Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions

Terrestrial ecosystems play a vital role in regulating the accumulation of carbon (C) in the atmosphere. Understanding the factors controlling land C uptake is critical for reducing uncertainties in projections of future climate. The relative importance of changing climate, rising atmospheric CO2, and other factors, however, remains unclear despite decades of research. Here, we use an ensemble of land models to show that models disagree on the primary driver of cumulative C uptake for 85% of vegetated land area. Disagreement is largest in model sensitivity to rising atmospheric CO2 which shows almost twice the variability in cumulative land uptake since 1901 (1 s.d. of 212.8 PgC vs. 138.5 PgC, respectively). We find that variability in CO2 and temperature sensitivity is attributable, in part, to their compensatory effects on C uptake, whereby comparable estimates of C uptake can arise by invoking different sensitivities to key environmental conditions. Conversely, divergent estimates of C uptake can occur despite being based on the same environmental sensitivities. Together, these findings imply an important limitation to the predictability of C cycling and climate under unprecedented environmental conditions. We suggest that the carbon modeling community prioritize a probabilistic multi-model approach to generate more robust C cycle projections.

. Cumulative net contribution in PgC of drivers to simulated global net land sink over the last 5 decades , along with the percent contribution of each driver to the total change (+ and -) in cumulative flux for that region. Attribution is done through simulation differencing. Thus, only models with that submitted simulations RG1 through BG1 (models with N-cycle) and RG1 through SG3 (models without N-cycle) are included. Models with a coupled carbon-nitrogen cycle are denoted in bold.

Cumulative contribution (PgC)
Percent contribution to total change (%) Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Table S2. Average land uptake and cumulative land uptake simulated by the MsTMIP models compared with the observation-based estimate from Global Carbon Project 1 and average uptake simulated by 11 coupled carbon-climate Earth System Models (ESM) 2 . Models that are in bold font include a dynamic nitrogen cycle. Also shown are the multi-model mean and range (1σ).
Net C uptake (PgC yr -1 ) Cumulative C uptake (Pg C) 1960-1969 1990-1999 2000-2009 1959-2010 1901-2010  Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Table S3. Cumulative net contribution in PgC of drivers to simulated net land sink over the last 5 decades  for the tropics (30S to 30N), along with the percent contribution of each driver to the total change (+ and -) in cumulative flux for that region. Attribution is done through simulation differencing. Thus, only models with that submitted simulations RG1 through BG1 (models with Ncycle) and RG1 through SG3 (models without N-cycle) are included. Models with a coupled carbonnitrogen cycle are denoted in bold.
Cumulative contribution (PgC) Percent contribution to total change (%) Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Table S4. Cumulative net contribution in PgC of drivers to simulated net land sink over the last 5 decades  for the northern mid-latitudes (>30N and <50N), along with the percent contribution of each driver to the total change (+ and -) in cumulative flux for that region. Attribution is done through simulation differencing. Thus, only models with that submitted simulations RG1 through BG1 (models with N-cycle) and RG1 through SG3 (models without N-cycle) are included. Models with a coupled carbon-nitrogen cycle are denoted in bold.
Cumulative contribution (PgC) Percent contribution to total change (%) Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Table S5. Cumulative net contribution in PgC of drivers to simulated net land sink over the last 5 decades  for the arctic-boreal region (>50N), along with the percent contribution of each driver to the total change (+ and -) in cumulative flux for that region. Attribution is done through simulation differencing. Thus, only models with that submitted simulations RG1 through BG1 (models with N-cycle) and RG1 through SG3 (models without N-cycle) are included. Models with a coupled carbon-nitrogen cycle are denoted in bold.
Cumulative contribution (PgC) Percent contribution to total change (%)  Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Table S7. Sensitivity of the global net land sink to atmospheric CO2 (β), temperature (γ), and temperature variability (γIAV) for each model and for a mass balance observational product based on the Global Carbon Project (GCP) estimate of the net land sink. Sensitivities are calculated from 1901 to 2010 (β and γ) and 1959 to 2010 (γIAV), excluding post volcano years (1963,1964,1982,1983,1991). Table also shows the associated carbon gain/loss since 1901 as a result of these sensitivities, as well as the relative contribution of each. Values in parentheses next to the sensitivities for each model and for the observational product (GCP) indicate the uncertainties (standard error) of each coefficient (β, γ, γIAV) based on the regression. The values in parentheses next to the ensemble mean values of each coefficient indicate the standard deviation across model sensitivities.
Global C gain/loss since 1901 due to: Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Figure S2. Steady-state tropical (a) live biomass pool size and (b) soil carbon pool size against cumulative land carbon uptake / loss in the tropics since 1959 attributed to driver. Circles show individual models (n=9); lines show trend across models. Models not included: Biome-BGC, SiB3.0, SiB-CASA, and VISIT. Biome-BGC did not submit sensitivity simulations for land-cover change and atmospheric CO2; SiB3.0 does not have carbon pools; VISIT did not provide stead-state soil carbon pools; and SiB-CASA reported very large total living biomass (>1,600 PgC) which skewed the bestfit line in (a).
Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Figure S3. Dominate driver of the cumulative net land sink over the simulation period 1959 to 2010 for each model, and whether that driver is associated with a enhancement (+) or weakening (-) of land carbon sink strength over that time period. Drivers include: climate (blue), land-cover change (brown), atmospheric CO2 (orange), and nitrogen deposition (green). Figure was created using Matlab version R2015a (http://www.mathworks.com/products/matlab/) with post processing done in Adobe Illustrator CS6 Version 16.04 (https://www.adobe.com/products/illustrator.html).
Supplemental for: Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions Figure S4. Impact of climate and atmospheric CO2 on (a,b) global, (c,d) tropical, (e,f) northern midlatitude, and (g,h) arctic-boreal land carbon uptake. (a,c,e,f) The response of the net land sink over the period 1901 to 2010 to rising atmospheric CO2 (β) and temperature (γ) estimated from multilinear regression for models with (green) and without (purple) a dynamic nitrogen cycle. (b,d,f,h) The long-term sensitivity of the net land sink to climate warming (γ) versus the short-term sensitivity of net uptake to interannual variability in temperature (γIAV). The error bars show uncertainty (s.e.) in the regression coefficients (β, γ, γIAV). The black lines show the best-fit, linear relationship between short-and long-term climate sensitivity. Figure was created using Matlab version R2015a