aceProductivity-based sampling for field-scale soil organic carbon mapping: from regional models to carbon farming needs
Abstract: This study evaluates an innovative field-scale targeted sampling strategy within a regional hybrid spatial prediction model that combines machine learning and geostatistics. The framework is designed so that newly collected field observations are incorporated only through the local residual kriging step, while the regional Random Forest trend model remains unchanged, allowing field-scale predictions to be refined without full model refitting. The proposed sampling approach integrates a Normalized Difference Vege…
"Grassy Field" by shaire productions is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.
Researchers tested a field-scale targeted sampling strategy inside a regional hybrid model that combines machine learning and geostatistics for soil organic carbon mapping. The method pairs a long-term satellite NDVI-based Productivity Index with regional prediction uncertainty to choose sampling locations, with new observations incorporated only via local residual kriging. Across 28 agricultural fields, the regional model alone averaged 0.24% SOC RMSE, while adding all field samples reached 0.17% and the four-sample targeted approach reached 0.18%.
Accurate, low-cost field-scale SOC maps matter for carbon farming verification because payments and climate claims depend on detecting real SOC changes above model error. The study shows four well-placed samples can match dense sampling, but also shows random sampling can miss informative locations and perform worse than no field sampling. Whether the productivity-based rule holds across different climates, soils, and cropping systems beyond the 28 test fields remains untested.
- Regional hybrid model combines Random Forest trend that remains unchanged with local residual kriging that incorporates new field observations without full refitting.
- Baseline regional model alone had mean RMSE = 0.24% SOC and was unable to adequately represent local spatial heterogeneity.
- Targeted sampling uses Normalized Difference Vegetation Index (NDVI)-based Productivity Index derived from long-term satellite NDVI time series plus regional model-derived prediction uncertainty.
- Evaluation used 28 test agricultural fields containing 7 to 19 in-field SOC observations to assess diverse sampling configurations and within-field variability.
- Minimum detectable change (MDC) was used to evaluate capability to detect meaningful SOC changes beyond prediction error for carbon farming monitoring.
Integrating an NDVI-based Productivity Index with regional prediction uncertainty to select four field samples reduced mean field-scale SOC prediction error from 0.24% to 0.18% RMSE, achieving accuracy comparable to using all available field observations.
The rundown
The framework keeps the regional Random Forest trend model unchanged and refines predictions only through the local residual kriging step when new field observations are added, avoiding full model refitting.
Beyond RMSE, the study applied minimum detectable change to assess whether predicted SOC changes exceed prediction error, which is directly relevant for verifying carbon farming outcomes.
Sources
- Peer-reviewedEnvironmental Monitoring and Assessment2026-08-17
How should this claim be treated?
The debate