Why is the strength of relationships between pairs of methods for estimating soil microbial biomass often so variable?

Physiological and biochemical methods for estimating soil microbial biomass are usually calibrated against other methods and parameters. However, while calibrations are usually made over soils with a very wide range of microbial biomass values (across a wide geographical range) they are often used to assess relatively small differences in microbial biomass across a narrower range of microbial biomass values (across a smaller geographical area, e.g. within a single field), where their reliability may be considerably less. We investigated the abilities of three methods of quantifying microbial biomass, i.e. substrate-induced respiration (SIR), fumigation-incubation (FI) and fumigation-extraction (FE) to serve as predictors of each other across two geographical scales, i.e. across 12 sites over an area of ≈ 100 × 100 km; and within each of these sites (12 samples per site) over an area of 0.3 ha each. Over the larger scale, relationships between pairs of methods were strong, with R 2 values always > 0.90. However, over the smaller scale, correlations between pairs of methods were variable and only significant for those sites in which spatial variability in soil organic matter was relatively high. Uncertain relationships between SIR and the fumigation-based methods may be expected because they apply to different subsets of the soil biomass (i.e. glucose-responsive vs chloroform-sensitive). However, we suggest that FI and FE are sometimes weakly correlated because the FI decomposition constant k c and the FE constant k EC vary differently relative to each other across underlying gradients. Calibration equations for estimating microbial biomass are most accurate when restricted to situations where the range of biomass values is comparable to that from which the calibration was first derived, and to similar soil types. In our study, it appears that calibration equations for predicting microbial biomass are only likely to provide reliable relative estimates in situations where the coefficient of variation (standard deviation/mean) of soil organic C is > 15%.