results but may nevertheless still provide valuable statistical signals for understanding drivers of fire processes across regions and fire severities.
Previous studies have demonstrated strong effects of moisture on boreal wildfire C emissions. For example, a study including several large North American fire complexes found C emissions to increase along gradients of topo-edaphic derived soil 385 moisture (due to its positive relation to total fuel) until reaching high moisture sites where the trend inverted and began to decrease due to the inhibiting effects of this increased moisture on fuel availability (Walker et al., 2018). The position of this point of inflection along the long term moisture curve is likely dictated by short term moisture levels, which are in turn controlled by the extremity of drying during a fire season. Accordingly, fuel availability, and therefore C emissions, are strongly dependent on drying processes specific to individual stand composition and structure and its local fire weather. By incorporating Boreal wildfire literature has tended to focus on highly flammable forest ecosystems with intense burning of several hundred hectares or more despite the vast majority of fires in the boreal region being less than 200 ha (Stocks et al., 2002;Valendik, 1996). This research bias may limit knowledge to a particular population by studying fire events only at their utmost extremes 400 and effectively masking important signals of ecosystem heterogeneity on fire severity at differing intensities. The only restriction this study placed on wildfire intensity and size was that the fire activity could be remotely confirmed using Sentinel-2 infrared data. This has the potential to affect comparability of results both in terms of total C and N losses as well as their drivers which may exhibit differing patterns and strengths over ranges of fire intensity. It may therefore be beneficial for wildfire studies to be examined and compared within categories of absolute severity and intensity. This methodology might be 405 particularly useful in gaining understanding of the drivers of fire severity as it increases in traditionally fire protected ecosystems such as wetlands (Zoltai et al., 1998), which tend to store vastly larger amounts of C per area than their dryer forested counterparts in the boreal region (Deluca and Boisvenue, 2012).

Conclusions
This study measured wildfire impacts across current climatic gradients of precipitation and temperature to show that climate 410 controls total releases of C and N during fire events mainly due to its effect on increasing organic layer fuel load. The role of MAP is focused on the total quantity of this fuel load whereas MAT has a more qualitative effect by influencing bulk density, C R and N R in the organic layer. When controlling for total organic layer fuel, increasing MAT, and to a lesser extent MAP, reduces C loss due to fire through preconditioning of the organic layer as measured by a lowered C:N ratio. Additionally, both C and 13 https://doi.org/10.5194/bg-2021-178 Preprint. Discussion started: 9 July 2021 c Author(s) 2021. CC BY 4.0 License. N losses are mitigated by increased MAT through the sequestering of fire affected fuel into a surface layer of charred material.

415
This conservation effect is stronger for N, which had no overall significant loss in stocks due to fire, and which also had stronger unexplained direct mitigating effects of MAT on its loss that were hypothesized to be related to time-of-fire properties such as fire intensity and temperature. While remaining ecosystem variables regarding fuel composition and arrangement could not be strongly linked to total C and N losses it is of interest to analyze their role in cross-regional comparisons and to investigate whether they influence other fire related properties such as ignition likelihood, fire propagability and intensity. Advancing 420 knowledge of the intricate ties between instantaneous processes of fire events and their long term effects on C and N cycling demands comprehensive research approaches that pay particular attention to climate sensitivity. This knowledge is imperative for producing accurate predictions of boreal forest functioning under future climate scenarios.
Data availability. All data used to produce the results in this document are original unless otherwise stated in the text. It is found freely available through Eckdahl et al. (2021).