Loss-on-ignition for scalable soil carbon monitoring: separating useful method transfer equations from misleading conversion factors

Abstract. A central challenge for monitoring soil organic carbon (SOC) to support soil health assessment, national inventories, and carbon markets is the cost of sampling required to detect significant change over time. Direct measurement by dry combustion (DC) is relatively expensive and can limit sampling intensity. Loss-on-ignition (LOI) offers a low-cost alternative, but confusion persists regarding the sources of measurement error in LOI-derived SOC estimates across diverse soils. This study evaluated three LOI protocols that are in regular use today for measuring SOM and predicting SOC as measured by dry combustion (DC): ignition at 360 °C for 2 h (LOI360), 400 °C for 16 h (LOI400), and 550 °C for 3 h (LOI550), across 423 soil samples from contrasting landform regions, horizons, textures, and inorganic carbon contents. Method transfer equations were developed to relate soil organic matter (SOM) measured by LOI to SOC measured by DC. Those models were evaluated across multiple aspects of performance using measures of precision (R²), accuracy (RMSE, MAE, and MedAE), and overall agreement (concordance correlation coefficient). The use of regression to relate method results, rather than estimating the SOC:SOM conversion ratio, is emphasized. The repeatability of the methods was also assessed by performing multiple measurements and calculating the coefficient of variation. LOI360 and LOI400 produced comparable and relatively strong relationships with DC-SOC, whereas LOI550 exhibited substantially lower precision and accuracy because of additional mineral mass losses during ignition. Only minor improvements over generalized models were achieved by localizing equations based on landform region or soil horizon, which suggests that broadly applicable transfer functions are sufficient for most applications. While DC provided the highest repeatability (CV = 2.60 %), LOI400 demonstrated acceptable repeatability (CV = 5.3 %) and accuracy (RMSE = 0.63 SOC%). LOI400 was also favourable because its method transfer equation had an intercept that was close to zero. These results demonstrate that LOI, when using an optimized and standardized protocol, is a practical and scientifically defensible approach for large-scale SOC monitoring. Although LOI measurements have more measurement error than DC analyses, their substantially lower cost enables much higher sampling densities, improving the characterization of spatial variability and increasing statistical power to detect SOC change. Additionally, LOI400 can serve as a preliminary screening tool to identify zones of contrasting SOC and guide targeted DC analyses, while also supporting the development of large ground-based SOC datasets for remote sensing and machine learning approaches to digital SOC mapping and monitoring.