========================================================================== HOW MUCH CARBON CAN FARMLAND ACTUALLY HOLD? Random-effects pooling of published soil organic carbon effect sizes Science Journaling Club -- our own code, written from the formulas ========================================================================== Read PART 1 before PART 3. The machinery is checked against a known answer first, because a pooling routine that has never been asked to recover a number you already know is a routine you have no reason to trust. ========================================================================== PART 1 -- DOES THE MACHINERY WORK? (two checks, before any real data) ========================================================================== Check A: recover a known true effect from synthetic data -------------------------------------------------------- true mu = 0.113329 lnRR (= +12.00% change) true tau^2 = 0.002500 k = 12 synthetic studies pooled (random) = 0.114106 [0.083567, 0.144645] ... as a percentage = +12.09% [+8.72%, +15.56%] estimated tau^2 = 0.001229 (true 0.002500) I^2 = 47.2% TRUTH INSIDE THE 95% INTERVAL: YES coverage over 2000 replicates = 91.5% (nominal 95%) mean estimated tau^2 over replicates = 0.002498 (true 0.002500) Two things to read off that pair of lines. The estimator recovers tau^2 almost exactly on average, so it is not biased. Coverage still lands below the nominal 95%, because the interval treats the estimated tau^2 as if it were known exactly, and it is not. Our intervals on the real data are therefore a little too narrow, and we would rather print that than hide it. Check B: at tau^2 = 0 the fixed and random estimates must coincide ------------------------------------------------------------------ Q = 0.000000000000 (df = 5) tau^2 (DL) = 0.000000000000 I^2 = 0.0000% fixed-effect theta = 0.067658648473815 random-effects theta = 0.067658648473815 |difference| = 0.000e+00 IDENTITY HOLDS: YES Why it must: at tau^2 = 0 the random-effects weight 1/(v+tau^2) IS the fixed-effect weight 1/v, so the two weighted means are the same weighted mean. Any code that fails this is wired wrong. second case (real but tiny spread): Q = 0.000657, df = 4, tau^2 = 0.000000000000 fixed 0.069980458398414 vs random 0.069980458398414 -> gap 0.000e+00 ========================================================================== PART 2 -- THE DATASET, WITH PROVENANCE FOR EVERY NUMBER ========================================================================== key source practice pct lo hi lnRR se depth ------------------------------------------------------------------------------------------------------------------ Du17-ESM Du et al. 2017 no-till (equivalent soil mass) 3.8 1.4 6.3 0.0373 0.0120 40 Du17-FD Du et al. 2017 no-till (fixed depth) 5.1 2.5 7.7 0.0497 0.0126 30 Bei23-NT Beillouin et al. 2023 no-till 9.3 5.6 13.0 0.0889 0.0173 n/s Bei23-RT Beillouin et al. 2023 reduced tillage 12.0 0.1 24.0 0.1133 0.0546 n/s Bei23-RES Beillouin et al. 2023 crop residue retention 13.0 9.8 16.0 0.1222 0.0140 n/s Bei23-ROT Beillouin et al. 2023 crop rotation 6.5 -0.9 14.0 0.0630 0.0357 n/s Jian20-CC Jian et al. 2020 cover cropping 15.5 13.8 17.3 0.1441 0.0077 30 Liu14-STR Liu et al. 2014 straw return 12.8 12.0 13.6 0.1204 0.0036 20 Han16-CFS Han et al. 2016 straw + chemical fertiliser 19.5 18.5 21.5 0.1781 0.0064 20 Han16-CFM Han et al. 2016 manure + chemical fertiliser 36.2 33.1 39.3 0.3090 0.0116 20 GG21-MAN Gross & Glaser 2021 manure application 35.0 32.0 39.0 0.3001 0.0132 30 Bei23-ORG Beillouin et al. 2023 organic for mineral substitution 34.0 20.0 49.0 0.2927 0.0552 n/s Where each number came from --------------------------- [1] Du17-ESM Abstract, verbatim: '3.8% [95% CI: 1.4%-6.3%; P=0.005]' (ESM) [1] Du17-FD Abstract, verbatim: '5.1% (95%CI: 2.5%-7.7%; P<0.001)' (FD) [2] Bei23-NT Results text: 'no-till farming (+9,3%, CI [+5,6, +13])' [2] Bei23-RT Results text: 'reduced tillage intensity (+12%, CI [+0.1, +24])' [2] Bei23-RES Results text: 'crop residue retention (+13%, CI [+9.8, +16])' [2] Bei23-ROT Results text: 'crop rotation resulted ... 6.5% (CI [-0.9, +14])' [3] Jian20-CC Abstract, verbatim: 'mean change of 15.5% (95% CI 13.8%-17.3%)' [4] Liu14-STR Abstract: '12.8 +/- 0.4%'; we read +/- as one standard error and widened it to 95% (12.8 +/- 1.96*0.4). FLAGGED: derived. [5] Han16-CFS Abstract: '2.0 (1.9-2.2) g/kg (19.5%)'. We scaled the reported absolute interval by 19.5/2.0. FLAGGED: derived. [5] Han16-CFM Abstract: '3.5 (3.2-3.8) g/kg (36.2%)'. Scaled by 36.2/3.5. FLAGGED: derived. [6] GG21-MAN Results text, verbatim: 'increase of SOC stocks of 35% (95% CI 32-39%)' [2] Bei23-ORG Results text: 'increase in SOC of +34%, CI [+20, +49]' Effect sizes we found but could NOT pool ---------------------------------------- 15 entries discarded. The discard pile matters: it is larger than the pool, and it is not a random sample of the evidence. Nulls and absolute-stock results are over-represented in it, which biases whatever survives UPWARD. [7] Luo et al. 2010 no-till, 0-10 cm +3.15 +/- 2.42 Mg C/ha absolute stock, no control stock given -> no response ratio [7] Luo et al. 2010 no-till, 20-40 cm -3.30 +/- 1.61 Mg C/ha absolute stock, same reason [7] Luo et al. 2010 no-till, 0-40 cm no significant increase reported as a null, no point estimate given in the abstract [8] Haddaway et al. 2017 no-till, 0-30 cm, >=10 yr +4.6 Mg/ha (0.78-8.43) absolute stock, no control stock given [8] Haddaway et al. 2017 no-till, full profile no effect detected reported as a null [9] Angers & Eriksen-Hamel 2008 no-till vs full-inversion, profile +4.9 Mg/ha no interval retrievable from the sources we could read [10] Virto et al. 2012 no-till, 0-30 cm ESM +6.7% (3.4 Mg C/ha) no interval in the abstract [11] McClelland et al. 2021 cover crops, 0-30 cm +12% (1.11 Mg C/ha) interval is in a figure we could not read numerically [12] Prairie et al. 2023 no-till, 0-20 cm +11.3% interval is in Fig. 2 only [12] Prairie et al. 2023 no-till, >20 cm not significant reported as a null [13] Bai et al. 2019 conservation tillage +5% no interval in the abstract [13] Bai et al. 2019 cover crops +6% no interval in the abstract [14] Crystal-Ornelas et al. 2021 conservation tillage, organic systems +14% no interval in the abstract [15] Emde et al. 2021 irrigation +5.9% intervals in Fig. 3 only [16] Sun et al. 2020 no-till, 260 paired studies no global summary figure regional results only ========================================================================== PART 3 -- RANDOM-EFFECTS POOL (all 12 effect sizes) ========================================================================== k = 12 fixed-effect pooled = 0.14085 lnRR = +15.13% RANDOM-EFFECTS POOL = 0.14982 lnRR = +16.16% 95% CI = [0.10711, 0.19252] = [+11.31%, +21.23%] 95% PREDICTION INT. = [-1.70%, +37.27%] z = 6.88, p = 6.15e-12 Cochran's Q = 571.03 on 11 df, p = 2.19e-115 tau^2 = 0.00514 (tau = 0.0717 on the log scale) I^2 = 98.1% H^2 = 51.9 READ THE PREDICTION INTERVAL, NOT THE CONFIDENCE INTERVAL. The confidence interval describes how well we know the AVERAGE of these studies. The prediction interval describes the range a NEW field would plausibly fall in, and it is the one a farmer or a credit buyer actually faces. Study weights in the random-effects pool ---------------------------------------- Du17-ESM 8.98% Du et al. 2017 / no-till (equivalent soil mass) Du17-FD 8.95% Du et al. 2017 / no-till (fixed depth) Bei23-NT 8.72% Beillouin et al. 2023 / no-till Bei23-RT 5.84% Beillouin et al. 2023 / reduced tillage Bei23-RES 8.89% Beillouin et al. 2023 / crop residue retention Bei23-ROT 7.40% Beillouin et al. 2023 / crop rotation Jian20-CC 9.13% Jian et al. 2020 / cover cropping Liu14-STR 9.21% Liu et al. 2014 / straw return Han16-CFS 9.16% Han et al. 2016 / straw + chemical fertiliser Han16-CFM 9.00% Han et al. 2016 / manure + chemical fertiliser GG21-MAN 8.93% Gross & Glaser 2021 / manure application Bei23-ORG 5.80% Beillouin et al. 2023 / organic for mineral substitution Note how flat these are. With tau^2 this large, tau^2 swamps the sampling variances and every study ends up weighted almost equally, so a huge synthesis of 592 comparisons counts about the same as one line lifted from a second-order review. ========================================================================== PART 4 -- SPLIT BY WHAT THE PRACTICE ACTUALLY DOES ========================================================================== REARRANGES carbon already in the field (tillage, residues, cover crops, rotation) k = 9, pooled = +10.95% [+7.63%, +14.37%], I^2 = 95.4%, tau^2 = 0.0018 Du17-ESM 3.8% no-till (equivalent soil mass) Du17-FD 5.1% no-till (fixed depth) Bei23-NT 9.3% no-till Bei23-RT 12.0% reduced tillage Bei23-RES 13.0% crop residue retention Bei23-ROT 6.5% crop rotation Jian20-CC 15.5% cover cropping Liu14-STR 12.8% straw return Han16-CFS 19.5% straw + chemical fertiliser IMPORTS carbon from outside the field (manure, organic substitution) k = 3, pooled = +35.63% [+33.36%, +37.94%], I^2 = 0.0%, tau^2 = 0.0000 Han16-CFM 36.2% manure + chemical fertiliser GG21-MAN 35.0% manure application Bei23-ORG 34.0% organic for mineral substitution Difference (add - move) = 0.2009 lnRR, se 0.0177, z = 11.34, p = 8.74e-30 On the percentage scale the adding practices run about 3.3x the rearranging ones. ========================================================================== PART 5 -- SAMPLING DEPTH ========================================================================== Of 12 pooled effect sizes, 7 state a maximum sampling depth. That is the first result of this section and it is not a good one. key depth cm pct se(lnRR) source Liu14-STR 20 12.8 0.0036 Liu et al. 2014 Han16-CFS 20 19.5 0.0064 Han et al. 2016 Han16-CFM 20 36.2 0.0116 Han et al. 2016 Du17-FD 30 5.1 0.0126 Du et al. 2017 Jian20-CC 30 15.5 0.0077 Jian et al. 2020 GG21-MAN 30 35.0 0.0132 Gross & Glaser 2021 Du17-ESM 40 3.8 0.0120 Du et al. 2017 Meta-regression of lnRR on maximum sampling depth (k = 7): intercept = +0.35558 (se 0.15647) slope = -0.007115 per cm (se 0.005586), z = -1.27, p = 0.203 residual tau^2 = 0.01059, Q_res = 5.00 on 5 df A slope of -0.007115 per cm means each extra 10 cm of required sampling depth moves the estimated effect by -6.87%. With k = 7 and one point beyond 30 cm, this regression is underpowered and we are not going to pretend otherwise. The number is printed so a reader can see how weak it is. Pooled effect under a minimum-depth requirement ----------------------------------------------- This is the calculation that matters for a credit protocol: if you refuse to count any study that did not sample at least D cm, what is left, and what does it say? require >= D cm k pooled % 95% CI I^2 >= 0 12 16.16 [+11.31, +21.23] 98.1% >= 20 7 17.65 [+11.23, +24.43] 98.9% >= 30 4 14.20 [+3.03, +26.58] 98.9% >= 40 1 3.80 [+1.38, +6.28] n/a Within-study depth contrasts (NOT pooled, and stronger for that) ---------------------------------------------------------------- Each row is one synthesis comparing its own shallow layer with its own deep layer. Everything except depth is held constant, which no between-study regression can manage. [7] Luo et al. 2010 no-till 0-10 cm: +3.15 Mg C/ha 20-40 cm: -3.30 Mg C/ha 0-40 cm: no increase [1] Du et al. 2017 no-till 0-20 cm: accumulation 30-40 cm: depletion FD 5.1% vs ESM 3.8% [12] Prairie et al. 2023 no-till 0-20 cm: +11.3% >20 cm: not significant 118 studies, 157 experiments [3] Jian et al. 2020 cover crops <=30 cm: significant increase >30 cm: not significant - [8] Haddaway et al. 2017 no-till 0-30 cm: +4.6 Mg/ha full profile: no effect >=10 yr comparisons [6] Gross & Glaser 2021 manure <=15 cm: +40% (conv. tillage) >30 cm: +23% (conv. tillage) CARBON-ADDING CONTROL Five of the six rows are practices that rearrange carbon, and all five lose the effect below the plough layer. The sixth is manure, which is carbon carted in from somewhere else, and it holds a significant effect below 30 cm. That contrast is the argument. ========================================================================== PART 6 -- FUNNEL ASYMMETRY AND EGGER'S TEST ========================================================================== Egger intercept = +1.607 (se 3.560) t = 0.45 on 10 df, p = 0.661 slope (the bias-adjusted effect) = +0.1296 lnRR = +13.83% Funnel coordinates (for Figure 4): key lnRR se 1/se y/se Du17-ESM 0.0373 0.0120 83.06 3.10 Du17-FD 0.0497 0.0126 79.21 3.94 Bei23-NT 0.0889 0.0173 57.88 5.15 Bei23-RT 0.1133 0.0546 18.31 2.07 Bei23-RES 0.1222 0.0140 71.36 8.72 Bei23-ROT 0.0630 0.0357 27.99 1.76 Jian20-CC 0.1441 0.0077 129.40 18.65 Liu14-STR 0.1204 0.0036 276.35 33.29 Han16-CFS 0.1781 0.0064 156.79 27.93 Han16-CFM 0.3090 0.0116 86.10 26.60 GG21-MAN 0.3001 0.0132 75.86 22.77 Bei23-ORG 0.2927 0.0552 18.11 5.30 CAUTION. Egger's test is close to meaningless at k = 12 with I^2 this high: heterogeneity alone generates funnel asymmetry, and with a dozen points the test has almost no power. Treat the number as a description of the plot, not as a hypothesis test. ========================================================================== PART 7 -- LEAVE-ONE-OUT SENSITIVITY ========================================================================== omitted k pooled % 95% CI I^2 tau^2 Du17-ESM 11 17.46 [+12.46, +22.67] 98.0% 0.0048 Du17-FD 11 17.31 [+12.23, +22.62] 98.1% 0.0050 Bei23-NT 11 16.84 [+11.68, +22.25] 98.2% 0.0053 Bei23-RT 11 16.43 [+11.40, +21.68] 98.2% 0.0052 Bei23-RES 11 16.48 [+11.25, +21.96] 98.2% 0.0055 Bei23-ROT 11 16.97 [+11.89, +22.28] 98.2% 0.0052 Jian20-CC 11 16.25 [+10.66, +22.12] 98.2% 0.0063 Liu14-STR 11 16.55 [+10.36, +23.08] 98.0% 0.0079 Han16-CFS 11 15.86 [+10.23, +21.78] 98.1% 0.0065 Han16-CFM 11 14.26 [+10.08, +18.59] 97.2% 0.0034 GG21-MAN 11 14.41 [+9.92, +19.09] 97.6% 0.0040 Bei23-ORG 11 15.15 [+10.21, +20.31] 98.2% 0.0051 Range across leave-one-out fits: 14.26% to 17.46% Full-data pool: 16.16% Widest single-study swing: 1.90 percentage points Pre-planned sensitivity: one effect size per publication -------------------------------------------------------- Beillouin et al. contribute four rows and Han et al. two, which double-counts those papers. Keeping only the first row from each publication gives: k = 6, pooled = +15.61% [+10.19%, +21.29%], I^2 = 98.3% kept: Du17-ESM, Bei23-NT, Jian20-CC, Liu14-STR, Han16-CFS, GG21-MAN Pre-planned sensitivity: drop the two derived intervals ------------------------------------------------------- k = 9, pooled = +14.03% [+7.34%, +21.14%], I^2 = 97.2% ========================================================================== PART 8 -- THE RATE POOL (Mg C per hectare per year) ========================================================================== A percentage is not a credit. Credits are issued in tonnes per hectare per year, so here is the small set of estimates reported in that unit, kept in a separate pool because you cannot average a percentage with a rate. key practice rate se 95% CI depth WP02 no-till 0.570 0.140 [+0.296, +0.844] mostly 0-30 cm PD15 cover crops 0.320 0.080 [+0.163, +0.477] 22 cm mean Du17-FDr no-till 0.300 0.126 [+0.053, +0.546] 0-30 cm, fixed depth Du17-ESMr no-till 0.141 0.124 [-0.102, +0.384] 0-30 cm, equivalent soil mass Pooled rate = 0.324 Mg C/ha/yr [0.175, 0.473] Q = 5.30 on 3 df, tau^2 = 0.00997, I^2 = 43.4% (No log transform here: these are already differences, not ratios.) The accounting pair -- the cleanest number in this whole review --------------------------------------------------------------- Same 95 comparisons, same 57 sites, same 0-30 cm, same authors. fixed depth : 0.300 Mg C/ha/yr [0.054, 0.547] significant equivalent soil mass : 0.141 Mg C/ha/yr [-0.102, 0.384] NOT significant ratio : 2.13x difference : 0.159 Mg C/ha/yr Nothing changed in the field. Only the arithmetic changed. No-till loosens soil, bulk density falls, and a fixed 30 cm core therefore scoops up less soil under no-till than under the plough. Correct for that by comparing equal masses of soil rather than equal depths and more than half the sequestration rate goes away. ========================================================================== PART 9 -- SATURATION: HOW LONG DOES A PRACTICE ACTUALLY BUY? ========================================================================== A soil is not a bucket you can keep filling. Carbon comes in from plant residues and roots and leaves again through microbial respiration, and the loss term grows with the stock itself. The simplest honest model is one line: dC/dt = I - k*C with C the stock in Mg C/ha, I the annual carbon input reaching the stable pool, and k the fractional loss rate per year. It has a closed solution: C(t) = C_eq + (C0 - C_eq) * exp(-k*t), C_eq = I / k Raise the input and the equilibrium rises proportionally, but the approach to it decays exponentially. That exponential is the whole argument about permanence: the practice does not stop working because the farmer stops trying, it stops working because the stock has climbed near its new ceiling and respiration has caught up with the extra input. C0 = 45.0 Mg C/ha (0-30 cm), k = 0.030/yr baseline input I0 = C0*k = 1.350 Mg C/ha/yr (holds C0 steady) practice adds dI = 0.50 Mg C/ha/yr new equilibrium C_eq = (I0+dI)/k = 61.67 Mg C/ha total gain available = 16.67 Mg C/ha, and not one tonne more year stock cumulative gain that year % of total 0 45.00 0.00 0.500 0.0% 5 47.32 2.32 0.430 13.9% 10 49.32 4.32 0.370 25.9% 20 52.52 7.52 0.274 45.1% 30 54.89 9.89 0.203 59.3% 40 56.65 11.65 0.151 69.9% 50 57.95 12.95 0.112 77.7% 75 59.91 14.91 0.053 89.5% 100 60.84 15.84 0.025 95.0% 150 61.48 16.48 0.006 98.9% Years until the annual gain falls below 0.30 Mg C/ha/yr: 17.0 Years until the annual gain falls below 0.20 Mg C/ha/yr: 30.5 Years until the annual gain falls below 0.10 Mg C/ha/yr: 53.6 Years until the annual gain falls below 0.05 Mg C/ha/yr: 76.8 Half the total gain is banked by year 23.1. Ninety per cent by year 76.8. Published durations, for comparison with the model: Liu et al. 2014 straw return saturates after about 12 years Han et al. 2016 28-73 yr (straw+NPK), 26-117 yr (manure+NPK) Poeplau & Don 2015 new steady state after ~155 yr, total gain 16.7 +/- 1.5 Mg C/ha at 22 cm depth Poulton et al. 2018 Rothamsted farmyard manure gave 18 and 43 per mille per year in the first 20 years, still above 7 per mille at 40-60 years, then fell Georgiou et al. 2022 accrual runs ~3x faster in soils at a tenth of their mineral capacity than at a half The Georgiou capacity check, done arithmetically ------------------------------------------------ Georgiou et al. report global mineral-associated carbon at 42% of mineralogical capacity in surface soil and 21% at depth, across 1144 profiles, and accrual roughly 3x faster at one tenth of capacity than at one half. In our one-line model the accrual rate at stock C is I - k*C, so the ratio of rates at two stocks C_a and C_b is: at 10% of a 100 Mg/ha capacity: 1.550 Mg C/ha/yr at 50% of the same capacity: 0.350 Mg C/ha/yr ratio = 4.43x, against their reported ~3x. The one-line model gets the sign and the rough size of the capacity effect without being told about it. That is a weak check, not a validation, and we are calling it weak. ========================================================================== PART 10 -- THE CREDIT ARITHMETIC, IN PLAIN NUMBERS ========================================================================== pooled rate, all four entries 0.324 Mg C/ha/yr = 1.186 t CO2e/ha/yr at $ 20/t CO2e: $ 23.73 per hectare per year at $ 50/t CO2e: $ 59.32 per hectare per year at $100/t CO2e: $ 118.64 per hectare per year Du et al. fixed depth 0.300 Mg C/ha/yr = 1.100 t CO2e/ha/yr at $ 20/t CO2e: $ 22.00 per hectare per year at $ 50/t CO2e: $ 55.00 per hectare per year at $100/t CO2e: $ 110.00 per hectare per year Du et al. equivalent soil mass 0.141 Mg C/ha/yr = 0.517 t CO2e/ha/yr at $ 20/t CO2e: $ 10.34 per hectare per year at $ 50/t CO2e: $ 25.85 per hectare per year at $100/t CO2e: $ 51.70 per hectare per year West & Post no-till 0.570 Mg C/ha/yr = 2.090 t CO2e/ha/yr at $ 20/t CO2e: $ 41.80 per hectare per year at $ 50/t CO2e: $ 104.50 per hectare per year at $100/t CO2e: $ 209.00 per hectare per year Poeplau & Don cover crops 0.320 Mg C/ha/yr = 1.173 t CO2e/ha/yr at $ 20/t CO2e: $ 23.47 per hectare per year at $ 50/t CO2e: $ 58.67 per hectare per year at $100/t CO2e: $ 117.33 per hectare per year A 200-hectare arable farm, at the equivalent-soil-mass rate and $50/t CO2e: 0.517 t CO2e/ha/yr x 200 ha x $50 = $5,170 per year, gross. Soil sampling to 30 cm on a grid, laboratory analysis, verification and registry fees are not free. We are not going to invent a cost figure, but the reader can see the size of the pot being divided. ========================================================================== PART 11 -- HEADLINE NUMBERS QUOTED IN THE ARTICLE ========================================================================== studies screened and read : 22 effect sizes extracted : 27 effect sizes with usable uncertainty: 12 distinct publications in the pool : 6 pooled SOC change (random effects) : +16.2% [+11.3%, +21.2%] 95% prediction interval : [-1.7%, +37.3%] I^2 : 98% tau^2 : 0.0051 Q : 571.0 on 11 df rearranging practices pooled : +10.9% [+7.6%, +14.4%] (k=9) importing practices pooled : +35.6% [+33.4%, +37.9%] (k=3) depth meta-regression slope : -0.007115 lnRR per cm (-6.87% per 10 cm), p = 0.20 effect sizes stating a max depth : 7 of 12 Egger intercept : +1.61, p = 0.66 leave-one-out range : 14.3% to 17.5% one-row-per-paper pool : +15.6% (k=6) fixed depth vs equivalent soil mass : 0.300 vs 0.141 Mg C/ha/yr (2.1x) saturation ceiling, default settings: 61.7 Mg C/ha (gain 16.7) gain banked by year 20 : 7.52 Mg C/ha gain banked by year 50 : 12.95 Mg C/ha years until gain < 0.10 Mg C/ha/yr : 53.6 ========================================================================== WHAT THIS NUMBER IS WORTH ========================================================================== Not much, as a number. I^2 near the ceiling means the studies are not estimating one common effect, so the pooled percentage is an average over a set of things that are not the same thing. The prediction interval is the honest summary and it is wide enough to include no change at all. What the review does establish, and we think it establishes it firmly: 1. Practices that rearrange existing carbon lose their effect below the plough layer. Practices that import carbon do not. 2. The measured effect depends on the accounting convention by a factor of two, in the same dataset, with nothing happening in the field. 3. Most of the literature does not sample deep enough to tell the difference, and a protocol that accepts 30 cm cores cannot either. 4. Whatever the rate is, it is temporary, and the model that says so is one line long. A club extracting effect sizes from abstracts is not a systematic review with two independent screeners. Weight this accordingly, and weight it less than the individual long-term experiments it is built from. Those trials ran for decades on budgets nobody wanted to renew, and they are the reason any of this is arguable at all. ==========================================================================