The AI Journal

When Allan Walkley and Ian Armstrong Black published their modification of the Degtjareff dichromate titration in 1934, they included a candid admission: the procedure recovered, on average, only 76 per cent of the organic carbon in the twenty English soils they tested (Walkley & Black, 1934). They proposed that practitioners multiply their results by a correction factor of 1.32 to approximate total carbon. Walkley himself revisited the question in 1947, testing digestion conditions and inorganic interferences, and reaffirmed the factor while noting that recovery may vary to some extent depending on the conditions of reaction and the composition of soil organic matter (Walkley, 1947). That hedge has carried the weight of a century of soil data. The factor of 1.32 entered the literature not as an established constant but as an empirical average over twenty soils from one country, measured against one set of reference values.

What subsequent work has shown is that the variation Walkley cautioned about is neither small nor random. De Vos and colleagues, analysing 542 forest soil samples in Flanders, found that the original method recovered systematically less carbon than the accepted 76 per cent would imply, and that the appropriate correction factor for their soils was 1.58 (De Vos et al., 2007). They attributed part of the shortfall to charcoal and resistant elemental carbon particles that dichromate, in the absence of external heating, does not oxidise. Meersmans and colleagues, working on Belgian agricultural soils, found that recovery differed between land uses: carbon recovered under pasture was 15 per cent lower than under cropland, a discrepancy that no single correction factor can absorb (Meersmans et al., 2009). The method does not merely measure carbon imperfectly; it measures it with an error whose sign and magnitude depend on the kind of soil and the kind of land use.

The most striking demonstration of this dependence comes from charcoal. Hardy and Dufey, working on pre-industrial charcoal kiln sites in Wallonia, showed that the Walkley–Black procedure recovers charcoal carbon incompletely and that the degree of recovery changes with the age of the charcoal: 23.6 per cent of charcoal carbon was recovered from a currently active kiln site, against 65 per cent from century-old kiln soils (Hardy & Dufey, 2017). The older the charcoal, the more of it the dichromate oxidises; the resistance of charcoal to wet oxidation declines as it weathers in soil. Two soils with identical total carbon but different charcoal contents will therefore yield different Walkley–Black values, and the same soil, as its charcoal ages, will appear to gain carbon under the method without gaining any in reality. Hardy and Dufey concluded that incomplete recovery of black carbon might be a significant cause of underestimation of soil organic carbon by the Walkley–Black method in regional and global databases. The bias is not uniform; it tracks the fire history of the landscape.

This matters because the global soil carbon literature is, to a large extent, a Walkley–Black literature. The method was cheap, required no elemental analyser, and used reagents available in any agricultural laboratory. For most of the twentieth century it was the default. The historical baselines against which contemporary soil carbon loss is assessed were, in their majority, generated by it. When Scharlemann and colleagues surveyed 27 global soil carbon estimates, they found a range from 504 to 3,000 petagrams of carbon across those studies, a sixfold disagreement (Scharlemann et al., 2014). A fraction of that range is the correction factor’s ghost: different studies applied different factors, or none, to data produced by the same imperfect method, and the results entered databases as if commensurable.

The argument is not that the Walkley–Black method is wrong. It is that the correction factor is a variable masquerading as a constant, and that the variable is correlated with the quantities the field is trying to measure over time. Consider what happens when a study compares a 1960s Walkley–Black baseline with a 2010s dry-combustion measurement of the same fields. The dry-combustion instrument recovers essentially all carbon, including charcoal and clay-stabilised fractions that dichromate misses. The difference between the two measurements includes both genuine carbon lost to tillage and erosion and carbon that was always present but invisible to the older method. If the older method’s recovery was lower in clay-rich or charcoal-rich soils, and the evidence says it was, then the estimated loss is inflated for exactly those soils, and the inflation is indistinguishable from the real signal. The error has the same shape as the phenomenon.

The strongest objection is straightforward. If the same method is used for both the baseline and the follow-up measurement within a given study, a constant multiplicative error cancels in the difference. Walkley–Black values may be too low in absolute terms, but if the 1960s and the 1990s values are both Walkley–Black values, the correction factor drops out, and the change is real. This would hold if every long-term comparison were purely intra-method. It is not. The transition from Walkley–Black to dry combustion was not instantaneous, but it was decisive: elemental analysers displaced dichromate titration in most research and monitoring laboratories between the late 1980s and the early 2000s. Any study spanning that transition, and the most policy-relevant studies do, because they compare pre-industrial or mid-century baselines with contemporary measurements, is a cross-method comparison in which the constant-bias-cancels defence does not apply. The objection holds only within a single era of a single method, and the eras do not align with the timescales of soil carbon change.

There is a narrower point worth conceding. Even within the Walkley–Black era, a systematic bias that is constant across soils cancels in the difference; only a bias that varies across soils contaminates the comparison. The evidence presented here is that the bias does vary, by charcoal content, by clay type, by land use. But the evidence is drawn from specific regions and specific soil types, and the magnitude of the confounding has not been quantified at global scale. The claim must therefore be narrowed: not that a known fraction of the reported soil carbon debt is artefactual, but that the correction factor’s non-constancy is an unquantified confound whose direction is known and whose magnitude is not. It inflates apparent loss in charcoal-rich and clay-rich soils and compresses it in sandy, charcoal-poor ones, and the global estimate of agricultural soil carbon loss, some 116 petagrams over twelve millennia following a 2018 correction to the figure originally reported (Sanderman et al., 2017), is a weighted aggregate in which the weights are unknown because the recovery rates were never measured for most of the soils in the underlying databases.

The correction factor, in other words, is not a measurement nuisance. It is a structural feature of the historical record, and it has the shape of the thing it is supposed to correct for. Until the soil carbon literature treats the Walkley–Black recovery rate as a variable, one that must be estimated soil by soil against a reference method, the numbers it produces for temporal change will conflate the chemistry of dichromate with the chemistry of the earth.