A warming climate will make Australian soil a net emitter of atmospheric CO₂

The Earth’s soil is a crucial component of the global carbon (C) cycle. Its immense C store and concerns about the effects of a warming climate on its stability have forced scientists and politicians to take serious note of the soil. Small changes in the soil’s C store can significantly affect the terrestrial C cycle. Understanding the effects of climate change on the soil’s C dynamics is therefore crucial for land management, adaptation to a warming climate, and mitigation in the medium (2045–2070) and long-term (2070–2100).

It is generally agreed that the Earth has warmed by approximately 1 °C in the last 100 years, mainly due to increased CO₂ emissions from industry and transport. A further increase to 1.5 °C above preindustrial temperature is expected within the foreseeable future unless we can reduce emissions and remove greenhouse gases from the atmosphere. Delegates at COP27, although wishing to limit the increase to 1.5 °C by 2050, failed to agree on cuts in emissions to achieve that goal. Worse, if emissions continue at the current rate, an increase to 2 °C is likely sometime this century. Such warming is predicted to have dire consequences and potentially catastrophic impacts on humankind, the environment, and the economy. The warnings have been repeated by scientists, and many politicians, too, are now treating them seriously. If we are to restrict global warming beyond 1.5 °C then we need to limit the net increase in CO₂ in the atmosphere to zero, together with substantial reductions in the emissions of other greenhouse gases such as methane and nitrous oxide. Therefore, scientists, stakeholders, and politicians are considering the capture and storage of greenhouse gases (GHG) and C sequestration to achieve ‘net zero’ emissions.

The capture and storage of CO₂ at sources from industry and power stations is a matter of technology. That from the atmosphere must depend on nature, by photosynthesis and on land by storing C in the soil. Soil and vegetation currently absorb roughly one-third of global anthropogenic emissions. The soil’s capacity to sequester C could be increased by changing land use, conservation, restoration, and sound soil management. These ‘Nature-based’ actions, aided by modern technologies, could be implemented immediately and cost-effectively to limit the effects of climate change. Unlike the C stored in vegetation, C in the soil is more resilient to fire, pests, and wind, providing a medium for plant growth and ecosystem services, such as greater water storage and reduced runoff, erosion and flooding, landscape rehabilitation, food security and sustainable development.

The study integrates more than 4000 site observations of 0–30 cm soil carbon fractions—total organic carbon (TOC), particulate organic carbon (POC), mineral-associated organic carbon (MAOC), and pyrogenic C (PyC)—with the Rothamsted Carbon Model (RothC v26.3), one of the constituent models in Australia’s Full Carbon Accounting Model (FullCAM). Model initialization leveraged measured POC and MAOC to represent RothC’s main pool structure (replacing RPM and HUM), and PyC for inert organic matter, enabling data-driven initialization without spin-up. Decomposition follows first-order kinetics in pools DPM, RPM/POC, BIO, HUM/MAOC with rate constants 10, 0.3, 0.15, and 0.02 year−1, respectively, modified by temperature, soil water deficit, and vegetation cover.

Baseline period and inputs: The 1991–2010 period served as baseline. Soil attributes (clay, bulk density, AWC) and climate (air temperature, precipitation, pan evaporation) were compiled from national datasets (SILO, TERN). Monthly C inputs were estimated by land use class (arable cropping, modified grazing, native grazing, nature conservation). For cropping, a water-limited crop growth model was used to simulate dry matter production based on soil water balance (AWC, pan evaporation, rainfall, soil properties) and crop parameters, allocating root:shoot ratio 0.3 and residue C content 42%. Crop and pasture rotations were determined at Statistical Area Level 2 (SA2) from agricultural activity data, selecting the three most common crops/grass species per SA2; modified/native pastures and conservation areas used representative grass species or fixed small residue inputs. Grazing assumptions included consumption and return fractions and drought-induced mortality thresholds. Initial C inputs were iteratively adjusted to fit simulated to measured TOC/POC/MAOC under a long-term steady-state assumption.

Future simulations (2010–2100): Land management was held constant to isolate climate effects. Climate forcing came from six CMIP6 Earth System Models (ACCESS-ESM1.5, CESM2, CNRM-ESM2-1, IPSL-CM6A-LR, MIROC-ES2L, NorESM2-LM) under three SSPs (SSP1-2.6, SSP2-4.5, SSP5-8.5). Future monthly climate was bias-corrected using a multivariate quantile mapping algorithm (R package MBC) with 1970–2010 as reference. Projected NPP from ESMs and MODIS (MOD17A3H) informed annual rates of change to account for CO₂ fertilization and transpiration efficiency changes; rangeland monthly C inputs were adjusted by annual NPP change. For cropping and coastal temperate areas, future C inputs were derived by running the crop model for the three most representative crop/grass types per SA2 under each ESM-SSP, implicitly capturing management effects (e.g., rotations, no-till prevalence) given limited fertilizer/stubble data.

Analyses and reporting: Median annual TOC, POC, MAOC were computed across ESMs per SSP, smoothed with an 11-year moving average. Changes were calculated relative to the 1990–2010 baseline for periods 2010–2020, 2020–2045, 2045–2070, 2070–2100. Units reported include t C ha−1 and Gt C; annualized rates in t C ha−1 yr−1; and CO₂-equivalents (Gt CO₂-e yr−1) using 3.67 conversion. Uncertainty is the interdecile range (10th–90th percentiles) across ESMs and C-input scenarios.

Mapping: Predictions (TOC, POC, MAOC) for each 25-year period and SSP were mapped at 1 km resolution by punctual kriging with external drift (KED). Cubist models related site responses to spatially explicit forcings (mean annual temperature, precipitation, pan evaporation, NPP, Prescott index) and static covariates (DEM, slope, TWI, gamma radiometric dose and K, clay minerals: kaolinite, illite, smectite), producing covariate surfaces used as external drift in KED. Models were validated by 10-fold cross-validation with R² between 0.73 and 0.92. Data and code sources for soil datasets, climate projections, RothC, and processing scripts are cited; implementation code available from the corresponding author on request.

• Australia becomes a net soil CO₂ source this century under warming, with losses driven by rangelands and coastal regions outweighing cropland gains.
• Australia-wide TOC change rates (t C ha−1 yr−1): 2020–2045 median losses of −0.0137 (SSP1-2.6), −0.0228 (SSP2-4.5), −0.0770 (SSP5-8.5); 2070–2100 losses of −0.0125, −0.0343, −0.0466, respectively (Table 2).
• Relative to baseline TOC (27.4 t ha−1): Australia-wide change remains positive to ~2050, then turns negative under SSP2-4.5 and SSP5-8.5. By 2070–2100 median Australia-wide TOC change is 0.28 t ha−1 (SSP1-2.6), −0.68 t ha−1 (SSP2-4.5), −1.96 t ha−1 (SSP5-8.5) (Table 1). Predicted C inputs would need to exceed ~1.2 t ha−1 to offset losses associated with ~1.4 °C warming.
• Carbon pools: MAOC loss and loss rates exceed those of POC at national scale and in rangelands; in croplands, MAOC accumulates more than POC during gains.
• Rangelands (≈80% of land area): Low per-area C inputs (>0.5 t ha−1 baseline) and persistent C losses. Median loss rates (t ha−1 yr−1): 2020–2045 −0.0450 (SSP1-2.6), −0.0499 (SSP2-4.5), −0.0442 (SSP5-8.5); losses remain of similar magnitude later, with some regional/period gains under SSP2-4.5 in mid–late century. Small per-area losses translate to large national impacts.
• Croplands (including modified grazing): Management-driven gains early, diminishing with warming. Median sequestration rates (t ha−1 yr−1): 2020–2045 0.1947 (SSP1-2.6), 0.0950 (SSP2-4.5), 0.1640 (SSP5-8.5); 2045–2070 0.0949, 0.0939, −0.0302; 2070–2100 0.0314, 0.0171, −0.0663. These rates align with literature (0.1–0.4 t ha−1 yr−1) for practices like stubble retention, reduced tillage, and rotations.
• Coastal temperate regions: Largest C inputs but net losses, strongest early. Median rates (t ha−1 yr−1): 2020–2045 −0.0925 (SSP1-2.6) to −0.1366 (SSP5-8.5); losses persist, with rates generally decreasing after 2045 but remaining more negative with greater warming.
• Emissions in CO₂-e (Gt CO₂-e yr−1, median): Under SSP2-4.5, Australia-wide soil emissions are 0.064 (2020–2045), 0.030 (2045–2070), 0.097 (2070–2100), equivalent to ~14% of 2022 national emissions in 2020–2045. Under SSP5-8.5, emissions are larger and more uncertain; 2045–2070 0.123, increasing further by 2070–2100. Under SSP1-2.6, emissions are smallest: 0.039 (2020–2045), 0.029 (2045–2070), 0.035 (2070–2100).
• Cropland sequestration potential (Gt CO₂-e yr−1, 2020–2045): Median 0.104 (SSP1-2.6), 0.051 (SSP2-4.5), 0.088 (SSP5-8.5). Gains decrease by roughly half in 2045–2070; under SSP2-4.5 and SSP5-8.5, soils become net emitters by late century (up to 0.036 Gt CO₂-e yr−1).
• Spatial patterns: Rangeland losses expand from northernmost areas south and west through the century, with MAOC lost more than POC; local POC increases in some southern rangelands do not offset broader losses. Croplands are net sinks under all SSPs, but gains shrink with warming and some losses emerge regionally by 2070–2100 under SSP2-4.5 and SSP5-8.5. Coastal temperate regions show widespread losses; by 2100 MAOC loss exceeds POC loss.
• Magnitudes over 25-year blocks (SSP1-2.6): Median organic C loss totals increase across periods (e.g., 0.293, 0.805, 1.239 Gt), with larger losses under SSP2-4.5 and SSP5-8.5.
• Climate trajectories: CMIP6 ensemble projects Australia warms and generally dries, especially rangelands; cropping warms more slowly; coastal regions warm least but become more prone to drying. C inputs decline with stronger forcing (SSP5-8.5).

The analysis indicates that, under a middle-of-the-road socioeconomic trajectory (SSP2-4.5), Australian soils are projected to be a net source of carbon to the atmosphere throughout the century, adding to emissions from other sectors. Early-century emissions are substantial relative to national totals and agricultural sector emissions. A sustainable trajectory (SSP1-2.6) reduces soil emissions but does not fully eliminate them, whereas a fossil-fueled development path (SSP5-8.5) leads to larger soil C losses and higher, more uncertain emissions.

Mechanistically, warming and drying enhance decomposition and reduce plant-derived C inputs in many regions, particularly in rangelands and coastal temperate zones, leading to persistent net losses. The more stable MAOC pool is disproportionately depleted in rangelands and coastal regions. Croplands can sequester carbon under prevailing rotations and practices, especially in early to mid-century, but their gains diminish with warming and cannot offset losses across the much larger rangeland and coastal extents.

Policy implications are significant. Anticipated soil C gains in national plans appear optimistic relative to these projections. To mitigate losses and enhance sequestration, especially in rangelands, management must improve: better grazing management, cultural burning, and regeneration of biodiverse native plant communities could help bolster C inputs and stabilize soils. Strengthening carbon crediting schemes (e.g., ACCU) and nature-positive incentives, while not a panacea, can support adoption. Achieving national emissions targets will also require broader deployment of GHG removal technologies and alignment with a more sustainable emissions pathway.

Uncertainty is inherent in long-term projections, but the study reduces parameter and input uncertainty via calibration to Australian conditions, measured pool initialization, region-specific C input derivation, CO₂ fertilization consideration, bias-corrected CMIP6 forcing, and uncertainty propagation across ESMs and inputs. Nonetheless, structural model uncertainty remains unquantified, and assumptions of constant management likely understate the range of possible outcomes under alternative practices.

This study combines extensive measurements of soil organic carbon fractions with a calibrated RothC framework and bias-corrected CMIP6 projections to map and quantify Australia’s soil C dynamics to 2100 under multiple socioeconomic pathways. Results indicate that Australian soils will, on balance, emit CO₂ this century, with rangelands and coastal temperate regions driving losses that overwhelm cropland gains. Losses are smallest under a sustainability pathway (SSP1-2.6) and largest under high forcing (SSP5-8.5). The more stable MAOC pool is particularly vulnerable in non-cropland regions, highlighting risks to long-term soil C storage.

Practically, Australia’s emissions-reduction strategy must account for projected net soil C losses, emphasizing rangeland conservation and improved management, continued support for scientifically robust carbon crediting, and integration of nature-based solutions alongside technological removals. Future research should: (i) quantify structural model uncertainty via multi-model comparisons; (ii) integrate dynamic management scenarios (e.g., altered rotations, residue and fertilizer management, grazing regimes, restoration) to assess mitigation potential; (iii) enhance spatial and temporal data on C inputs and management; and (iv) refine regional projections by improving climate, vegetation, and soil process representations.

• Structural model uncertainty was not assessed; RothC’s relatively simple structure was chosen for large-scale application, but different model structures can alter climate response and uncertainty.
• Land management was assumed constant through 2100, excluding potential changes in practices, policies, technology, disturbances (e.g., fires), or land use that could alter C inputs and turnover.
• Limited national data on fertilizers, residue management, and specific agronomic practices necessitated implicit treatment of management via crop/pasture rotations, potentially underrepresenting management variability.
• Climate inputs derive from six CMIP6 ESMs at coarse native resolution (2°) with bias correction; residual biases and inter-model spread contribute to uncertainty.
• C input derivations for rangelands depend on ESM NPP and MODIS trends to capture CO₂ fertilization and water-use efficiency; these effects remain uncertain and regionally variable.
• Mapping and interpolation, while validated (R² 0.73–0.92), introduce spatial modeling uncertainty; 1 km outputs may not capture fine-scale heterogeneity or local management effects.