AgriCarbon-EO v1.0.1: large-scale and high-resolution simulation of carbon fluxes by assimilation of Sentinel-2 and Landsat-8 reflectances using a Bayesian approach

Soil organic carbon storage is a well-identified climate change mitigation solution. Quantification of the soil carbon storage in cropland for agricultural policy and offset carbon markets using in situ sampling would be excessively costly, especially at the intrafield scale. For this reason, comprehensive monitoring, reporting, and verification (MRV) of soil carbon and its explanatory variables at a large scale need to rely on hybrid approaches that combine remote sensing and modelling tools to provide the carbon budget components with their associated uncertainties at intrafield scale. Here, we present AgriCarbon-EO v1.0.1: an end-to-end processing chain that enables the estimation of carbon budget components for major and cover crops at intrafield resolution (10 m) and regional extents (e.g. 10 000 km 2 ) by assimilating remote sensing data (e.g. Sentinel-2 and Landsat8) in a physically based radiative transfer (PROSAIL) and agronomic models (SAFYE-CO2). The data assimilation in AgriCarbon-EO is based on a novel Bayesian approach that combines normalized importance sampling and look-up table generation. This approach propagates the uncertainties across the processing chain from the reflectances to the output variables. After a presentation of the chain, we demonstrate the accuracy of the estimates of AgriCarbon-EO through an application over winter wheat in the southwest of France during the cropping seasons from 2017 to 2019. We validate the outputs with flux tower data for net ecosystem exchange, biomass destructive samples, and combined harvester yield maps. Our results show that the scalability and uncertainty estimates proposed by the approach do not hinder the accuracy of the estimates (net ecosystem exchange, NEE: RMSE = 1.68 -2.38 gC m - 2 , R 2 = 0.87 -0.77; biomass: RMSE = 11.34 g m - 2 , R 2 = 0.94 ). We also show the added value of intrafield simulations for the carbon components through scenario testing of pixel and field simulations (biomass: bias = - 47 g m - 2 , - 39 % variability). Our overall analysis shows satisfying accuracy, but it also points out the need to represent more soil processes and include synthetic aperture radar data that would enable a larger coverage of AgriCarbon-EO. The paper's findings confirm the suitability of the choices made in building AgriCarbon-EO as a hybrid solution for an MRV scheme to diagnose agro-ecosystem carbon fluxes.