Cross-scale Sensing of Field-level Essential Agroecosystem Variables for the EU Climate-smart Agriculture

Climate-smart agriculture aims to implement a suite of conservation management practices, such as cover crops, reduced tillage, smart irrigation and crop rotations, to maximize agroecosystem productivity and reduce greenhouse gas emissions. Timely and high-resolution agriculture data are crucial for measuring, reporting and verifying the implementation and benefits of climate-smart agriculture practices. However, agricultural data collection through field sampling, laboratory analysis, and/or grower surveys is time-consuming and costly. To address these challenges, we developed an artificial intelligence-empowered cross-scale sensing framework to integrate multi-source ground truth data with multi-modal satellite Earth observations to quantify high spatial and temporal information of essential agroecosystem variables in the EU. Specifically, these essential variables include crop types, harvest time, tillage practices, cover crop adoption and biomass, crop yield, soil moisture, ecosystem gross primary productivity and evapotranspiration. We developed computer vision and machine learning algorithms to obtain ground truth data from in-situ measurements, citizen sciences, census surveys, and ground or aerial vehicle system data. Through process-guided machine learning (PGML), we integrated the domain knowledge of soil-vegetation radiative transfer models and ground truth data to accurately quantify these essential variables from Sentinel-1, 2, 3 and SMAP satellite data. This study highlights the potential of integrating cross-scale sensing and PGML to quantify essential ecosystem variables to support climate-smart agriculture.