Multiscale responses and recovery of soils to wildfire in a sagebrush steppe ecosystem

Ecological theory predicts a pulse disturbance results in loss of soil organic carbon and short-term respiration losses that exceed recovery of productivity in many ecosystems. However, fundamental uncertainties remain in our understanding of ecosystem recovery where spatiotemporal variation in structure and function are not adequately represented in conceptual models. Here we show that wildfire in sagebrush shrublands results in multiscale responses that vary with ecosystem properties, landscape position, and their interactions. Consistent with ecological theory, soil pH increased and soil organic carbon (SOC) decreased following fire. In contrast, SOC responses were slope aspect and shrub-microsite dependent, with a larger proportional decrease under previous shrubs on north-facing aspects compared to south-facing ones. In addition, respiratory losses from burned aspects were not significantly different than losses from unburned aspects. We also documented the novel formation of soil inorganic carbon (SIC) with wildfire that differed significantly with aspect and microsite scale. Whereas pH and SIC recovered within 37 months post-fire, SOC stocks remained reduced, especially on north-facing aspects. Spatially, SIC formation was paired with reduced respiration losses, presumably lower partial pressure of carbon dioxide (pCO2), and increased calcium availability, consistent with geochemical models of carbonate formation. Our findings highlight the formation of SIC after fire as a novel short-term sink of carbon in non-forested shrubland ecosystems. Resiliency in sagebrush shrublands may be more complex and integrated across ecosystem to landscape scales than predicted based on current theory.

soil carbon [13][14][15] . Seyfried et al. 19 , for example, showed that the mean annual soil temperature difference (5.2 °C) between a north-and southwest-facing aspects was greater in a 0.25 km 2 catchment than the 4.4 °C difference associated with a 1000 m elevation change across an entire watershed (239 km 2 ). Patton et al. 13 also showed that north-facing aspects contained approximately 3 times more SOC than south-facing aspects. These landscape-level aspect effects may interact with or even trump ecosystem-scale shrub microsite effects to determine recovery trajectories following fire but these effects remain understudied.
Disturbance trajectory models have been proposed for shrublands that incorporate landscape-level properties (i.e. aspect and slope) to assess risk to disturbance as well as predict ecosystem recovery trajectories. In particular, Chambers et al. 20 have posited high elevation and north-facing aspects to be more resistant to cheatgrass and also more resilient to disturbance compared to lower elevation and south-facing aspects. While numerous studies have focused on plant community responses to fire and generally confirmed lower resistance to cheatgrass invasion at lower elevations and on warmer south-facing aspects compared to north-facing ones 21,22 , comprehensive tests of aspect-driven recovery of soil properties and processes following fire remain lacking. Most studies have focused on ecosystem microsite (plant and interplant) responses within a year of a fire. In a meta-analysis of 39 studies from the Great Basin, USA, Sankey et al. 23 showed no significant effect of fire on most soil properties, including SOC, suggesting soils in this region are relatively resilient to fire. However, there was no apparent consideration of landscape controls on fire influencing these microsite effects such that questions remain about transferability of these findings to complex terrain. One of the few experimental studies examining prescribed fire effects on C fluxes in a more mesic shrubland ecosystem documented rapid recovery of gross ecosystem production (GEP) and net carbon uptake as measured by net ecosystem exchange, supporting the hypothesis that higher elevation, more mesic sites are more resilient to fire disturbance 24 . Vegetation surveys and remote sensing supported rapid recovery of primary productivity associated with growth of the grass community immediately after fire. However, Fellows et al. 24 also showed little to marginal responses in terms of ecosystem respiration following prescribed fire. This apparent lack of respiratory pulse after disturbance appears to be at odds with ecological theory. This finding indicates a need to revisit and test these models in non-forested ecosystems and possibly incorporate interactions across scales to adequately represent them.
Here we used a state factor approach to evaluate the response and recovery of sagebrush shrublands to a wildfire event across a range of scales at the Reynolds Creek Experimental Watershed and Critical Zone Observatory (RCEW-CZO) in southwestern Idaho. Specifically, we selected two catchments, one that had been burned in the Soda Fire in August 2015 and one that had remained unburned, holding other state factors such as granite parent material, climate, topography and vegetation relatively constant. In each of these catchments, we measured soil surface (to 2 cm depth) properties and processes at four paired plant-interplant locations on north-and south-facing aspects at 2, 3, 6, and 37 months after fire (Fig. 2b). Aboveground biomass was largely eliminated by the fire (Fig. 2a). We hypothesized that soil pH would increase, SOC would decrease, and pyrogenic carbon (PyC) would increase following fire disturbance, but that these patterns would be aspect, microsite, and time Predicted plant and soil carbon pools and processes following disturbance. Immediately after disturbance, loss of soil organic carbon and plant material are predicted (above panel). Prior to disturbance, net primary productivity (NPP) is in balance with hetetrophic respiration (R H ) such that NEP equals zero. A pulse in respiration (R H ) (bottom panel) is predicted after disturbance such that NPP < R H and NEP is negative (source of carbon to the environment) and eventually the ecosystem comes back into quasi-steady state (modified from Gorham et al. 2 20 , we expected north-facing aspects with higher SOC to be more resilient to fire and lose proportionally less carbon than south-facing ones and recover faster following fire. We also expected under plant microsites to have higher SOC than interplant ones and therefore be more resilient to fire. Finally, we expected a marginal respiration pulse immediately following fire, similar to Fellows et al. 24 .

Results
Altered soil properties. Consistent with expectations, soil pH increased significantly, SOC decreased, and PyC increased following fire, but the burn effects often varied with aspect and/or microsite (Fig. 3). In the case of soil pH, only a main burn effect was observed with average soil pH increasing by an order of magnitude on burned sites immediately after fire compared to controls (burn: F 1,12 374, p < 0.0001), 7.08 ± 0.11 compared to 5.67 ± 0.17 (Supplementary Table S1). No microsite effect was observed for soil pH based on repeated measures multivariate analysis of variance (RMANOVA) test on paired plant-interplant differences (p = 0.562) (see Statistical Analyses in Methods for more detail). In addition, there was no significant difference in soil pH between north-and south-facing aspects (p = 0.214) and no burn-aspect interaction (p = 0.8140). Soil pH decreased significantly with time in the burn treatments (F 3,10 22.81, p < 0.001) and returned to control levels under previous plant canopies within 37 months. Burned interplant spaces remained significantly elevated after 37 months, regardless of aspect (p < 0.05) (Fig. 3a). SOC decreased significantly following fire, but the patterns were aspect and microsite dependent. In contrast to pH, a significant microsite effect was identified based on a RMANOVA performed on paired plant-interplant differences (F 1,12 6.34, p = 0.027) (Fig. 3b, Supplementary Table S1). Furthermore, RMANOVA tests on plantspecific SOC stocks showed that average SOC stocks under previous plants on the burned aspects were significantly lower compared to unburned controls (F 1,12 25.4, p = 0.003), with a significant aspect (F 1,12 9.46, p = 0.0096) but no significant burn-aspect, time, and interaction effects (Fig. 3b). Indeed, SOC stocks under previous plants on burned north-facing aspects were reduced by half 2 months after fire, 0.42 ± 0.011 kg C m −2 on north burn compared to 0.96 ± 0.018 kg C m −2 on north control. This contrasted with burned south-facing aspects where SOC stocks were reduced but were not significantly different than south controls, 0.34 ± 0.077 compared to 0.448 ± 0.010 kg C m −2 (p < 0.05). For interplant microsites, SOC decreased significantly with burn, aspect, and its interaction (burn: F 1,12 11.29, p = 0.005, aspect: 27.19, p = 0.0002, burn x aspect: 5.08, p = 0.04, and time: 5.04, p = 0.02) (Fig. 3b). Thirty-seven months post-fire, SOC stocks under previous plants remained significantly lower on burned compared to control (p < 0.05), especially on north-facing aspects, whereas SOC stocks in interplant spaces on both burned aspects were not significantly different from controls. Emergent scale dependent soil properties. While SOC stocks decreased more on north-than southfacing aspects after fire, the fraction of SOC as PyC increased significantly under previous plant canopies, most notably on south-facing aspects (burn: F 1,12 201.28, p < 0.0001, aspect: 17.37, p = 0.0013, burn × aspect, 6.55, p = 0.025) (Fig. 3c). Post-hoc Tukey tests showed that mean PyC fraction on burned south-facing aspects was significantly higher than on burned north-facing aspects, 0.36 ± 0.012 compared to 0.30 ± 0.006, which were  In addition to PyC increases, we found that soil inorganic carbon (SIC) formed in largely carbonate-free soils either during or immediately after fire (Fig. 3d). Specifically, SIC increased significantly and preferentially under previous plants compared to interplant spaces, and it differed significantly with aspect, with significantly more SIC on north-than south-facing aspects (burn: F 1,12 31.61, p < 0.0001, aspect: 24.51, p = 0.003, burn-aspect, 16.08, p = 0.0017). 2 months after fire, SIC stocks in plant microsites were as high as 4.41 ± 0.90 g C m −2 and 1.4 ± 0.54 g C m −2 on burned north-and south-facing aspects, respectively. SIC was only marginally significantly elevated in burned interplant microsites (burn: F 1,12 4.38, p < 0.058, aspect: 13.3, p = 0.003, burn-aspect, 0.0003, p = 0.98) (Fig. 3d). SIC in plant microsites decreased significantly with time since fire and returned to background levels within 37 months (F 3,10 time: 17.92, p = 0.003, time x burn: F 3,10 21.17, p = 0.0001, time × aspect: F 3,10 7.63 p = 0.006, time × burn × aspect: F 3,10 5.67, p = 0.015). SIC appeared to form above a pH threshold of 6.5 (Fig. 4a). Post-fire carbonates (with enough mass to be run for isotopes, n = 6) revealed that the carbonate isotopes were markedly depleted in both δ 13 C and δ 18 O values compared to RCEW-CZO pedogenic soil carbonates formed at depth (Fig. 4b). Post -fire δ 13 C-carbonate values were correlated with SOC-δ 13 C values (Fig. 4c). to capture CO 2 effluxes along with controlled laboratory incubations to evaluate potential C mineralization.
All approaches generally showed reduced and aspect-dependent CO 2 effluxes immediately after fire. Consistent with expectations, rates of potential C mineralization were significantly reduced on burn aspects relative to controls immediately after fire, with marginal aspect and no microsite effects (F 1,12 burn: 12.25 p = 0.004, aspect: 4.42, p = 0.057) (Fig. 5, Supplementary Table 1). Carbon mineralization rates varied significantly with time (F 3,10 20.43, p = 0.0001) and returned to control levels within 37 months (Fig. 5a). Cumulative C mineralization losses on burned north and south aspects were significantly lower than north controls but not the south controls (p < 0.05) (Fig. 5b). In line with these observations, continuous field CO 2 fluxes measured by forced diffusion (FD) chambers (Eosense, Dartmouth, NS, CAN) showed fluxes from burn aspects were similar or lower (and even negative) compared to controls immediately following fire (within a month of fire containment) (Fig. 5c). A respiratory pulse was observed from all sites after the first rains in October, with larger total R H losses from the burned north-facing aspect than its control, although losses were comparable to the south control (Fig. 5d). Average total R H losses from the burn aspects were reduced but not significantly different than controls (Fig. 5e).

Discussion
We found significant changes in soil properties and processes following fire, some of which varied with landscapeand ecosystem-scale and their cross-scale interactions. Others such as PyC and SIC were emergent properties following fire. Some of these properties and processes recovered and returned to ambient conditions within 37 months, while others did not (Fig. 6). We describe these changes in scale dependent structure and function and explore possible mechanisms explaining observed SIC formation.
Changes in scale dependent structures. Consistent with our expectations, we found that soil pH increased significantly following fire. In contrast to our expectations, the burn effects on pH were not shrub microsite or aspect scale dependent (Figs. 3a, 6). Meanwhile, SOC pools were significantly affected by burn, aspect, microsite, and their interaction. These findings contrast with a meta-analysis conducted by Sankey et al. 23 showing no significant shrub microsite effect on SOC pools. We also found that emergent properties, PyC and SIC, were strongly aspect and microsite-scale dependent (Fig. 3c). Differences in fire severity and/or lower initial stocks of SOC on the south aspects may explain the higher fraction of PyC formed under previous plants on the south-facing aspects. PyC also persisted in plant sites after 37 months suggesting a structural shift in the SOC composition. While Santin et al. 4 have incorporated PyC into landscape recovery dynamics, our findings point to contrasts in PyC accumulation at the plant-interplant scale that as of yet, have received little to no attention in models of nonforested systems. Finally, in contrast to PyC, SIC formed preferentially on north-facing aspects under former shrubs. To our knowledge, this scale dependent formation of SIC structure after fire has not been quantified, and we expand on this more below. These conflicting results showing microsite effects might in part be explained by the fact that landscape position was considered in this study. In addition, sampling methodology differences (timing of sampling, depth of sampling) could help to explain differences in microsite responses to fire as the ability to detect change in soils collected 0-15 cm was likely lower compared to shallow surface soils (0-2.5 cm) in this study. These results raise questions about previously assumed microsite responses and highlight the need to consider landscape-level processes. Together, these findings showing ecosystem to landscape scale dependent responses in ecosystem structure indicate that theoretical disturbance models need to be revisited to incorporate these multiscale responses.
Recovery time following disturbance. Based on Chambers et al. 20 , we expected north-facing aspects with higher SOC to be more resilient and recover faster following fire. While soil pH mostly recovered on both www.nature.com/scientificreports/ aspects and microsites within 37 months of fire, SOC stocks were slower to recover under previous plants, especially on the north-facing aspect (Fig. 3b). This slower recovery on north-facing aspects may in part reflect preferential wind erosion and redistribution of surficial organic matter and fine fraction on north-facing slopes into swales and then subsequent loss of this material via water erosion 25 . It is also likely that reduction of soil cohesion after fire would have preferentially affected the north-facing aspect because it has finer textured soils 16 and steeper gradients than the south-facing slopes 17,18 . Our results provide little support that soils on north-facing aspects are more resilient to disturbance relative to soils on south-facing aspects as predicted by Chambers et al. 20 .

Marginal respiration pulse.
A large R H pulse was not observed following fire (Fig. 1). Consistent with Fellows et al. 24 , we found lower rates of respiration immediately after fire in both incubation and field studies.
Rates of potential C mineralization were significantly reduced on both burn aspects, with marginal aspect and no microsite effects. Rates returned to ambient levels within 37 months indicating the more soluble carbon pools were able to recover, while other pools were not. Field R H was more variable (Fig. 5d), but average total R H after fire was comparable to unburned R H (Fig. 5e). Interestingly, R H was low and even negative on the north-facing Possible mechanisms contributing to carbonate formation. We found preferential and ephemeral formation of soil carbonates under previous plant canopies on burn north-facing compared to south-facing aspects and no carbonate formation in interplant microsites. Previous studies at the RCEW-CZO have shown that pedogenic carbonate formation is restricted to sites with annual precipitation less than 500 mm 13,27,28 , yet these catchments were above this threshold (550-600 mm) and little to no carbonate was observed previously 13 . Further examination of the controls on carbonate formation showed that carbonates increased nonlinearly as a function of soil pH (Fig. 4a) and that post-fire carbonate isotopes were markedly depleted in both δ 13 C and δ 18 O compared to pedogenic soil carbonates formed at depth at the RCEW-CZO (Fig. 4b). Furthermore, the formation of SIC under plants compared to the marginal formation in interplant spaces provided clues as to the possible mechanisms allowing for the formation of carbonates under some post burn conditions. Equilibrium theory indicates that carbonates can form when solution becomes supersaturated with calcite 29 where Possible mechanisms explaining carbonate formation include a decrease in partial pressure of CO 2 (pCO 2 ) in the soil or an increase in the relative concentrations and activities of dissolved calcium [a(Ca +2 )] and bicarbonate ([a(HCO 3 − )] 2 , possibly due to decreased water content where such that K cal is the equilibrium constant for calcite. The relative distribution of dissolved inorganic carbon species is mainly controlled by pH and pCO 2 where the concentration ratio [CO 3 2-]/[CO 2 ] is a function of the [H + ]. Thus, higher pH following fire as observed in Figs. 3 and 4 apparently favored the formation of HCO 3 and carbonates as SIC (Fig. 4). Calcium previously stored in plant biomass would have also been made available as ash in the soil and promoted formation of carbonates under canopies compared to interplant spaces (Figs. 2, 4a, b). Finally, a reduction in carbon mineralization or respiration and decrease pCO 2 post fire, both of which were observed in lab and field, could further promote the formation of carbonate.
Carbonate isotopes also provided a clue about their origin. δ 13 C-carbonate values of burned soils were similar to the δ 13 C of CO 2 of biogenic origin whereas δ 18 O values resembled groundwater or cold-season precipitation inputs 30 . It is typical that soil carbonate formation results in a systematic enrichment of 13.5-16.5 δ 13 C relative to the δ 13 C of respired CO 2 31,32 , both as a function of phase transformation from CO 2 to CaCO 3 , and preferential 12 CO 2 diffusion. δ 13 C values of pedogenic carbonates derived solely from respired CO 2 would range between − 13.6 and − 7.6 ‰, or with an equal mixture with atmospheric CO 2 closer to − 4.0 ‰. Thus, the unexpected depletion of post-fire carbonate-δ 13 C values represents a potential unaccounted for carbon fractionation mechanism post fire. Other mechanisms explaining the formation of carbonates includes formation from combustion of calcium oxalates in stems, leaves, and roots under lower temperatures < 500 °C and the hydration of calcium oxide abiotically to calcium carbonate or bicarbonate 33 . However, these mechanisms would not induce depletion of carbonate-δ 13 C values (Fig. 4b). Similarly, complete consumption of the SOC and conversion of all the (1) CaCO 3 + CO 2 + H 2 O = Ca 2+ + 2 HCO − 3 .
(2) K cal = a Ca +2 a HCO − A possible explanation for such carbonate isotope values is an extracellular oxidative metabolism (EXOMET) 34 . EXOMET is C mineralization devoid of active biological metabolism via soil-stabilized enzymes or soil mineral, metal, and PyC catalysts, and known to induce a strong carbon isotope fractionation in evolved CO 2 (≤ 50 ‰) for several months. Evidence for EXOMET activity in burn soil contributing to carbonate formation immediately after fire includes the correlation (r 2 = 0.9) between δ 13 C-SOC and depletion of carbonate-δ 13 C values (Fig. 4c). Our results collectively indicate that a combination of conditions, high fire temperature under shrubs, increases in soil pH, a reduction in respiration and decrease pCO 2 , and increase in the availability of calcium from ash were likely mechanisms resulting in the formation of carbonates and that EXOMET activity might have catalyzed this formation. Together, these findings indicate a previously overlooked, short-term sink or storage of carbon as SIC on the landscape, followed by an additional medium-term C loss pathway. Further experimental study of SIC formation after fire beyond these field observations is merited to understand the formation and fate of carbonates. The scope of our findings is limited given that our samples only represented shallow soil depths and two watersheds in the RCEW-CZO in Idaho, though we note that Goforth et al. 35 documented effervescence after a wildfire in the San Gabriel Mountains suggesting the formation of carbonates elsewhere after fire. We anticipate that the controls on the proposed pathway of post-fire SIC formation to further depend on fire intensity, soil climate, and physical and chemical properties.
Other loss pathways not considered in this study but documented in other studies such as wind and water erosional losses may also need consideration in dryland disturbance models. In particular, large losses as sediment and associated particular organic carbon were documented in the two years following the fire, a total of 1.17 and 0.0625 kg m −2 sediment and carbon, respectively 36 . These losses were an order of magnitude higher than unburned catchments 36 . Vega et al. 25 further showed that the fire response was a two-step process with wind erosion loading leeward (north) aspects preferentially in swales, and these swales were emptied to streams by water erosion 36 . These results again underscore the importance of incorporating multi-scale domains into disturbance models in these regions.
In conclusion, our findings confirm that these non-forested ecosystems do not respond with a large respiratory pulse immediately after disturbance as found in non-forested ecosystems; rather they appear to form short-term carbonate structures preferentially under shrubs. In contrast to our expectations and others, we found little evidence of north-facing aspects being more resilient than south-facing ones as north facing aspects lost proportionally more SOC than south facing ones, and both aspects did not recover back to ambient SOC levels. While fire significantly reduced mineralizable C, these processes recovered within 37 months. PyC under burned plants appears to be a persistent shift in the structural component of SOC that merits more study, particularly in regards to ecosystem resilience. We conclude that non-forested systems have multi-scale structure and functions, and conditions during and following fire result in cross-scale interactions that amplify some structures but not others that are not represented in current theoretical disturbance models.

Study site. This study was conducted at the Reynolds Creek Experimental Watershed and Critical Zone
Observatory (RCEW-CZO), a semiarid catchment representative of Intermountain West United States 37 . Located in southwestern Idaho, the RCEW (239 km 2 ) extends over a steep climatic gradient (mean annual precipitation 250-1100 mm year −1 , mean annual temperature 5.5-11 °C) (Fig. 2). The environmental variability is driven by the nearly 1000 m elevation range and contains variable geology as described below. Rain is the dominant form of precipitation in the RCEW, with snow dominating in the highest elevations. Corresponding vegetation types include Wyoming big sagebrush steppe (Artemesia tridentata wyomingensis) in the lower elevations, transitioning to mountain sagebrush (Artemesia tridentata ssp. vaseyana), western juniper (Juniperus occidentalis), aspen (Populus tremuloides)) and Douglas Fir (Pseudotsuga menziesii) at higher elevations. The geology of the Reynolds Creek basin consists of Neogene volcanic and sedimentary rocks overlying Cretaceous granitic basement 38 .
Starting August 10, 2015, the Soda Fire burned 113,000 hectares (279,000 acres) of sagebrush steppe habitat in southwestern Idaho and was contained by August 22, 2015. The fire severity was rated predominately moderate severity (yellow) in the areas of study with pockets of high severity (red) (USGS image source, Landsat 7) 39 . Approximately 29% or 68 km 2 of the RCEW-CZO was burned. Given the focus of the RCEW-CZO on improving prediction and understanding of soil carbon at the pedon to landscape scale 37 , the Soda Fire provided a unique opportunity to examine how fire alters soil carbon stocks and its recovery.

Experimental design. Two catchments, Babbington
Creek and Mack's Creek, were selected for study (Fig. 2). Mack's Creek was located within the burn perimeter of the Soda Fire, and Babbington Creek was selected as the control due to its similarity to Mack's Creek (Fig. 2). Specifically, catchments were selected such that state factors (i.e. lithology, vegetation, topography, and climate) were held relatively constant 40 . Both catchments were underlain by granodiorite lithology 38 and experienced a mean annual precipitation of 640 mm and a mean annual temperature of 6.6 °C based on available climate data 41 . Elevation locations were selected to match climate. Catchments have predominantly north-and south-facing aspects (Fig. 2) with Wyoming Big Sagebrush (Artemisia tridentata ssp. wyomingensis) as the dominant plant species on both aspects representing 50-75% and bitterbrush (Purshia stansburyana), grasses, and western juniper (Juniperus occidentalis) making up the remaining major vegetation.
Sites were established on the north-and south-facing aspect slopes using a wheel-and-axle sampling design. Each aspect contained four shrub sites with paired interplant space (microsite). Soil sample collection occurred 2, 3, and 6, and 37 months after the fire. Surface samples were collected from 0 to 2.5 cm depth. Others at 2.5-5, www.nature.com/scientificreports/ 5-10, and 10-20 cm were also collected but not the focus of this study. The fine fraction (< 2 mm) was separated from the coarse fraction (> 2 mm) to provide percent coarse fraction by weight. Soils were analyzed for pH, SOC, pyrogenic carbon (PyC), and SIC.
Soil methods. Soil pH. Soil pH was determined following standard methods 42 . Briefly, we determine soil pH on 1:1 soil to 18.2 Mohm water slurry paste after 1 h to allow equilibration. The pH meter was calibrated using the pH 4, 7, and 10 buffers and a check calibrant was checked every 10 samples and accepted within 0.02 units.
SOC and stable isotopes. SOC and isotopes were determined on an Elemental Analysis-Isotope Ratio Mass Spectrometry (EA-IRMS) at Idaho State University's Center for Archaeology, Materials and Applied Spectroscopy (CAMAS) laboratory. Soils were pretreated with 5% v/v hydrochloric acid (HCl) to remove soil inorganic carbon 43 , ball-milled, and dried prior to packing. Carbon stocks (kg m −2 ) were estimated from bulk density using Patton et al. 44 adjusting for measured differences in coarse fraction 44 .
Pyrogenic carbon (PyC). Pyrogenic carbon (PyC) was estimated using a chemometric model relating Fourier-transform mid-infrared spectra of the soil samples to lab-based estimates of PyrC content as described in McGuire et al. 45 . PyC are compounds produced from incomplete combustion and include soot, charcoal, biochar, black carbon, elemental carbon 46 . The training set consisted of 287 samples from Australia 47 and 99 samples from the United States 48 that underwent the same carbon fractionation procedure 47 which isolates particulate, mineral-associated and pyrogenic carbon fractions using a combination of physical size fractionation and solidstate 13 C nuclear magnetic resonance (NMR) spectroscopy. Finely milled subsamples of the same samples along with all of the samples in this present study were scanned on a Bruker Vertex 70 FTIR with a Pike Autodiff diffuse reflectance accessory. Spectra were acquired in the range of 6000-180 cm −1 at a 4 cm −1 resolution. A partial least squares regression model built on standard normal variate transformed spectra could explain 86% of the variation in observed PyC data with a root mean square error of 0.32 g kg −145 .
Soil inorganic carbon (SIC). Soil inorganic carbon (SIC) was measured initially using effervescence tests with a 1 N HCl drops. The expectation was that there would be little to no effervescence given the lack of it prior to the fire. However, tests revealed moderate to high effervescence, which warranted further testing on the samples in order to quantify inorganic carbon by modified pressure-calcimeter method 49 . In brief, 1 g soil or standard was added to a 20 mL or 100 mL serum vial and 2 mL of 6 M hydrochloric acid (HCl) with 3% by weight ferrous chloride tetrahydrate chloride (FeCl 2 •4H 2 O) was added in a dram and sealed with a Wheaton rubber stoppers, (no. 224100-192) and aluminum cap and crimper. The bottles were agitated to release the acid and allowed to sit overnight before measurement on a pressure transducer (Setra model 280, 0-15 PSIG, 0.03-5.03 VDC output). Six standards containing incrementally increasing calcium carbonate (CaCO 3 ) in a sand mesh matrix. CaCO 3 was converted to SIC using the following Finally, a select set of samples were run for carbonate isotopes and compared to samples analyzed at a select set of sites.
Carbonate isotopes. Isotopic analysis of SIC pools was done separately from analysis of SIC pools and involved removing organic material from air-dried soils by adding 10 mL of 5% sodium hypochlorite to each sample and agitating for 24 h 50 . Following digestion, each sample was centrifuged for 5 min at 3600 rpm to settle clay particles, and then decanted. To remove trace amounts of sodium hypochlorite, samples were rinsed with 10 mL of deionized water, and the centrifuging-decanting process repeated twice. Soil samples were then freeze-dried prior to analysis at CAMAS. Samples were analyzed on a GasBench II. Analysis consisted of purging 12 mL screw top Exetainer™ sample vials using helium and subsequent acidification with 0.2 mL of 19.3 M phosphoric acid (H 3 PO 4 ), volatilizing SIC as CO 2 . The evolved CO 2 was routed through a ThermoScientific Delta Advantage stable isotope ratio mass spectrometer, as above. The ratio of 13 C to 12 C is reported as δ 13 C using standard delta notation. Isotopic ratios from SOC pools were used to constrain source pool for SIC-δ 13 C values. Carbon values determined by calcite volatilization were compared to molar C content determined by the calcimeter method to assess losses of calcite to SOC removal treatments; change in percent SIC-C was not significant between methods.
Field measured carbon dioxide efflux. We monitored field carbon dioxide efflux using four automated forced diffusion (FD) chambers (Eosense, Dartmouth, NS, CAN) that were deployed within one month of the fire on north and south facing aspects on burned and unburned catchments. The chambers operate similarly to surface chambers, but require no flow or flushing of atmospheric air. Instead, they measured purely diffusive transport by monitoring concentration gradients across membranes of known permeability and can monitor CO 2 efflux every 15 min 51 . Given gaps in data, we summed fluxes where all four chambers had data present during the study period. Owing to major challenges with access (16 km roundtrip) and maintaining chamber power and stability on burned hillslopes in winter, we removed the chambers.
Potential soil carbon mineralization. We compared these field fluxes to potential rates of carbon mineralization on air-dried soils following Fierer and Schimel 52 . While a less conventional approach, air-dried soils provided www.nature.com/scientificreports/ the opportunity to control for environmental conditions and moisture content and evaluate the potential or relative rates of carbon mineralization following fire. Carbon mineralization rates were measured regularly once a day over a 4 day period. In brief, we added 20 g of fine-fraction (< 2 mm) soil into 235 mL mason jars and brought the soils up to 60% water filled pore space, based on prior determination of water holding capacity 53 . Jars were sealed with a rubber septa in the lid at the start of the incubation, and 5 mL gas sample was drawn at time zero (t 0 ) and after 4 h (t f ), and injected into a EMG 4 (PP Systems, Amesbury, MA) infrared gas analyzer (IRGA). Temperature and barometric pressure were also monitored. Between the daily 4-h snap shot incubations, lids were removed from the jars, jars were vented to the atmosphere, jars were covered with saran wrap to allow to exchange gas with atmosphere but minimize water loss and sealed with rubber bands, and stored in the dark. Carbon mineralization (ug C g dry soil −1 h 1 ) was calculated from the difference in concentrations and divided by the g dry soil and incubation time (4 h). We determine the average rate of carbon mineralization based on the mean of the 4 daily snap shot incubations. We converted these values g C m −2 h −1 using Patton et al. 44 as described above. We also calculated cumulative carbon mineralized by multiplying the hourly rates by 24 and summing the total losses (ug g dry soil −1 ) and converted these an areal basis (g C m −2 ).
Statistical analyses. Given the balanced, repeated, paired plant-interplant design, we first tested for influences of burn (B), aspect (A), and its interaction (B*A) on plant-interspace properties by conducting a Repeated measures multivariate analysis of variance (RMANOVA) in JMP Pro 16 (Carey, NJ) on the difference between the pairs. If no significant differences were detected, we pooled the plant and interplant data and averaged it and then performed a RMANOVA, to retain correct degrees of freedom. If there were significant effects based on plant-interplant differences, we conducted a RMANOVA on separate plant or interplant properties.