全球变化知识资源中心

Crop residues are abundant feedstocks that are used for biofuel production globally^{17, 18}. By 2022, the US Energy Independence and Security Act (EISA) mandates production capacity for cellulosic ethanol and advanced biofuels to be 61 billion litres per year (bly) and 19 bly, respectively¹⁷. Corn residue is predominantly used in US cellulosic ethanol biorefineries, with a combined capacity of 0.38 bly in 2014 (ref. 19). An additional 0.42 bly of US hydrocarbon biofuels mostly uses wood¹⁹, but could also be derived from crop residue²⁰. Absolute changes in soil organic carbon (SOC) from corn residue removal have been estimated in LCA (ref. 6), but few have estimated net changes in SOC and CO₂ emissions compared with no residue removal^{7, 8, 21, 22}, as required by consequential LCA (ref. 23).

Recent research suggests soil CO₂ emissions from residue removal could produce life cycle GHG emissions for cellulosic ethanol that exceed the mandated emissions reduction⁸. Incubation experiments with soil and corn residue showed that SOC is oxidized to CO₂ at 0.54–0.80 Mg C ha⁻¹ per season when residues are completely removed³. Modelled removal of all corn residue in Austria projected an SOC loss of 0.35 Mg C ha⁻¹ yr⁻¹, which represents nearly 50% of life cycle GHG emissions from a biorefinery system²⁴. Modelled SOC oxidation to CO₂ from removal of sweet sorghum residue showed these emissions could eliminate all GHG emissions benefits of the resulting biofuel compared with gasoline²⁵. Similar net losses of C stocks have also been projected for biofuels from forestry in some cases²⁶.

Changes in SOC occur by two dominant processes: soil erosion by water and wind, and soil respiration where SOC is oxidized to CO₂ (refs 4, 5). Soil erosion has significantly depleted SOC across the US Corn Belt, with a recent loss of 1.7 billion tons of soil in the US in 2007 (ref. 27). Crop residue has conventionally been left on the field after harvest to reduce soil erosion and maintain the SOC stocks and soil fertility of the Corn Belt¹. Although some soil measurements in the Corn Belt have shown that complete residue removal reduces SOC compared with no removal^{28, 29}, other studies found no significant differences¹⁶. Measuring SOC change accurately is limited owing to the high spatial variability in SOC stocks, the inability to detect a small annual percentage change, short-term studies, and failure to express SOC results in an equivalent mass basis to account for changes in soil bulk density^{30, 31}. Furthermore, when crop residue is removed, it is essential to determine whether SOC loss is due to erosion or respiration, to accurately estimate the resulting net CO₂ emissions.

Models are necessary to confidently estimate small percentage annual changes in regional SOC stocks due to respiration^{30, 31}, as extensive gas exchange measurements are too costly. Although soil moisture and texture are often used in SOC models⁴, a robust model can estimate daily changes in SOC due to oxidation to CO₂ based on initial SOC (C₀), C inputs from agricultural crops (C_i), and average daily temperature (T_a), as shown below^{9, 10, 11}. The SOC model used here is based on exponential oxidation coefficients for SOC (k_s, S_s) and cereal crop residues (k_r, S_r) from 36 field studies across North America, Europe, Africa and Asia¹⁰ (see Supplementary Table 1 and Methods). An additional term in the equation is added for each year of new C inputs to the soil from residue and roots.

To test the model in the central US, we compared model results with measured CO₂ emissions, residue biomass, and SOC from an irrigated no-till continuous corn field experiment in eastern Nebraska (Mead) from 2001 to 2010 (refs 12, 13, 14). The model estimated that 83% of initial residue C input was oxidized during the first three years, which closely agreed with field measurements that found an average of 20% remained¹⁴ (Supplementary Fig. 1) Cellulose, hemicellulose and protein in residue rapidly oxidize, whereas the more recalcitrant lignin fraction (~18% dry matter⁶) undergoes a slower oxidation process and contributes to SOC (ref. 4). The model estimated 80.9% of initial SOC remained after nine years (56.1 of 69.4 Mg C ha⁻¹) in the 0–30 cm depth, and net C from residue (8.53 Mg C ha⁻¹) contributed to the maintenance of a total of 93.2% of the initial SOC stock (Fig. 1). When compared with soil measurements, the model predicted net SOC loss within 17% accuracy during the first four years of the experiment (Supplementary Table 2). Eddy covariance was used to measure net CO₂ fluxes to the atmosphere to estimate ecosystem respiration, which was partitioned into emissions from crop respiration and from soil and residue³² (Methods). The model predicted annual measured net CO₂ emissions to the atmosphere from soil and residue with an error of 12.4% on average (range 34 to −22%; Supplementary Tables 3 and 4). While using coefficients for SOC oxidation derived from a global span of field measurements, the modeled SOC dynamics agreed well with the field measurements of CO₂ emissions, residue remaining, and SOC. The global character of the model assumptions combined with these regional tests indicates the model has enough accuracy to confidently estimate the average direction of change in net CO₂ emissions and SOC from residue removal across the Corn Belt.

Daily modelled oxidation of soil organic carbon (SOC) and residue to CO₂ is based on field measurements of initial SOC (0–30 cm soil depth), corn residue input, and temperature at Mead, Nebraska. The average annual net loss of SOC is 0.47 Mg C ha⁻¹ yr⁻¹, but declines exponentially from 1.13 to 0.25 Mg C ha⁻¹ yr⁻¹ over the first eight years.

Soil organic carbon model.

Oxidation rate coefficients were estimated for soil organic matter (SOM) and plant residue (k_S and k_r, respectively) and the rate of ageing of SOM and plant residue (S_S and S_r, respectively) from 306 datasets from 36 studies covering a wide range of residue substrates, soil types and climatic conditions globally¹⁰ (Supplementary Table 1). Average oxidation response due to temperature (Q₁₀) is based on previous research. Decomposition rates were modelled for all C components (nine years of residue inputs and initial SOC) at the field site based on daily average temperature data and measured C₀ and C_i values (Supplementary Fig. 1 and Tables 2,3). If T_a is greater than the reference temperature (T_r, 10 °C), T_a is subtracted from T_r and divided by 10, and placed as an exponent on Q₁₀ in the model; this term is the temperature coefficient (T_co). If T_a is less than T_r, then T_co is assumed to change linearly with T_a, with a rate of 0.1 per degree of T_a; no oxidation occurs below 0 °C. The sum of T_co (total heat accumulated) determines the amount of C remaining at time t.

Comparison of model with field CO₂ measurements.

Fluxes of CO₂ were measured using tower eddy covariance above continuous corn from 2001 to 2010 at Mead, Nebraska. Inputs of C to soil at Mead were estimated based on measured grain and residue yield, and estimated root biomass (Supplementary Table 3). Measured ecosystem total respiration was partitioned into emissions from: live root and aboveground biomass of the growing crop, irrigation water, and SOC and crop residue (Supplementary Table 4). The gas measurements account for net CO₂ flux from the entire soil profile depth, and modelling of CO₂ emissions from the top 0–30 cm is expected to underestimate measured flux emissions; but as the majority of SOC is often in the top 30 cm in the Corn Belt, modelling the dynamics of this zone would probably account for the majority of emissions.

Geospatial data and supercomputer simulations.

A 10 m Soil Survey Geographic grid (gSSURGO) of representative 30 cm depth SOC values was resampled to 30 × 30 m and converted to Mg C ha⁻¹ (30 cm)⁻¹ (Supplementary Fig. 2). All other spatial inputs were resampled to 30 m and aligned with the SOC grid space using zero-valued SOC masks of the area planted in corn or soybean in 2010. Monthly maximum and minimum average temperatures from the PRISM database (2001–2010) were used. Rainfed county corn grain yield estimates from NASS (2001–2010) were converted to Mg C ha⁻¹ yr⁻¹ using a harvest index (0.53), and estimated C from roots was added (Supplementary Fig. 2 and Table 3). Simulated removal of C was limited to the actual amount of aboveground C estimated in each grid per year.

A massive amount of data was used to produce these results. Processing on a PC with ESRI’s ArcGIS 9.3 limited input file size to the state level (1-2 gigabytes (GB)). Data were analysed using high-performance computer clusters in the Holland Computing Center (HCC) at University of Nebraska-Lincoln (http://hcc.unl.edu) that employ parallel programs to speed up computation. The uncompressed input data totalled ~3 terabytes (TB) and the uncompressed output data totalled > 30 TB. The program split each state’s input file into ~40 megabyte (MB) files, and then executed computations on the smaller files in parallel. The output files were then joined together in a single state file, for each of the 12 states. If input files had not been split, the computational speed would have been significantly reduced owing to opening and closing of files and because loading an entire large disk file into memory at once is infeasible.

1. Wilhelm, W. W., Johnson, J. M. F., Karlen, D. L. & Lightle, D. T. Corn stover to sustain soil organic carbon further constrains biomass supply. Agronomy J. 99, 1665–1667 (2007).
   • CAS
   • Article
2. Anderson-Teixeira, K. J., Davis, S. C., Masters, M. D. & DeLucia, E. H. Changes in soil organic carbon under biofuel crops. GCB Bioenergy 1, 75–96 (2009).
   • CAS
   • Article
3. Kochsiek, A. E. & Knops, M. H. Maize cellulosic biofuels: Soil carbon loss can be a hidden cost of residue removal. GCB Bioenergy 4, 299–233 (2012).
   • CAS
   • Article
4. Kutsch, W. L., Bahn, M. & Heinemeyer, A. Soil Carbon Dynamics: An Integrated Methodology (Cambridge Univ. Press, 2009).