Prompted by the European Commission’s goal to improve soil health by 2030, also known as “Mission Soil”, the AI4SoilHealth project directly aims to implement and improve soil health monitoring and modelling services by leveraging AI technology. Central to the initiative is the development of a Soil Digital Twin, which will offer real-time forecasting and assessment of soil health indicators. These predictive tools will be integrated into practical applications, such as a Soil Health Index and an AI-powered app, enabling farmers and policymakers to make informed decisions. By improving soil health management through these services, the project supports the EU's goal of healthier soils by 2030. IIASA, as a facilitator of policy solutions to environmental challenges, supports this initiative by providing efficient methods to predict how soil health changes in the future as a function of climate change and management decisions. Biophysical crop models are vital for answering such question, as purely data- driven approaches lack the extrapolation capabilities for predictions far into the future. However, their complexity often leads to significant computational demands and the need for location-specific parameters. To address these challenges, we propose a data-driven approach that emulates crop model simulations using the EPIC-IIASA model. The emulator, based on third-order polynomial regression models, is trained on diverse climate, soil, and management data. It predicts average crop yield and carbon sequestration potentials by breaking down the problem into an ensemble of smaller models, allowing for efficient, large-scale predictions into the future with well- defined uncertainty intervals and explainable results. The approach is demonstrated through two use cases: a Farmer’s Advice app that informs users on how management practices affect yield now and up to the year 2100, and a broader application across 86,000 units in Europe. By emulating crop models with data-driven techniques, we can drastically reduce computational effort while opening up new applications, including citizen science and integration into larger model ensembles.