Biomass offsets little or none of permafrost carbon release from soils, streams, and wildfire: an expert assessment

The United Nations has set a target of limiting warming to 2 °C above pre-industrial temperatures to mitigate risk of the most damaging consequences of climate change (UNEP 2013). Maintaining global climate within this target depends on understanding ecosystem feedbacks to climate change so that adequate limits on human emissions can be set. As high latitudes warm, more of the large permafrost carbon pool will be exposed to decomposition, combustion, and hydrologic export (Harden et al 2012, Schuur et al 2015). Up to 220 Petagrams (Pg) carbon could be released from permafrost-region soil by 2100, and 500 Pg by 2300 (MacDougall et al 2012, Schuur et al 2013), representing 10%–30% of greenhouse gas emissions required to push the global climate system beyond the 2 °C target (Schaefer et al 2014). Models project that some permafrost carbon release will be offset by increases in Arctic and boreal primary productivity due to extended growing season, CO₂ fertilization, and nutrient release from decomposing soil organic matter. However, many processes and dynamics known to influence biomass accumulation, such as ecosystem disturbance and nutrient limitation, are incompletely represented or absent in current models (Qian et al 2010, Koven et al 2011, Schaefer et al 2011, Koven et al 2015b). Likewise, only a few models projecting future permafrost carbon release consider wildfire emissions, and none include hydrologic carbon flux (Qian et al 2010, Koven et al 2011, Schaefer et al 2011, MacDougall et al 2012, Schaefer et al 2014), though past hydrologic flux has been simulated (McGuire et al 2010, Laudon et al 2012, Kicklighter et al 2013). Despite clear policy implications of this climate feedback, considerable uncertainty of both carbon inputs and outputs limits our ability to model carbon balance of the permafrost region. To bring to bear the best available quantitative and qualitative scientific information (Joly et al 2010) on this climate feedback, we present results from expert assessment surveys indicating that there is little consensus on the magnitude and even sign of change in high-latitude biomass, whereas most researchers expect fire emissions and hydrologic organic carbon flux to substantially increase by the end of the century.

When data are sparse but management decisions are pressing, expert judgements have long been used to constrain possible system response and risk of dangerous or undesired outcomes (Zickfeld et al 2010, Morgan 2014). There are multiple methods for collecting and combining expert opinion including formal expert elicitation interviews, interactive software, and surveys (Aspinall 2010, Javeline et al 2013, Morgan 2014). While expert assessment cannot definitively answer questions of future system response, it complements modeling and empirical approaches by allowing the synthesis of formal and informal system information and by identifying research priorities (figure 1; Sutherland et al 2013, Morgan 2014). The approach is similar to the concept of ensemble models where multiple estimates built on different assumptions and data provide a more robust estimate and measure of variance. Because the experimental unit is an individual researcher, each data point represents an integration of quantitative knowledge from modeling, field, and laboratory studies as well as qualitative information based on professional opinion and personal experience with the system. Expert assessment has been used in risk assessment and forecasting of natural disasters, human impacts on ecosystems, and tipping points in the climate system (Halpern et al 2008, Lenton et al 2008, Aspinall 2010). In a data-limited environment such as the permafrost region, expert assessment allows formal consideration of a range of factors known to affect carbon balance but insufficiently quantified for inclusion in models. For permafrost carbon balance, these factors include nutrient dynamics, nonlinear shifts in vegetation community, human disturbance, land–water interactions, and the relationship of permafrost degradation with water balance.

Zoom In Zoom Out Reset image size

Because precise empirical or model-based assessments of the critical factors driving permafrost-region carbon balance are unlikely in the near future (Harden et al 2012), we collected estimates of the components of net ecosystem carbon balance from 98 permafrost-region experts (table 1). We had two major goals: (1) Assess current understanding of the timing and magnitude of non-soil biomass accumulation, hydrologic organic carbon flux, and wildfire carbon emissions, and (2) Identify major sources of uncertainty in high-latitude carbon balance to inform future research.

In the fall of 2013 we administered three expert assessments to address knowledge gaps concerning the response of permafrost-region biomass, wildfire, and hydrologic carbon flux to climate change. Development of assessment methodology began in early 2009 as a part of the Dangerous Climate Change Assessment Project administered by the University of Oxford. We iteratively revised questions, response format, and background information based on four rounds of input from participants, including at the Vulnerability of Permafrost Carbon Research Coordination Network meeting in Seattle 2011 (Schuur et al 2013). To help survey participants consider all of the evidence available from field and modeling studies, we distributed a system summary document for each questionnaire including regional and pan-Arctic estimates of current carbon pools and fluxes, a brief treatment of historical trends, and a summary of model projections where available (table 2; supplementary information questionnaires and system summaries).

Participants were selected based on contribution to peer-reviewed literature or referrals from other experts and had experience in all major boreal and Arctic regions (table 1). We identified potential participants by querying Thomas Reuters Web of Science (webofknowledge.com) with applicable search terms (e.g. Arctic, boreal, biomass, dissolved organic carbon, fire, permafrost). To reach researchers with applicable expertise who were underrepresented in the literature, we supplemented the list with personal referrals from lead experts and all participants. In total 256 experts were invited to participate. We distributed the surveys and system summaries via email with a two-week deadline. After sending out three reminders and accepting responses for three months after initial invitation, we received 115 responses from 98 experts (38% response rate), with 15 experts participating in more than one survey (supplementary information list of experts). Experts who provided estimates and input to this paper are coauthors.

Experts provided quantitative estimates of change in biomass, hydrologic flux, or wildfire for three time points (2040, 2100, and 2300), and four regional warming scenarios based on representative concentration pathway (RCP) scenarios from the IPCC Fifth Assessment Report (Moss et al 2010). Warming scenarios ranged from cessation of human emissions before 2100 (RCP2.6) to sustained human emissions (RCP8.5) and corresponded to permafrost-region mean annual warming of 2 °C–7.5 °C by 2100. All surveys were driven by the same scenarios of high-latitude warming generated from RCP2.6, 4.5, 6.0, and 8.5 with the National Center for Atmospheric Research's Community Climate System Model 4 (Lawrence et al 2012). For the purposes of this survey, warming was assumed to stabilize at 2100 levels for all scenarios so that estimates for the 2300 time point would account for lags in ecosystem responses to climate drivers. While climate scenarios were defined by temperature, we asked experts to consider all accompanying direct climate effects (e.g. temperature, precipitation, and atmospheric CO₂) and indirect effects (e.g. vegetation shifts, permafrost degradation, invasive species, and disturbance). Experts were encouraged to consider all available formal and informal information when generating their estimates including published and unpublished modeled and empirical data as well as professional judgment. Participants listed the major sources of uncertainty in their estimates, self-rated their confidence and expertise for each question, described rationale for their estimates, and provided background information (table 1 and S1).

The biomass survey consisted of a single question asking for cumulative change in tundra and boreal non-soil biomass including above and belowground living biomass, standing deadwood, and litter. The wildfire survey asked for estimates of change in wildfire extent and CO₂ emissions for the boreal and tundra regions to assess changes in both fire extent and severity. The hydrologic flux survey asked for estimates of dissolved and particulate organic carbon (DOC and POC, respectively) delivery to freshwater ecosystems in the pan-Arctic watershed and delivery to the Arctic Ocean and surrounding seas via riverine flux and coastal erosion, allowing the calculation of losses during transport due to burial or mineralization. Dissolved inorganic carbon fluxes were not included in this survey.

The original questionnaires in 2009 asked for participants to estimate subjective 95% confidence intervals of the whole system response (e.g. total change in high-latitude biomass). Based on expert input during subsequent testing we disaggregated the system into different components to encourage detailed consideration of possibly competing dynamics (e.g. asking for separate estimates of boreal forest and Arctic tundra response; Morgan 2014). This resulted in a large response table for each question (72–102 quantitative estimates), which we found caused respondent fatigue and decreased the number of experts willing to participate. As a compromise, we asked respondents to provide a single best estimate and indicate confidence with a five-point scale (table S1). While analysis of best estimates can return narrower uncertainty ranges than subjective probability distributions (Morgan 2014), we believe this tradeoff resulted in broader expert participation, better representing diversity of opinion across disciplines and compensating for possible underestimation of variability and uncertainty.

We calculated basic summary statistics, using median values to estimate center and interquartile ranges (IQR) to estimate spread. To calculate the portion of permafrost carbon release offset by biomass accumulation, we combined estimates from this study with reanalyzed data from Schuur et al (2013). The low IQR for carbon release offset by biomass growth was calculated by dividing the low IQR of uptake by the upper IQR of carbon release and conversely for the high IQR (figure 3). All analyses were performed in R 3.0.2. The complete dataset of quantitative estimates and comments from survey participants stripped of personal identifiers is available at www.aoncadis.org/dataset/Permafrost_carbon_balance_survey.html.

Expert estimates revealed diverging views on the response of boreal biomass to warming, with over a third of estimates predicting a decrease or no change in boreal biomass across scenarios and time periods (figure 2). While median change in boreal biomass was similar across warming scenarios for each time step (3%, 9%, and 11% increases by 2040, 2100, and 2300, respectively; figure 2 and S1), variability was much higher for warmer scenarios. Consequently, all of the IQR of change in boreal biomass for RCP6.0 and RCP8.5 included zero. Experts projecting a decrease in boreal biomass attributed their estimates primarily to water-stress and disturbance such as fire and permafrost degradation. In contrast, there was general agreement that tundra biomass would respond positively to warming, with end-of-century increases of 6%–30% projected for RCP2.6 and 10%–90% for RCP8.5. Because of these contrasting responses to increased warming, tundra accounted for 40% of total biomass gain by 2300 for RCP8.5, though it currently constitutes less than 10% of total permafrost region biomass (based on median values in figures 2, 3(a) and table 2). Estimates of boreal biomass were generally symmetrically distributed while tundra biomass estimates were right-skewed, and most datasets had 1–4 estimates beyond 1.5 times the interquartile range (figure S2). Self-rated confidence was higher for tundra than for boreal forest, but was below 3 (moderately confident) in both cases (table S1), highlighting considerable uncertainty of individual estimates in addition to variability among respondents.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Experts projected major shifts in both fire and hydrologic carbon regimes, with up to a 75% increase of riverine organic carbon flux to the ocean and a four-fold increase in fire emissions by 2100 for RCP8.5 based on IQR (figure 2 and S1). Fire and hydrologic carbon release estimates peaked at 2100, followed by a 10%–40% decrease by 2300. In contrast to biomass, the response of both fire-driven and hydrologic carbon flux varied strongly by warming scenario, with RCP8.5 resulting in 2–6 times more carbon release than RCP2.6. While the boreal forest dominated total wildfire emissions, the relative change in tundra fire emissions was 1.5- and 2-fold greater than the relative boreal response for 2100 and 2300, respectively (figure S1). Increases in fire emissions were attributed to changes in fire extent rather than severity, which varied less than 5% among scenarios and time periods. Though dissolved organic carbon (DOC) represented the majority of total hydrologic organic carbon release, experts projected higher relative increases for coastal POC, with end-of-the-century increases of 6%–50% for RCP2.6 and 13%–190% for RCP8.5. There was a lack of consensus on the response of DOC delivery to the ocean, with 21% of estimates predicting a decrease or no change. Experts predicting a decrease attributed their estimates to increased mineralization, changes in hydrologic flowpath, and changes in DOC photo- and bio-lability (Cory et al 2014, Abbott et al 2014). Responses indicated no change in the proportion of organic carbon mineralized or trapped in sediment before reaching the ocean, with 63%–69% of DOC and 68%–74% of POC lost in transport. Fire and hydrologic carbon flux estimates were strongly right-skewed with a few experts projecting extreme change well beyond 1.5 times the interquartile range for each timestep and warming scenario combination (figures S3 and S4). Average self-rated confidence was between 2 and 3 for all questions except tundra fire emissions which had average confidence of 2.0 and 1.7 (table S1).

Along with quantitative estimates of carbon balance, experts identified sources of uncertainty currently limiting the prediction of system response to climate change (table 3). Water balance, including precipitation, soil moisture, runoff, infiltration, and discharge, was the most frequently mentioned source of uncertainty for both biomass and hydrologic organic carbon flux, and the second most mentioned for wildfire. Many experts noted that water balance is as or more important than temperature in controlling future carbon balance, yet projections of water balance are less well constrained (Zhang et al 2013, Bintanja and Selten 2014). Almost three-quarters of wildfire experts identified the future distribution of vegetation as the primary source of uncertainty in projecting wildfire, noting strong differences in flammability between different boreal and tundra species. Permafrost degradation was identified as an important source of uncertainty for biomass, hydrologic flux, and wildfire, due to both disturbance from ground collapse (thermokarst) and interactions with water-table dynamics and surface soil moisture as deeper thaw affects soil drainage.

Arctic tundra and boreal forest have accumulated a vast pool of organic carbon, twice as large as the atmospheric carbon pool and three times as large as the carbon contained by all living things (Hugelius et al 2014, Schuur et al 2015). Over the past several decades, the permafrost region has removed an average of 500 Tg carbon yr⁻¹ from the atmosphere (McGuire et al 2009, Pan et al 2011, Hayes et al 2011). Combining our estimates of biomass uptake with a recent projection of permafrost soil carbon release (Schuur et al 2013) suggests that the permafrost region will become a carbon source to the atmosphere by 2100 for all warming scenarios (figure 3(b)). Experts predicted that boreal and Arctic biomass could respond more quickly to warming than soil carbon release, offsetting −33% to 200% of mid-century emissions from permafrost-region soil (figure 3(b)). However, because estimates of change in biomass are similar across warming scenarios but permafrost carbon release is strongly temperature-sensitive, the emissions gap widens for warmer scenarios, resulting in five-times more net carbon release under RCP8.5 than RCP2.6. This suggests that 65 to 85% of permafrost carbon release could be avoided if human emissions are actively reduced—i.e. if emissions follow RCP2.6 instead of RCP8.5 (figure 4).

Zoom In Zoom Out Reset image size

Model projections of future boreal and Arctic biomass agree in sign but vary widely in magnitude, with increases of 9–61 Pg carbon projected by 2100 (Qian et al 2010, Koven et al 2011, Schaefer et al 2011, Falloon et al 2012). While some of these models fall within the range estimated here of −20 to 28 Pg carbon by 2100, none include zero or negative change in biomass as predicted by over a third of participants in our expert assessment. Two potential reasons for this disagreement are an overestimation of the effect of CO₂ fertilization or an underestimation of the role of disturbance in some models. Firstly, CO₂ fertilization exerts a larger effect on carbon balance than all other climate effects in many models (Balshi et al 2009), with up to 88 Pg carbon difference between model runs with and without CO₂ fertilization effects for some models (Koven et al 2011). However, there is little field evidence that CO₂ fertilization results in long-term biomass accumulation in tundra and boreal ecosystems (Hickler et al 2008, Gedalof and Berg 2010, Peñuelas et al 2011). Additionally, many models with large CO₂ effects do not include other limiting factors, such as nutrients and water, known to interact with CO₂ fertilization (Hyvonen et al 2007, Thornton et al 2007, Yarie and Van Cleve 2010, Maaroufi et al 2015, Koven et al 2015a). Secondly, models that do not account for disturbance such as wildfire, permafrost collapse, insect damage, and human resource extraction likely overestimate the positive response of biomass to climate change (Kurz et al 2008, Abbott and Jones 2015, Hewitt et al 2015).

Considering the scenario of a complete biome shift is useful in evaluating both model projections of change and estimates from our expert assessment. If all boreal forest became temperate forest, living biomass would increase by 27%, resulting in the uptake of 16 Pg carbon based on average carbon densities from both ecosystems (Pan et al 2011). However, 22 Pg carbon would be lost due to decreases in dead wood and litter, resulting in a net circumboreal loss of 6 Pg carbon. If all tundra became boreal forest, non-soil biomass would increase by 205% (Saugier et al 2001, Epstein et al 2012, Raynolds et al 2012), taking up 17 Pg carbon. This scenario may not represent the upper limit of possible carbon uptake if other unforeseen shifts in C allocation take place; however, it highlights the relatively modest carbon gains probable on century timescales.

While regional projections from models of boreal wildfire vary in sign and magnitude (supplementary information system summaries), most models agree that at the circumboreal scale, fire emissions will increase several-fold, with increases of 200%–560% projected by the end of the century (Flannigan et al 2009, Kloster et al 2012). IQR from our study are somewhat lower (40% to 300%, median 170%), but participant confidence in these estimates was low, suggesting considerable uncertainty in the future response of boreal fire. The 60%–480% increase in tundra fire projected by our study would represent an even larger ecological shift than experienced by the boreal forest, with implications for regional biomass, habitat, and carbon balance, though there are few models that project changes in tundra fire (Rupp et al 2000) and none at a circumarctic scale (Mack et al 2011).

The production of Arctic DOC and POC depends on abundance of carbon sources in terrestrial ecosystems (influenced by biomass, wildfire, temperature, and permafrost degradation) and the ability of hydrologic flow to transport that carbon (determined by factors such as precipitation, runoff, depth of flow through soil, and coastal erosion; Guo et al 2007, Kicklighter et al 2013, Abbott et al 2015, Larouche et al 2015). Due to these complexities and others, there are currently no quantitative projections of future DOC and POC flux from the circumarctic. However, estimates from our study suggest a substantial departure from historical rates of change. For RCP8.5, hydrologic organic carbon loading would increase 4–20 times faster in the 21st century than it did in the 20th (Kicklighter et al 2013), representing a nonlinear response to high-latitude warming. The lack of consensus on the response of DOC, the largest component of hydrologic organic carbon flux, highlights the importance of developing and testing conceptual frameworks to be incorporated into models (Laudon et al 2012).

An alternative explanation for differences between expert estimates and modeled projections is the possibility of bias in the group of experts. Participants in our assessment tended to have more field than modeling experience (table 1) and may have therefore been skeptical of simulated ecosystem responses that have not been observed in the field such as CO₂ fertilization and rapid migration of treeline (McGuire et al 2009). Because future dynamics cannot always reliably be predicted on the basis of past system behavior, this bias may or may not result in overly conservative estimates. Furthermore, because experts are likely to base projections on the study areas with which they are most familiar, regional differences could be a source of bias. Fundamental differences among regions in the response of DOC flux and fire-regime to warming have been observed (Kicklighter et al 2013, de Groot et al 2013; supplementary information system summaries). Asia, which represents more than half of the total permafrost region, was under-represented in all three surveys, particularly wildfire (table 1). However, the regional bias in this study may not be greater than that of model projections, which depend on observational and experimental data that are not evenly distributed throughout the permafrost region.

Experts identified water balance, vegetation distribution, and permafrost degradation as the most important sources of uncertainty in predicting the timing and magnitude of the permafrost carbon feedback (table 3). These three processes are closely interconnected by several internal feedbacks (Anisimov and Reneva 2006, Shur and Jorgenson 2007, Jorgenson et al 2013, Girardin et al 2016). For example, wildfire or drought can trigger a transition from coniferous to deciduous dominance, warming permafrost by up to 7 °C due to loss of insulating moss and associated changes (Sturm et al 2001, Shur and Jorgenson 2007, Yarie and Van Cleve 2010). The subsequent recovery trajectories of vegetation and permafrost, as well as the proportion of thawed carbon released CO₂ or CH_4, then depend largely on near-surface hydrologic conditions (Payette et al 2004, Myers-Smith et al 2008, Jorgenson et al 2010, Chapin et al 2010, O'Donnell et al 2011, Lawrence et al 2015). These interdepencies mean that improving projections of the permafrost carbon feedback will require conceptualizing these parameters together. The question of water balance is additionally important in Arctic and boreal ecosystems where hydrologic carbon flux can be the determining factor causing net carbon uptake or release (Kling et al 1991, Aufdenkampe et al 2011, Raymond et al 2013). The lack of model projections of hydrologic carbon fluxes is a major gap in our ability to estimate the permafrost carbon feedback.

The permafrost region has responded differently to various climatic perturbations in the past, representing another tool to constrain possible future response (Zachos et al 2008). During the Paleocene–Eocene thermal maximum, high-latitude temperature warmed more than 10 °C, causing almost complete loss of permafrost and the mineralization of most permafrost soil organic matter (Bowen and Zachos 2010, DeConto et al 2012). More recently, the 2 °C–4 °C warming at high-latitudes during the early Holocene caused active-layer deepening throughout the permafrost region but did not trigger complete permafrost loss or widespread carbon release (French 1999, Schirrmeister et al 2002, Jorgenson et al 2013). While there are many differences between the Paleozoic and Holocene warming events, one clear distinction is the degree of warming. There may have been a threshold between 4 °C and 10 °C high-latitude warming due to positive feedbacks such as a shift from a coniferous to a deciduous dominated system or an abrupt change in hydrology. If a tipping point does exist between 4 and 10 °C high-latitude warming, it would fall between scenarios RCP4.5 and RCP8.5, representing maximum atmospheric CO₂ of 650 ppm and 850 ppm, respectively (Moss et al 2010, Lawrence et al 2012). RCP4.5 is still widely accepted as politically and technically attainable, though it assumes global CO₂ emissions peak before 2050 and decrease by half by 2080 (Moss et al 2010).