Fusion of in-situ data, remote sensing, and soil carbon modeling for accurate parcel scale carbon balance monitoring

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Demonstrates a method to fuse multiple data streams for soil carbon accounting, with validation across European sites.

🏢実務担当者:Offers a practical approach for field-level carbon monitoring that could be adopted by agricultural carbon projects.

🏛政策担当者:Supports development of monitoring, reporting, and verification (MRV) frameworks for agricultural carbon credits.

The carbon balance of an agricultural ecosystem depends on the difference between the net carbon assimilation by plants and carbon respired from organic matter decomposition. Accurate estimates of the soil carbon balance therefore require simultaneously constraining both components. This poster describes how satellite-derived crop productivity estimates, in-situ measurements of soil organic carbon (SOC) and predictions of a soil carbon model can be combined to derive data-driven estimates of annual carbon exchange for agricultural fields. The approach consists of an empirical model for gross primary productivity (GPP) and the soil carbon model YASSO15 which simulates the evolution of the SOC stock on a yearly time step. The GPP model is based on a red edge vegetation index measured by the Sentinel-2 satellites and is trained on eddy covariance measurements recorded on five agricultural sites in Finland. The GPP estimates are converted to the net primary productivity (NPP) and combined with records of lateral carbon fluxes (harvest yields and organic fertilization) to estimate the annual carbon inputs to the soil. The carbon input estimates are then linked to the SOC stocks derived from soil samples using a data fusion algorithm based on Bayesian inference. The algorithm produces SOC time series which are constrained by the SOC model and statistically consistent with both observed stocks and the estimated carbon inputs. We validated the approach against multiple data sources. We first cross-validated the GPP model with a leave-one-site-out approach over the five Finnish sites and subsequently tested the model against an independent set of in-situ biomass measurements. The model captured 60-70 % of the observed variability in both cases. We then evaluated the full data fusion approach against a continental scale data set consisting of CO2 flux measurements on 10 eddy covariance sites across Europe. The GPP model showed reduced performance (R2 = 0.38) compared to the cross-validation over Finnish sites, perhaps due to differences in climate conditions or crop types. However, the estimates of net ecosystem exchange (the balance between GPP and ecosystem respiration), which were informed by both in-situ SOC data and the GPP estimates, remained skillful at capturing inter-annual and geographical variability (R2 = 0.71). These results demonstrate the significance of historical SOC data for estimating the current stock change, but also highlight the added value from combining independent data streams for monitoring changes in soil carbon stocks in agriculture.

• openalex https://doi.org/10.5281/zenodo.20440836first seen 2026-06-18 05:12:56 · last seen 2026-06-18 05:16:57