Modeling Soil Organic Carbon Changes Using Signal-To-Noise Analysis: A Case Study Using European Soil Survey Datasets

Area of Science:

• Environmental Science
• Soil Science
• Remote Sensing

Background:

• Soil organic carbon (SOC) is vital for soil health and climate mitigation.
• Digital Soil Mapping (DSM) using Earth Observation (EO) data generates SOC time series, but detecting changes is hampered by prediction uncertainties.
• Existing methods rarely account for uncertainties when assessing SOC change.

Purpose of the Study:

• Introduce a novel signal-to-noise ratio (SNR) framework to evaluate the detectability and reliability of SOC change from DSM predictions.
• Assess SOC change modeling reliability across land-cover types using both state-first and change-first approaches.
• Provide a practical diagnostic for change-model confidence, especially when ground-truth data is limited.

Main Methods:

• Developed a model-based SNR framework defining SNR as the ratio of predicted SOC change to its modeled uncertainty.
• Applied the SNR framework to pan-European SOC observations using Random Forest and Quantile Regression Forests.
• Evaluated SNR at pixel levels and assessed the impact of spatial aggregation on SOC change assessment.

Main Results:

• At the site level, prediction accuracy was low with consistently low SNR values.
• Spatial averaging of predictions demonstrably improved SNR, enhancing the reliability of SOC change assessments at broader scales.
• The SNR framework proved effective in evaluating change-model confidence using model predictions and their uncertainties.

Conclusions:

• The SNR framework offers a practical internal metric for assessing the credibility of DSM-based SOC change monitoring.
• Routine reporting of SNR can significantly enhance transparency and trust in SOC change assessments.
• Further research should integrate land use/management data and explore aggregation effects for policy-relevant outcomes.